Changes in Respiratory Muscle Strength Following Cardiac Rehabilitation for Prognosis in Patients with Heart Failure.

Respiratory muscle weakness, frequently observed in patients with heart failure (HF), is reported as a predictor for poor prognosis. Although increased respiratory muscle strength ameliorates exercise tolerance and quality of life in HF patients, the relationship between changes in respiratory muscle strength and patient prognosis remains unclear. A total of 456 patients with HF who continued a 5-month cardiac rehabilitation (CR) were studied. We measured maximal inspiratory pressure (PImax) at hospital discharge as the baseline and five months thereafter to assess the respiratory muscle strength. Changes in PImax during the 5-month observation period (⊿PImax) were examined. We investigated the composite multiple incidence of all-cause death or unplanned readmission after 5-month CR. The relationship between ⊿PImax and the incidence of clinical events was analyzed. Over a median follow-up of 1.8 years, 221 deaths or readmissions occurred, and their rate of incidence was 4.3/100 person-years. The higher ⊿PImax was significantly associated with lower incidence of clinical event. In multivariate Poisson regression model after adjustment for clinical confounding factors, ⊿PImax remained a significant and independent predictor for all-cause death/readmission (adjusted incident rate ratio for ⊿PImax increase of 10 cmH2O: 0.77, 95% confidence interval: 0.70–0.86). In conclusion, the changes in respiratory muscle strength independently predict the incidence of clinical events in patients with HF.

Abstract: Respiratory muscle weakness, frequently observed in patients with heart failure (HF), is reported as a predictor for poor prognosis. Although increased respiratory muscle strength ameliorates exercise tolerance and quality of life in HF patients, the relationship between changes in respiratory muscle strength and patient prognosis remains unclear. A total of 456 patients with HF who continued a 5-month cardiac rehabilitation (CR) were studied. We measured maximal inspiratory pressure (PI max ) at hospital discharge as the baseline and five months thereafter to assess the respiratory muscle strength. Changes in PI max during the 5-month observation period ( J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month paired Student's t-test or the Wilcoxon signed rank PImax from baseline to five months post-rehabilitatio based on whether ⊿PImax was positive, and differe between the two groups using the Student's unpai if non-normally distributed for continuous variab categorical variables, as appropriate. The Kruskalto assess the differences in baseline characteristics the time point (years) of study participation. Rela were analyzed using the Kaplan-Meier method w between ⊿ PImax and a composite of multiple h cardiovascular events, univariate and multivariate incident rate ratios (IRRs) were estimated by analy continuous variable (unit increase in 10 cmH2O of confounders at the end of the 5-month cardiac reha analyses: age, sex, BMI, NYHA class, AHEAD score For missing data on confounders, we performed method, assuming that analyzed data were miss imputed datasets for analysis, Rubin's formula w started at the end of the 5-month cardiac rehabili knots were also used to determine the associatio analyses of ⊿PImax in various subgroups relevant t potential effect modification on the association of ⊿ clinical events with categories of ⊿PImax per 10 cmH analysis. We also estimated the association betwee rehabilitation and all-cause clinical events using variable. The IRRs were analyzed with meaningfu predictive capability of ⊿PImax and the other signifi clinical events, the C-index was calculated using m PI max ) were examined. We investigated the composite multiple incidence of all-cause death or unplanned readmission after 5-month CR. The relationship between J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were comp paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided in based on whether ⊿PImax was positive, and differences in baseline clinical variables w between the two groups using the Student's unpaired t-test if parametric, the Mann-W if non-normally distributed for continuous variables, and the Chi-square or Fisher's categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test w to assess the differences in baseline characteristics and outcomes during the study pe the time point (years) of study participation. Relationships between ⊿PImax and clini were analyzed using the Kaplan-Meier method with the log-rank test. To estimate t between ⊿ PImax and a composite of multiple hospitalization and/or death due t cardiovascular events, univariate and multivariate Poisson regression models were u incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positiv continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The foll confounders at the end of the 5-month cardiac rehabilitation were used as covariates i analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participa For missing data on confounders, we performed multiple imputation using the cha method, assuming that analyzed data were missing at random. To combine the re imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regr started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline cur knots were also used to determine the association between ⊿PImax and clinical eve analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performe potential effect modification on the association of ⊿PImax with clinical events. Trend r clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Coch analysis. We also estimated the association between changes in clinical variables foll rehabilitation and all-cause clinical events using multivariate Poisson regression m variable. The IRRs were analyzed with meaningful unit change in each variable. To PI max and the incidence of clinical events was analyzed. Over a median follow-up of 1.8 years, 221 deaths or readmissions occurred, and their rate of incidence was 4.3/100 person-years. The higher J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compar paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined th PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into based on whether ⊿PImax was positive, and differences in baseline clinical variables wer between the two groups using the Student's unpaired t-test if parametric, the Mann-Wh if non-normally distributed for continuous variables, and the Chi-square or Fisher's e categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test we to assess the differences in baseline characteristics and outcomes during the study peri the time point (years) of study participation. Relationships between ⊿PImax and clinica were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the between ⊿ PImax and a composite of multiple hospitalization and/or death due to cardiovascular events, univariate and multivariate Poisson regression models were use incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The follow confounders at the end of the 5-month cardiac rehabilitation were used as covariates in analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participatio For missing data on confounders, we performed multiple imputation using the chain method, assuming that analyzed data were missing at random. To combine the resu imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regres started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curve knots were also used to determine the association between ⊿PImax and clinical event analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed potential effect modification on the association of ⊿PImax with clinical events. Trend rela clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochra analysis. We also estimated the association between changes in clinical variables follow PI max was significantly associated with lower incidence of clinical event. In multivariate Poisson regression model after adjustment for clinical confounding factors, J. Clin. Med. 2020, 9, x FOR PEER REVIEW 4 of 13

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the chang PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two gro based on whether ⊿PImax was positive, and differences in baseline clinical variables were comp between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact tes categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also to assess the differences in baseline characteristics and outcomes during the study period base the time point (years) of study participation. Relationships between ⊿PImax and clinical end-po were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the associa between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-caus cardiovascular events, univariate and multivariate Poisson regression models were used. Adju incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PIma continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following cli confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multiva analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and P For missing data on confounders, we performed multiple imputation using the chained equa method, assuming that analyzed data were missing at random. To combine the results from imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression mo started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with t knots were also used to determine the association between ⊿PImax and clinical events. Subgr analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to asses

Introduction
Patients with heart failure (HF) frequently suffer from breathlessness during exercise, leading to exercise intolerance and decreased quality of life [1]. Exercise-related breathlessness partly results from respiratory muscle weakness, observed in approximately 30-50% of patients with HF [2]. The respiratory muscle weakness is caused by muscular atrophy and/or decreased number of cross-bridges, resulting from the activation of inflammatory and neuroendocrine factors due to HF [2][3][4][5]. Additionally, several studies have demonstrated that respiratory muscle weakness is an independent predictor of exercise intolerance, ventilatory inefficiency during exercise, and poor prognosis of patients with HF [6,7]. However, the influence of longitudinal changes in respiratory muscle strength on prognostic outcomes remains unclear.
Conversely, comprehensive cardiac rehabilitation including exercise training, medication administration, nutritional management, and correcting lifestyle behavior is recognized as one of the key treatment strategies to prevent HF recurrence and to improve prognosis of patients with HF [8]. Furthermore, respiratory muscle training is reported to increase respiratory muscle strength and consequently improve exercise tolerance and quality of life in HF patients [9]. Hence, the current guideline on preventive cardiology and rehabilitation has recommended that respiratory training should be prescribed to patients with HF with exercise intolerance and respiratory muscle weakness in addition to the usual exercise training [10]. However, the effect of increased respiratory muscle strength on prognosis is unclear, although decreased respiratory muscle strength is reportedly associated with HF severity [11]. Our hypothesis was that the changes in respiratory muscle strength might potentially be a useful marker to assess the clinical status of HF as well as a surrogate marker to predict outcomes in patients with HF.
Therefore, this study aimed to investigate the relationship between changes in respiratory muscle strength following cardiac rehabilitation and the incidence of adverse clinical events in patients with HF.

Study Design and Population
This is a single-center observational study conducted to review a cohort of consecutive patients with HF who were admitted to Kitasato University Hospital for HF treatment and underwent a 5-month cardiac rehabilitation during hospitalization and after hospital discharge from May 2009 to December 2017. Patients who had received thoracic or open-heart surgery within the last three months or had chronic diseases of respiratory systems were excluded from the study. Comprehensive cardiac rehabilitation consisted of supervised exercise training and education on self-management including medication, nutrition, and physical activity, based on the statement from the Japanese Circulation Society [12]. Blood examinations and echocardiograms at hospital discharge were considered as baseline data. We also assessed pulmonary and respiratory muscle functions at baseline and at the end of the 5-month cardiac rehabilitation. Events of all-cause mortality or all-cause unplanned readmission after the 5-month cardiac rehabilitation were considered as the primary end-point of this study. Data on all variables were obtained from an electronic database. The study protocol was approved by the Kitasato Institute Clinical Research Review Board (KMEO B18-075, September 4, 2018) and was performed according to the ethical guidelines of the Declaration of Helsinki.

Patient Characteristics
Data on age, gender, body mass index (BMI), HF severity assessed by the New York Heart Association functional classification (NYHA class), smoking history, medications, and medical history such as hypertension, diabetes mellitus, dyslipidemia, chronic kidney disease, or atrial fibrillation were obtained from medical records upon study participation. Routine laboratory analysis included hemoglobin and serum albumin as well as plasma brain natriuretic peptide (BNP). The estimated glomerular filtration rate (eGFR) was determined by serum creatinine levels. The left ventricular ejection fraction (LVEF) was also measured on echocardiograms using the 2D method. The AHEAD score was used to assess the patients' risk stratification and was calculated by assigning one point to the patient for each of the following factors: A: atrial fibrillation, H: hemoglobin <13 g/dL for men and 12 g/dL for women, E: elderly (age >70 years), A: abnormal renal parameters (creatinine >130 µmol/dL), and D: diabetes mellitus [13]. The functional capacity was measured using the 6-min walk distance according to standard guidelines [14].

Pulmonary and Respiratory Muscle Functions
To assess the pulmonary function, spirometry without bronchodilator was performed to measure forced vital capacity (FVC) and forced expiratory volume in 1 s (FEV 1 ) using a spirometer (Autospiro AS-507, Minato Medical Science, Osaka, Japan), and their percentages were calculated relative to the predictive values issued by the Japanese Respiratory Society [15]. To assess the respiratory muscle function, we measured the maximal inspiratory pressure (PI max ) using a pressure transducer (Autospiro AAM-377, Minato Medical Science, Osaka, Japan) connected to the spirometer, according to the joint statement of the American Thoracic Society and European Respiratory Society [16]. In the measurement of PI max , patients in a sitting position were instructed to hold a 25-mm-diameter flanged mouthpiece in their mouth and perform a 3-s forced inspiration from the level of maximal expiration. PI max was determined by the average value of the maximum pressure over a 1-s period during the 3-s forced inspiration. In this study, PI max was expressed as its absolute value, although it showed negative pressure for atmospheric pressure. The measurements were performed three times, and the maximum value in PI max was accepted for analysis. We also calculated percentage PI max (% PI max ) relative to the predictive value that was estimated using each age, gender, height, and body weight [17]. The % PI max of <70% was defined as respiratory muscle weakness based on previous reports in patients with HF [2,7,18].

End-Points
The primary end-point of this study was a composite of multiple all-cause clinical events including all-cause death and/or all-cause unplanned readmission identified through medical chart review. The secondary end-point was the composite of multiple cardiovascular events including cardiovascular death and/or unplanned readmission due to cardiovascular disease. We counted the number of these events after the end of the 5-month cardiac rehabilitation. The time period for these events was also calculated as the number of days from the end of the 5-month cardiac rehabilitation to the date of the events.

Cardiac Rehabilitation Program
All patients received comprehensive cardiac rehabilitation during hospitalization and after hospital discharge for five months. Cardiac rehabilitation was initiated when the HF condition was stabilized from the intensive care unit or general wards [12]. The median duration of cardiac rehabilitation initiation from hospitalization for all of studied patients was three days. In the acute phase, we facilitated mobilization and/or ambulation under the monitoring of electrocardiogram (ECG) and vital signs [12]. If patients were able to walk approximately 200 m with independence, they proceeded to inpatient exercise training with the low-moderate intensity [19,20]. Before hospital discharge, cardiologists and medical staff educated patients about self-management including medication, nutrition, and physical activity, and instructed encouragement to participate in outpatient cardiac rehabilitation at least once a week and perform 3-5/week of self-exercise. In outpatient cardiac rehabilitation, exercise training included a 5-min warm-up, 20-40-min aerobic training using a treadmill or bicycle ergometer, and 3-min cool-down periods [19]. Patients also received counselling about lifestyle in ambulatory visits. All exercise training sessions, both inpatient and outpatient, were supervised by trained nurses or physiotherapists, with continuous monitoring assessment.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PI max from baseline to five months post-rehabilitation ( Clinical variables before and after the 5-month cardiac rehabilitation paired Student's t-test or the Wilcoxon signed rank test, as appropriate. W PImax from baseline to five months post-rehabilitation (⊿PImax). Patients we based on whether ⊿PImax was positive, and differences in baseline clinica between the two groups using the Student's unpaired t-test if parametric, if non-normally distributed for continuous variables, and the Chi-squar categorical variables, as appropriate. The Kruskal-Wallis test and Fisher' to assess the differences in baseline characteristics and outcomes during the time point (years) of study participation. Relationships between ⊿PI were analyzed using the Kaplan-Meier method with the log-rank test. T between ⊿ PImax and a composite of multiple hospitalization and/or d cardiovascular events, univariate and multivariate Poisson regression mo incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a catego continuous variable (unit increase in 10 cmH2O of PImax) in separate mod confounders at the end of the 5-month cardiac rehabilitation were used as analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of stu For missing data on confounders, we performed multiple imputation u method, assuming that analyzed data were missing at random. To com imputed datasets for analysis, Rubin's formula was used. Time for the P started at the end of the 5-month cardiac rehabilitation. Restricted cubi knots were also used to determine the association between ⊿PImax and analyses of ⊿PImax in various subgroups relevant to the HF prognosis we potential effect modification on the association of ⊿PImax with clinical eve clinical events with categories of ⊿PImax per 10 cmH2O were examined us analysis. We also estimated the association between changes in clinical v rehabilitation and all-cause clinical events using multivariate Poisson re variable. The IRRs were analyzed with meaningful unit change in each predictive capability of ⊿PImax and the other significant variables in multiv clinical events, the C-index was calculated using multivariate logistic regr confounders used in the Poisson regression model. Continuous variables ± standard deviation or median with interquartile range, and categorical v patient numbers and their percentages. A two-tailed P value of <0.05 was analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata versio Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, V

Patient Characteristics
The potential study population consisted of 1570 consecutive patie month cardiac rehabilitation, and those who had received thoracic sur months (n = 393) or had chronic respiratory diseases (n = 140) were exclude who could not perform the respiratory muscle function test during the ob were also excluded. Consequently, 456 patients with HF were included fo Overall PImax increased significantly after the 5-month cardiac reha positive changes in PImax observed in 326 patients (71.5%). Table 1 sh characteristics in the two groups based on positive changes in PImax. The p significantly associated with lower BNP and PImax and higher prevale weakness at the baseline. However, no statistical differences in other base were observed between the two groups. Table S2 shows the differences treatment, and outcomes during the study period based on the tim participation. The time point was significantly associated with age, use PI max ). Patients were divided into two groups based on whether

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5month cardiac rehabilitation, and those who had received thoracic surgery within the last three months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. Patients who could not perform the respiratory muscle function test during the observation period (n = 581) were also excluded. Consequently, 456 patients with HF were included for analysis in this study.
Overall PImax increased significantly after the 5-month cardiac rehabilitation (Table S1) with positive changes in PImax observed in 326 patients (71.5%). Table 1 shows the baseline patient characteristics in the two groups based on positive changes in PImax.
The positive change in PImax was significantly associated with lower BNP and PImax and higher prevalence of respiratory muscle weakness at the baseline. However, no statistical differences in other baseline patient characteristics were observed between the two groups. Table S2 shows the differences in baseline characteristics, treatment, and outcomes during the study period based on the time point (years) of study participation. The time point was significantly associated with age, use of diuretics, frequency of PI max was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac reh paired Student's t-test or the Wilcoxon signed rank test, as appr PImax from baseline to five months post-rehabilitation (⊿PImax). P based on whether ⊿PImax was positive, and differences in basel between the two groups using the Student's unpaired t-test if p if non-normally distributed for continuous variables, and the categorical variables, as appropriate. The Kruskal-Wallis test a to assess the differences in baseline characteristics and outcom the time point (years) of study participation. Relationships be were analyzed using the Kaplan-Meier method with the log-r between ⊿ PImax and a composite of multiple hospitalizatio cardiovascular events, univariate and multivariate Poisson reg incident rate ratios (IRRs) were estimated by analyzing ⊿PImax a continuous variable (unit increase in 10 cmH2O of PImax) in sep confounders at the end of the 5-month cardiac rehabilitation w analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the y For missing data on confounders, we performed multiple im method, assuming that analyzed data were missing at rando imputed datasets for analysis, Rubin's formula was used. Tim started at the end of the 5-month cardiac rehabilitation. Restr knots were also used to determine the association between ⊿ analyses of ⊿PImax in various subgroups relevant to the HF pro potential effect modification on the association of ⊿PImax with c clinical events with categories of ⊿PImax per 10 cmH2O were ex analysis. We also estimated the association between changes i rehabilitation and all-cause clinical events using multivariate variable. The IRRs were analyzed with meaningful unit chan predictive capability of ⊿PImax and the other significant variable clinical events, the C-index was calculated using multivariate lo confounders used in the Poisson regression model. Continuous ± standard deviation or median with interquartile range, and ca patient numbers and their percentages. A two-tailed P value of analyses were performed using SPSS 25.0 (IBM, Armonk, NY), S Station, TX) and R version 3.1.2 (R Foundation for Statistical Co

Patient Characteristics
The potential study population consisted of 1570 consec month cardiac rehabilitation, and those who had received th months (n = 393) or had chronic respiratory diseases (n = 140) w who could not perform the respiratory muscle function test du were also excluded. Consequently, 456 patients with HF were i Overall PImax increased significantly after the 5-month c positive changes in PImax observed in 326 patients (71.5%). characteristics in the two groups based on positive changes in P significantly associated with lower BNP and PImax and high weakness at the baseline. However, no statistical differences in were observed between the two groups. Table S2 shows the d treatment, and outcomes during the study period based o participation. The time point was significantly associated wit PI max and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between J. Clin. Med. 2020, 9, x FOR PEER REVIEW 4 of 13

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5month cardiac rehabilitation, and those who had received thoracic surgery within the last three months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. Patients who could not perform the respiratory muscle function test during the observation period (n = 581) were also excluded. Consequently, 456 patients with HF were included for analysis in this study.
Overall PImax increased significantly after the 5-month cardiac rehabilitation (Table S1) with positive changes in PImax observed in 326 patients (71.5%). Table 1 shows the baseline patient characteristics in the two groups based on positive changes in PImax.
The positive change in PImax was significantly associated with lower BNP and PImax and higher prevalence of respiratory muscle weakness at the baseline. However, no statistical differences in other baseline patient characteristics were observed between the two groups. Table S2 shows the differences in baseline characteristics, treatment, and outcomes during the study period based on the time point (years) of study participation. The time point was significantly associated with age, use of diuretics, frequency of PI max and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared u paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the ch PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two based on whether ⊿PImax was positive, and differences in baseline clinical variables were co between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitne if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were a to assess the differences in baseline characteristics and outcomes during the study period b the time point (years) of study participation. Relationships between ⊿PImax and clinical en were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the ass between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-c cardiovascular events, univariate and multivariate Poisson regression models were used. A incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿ continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following confounders at the end of the 5-month cardiac rehabilitation were used as covariates in mul analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, an For missing data on confounders, we performed multiple imputation using the chained e method, assuming that analyzed data were missing at random. To combine the results imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves wi knots were also used to determine the association between ⊿PImax and clinical events. Su analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to as potential effect modification on the association of ⊿PImax with clinical events. Trend relation clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-A analysis. We also estimated the association between changes in clinical variables following rehabilitation and all-cause clinical events using multivariate Poisson regression models variable. The IRRs were analyzed with meaningful unit change in each variable. To com predictive capability of ⊿PImax and the other significant variables in multivariate Poisson ana clinical events, the C-index was calculated using multivariate logistic regression models adju confounders used in the Poisson regression model. Continuous variables were reported as th ± standard deviation or median with interquartile range, and categorical variables were expr patient numbers and their percentages. A two-tailed P value of <0.05 was considered signifi analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continue month cardiac rehabilitation, and those who had received thoracic surgery within the la months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. who could not perform the respiratory muscle function test during the observation period ( were also excluded. Consequently, 456 patients with HF were included for analysis in this st Overall PImax increased significantly after the 5-month cardiac rehabilitation (Table S positive changes in PImax observed in 326 patients (71.5%). Table 1 shows the baseline characteristics in the two groups based on positive changes in PImax.
The positive change in P significantly associated with lower BNP and PImax and higher prevalence of respiratory weakness at the baseline. However, no statistical differences in other baseline patient charac were observed between the two groups.

Statistical Analysis
Clinical variables before and after the 5-month cardi paired Student's t-test or the Wilcoxon signed rank test, as PImax from baseline to five months post-rehabilitation (⊿PI based on whether ⊿PImax was positive, and differences in between the two groups using the Student's unpaired t-te if non-normally distributed for continuous variables, an categorical variables, as appropriate. The Kruskal-Wallis to assess the differences in baseline characteristics and ou the time point (years) of study participation. Relationshi were analyzed using the Kaplan-Meier method with the between ⊿ PImax and a composite of multiple hospital cardiovascular events, univariate and multivariate Poisso incident rate ratios (IRRs) were estimated by analyzing ⊿ continuous variable (unit increase in 10 cmH2O of PImax) confounders at the end of the 5-month cardiac rehabilitati analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, For missing data on confounders, we performed multip method, assuming that analyzed data were missing at imputed datasets for analysis, Rubin's formula was used started at the end of the 5-month cardiac rehabilitation. knots were also used to determine the association betw analyses of ⊿PImax in various subgroups relevant to the H potential effect modification on the association of ⊿PImax w clinical events with categories of ⊿PImax per 10 cmH2O we analysis. We also estimated the association between chan rehabilitation and all-cause clinical events using multiv variable. The IRRs were analyzed with meaningful unit predictive capability of ⊿PImax and the other significant va clinical events, the C-index was calculated using multivar confounders used in the Poisson regression model. Contin ± standard deviation or median with interquartile range, a patient numbers and their percentages. A two-tailed P va analyses were performed using SPSS 25.0 (IBM, Armonk, N Station, TX) and R version 3.1.2 (R Foundation for Statisti

Patient Characteristics
The potential study population consisted of 1570 c month cardiac rehabilitation, and those who had receiv months (n = 393) or had chronic respiratory diseases (n = 1 who could not perform the respiratory muscle function te were also excluded. Consequently, 456 patients with HF w Overall PImax increased significantly after the 5-mo positive changes in PImax observed in 326 patients (71 characteristics in the two groups based on positive change significantly associated with lower BNP and PImax and weakness at the baseline. However, no statistical differen were observed between the two groups. Table S2 shows treatment, and outcomes during the study period ba participation. The time point was significantly associate PI max ) or continuous variable (unit increase in 10 cmH 2 O of PI max ) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PI max . For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared us paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the cha PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two based on whether ⊿PImax was positive, and differences in baseline clinical variables were com between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were als to assess the differences in baseline characteristics and outcomes during the study period ba the time point (years) of study participation. Relationships between ⊿PImax and clinical end were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the asso between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-ca cardiovascular events, univariate and multivariate Poisson regression models were used. A incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿P continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multi analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, an For missing data on confounders, we performed multiple imputation using the chained eq method, assuming that analyzed data were missing at random. To combine the results f imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves wit knots were also used to determine the association between ⊿PImax and clinical events. Su analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to ass potential effect modification on the association of ⊿PImax with clinical events. Trend relations clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Ar analysis. We also estimated the association between changes in clinical variables following rehabilitation and all-cause clinical events using multivariate Poisson regression models fo variable. The IRRs were analyzed with meaningful unit change in each variable. To comp predictive capability of ⊿PImax and the other significant variables in multivariate Poisson anal clinical events, the C-index was calculated using multivariate logistic regression models adjus confounders used in the Poisson regression model. Continuous variables were reported as th ± standard deviation or median with interquartile range, and categorical variables were expre patient numbers and their percentages. A two-tailed P value of <0.05 was considered signific analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., C Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued month cardiac rehabilitation, and those who had received thoracic surgery within the las months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. P who could not perform the respiratory muscle function test during the observation period (n were also excluded. Consequently, 456 patients with HF were included for analysis in this stu Overall PImax increased significantly after the 5-month cardiac rehabilitation (

Statistical Analysis
Clinical variables before and after the 5 paired Student's t-test or the Wilcoxon signe PImax from baseline to five months post-rehab based on whether ⊿PImax was positive, and d between the two groups using the Student's if non-normally distributed for continuous categorical variables, as appropriate. The Kr to assess the differences in baseline characte the time point (years) of study participation were analyzed using the Kaplan-Meier met between ⊿ PImax and a composite of mult cardiovascular events, univariate and multiv incident rate ratios (IRRs) were estimated by continuous variable (unit increase in 10 cmH confounders at the end of the 5-month cardia analyses: age, sex, BMI, NYHA class, AHEAD For missing data on confounders, we perfo method, assuming that analyzed data were imputed datasets for analysis, Rubin's form started at the end of the 5-month cardiac r knots were also used to determine the asso analyses of ⊿PImax in various subgroups rele potential effect modification on the associati clinical events with categories of ⊿PImax per 1 analysis. We also estimated the association rehabilitation and all-cause clinical events u variable. The IRRs were analyzed with mea predictive capability of ⊿PImax and the other clinical events, the C-index was calculated us confounders used in the Poisson regression m ± standard deviation or median with interqu patient numbers and their percentages. A tw analyses were performed using SPSS 25.0 (IB Station, TX) and R version 3.1.2 (R Foundatio

Patient Characteristics
The potential study population consist month cardiac rehabilitation, and those wh months (n = 393) or had chronic respiratory d who could not perform the respiratory musc were also excluded. Consequently, 456 patie Overall PImax increased significantly af PI max in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of J. Clin. Med. 2020, 9, x FOR PEER REVIEW 4 of 13

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5month cardiac rehabilitation, and those who had received thoracic surgery within the last three months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. Patients who could not perform the respiratory muscle function test during the observation period (n = 581)

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5month cardiac rehabilitation, and those who had received thoracic surgery within the last three months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. Patients PI max per 10 cmH 2 O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after paired Student's t-test or the Wilcoxon PImax from baseline to five months post based on whether ⊿PImax was positive, between the two groups using the Stud if non-normally distributed for contin categorical variables, as appropriate. T to assess the differences in baseline ch the time point (years) of study partici were analyzed using the Kaplan-Meie between ⊿ PImax and a composite of cardiovascular events, univariate and incident rate ratios (IRRs) were estima continuous variable (unit increase in 1 confounders at the end of the 5-month analyses: age, sex, BMI, NYHA class, A For missing data on confounders, we method, assuming that analyzed data imputed datasets for analysis, Rubin's started at the end of the 5-month card knots were also used to determine th analyses of ⊿PImax in various subgroup potential effect modification on the ass clinical events with categories of ⊿PIma analysis. We also estimated the associ rehabilitation and all-cause clinical ev variable. The IRRs were analyzed wit predictive capability of ⊿PImax and the clinical events, the C-index was calcula confounders used in the Poisson regres ± standard deviation or median with in patient numbers and their percentages analyses were performed using SPSS 25 Station, TX) and R version 3.1.2 (R Fou

Patient Characteristics
PI max and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5-month cardiac rehabilitation, and those who had received thoracic surgery within the last three months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. Patients who could not perform the respiratory muscle function test during the observation period (n = 581) were also excluded. Consequently, 456 patients with HF were included for analysis in this study.
Overall PI max increased significantly after the 5-month cardiac rehabilitation (Table S1) with positive changes in PI max observed in 326 patients (71.5%). Table 1 shows the baseline patient characteristics in the two groups based on positive changes in PI max . The positive change in PI max was significantly associated with lower BNP and PI max and higher prevalence of respiratory muscle weakness at the baseline. However, no statistical differences in other baseline patient characteristics were observed between the two groups. Table S2 shows the differences in baseline characteristics, treatment, and outcomes during the study period based on the time point (years) of study participation. The time point was significantly associated with age, use of diuretics, frequency of outpatient cardiac rehabilitation, and incidence of all-cause events, but not with use of ACE-I/ARB and beta-blockers, change in PI max , and incidence of cardiovascular events. Table 1. Baseline patient characteristics in the two groups based on change in PI max.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We exam PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divi based on whether ⊿PImax was positive, and differences in baseline clinical varia between the two groups using the Student's unpaired t-test if parametric, the M if non-normally distributed for continuous variables, and the Chi-square or F categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exac to assess the differences in baseline characteristics and outcomes during the stu the time point (years) of study participation. Relationships between ⊿PImax and were analyzed using the Kaplan-Meier method with the log-rank test. To esti between ⊿ PImax and a composite of multiple hospitalization and/or death cardiovascular events, univariate and multivariate Poisson regression models w incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (p continuous variable (unit increase in 10 cmH2O of PImax) in separate models. T confounders at the end of the 5-month cardiac rehabilitation were used as covar analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study par For missing data on confounders, we performed multiple imputation using th method, assuming that analyzed data were missing at random. To combine imputed datasets for analysis, Rubin's formula was used. Time for the Poisso started at the end of the 5-month cardiac rehabilitation. Restricted cubic splin knots were also used to determine the association between ⊿PImax and clinic analyses of ⊿PImax in various subgroups relevant to the HF prognosis were per potential effect modification on the association of ⊿PImax with clinical events. Tr clinical events with categories of ⊿PImax per 10 cmH2O were examined using the analysis. We also estimated the association between changes in clinical variabl rehabilitation and all-cause clinical events using multivariate Poisson regress variable. The IRRs were analyzed with meaningful unit change in each varia predictive capability of ⊿PImax and the other significant variables in multivariate clinical events, the C-index was calculated using multivariate logistic regression confounders used in the Poisson regression model. Continuous variables were r ± standard deviation or median with interquartile range, and categorical variable patient numbers and their percentages. A two-tailed P value of <0.05 was consid analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna

Patient Characteristics
The potential study population consisted of 1570 consecutive patients w month cardiac rehabilitation, and those who had received thoracic surgery w months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from who could not perform the respiratory muscle function test during the observa were also excluded. Consequently, 456 patients with HF were included for analy Overall PImax increased significantly after the 5-month cardiac rehabilitat positive changes in PImax observed in 326 patients (71.5%). Table 1 shows characteristics in the two groups based on positive changes in PImax. The positive significantly associated with lower BNP and PImax and higher prevalence of weakness at the baseline. However, no statistical differences in other baseline p were observed between the two groups. Table S2 shows the differences in bas treatment, and outcomes during the study period based on the time poi participation. The time point was significantly associated with age, use of diu 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac paired Student's t-test or the Wilcoxon signed rank test, as a PImax from baseline to five months post-rehabilitation (⊿PIma based on whether ⊿PImax was positive, and differences in b between the two groups using the Student's unpaired t-test if non-normally distributed for continuous variables, and categorical variables, as appropriate. The Kruskal-Wallis te to assess the differences in baseline characteristics and outc the time point (years) of study participation. Relationships were analyzed using the Kaplan-Meier method with the lo between ⊿ PImax and a composite of multiple hospitaliza cardiovascular events, univariate and multivariate Poisson incident rate ratios (IRRs) were estimated by analyzing ⊿PI continuous variable (unit increase in 10 cmH2O of PImax) in confounders at the end of the 5-month cardiac rehabilitation analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, th For missing data on confounders, we performed multiple method, assuming that analyzed data were missing at ra imputed datasets for analysis, Rubin's formula was used. started at the end of the 5-month cardiac rehabilitation. R knots were also used to determine the association betwee analyses of ⊿PImax in various subgroups relevant to the HF potential effect modification on the association of ⊿PImax wi clinical events with categories of ⊿PImax per 10 cmH2O were analysis. We also estimated the association between chang rehabilitation and all-cause clinical events using multivari variable. The IRRs were analyzed with meaningful unit ch predictive capability of ⊿PImax and the other significant varia clinical events, the C-index was calculated using multivariat confounders used in the Poisson regression model. Continu ± standard deviation or median with interquartile range, and patient numbers and their percentages. A two-tailed P value analyses were performed using SPSS 25.0 (IBM, Armonk, NY Station, TX) and R version 3.1.2 (R Foundation for Statistical

Patient Characteristics
The potential study population consisted of 1570 con month cardiac rehabilitation, and those who had received months (n = 393) or had chronic respiratory diseases (n = 140 who could not perform the respiratory muscle function test were also excluded. Consequently, 456 patients with HF we Overall PImax increased significantly after the 5-mont positive changes in PImax observed in 326 patients (71.5% characteristics in the two groups based on positive changes significantly associated with lower BNP and PImax and h weakness at the baseline. However, no statistical difference were observed between the two groups. Values are mean ± SD, or median (interquartile range). ACE-I, angiotensin convertor enzyme inhibitor; ARB, angiotensin II receptor blocker; BMI, body mass index; BNP, brain natriuretic peptide; dBP, diastolic blood pressure; eGFR, estimated glomerular filtration rate; FEV 1 , forced expiratory volume in 1-s; FVC, forced vital capacity; HF, heart failure; HR, heart rate; LVEF, left ventricular ejection fraction; NYHA, New York Heart Association; PI max , maximal inspiratory pressure; sBP, systolic blood pressure.

Relationships between Change in Respiratory Muscle Strength and Adverse Clinical Events
A total of 221 all-cause clinical events and 132 cardiovascular events occurred during the median follow-up period of 1.8 years, and the incidence rate of all-cause events and cardiovascular events was 4.3/100 and 2.4/100 person-years, respectively. Figure 1 shows the Kaplan-Meier survival curves for the two groups. Positive changes in PI max were significantly associated with a lower incidence of all-cause clinical events (log-rank: p = 0.021) and cardiovascular events (log-rank: p = 0.003).

Relationships between Change in Respiratory Muscle Strength and Adverse Clinical Events
A total of 221 all-cause clinical events and 132 cardiovascular events occurred during the median follow-up period of 1.8 years, and the incidence rate of all-cause events and cardiovascular events was 4.3/100 and 2.4/100 person-years, respectively. Figure 1 shows the Kaplan-Meier survival curves for the two groups. Positive changes in PImax were significantly associated with a lower incidence of all-cause clinical events (log-rank: p = 0.021) and cardiovascular events (log-rank: p = 0.003).  Table 2 shows the results of the Poisson regression models of ⊿PImax for all-cause clinical events and cardiovascular events. In the univariate Poisson regression models, ⊿PImax was significantly associated with all-cause clinical events (IRR: 0.75, 95% confidence interval (CI): 0.69-0.82, p < 0.001) and cardiovascular events (IRR: 0.71, 95% CI: 0.63-0.79, p < 0.001). In the multivariate Poisson regression models adjusted for clinical confounding factors including age, gender, BMI, NYHA class, AHEAD score, BNP, time point of study participation, and PImax values at the end of the 5-month cardiac rehabilitation, ⊿PImax was detected as a significant and independent predictor for all-cause clinical events (adjusted IRR: 0.77, 95% CI: 0.70-0.86, p < 0.001) and cardiovascular events (adjusted IRR: 0.72, 95% CI: 0.63-0.82, p < 0.001). Positive changes in PImax were also independently associated with decreased all-cause and cardiovascular clinical events ( Table 2). Cubic spline analyses clarified linear relationships between changes in PImax and all-cause or cardiovascular events ( Figure 2).

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5month cardiac rehabilitation, and those who had received thoracic surgery within the last three months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the study. Patients who could not perform the respiratory muscle function test during the observation period (n = 581)

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were com paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examine PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided based on whether ⊿PImax was positive, and differences in baseline clinical variables between the two groups using the Student's unpaired t-test if parametric, the Mann if non-normally distributed for continuous variables, and the Chi-square or Fishe categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact tes to assess the differences in baseline characteristics and outcomes during the study the time point (years) of study participation. Relationships between ⊿PImax and cl were analyzed using the Kaplan-Meier method with the log-rank test. To estimat between ⊿ PImax and a composite of multiple hospitalization and/or death due cardiovascular events, univariate and multivariate Poisson regression models wer incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (posi continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The f confounders at the end of the 5-month cardiac rehabilitation were used as covariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study particip For missing data on confounders, we performed multiple imputation using the c method, assuming that analyzed data were missing at random. To combine the imputed datasets for analysis, Rubin's formula was used. Time for the Poisson re started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline c knots were also used to determine the association between ⊿PImax and clinical e analyses of ⊿PImax in various subgroups relevant to the HF prognosis were perform potential effect modification on the association of ⊿PImax with clinical events. Trend clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Co analysis. We also estimated the association between changes in clinical variables f rehabilitation and all-cause clinical events using multivariate Poisson regression variable. The IRRs were analyzed with meaningful unit change in each variable. predictive capability of ⊿PImax and the other significant variables in multivariate Poi clinical events, the C-index was calculated using multivariate logistic regression mo confounders used in the Poisson regression model. Continuous variables were repo ± standard deviation or median with interquartile range, and categorical variables w patient numbers and their percentages. A two-tailed P value of <0.05 was considere analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Sta Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, A

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who month cardiac rehabilitation, and those who had received thoracic surgery with months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from th who could not perform the respiratory muscle function test during the observation PI max ≥ 0 cmH 2 O. PI max , maximal inspiratory pressure.

Statistical Analysis
Clinical variables before and after the 5-month cardiac r paired Student's t-test or the Wilcoxon signed rank test, as app PImax from baseline to five months post-rehabilitation (⊿PImax). based on whether ⊿PImax was positive, and differences in bas between the two groups using the Student's unpaired t-test if if non-normally distributed for continuous variables, and th categorical variables, as appropriate. The Kruskal-Wallis test to assess the differences in baseline characteristics and outco the time point (years) of study participation. Relationships b were analyzed using the Kaplan-Meier method with the log between ⊿ PImax and a composite of multiple hospitalizati cardiovascular events, univariate and multivariate Poisson re incident rate ratios (IRRs) were estimated by analyzing ⊿PIma continuous variable (unit increase in 10 cmH2O of PImax) in s confounders at the end of the 5-month cardiac rehabilitation w analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the For missing data on confounders, we performed multiple im method, assuming that analyzed data were missing at rand imputed datasets for analysis, Rubin's formula was used. Ti started at the end of the 5-month cardiac rehabilitation. Res knots were also used to determine the association between analyses of ⊿PImax in various subgroups relevant to the HF p potential effect modification on the association of ⊿PImax with clinical events with categories of ⊿PImax per 10 cmH2O were e analysis. We also estimated the association between changes rehabilitation and all-cause clinical events using multivariat variable. The IRRs were analyzed with meaningful unit cha predictive capability of ⊿PImax and the other significant variab clinical events, the C-index was calculated using multivariate confounders used in the Poisson regression model. Continuou ± standard deviation or median with interquartile range, and c patient numbers and their percentages. A two-tailed P value o analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Station, TX) and R version 3.1.2 (R Foundation for Statistical C

Patient Characteristics
The potential study population consisted of 1570 conse

Statistical Analysis
Clinical variables before and after the 5-month paired Student's t-test or the Wilcoxon signed rank t PImax from baseline to five months post-rehabilitation based on whether ⊿PImax was positive, and differenc between the two groups using the Student's unpaire if non-normally distributed for continuous variable categorical variables, as appropriate. The Kruskal-W to assess the differences in baseline characteristics a the time point (years) of study participation. Relati were analyzed using the Kaplan-Meier method wit between ⊿ PImax and a composite of multiple ho cardiovascular events, univariate and multivariate P incident rate ratios (IRRs) were estimated by analyz continuous variable (unit increase in 10 cmH2O of P confounders at the end of the 5-month cardiac rehab analyses: age, sex, BMI, NYHA class, AHEAD score, For missing data on confounders, we performed m method, assuming that analyzed data were missin imputed datasets for analysis, Rubin's formula was started at the end of the 5-month cardiac rehabilita knots were also used to determine the association analyses of ⊿PImax in various subgroups relevant to potential effect modification on the association of ⊿P clinical events with categories of ⊿PImax per 10 cmH analysis. We also estimated the association between rehabilitation and all-cause clinical events using m variable. The IRRs were analyzed with meaningful predictive capability of ⊿PImax and the other significa clinical events, the C-index was calculated using mu confounders used in the Poisson regression model. C ± standard deviation or median with interquartile ra patient numbers and their percentages. A two-tailed analyses were performed using SPSS 25.0 (IBM, Arm Station, TX) and R version 3.1.2 (R Foundation for St

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College PI max was detected as a significant and independent predictor for all-cause clinical events (adjusted IRR: 0.77, 95% CI: 0.70-0.86, p < 0.001) and cardiovascular events (adjusted IRR: 0.72, 95% CI: 0.63-0.82, p < 0.001). Positive changes in PI max were also independently associated with decreased all-cause and cardiovascular clinical events ( Table 2). Cubic spline analyses clarified linear relationships between changes in PI max and all-cause or cardiovascular events (Figure 2).

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted Figure 3 shows subgroup analyses of ⊿PImax for all-cause clinical events in various subgroups relevant to the HF prognosis. There were no significant interactions in the association of ⊿PImax with the incidence of adverse clinical events across the subgroups of aged >75 years, sex, NYHA class, and6MWD of <400 m at the baseline. Conversely, subgroups with baseline BNP of >200 pg/mL and respiratory muscle weakness showed significant interactions in the association of ⊿PImax with incidence of adverse clinical events. However, higher ⊿PImax was significantly associated with decreased adverse clinical events in all subgroups, even after adjusting for confounding factors used in the multivariate Poisson regression model.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were com paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided based on whether ⊿PImax was positive, and differences in baseline clinical variables between the two groups using the Student's unpaired t-test if parametric, the Mannif non-normally distributed for continuous variables, and the Chi-square or Fisher categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test to assess the differences in baseline characteristics and outcomes during the study p the time point (years) of study participation. Relationships between ⊿PImax and clin were analyzed using the Kaplan-Meier method with the log-rank test. To estimate between ⊿ PImax and a composite of multiple hospitalization and/or death due cardiovascular events, univariate and multivariate Poisson regression models were incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (posit continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The fo confounders at the end of the 5-month cardiac rehabilitation were used as covariates analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study particip For missing data on confounders, we performed multiple imputation using the ch method, assuming that analyzed data were missing at random. To combine the imputed datasets for analysis, Rubin's formula was used. Time for the Poisson reg started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline cu knots were also used to determine the association between ⊿PImax and clinical ev analyses of ⊿PImax in various subgroups relevant to the HF prognosis were perform potential effect modification on the association of ⊿PImax with clinical events. Trend clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Coc analysis. We also estimated the association between changes in clinical variables fo rehabilitation and all-cause clinical events using multivariate Poisson regression m variable. The IRRs were analyzed with meaningful unit change in each variable. T predictive capability of ⊿PImax and the other significant variables in multivariate Pois clinical events, the C-index was calculated using multivariate logistic regression mod confounders used in the Poisson regression model. Continuous variables were repor ± standard deviation or median with interquartile range, and categorical variables we patient numbers and their percentages. A two-tailed P value of <0.05 was considered analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stat Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Au

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who c month cardiac rehabilitation, and those who had received thoracic surgery withi months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the who could not perform the respiratory muscle function test during the observation were also excluded. Consequently, 456 patients with HF were included for analysis i Overall PImax increased significantly after the 5-month cardiac rehabilitation positive changes in PImax observed in 326 patients (71.5%). Table 1 shows the b characteristics in the two groups based on positive changes in PImax. The positive cha significantly associated with lower BNP and PImax and higher prevalence of resp weakness at the baseline. However, no statistical differences in other baseline patien were observed between the two groups.

Statistical Analysis
Clinical variables before and after paired Student's t-test or the Wilcoxon PImax from baseline to five months postbased on whether ⊿PImax was positive, between the two groups using the Stud if non-normally distributed for continu categorical variables, as appropriate. Th to assess the differences in baseline cha the time point (years) of study particip were analyzed using the Kaplan-Meie between ⊿ PImax and a composite of cardiovascular events, univariate and m incident rate ratios (IRRs) were estimat continuous variable (unit increase in 1 confounders at the end of the 5-month analyses: age, sex, BMI, NYHA class, A For missing data on confounders, we method, assuming that analyzed data imputed datasets for analysis, Rubin's started at the end of the 5-month card knots were also used to determine th analyses of ⊿PImax in various subgroup potential effect modification on the ass clinical events with categories of ⊿PImax analysis. We also estimated the associa rehabilitation and all-cause clinical ev variable. The IRRs were analyzed with predictive capability of ⊿PImax and the o clinical events, the C-index was calculat confounders used in the Poisson regres ± standard deviation or median with in patient numbers and their percentages. analyses were performed using SPSS 25 Station, TX) and R version 3.1.2 (R Foun

Patient Characteristics
The potential study population co month cardiac rehabilitation, and thos months (n = 393) or had chronic respirat who could not perform the respiratory were also excluded. Consequently, 456 Overall PImax increased significan positive changes in PImax observed in characteristics in the two groups based significantly associated with lower BN weakness at the baseline. However, no PI max with the incidence of adverse clinical events across the subgroups of aged >75 years, sex, NYHA class, and6MWD of <400 m at the baseline. Conversely, subgroups with baseline BNP of >200 pg/mL and respiratory muscle weakness showed significant interactions in the association of J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-mo paired Student's t-test or the Wilcoxon signed ran PImax from baseline to five months post-rehabilita based on whether ⊿PImax was positive, and diffe between the two groups using the Student's unp if non-normally distributed for continuous vari categorical variables, as appropriate. The Kruska to assess the differences in baseline characteristi the time point (years) of study participation. Re were analyzed using the Kaplan-Meier method between ⊿ PImax and a composite of multiple cardiovascular events, univariate and multivaria incident rate ratios (IRRs) were estimated by ana continuous variable (unit increase in 10 cmH2O confounders at the end of the 5-month cardiac re analyses: age, sex, BMI, NYHA class, AHEAD sco For missing data on confounders, we performe method, assuming that analyzed data were mi imputed datasets for analysis, Rubin's formula started at the end of the 5-month cardiac rehab knots were also used to determine the associat analyses of ⊿PImax in various subgroups relevan potential effect modification on the association o clinical events with categories of ⊿PImax per 10 cm analysis. We also estimated the association betw rehabilitation and all-cause clinical events using variable. The IRRs were analyzed with meanin predictive capability of ⊿PImax and the other sign clinical events, the C-index was calculated using confounders used in the Poisson regression mod ± standard deviation or median with interquartile patient numbers and their percentages. A two-ta analyses were performed using SPSS 25.0 (IBM, A Station, TX) and R version 3.1.2 (R Foundation fo

Patient Characteristics
The potential study population consisted o month cardiac rehabilitation, and those who h months (n = 393) or had chronic respiratory disea who could not perform the respiratory muscle fu were also excluded. Consequently, 456 patients w Overall PImax increased significantly after t positive changes in PImax observed in 326 pati PI max with incidence of adverse clinical events. However, higher J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were comp paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided i based on whether ⊿PImax was positive, and differences in baseline clinical variables w between the two groups using the Student's unpaired t-test if parametric, the Mannif non-normally distributed for continuous variables, and the Chi-square or Fisher categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test to assess the differences in baseline characteristics and outcomes during the study p the time point (years) of study participation. Relationships between ⊿PImax and clin were analyzed using the Kaplan-Meier method with the log-rank test. To estimate between ⊿ PImax and a composite of multiple hospitalization and/or death due cardiovascular events, univariate and multivariate Poisson regression models were incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positi continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The fo confounders at the end of the 5-month cardiac rehabilitation were used as covariates analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participa For missing data on confounders, we performed multiple imputation using the ch method, assuming that analyzed data were missing at random. To combine the r imputed datasets for analysis, Rubin's formula was used. Time for the Poisson reg started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline cur knots were also used to determine the association between ⊿PImax and clinical ev analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performe potential effect modification on the association of ⊿PImax with clinical events. Trend r clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Coc analysis. We also estimated the association between changes in clinical variables fol rehabilitation and all-cause clinical events using multivariate Poisson regression m variable. The IRRs were analyzed with meaningful unit change in each variable. T predictive capability of ⊿PImax and the other significant variables in multivariate Poiss clinical events, the C-index was calculated using multivariate logistic regression mod confounders used in the Poisson regression model. Continuous variables were report ± standard deviation or median with interquartile range, and categorical variables we patient numbers and their percentages. A two-tailed P value of <0.05 was considered analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Au

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who co month cardiac rehabilitation, and those who had received thoracic surgery within months (n = 393) or had chronic respiratory diseases (n = 140) were excluded from the who could not perform the respiratory muscle function test during the observation p were also excluded. Consequently, 456 patients with HF were included for analysis in Overall PImax increased significantly after the 5-month cardiac rehabilitation (  . Forest plots of hazard ratios for the association of change in respiratory muscle strength with all-cause clinical events according to major subgroups. Hazard ratios were adjusted for age, sex, BMI, AHEAD score, NYHA class, and BNP at the end of the 5-month cardiac rehabilitation. BMI, body mass index; BNP, brain natriuretic peptide; IRR, incident rate ratio; NYHA, New York Heart Association functional classification; PImax, maximal inspiratory pressure; 6MWD, 6-min walk distance.

Unadjusted Rates of Clinical Events
The unadjusted event rate for ⊿PImax categories per 10 cmH2O is shown in Figure 4. The increase in PImax per 10 cmH2O had significant trend relationships with a decreased rate of all-cause clinical events (Z = 2.975, p = 0.003). There was also a statistically significant trend relationship between the PImax increase and decreased rate of cardiovascular events (Z = 2.906, p = 0.004), but the rate showed a fall with increased PImax and subsequent slight rise. . Forest plots of hazard ratios for the association of change in respiratory muscle strength with all-cause clinical events according to major subgroups. Hazard ratios were adjusted for age, sex, BMI, AHEAD score, NYHA class, and BNP at the end of the 5-month cardiac rehabilitation. BMI, body mass index; BNP, brain natriuretic peptide; IRR, incident rate ratio; NYHA, New York Heart Association functional classification; PI max , maximal inspiratory pressure; 6MWD, 6-min walk distance.

Unadjusted Rates of Clinical Events
The unadjusted event rate for J. Clin. Med. 2020, 9, x FOR PEER REVIEW

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared us paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the cha PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two based on whether ⊿PImax was positive, and differences in baseline clinical variables were com between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were als to assess the differences in baseline characteristics and outcomes during the study period ba the time point (years) of study participation. Relationships between ⊿PImax and clinical end were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the asso between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-ca cardiovascular events, univariate and multivariate Poisson regression models were used. Ad incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿P continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multi analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, an For missing data on confounders, we performed multiple imputation using the chained eq PI max categories per 10 cmH 2 O is shown in Figure 4. The increase in PI max per 10 cmH 2 O had significant trend relationships with a decreased rate of all-cause clinical events (Z = 2.975, p = 0.003). There was also a statistically significant trend relationship between the PI max increase and decreased rate of cardiovascular events (Z = 2.906, p = 0.004), but the rate showed a fall with increased PI max and subsequent slight rise.

Unadjusted Rates of Clinical Events
The unadjusted event rate for ⊿PImax categories per 10 cmH2O is shown in Figure 4. The increase in PImax per 10 cmH2O had significant trend relationships with a decreased rate of all-cause clinical events (Z = 2.975, p = 0.003). There was also a statistically significant trend relationship between the PImax increase and decreased rate of cardiovascular events (Z = 2.906, p = 0.004), but the rate showed a fall with increased PImax and subsequent slight rise.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5-

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Patient Characteristics
The potential study population consisted of 1570 consecutive patients who continued the 5-creatinine were significantly and independently associated with all-cause clinical events, but the changes in the other variables were not. Figure 5 shows the C-index of the predictive models for clinical events in J. Clin. Med. 2020, 9, x FOR PEER REVIEW 4 of

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using t paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two grou based on whether ⊿PImax was positive, and differences in baseline clinical variables were compar between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-te if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test f categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also us to assess the differences in baseline characteristics and outcomes during the study period based the time point (years) of study participation. Relationships between ⊿PImax and clinical end-poin were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the associati between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause cardiovascular events, univariate and multivariate Poisson regression models were used. Adjust incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinic confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivaria analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PIm For missing data on confounders, we performed multiple imputation using the chained equati method, assuming that analyzed data were missing at random. To combine the results from imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression mode started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with thr knots were also used to determine the association between ⊿PImax and clinical events. Subgrou analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess t potential effect modification on the association of ⊿PImax with clinical events. Trend relationships clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armita analysis. We also estimated the association between changes in clinical variables following cardi rehabilitation and all-cause clinical events using multivariate Poisson regression models for ea variable. The IRRs were analyzed with meaningful unit change in each variable. To compare t predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis f clinical events, the C-index was calculated using multivariate logistic regression models adjusted f confounders used in the Poisson regression model. Continuous variables were reported as the me ± standard deviation or median with interquartile range, and categorical variables were expressed patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. A analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., Colle Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria).

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into t based on whether ⊿PImax was positive, and differences in baseline clinical variables were between the two groups using the Student's unpaired t-test if parametric, the Mann-Whit if non-normally distributed for continuous variables, and the Chi-square or Fisher's exa categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were to assess the differences in baseline characteristics and outcomes during the study period the time point (years) of study participation. Relationships between ⊿PImax and clinical e were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the a between ⊿ PImax and a composite of multiple hospitalization and/or death due to al cardiovascular events, univariate and multivariate Poisson regression models were used incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The followi confounders at the end of the 5-month cardiac rehabilitation were used as covariates in m analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation For missing data on confounders, we performed multiple imputation using the chained method, assuming that analyzed data were missing at random. To combine the result imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regressi started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves knots were also used to determine the association between ⊿PImax and clinical events. analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to potential effect modification on the association of ⊿PImax with clinical events. Trend relati clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran analysis. We also estimated the association between changes in clinical variables followi rehabilitation and all-cause clinical events using multivariate Poisson regression model variable. The IRRs were analyzed with meaningful unit change in each variable. To co predictive capability of ⊿PImax and the other significant variables in multivariate Poisson a clinical events, the C-index was calculated using multivariate logistic regression models ad confounders used in the Poisson regression model. Continuous variables were reported as ± standard deviation or median with interquartile range, and categorical variables were ex patient numbers and their percentages. A two-tailed P value of <0.05 was considered sign analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Cor Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria)

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation w paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were based on whether ⊿PImax was positive, and differences in baseline clinical v between the two groups using the Student's unpaired t-test if parametric, t if non-normally distributed for continuous variables, and the Chi-square categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's to assess the differences in baseline characteristics and outcomes during th the time point (years) of study participation. Relationships between ⊿PIma were analyzed using the Kaplan-Meier method with the log-rank test. To between ⊿ PImax and a composite of multiple hospitalization and/or de cardiovascular events, univariate and multivariate Poisson regression mod incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categori continuous variable (unit increase in 10 cmH2O of PImax) in separate mode confounders at the end of the 5-month cardiac rehabilitation were used as c analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study For missing data on confounders, we performed multiple imputation usi method, assuming that analyzed data were missing at random. To com imputed datasets for analysis, Rubin's formula was used. Time for the Po started at the end of the 5-month cardiac rehabilitation. Restricted cubic knots were also used to determine the association between ⊿PImax and c analyses of ⊿PImax in various subgroups relevant to the HF prognosis were potential effect modification on the association of ⊿PImax with clinical even clinical events with categories of ⊿PImax per 10 cmH2O were examined usin analysis. We also estimated the association between changes in clinical va rehabilitation and all-cause clinical events using multivariate Poisson reg variable. The IRRs were analyzed with meaningful unit change in each v predictive capability of ⊿PImax and the other significant variables in multiva clinical events, the C-index was calculated using multivariate logistic regres confounders used in the Poisson regression model. Continuous variables w ± standard deviation or median with interquartile range, and categorical va patient numbers and their percentages. A two-tailed P value of <0.05 was c analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, V

Results
creatinine. Models were adjusted for age, sex, BMI, AHEAD score, NYHA class, and BNP at the end of 5-month cardiac rehabilitation. The C-index of J. Clin. Med. 2020, 9, x FOR PEER REVIEW 4 of 13

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean ± standard deviation or median with interquartile range, and categorical variables were expressed as patient numbers and their percentages. A two-tailed P value of <0.05 was considered significant. All analyses were performed using SPSS 25.0 (IBM, Armonk, NY), Stata version 15.1 (Stata Corp., College Station, TX) and R version 3.1.2 (R Foundation for Statistical Computing, Vienna, Austria). creatinine were 0.72 (95% CI: 0.66-0.78), 0.71 (95% CI: 0.65-0.77), and 0.70 (0.65-0.76), respectively, and there were no statistical differences between the three predictive models.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for confounders used in the Poisson regression model. Continuous variables were reported as the mean

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the predictive capability of ⊿PImax and the other significant variables in multivariate Poisson analysis for clinical events, the C-index was calculated using multivariate logistic regression models adjusted for

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each variable. The IRRs were analyzed with meaningful unit change in each variable. To compare the

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac rehabilitation and all-cause clinical events using multivariate Poisson regression models for each

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage analysis. We also estimated the association between changes in clinical variables following cardiac

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events. Trend relationships of clinical events with categories of ⊿PImax per 10 cmH2O were examined using the Cochran-Armitage

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the potential effect modification on the association of ⊿PImax with clinical events.

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup analyses of ⊿PImax in various subgroups relevant to the HF prognosis were performed to assess the

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three knots were also used to determine the association between ⊿PImax and clinical events. Subgroup

Statistical Analysis
Clinical variables before and after the 5-month cardiac rehabilitation were compared using the paired Student's t-test or the Wilcoxon signed rank test, as appropriate. We examined the changes in PImax from baseline to five months post-rehabilitation (⊿PImax). Patients were divided into two groups based on whether ⊿PImax was positive, and differences in baseline clinical variables were compared between the two groups using the Student's unpaired t-test if parametric, the Mann-Whitney U-test if non-normally distributed for continuous variables, and the Chi-square or Fisher's exact test for categorical variables, as appropriate. The Kruskal-Wallis test and Fisher's exact test were also used to assess the differences in baseline characteristics and outcomes during the study period based on the time point (years) of study participation. Relationships between ⊿PImax and clinical end-points were analyzed using the Kaplan-Meier method with the log-rank test. To estimate the association between ⊿ PImax and a composite of multiple hospitalization and/or death due to all-cause or cardiovascular events, univariate and multivariate Poisson regression models were used. Adjusted incident rate ratios (IRRs) were estimated by analyzing ⊿PImax as a categorical (positive of ⊿PImax) or continuous variable (unit increase in 10 cmH2O of PImax) in separate models. The following clinical confounders at the end of the 5-month cardiac rehabilitation were used as covariates in multivariate analyses: age, sex, BMI, NYHA class, AHEAD score, BNP, the years of study participation, and PImax. For missing data on confounders, we performed multiple imputation using the chained equation method, assuming that analyzed data were missing at random. To combine the results from 20 imputed datasets for analysis, Rubin's formula was used. Time for the Poisson regression models started at the end of the 5-month cardiac rehabilitation. Restricted cubic spline curves with three FEV1/FVC 1.04 5% 0.95-1.14 0.344 IRRs were adjusted for age, gender BMI, time point of study participation, AHEAD score, NYHA class, and BNP. BMI, body mass index; BNP, brain natriuretic peptide; BP, blood pressure; CI, confidence interval; eGFR, estimated glomerular filtration rate; FEV 1 , forced expiratory volume in 1-s; FVC, forced vital capacity; IRR, incident rate ratio; HR, heart rate; 6MWD, 6-min walk distance.

Discussion
The novel findings in the present study are as follows. First, changes in respiratory muscle strength following the cardiac rehabilitation significantly and independently predicted the incidence of adverse clinical events in patients with HF. Second, positive changes in PI max of 10 cmH 2 O following the cardiac rehabilitation were associated with 23% decrease of adverse clinical events.
To the best of our knowledge, this study is the first to demonstrate that the longitudinal change of respiratory muscle strength is a significant indicator of prognosis in patients with HF. Our previous study reported on the respiratory muscle strength as a significant predictor for prognosis in patients with HF with both reduced ejection fraction (HFrEF) and preserved ejection fraction (HFpEF) [7]. In general, decreased respiratory muscle strength is associated with reduced pulmonary function [6,21], a known risk factor for cardiovascular event including HF [22]. Conversely, Habedank and colleagues showed that PI max , generally measured as a respiratory muscle function, was not an independent predictor of prognosis because it varied according to gender, BMI, and cachexia in patients with severe HFrEF [11]. A recent study has shown impaired respiratory muscle oxygenation during exercise in patients with HF, leading to breathlessness and decreased quality of life [23]. In the present study, the higher change in PI max was associated with lower respiratory muscle strength at the baseline. In addition, positive changes in PI max showed a trend relationship with declined clinical events, even after adjusting for gender, BMI, comorbidities, and HF severity. These relationships remained significant even in subgroups including elderly patients, those with lower functional capacity, or respiratory muscle weakness at baseline. These results suggest that the respiratory muscle weakness is likely to be improved by cardiac rehabilitation, and the repeated measurement of respiratory muscle strength might be more important than a single measurement to assess the clinical condition in patients with HF.
Several studies have indicated that increased respiratory muscle strength contributes to the improvement of respiratory muscle fatiguability, exercise tolerance, and quality of life [9,24]. Chiappa et al. reported the effects of inspiratory muscle training on peripheral blood flow during respiratory muscle fatigue stress in patients with HF [25]. They demonstrated that inspiratory muscle training improved peripheral muscle blood flow with decreased peripheral vascular resistance during respiratory muscle fatigue. In general, respiratory muscle fatigue induces sympathetic vasomotor outflow, resulting in the increased peripheral vascular resistance [26]. Conversely, increased inspiratory muscle strength augments tidal volume and consequently improves input to the pulmonary stretch receptor and autonomic nervous activity [27]. This improvement is documented in combination of reduced sympathetic activity and elevated parasympathetic activity, and thereby attenuates vascular resistance and increases peripheral blood flow [27]. These results are potential mechanisms of inspiratory muscle training to improve exercise tolerance and are likely correlated with our finding that increased respiratory muscle strength may improve the prognosis of patients with HF.
This study provides clinical implications that change in respiratory muscle strength is identified as a significant clinical marker for patients with HF. Previous studies have documented that the trajectory of 6MWD or renal function are associated with morbidity and mortality in these patients [28,29]. We revealed that the predictive capability of changes in respiratory muscle strength following cardiac rehabilitation was relatively higher or comparable to that of changes in 6MWD or renal function. In general, measurement of respiratory muscle strength is easy to perform in clinical practice. Therefore, respiratory muscle strength changes might be a useful marker to assess the effects of HF treatment. Furthermore, the respiratory muscle strength can be modified with exercise training including inspiratory muscle training in patients with HF [9,30]. Our results suggest the potential benefits of an increased respiratory muscle strength on the prognosis in patients with HF.
However, some limitations remain to be considered in the present study. First, as this was a single-center study that only included Japanese patients and the sample size was relatively small, whether these results can be applied to patients with HF in other hospitals or other populations remains to be elucidated. In addition, external validity could have been reduced, given that half of the potential study population was excluded from the analysis. Multivariate analyses were also performed using multiple confounders, which might increase the false-positive rates (type I error). Therefore, future multicenter studies are required to reveal the validity and reliability of change in respiratory muscle strength as predictors of prognosis in patients with HF. Second, this was not a randomized control trial. Hence, whether exercise training per se increased the respiratory muscle strength and decreased clinical events remains to be investigated. Further interventional study is required to investigate whether increased respiratory muscle strength due to exercise training improves the prognosis of patients with HF. Third, the period of data review was relatively long-term, 8.5 years, and the time point of study participation was associated with age, use of diuretics, and frequency of outpatient CR. Nevertheless, in the multivariate regression analyses, impact of the time point of study participation on relationships between change in respiratory muscle strength and outcomes was not observed. Fourth, in this study, while the respiratory muscle weakness was defined with 70% predicted value of PI max , there was limited evidence using the % PI max as a cut-off value for the weakness. As the more robust lower limit of normal values of PI max has been suggested by the ATS/ERS statement in other patient groups [16], we consider it important to clarify the meaningful cut-off value of respiratory muscle weakness for patients with HF in the future.

Conclusions
Change in respiratory muscle strength following cardiac rehabilitation significantly and independently predicted clinical events in patients with HF. The 10 cmH 2 O increase of PI max was significantly associated with a 23% decreased incidence of clinical events.
Supplementary Materials: The following are available online at http://www.mdpi.com/2077-0383/9/4/952/s1, Table S1: Changes in patient characteristics before and after the 5-month cardiac rehabilitation; Table S2: Differences in baseline characteristics, treatment, and outcomes during the study period based on the time point of study participation.