The Impact of Sepsis on the Outcomes of COPD Patients: A Population-Based Cohort Study

This study aims to identify the impact of new-onset sepsis in patients with chronic obstructive pulmonary disease (COPD) including the effects on acute exacerbations, pneumonia and mortality. Using the National Health Insurance Research Database of Taiwan, all patients with COPD older than 40 years between 1988 and 2010 were recruited. After propensity score matching, each of the 8774 COPD patients with and without sepsis were identified to have similar characteristics. The primary outcome was severe exacerbations of COPD, with a severe exacerbation being defined as a patient requiring hospital admission or an emergency department visit due to COPD. The secondary outcomes were pneumonia, serious pneumonia, and all-cause mortality. The post-index overall cumulative incidence rates of total acute exacerbations were 11.2/person-years in the sepsis group and 6.2/person-years in the non-sepsis group (adjusted hazard ratio (HR) = 1.38, 95% confidence interval (CI), 1.38–1.40). The sepsis group also had higher risks of severe exacerbations (adjusted HR = 2.05, 95% CI, 2.02–2.08), severe exacerbations requiring hospitalization (adjusted HR = 2.30, 95% CI, 2.24–2.36), and severe exacerbations leading to an emergency room visit (adjusted HR = 1.91, 95% CI, 1.87–1.94). Regarding the effect on secondary outcomes, the sepsis group had higher risks of mortality (incidence rate: 23.7/person-years vs. 11.34/person-years, adjusted HR = 2.27, 95% CI, 2.14–2.41), pneumonia (incidence rate: 26.41 per person-days vs. 10.34 per person-days, adjusted HR = 2.70, 95% CI, 2.5–2.91), and serious pneumonia (incidence rate: 5.84 per person-days vs. 1.98 per person-days, adjusted HR = 2.89, 95% CI, 2.5–3.33) compared with the non-sepsis group. Sepsis survivors among patients with COPD had a higher risk of severe exacerbations, pneumonia, serious pneumonia, and mortality compared to patients with COPD without sepsis.


Introduction
Chronic obstructive pulmonary disease (COPD) is a common disease characterized by persistent airflow limitation and associated respiratory symptoms. It is currently the fourth leading cause of death worldwide [1]. Globally, the prevalence of COPD is expected to increase due to a prolonged lifespan and increased exposure to the risk factors of COPD. An exacerbation of COPD is a critical acute event characterized by the worsening of respiratory symptoms, and it is associated with an accelerating decline in lung function. This has a negative effect on the quality of life and increases the rates of hospitalization and mortality [2]. Several factors may induce acute exacerbations in patients with COPD, including air pollution, poor compliance, and most commonly, respiratory tract infections [3,4]. The early identification of the predisposing factors is an important issue in the prevention of acute exacerbations and to improve outcomes in COPD.
Sepsis is a life-threatening condition defined by organ dysfunction caused by a dysregulated host response to infection [5]. Patients with COPD have been reported to be at a higher risk of developing sepsis due to the use of corticosteroids, underlying comorbidities, and possibly impaired barrier function [6,7]. However, the role of sepsis in increasing the risk of acute exacerbations and the effect on the outcomes in COPD has yet to be well defined.
The National Health Insurance (NHI) program in Taiwan was initiated in March 1995. It is a public health insurance system for the general population in Taiwan, and it provides medical care for up to 99% of the 23 million residents of Taiwan [8]. In this study, the National Health Insurance Research Database (NHIRD) of Taiwan was used to identify the impact of new-onset sepsis in patients with COPD including the effects on acute exacerbations, pneumonia and mortality.

Study Design and Patient Selection
This study was conducted using the NHIRD, which is a database constructed by the National Health Research Institutes (NHRI) and includes comprehensive medical care records of more than 97% of the hospitals and clinics in Taiwan [9]. All claims data of ambulatory care records, outpatient visits, prescriptions, inpatient care records, registration files, and disease and vital status data were retrieved from the NHIRD. Ethical approval was obtained from the Institutional Review Board of the Cardinal Tien Hospital (Number: CTH-106-3-5-060).
All patients with COPD older than 40 years between 2000 and 2010 were recruited for this study. The diagnosis of COPD was defined when the patients had an International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) diagnostic code for COPD (491, 492, or 496) in at least one hospital admission or three outpatient visits. Although the diagnosis of COPD is based on the clinical signs/symptoms and the findings of spirometry, these data are not available from NHIRD database. Therefore, like other researches based on health claims databases, we used ICD-9-CM codes of COPD and spirometry to identify patients with COPD [10]. To ensure the accuracy and reliability of the COPD diagnosis, the exclusion criteria were: (1) incomplete demographic data; (2) no pulmonary function test within one year before or after the diagnosis of COPD; and (3) COPD was not diagnosed after the lung function test. We also excluded those who died or were diagnosed with sepsis prior to being indexed.
The patients with sepsis were identified by ICD-9-CM codes for both an infectious process and acute organ dysfunction, which is similar to the method previously reported by Lai et al. [11]. For the patients with more than one episode of sepsis, only the first episode was included. The non-sepsis cohort was sampled as the reference group, and patients with a previous history of sepsis were excluded.

Demographic Characteristics and Comorbidities
The patients' demographic characteristics including age, sex, monthly income (less than NT$  19,100, NT$ 19,100-NT$ 41,999, and more than NT$ 42,000), hospital level at admission (medical center, regional, district, and others), number of severe exacerbations of COPD in the one year prior to the index date (never, 1, or ≥2 times/year), and the index year of sepsis (2000-2011) were extracted. Underlying comorbid conditions were identified according to ICD-9-CM codes and grouped into the following categories: myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, rheumatologic disease, peptic ulcer disease, hemiplegia or paraplegia, renal disease, diabetes, moderate/severe liver disease, and tumors. The Charlson Comorbidity Index (CCI) was used to determine the severity of the comorbidities in each patient. Important medications including aspirin, statins, non-steroidal anti-inflammatory drugs (NSAIDs), anti-hyperglycemic drugs, proton-pump inhibitors (PPIs), medications for hypertension and for COPD (long-acting beta agonist (LABAs), short-acting beta agonists (SABAs), long-acting muscarinic antagonists (LAMAs), and inhaled corticosteroids (ICSs)), were recorded.

Outcomes
The primary outcome was severe exacerbations of COPD, which was defined as a patient requiring hospital admission or an emergency department visit due to COPD. The secondary outcomes were pneumonia, serious pneumonia, and all-cause mortality. Pneumonia is identified according to ICD-9-CMcodes 480-486, and 507. Serious pneumonia was defined as pneumonia requiring invasive or non-invasive mechanical ventilation. Because of the high mortality rate and older-age in COPD patients, competing risk analysis using the Fine and Gray model was also performed [12]. All subjects were followed until the occurrence of events of interest, death or the end of the study (31 December 2011).

Statistical Analysis
Data analysis was performed using the SAS statistical package, version 9.4 (SAS Institute, Inc., Cary, NC, USA). The baseline characteristics are presented as descriptive statistics (means, standard deviations, counts, and percentages). Categorical baseline variables were compared using Pearson's chi-square test or Fisher exact test. For continuous variable, the Kolmogorov-Smirnov test was applied to test for a normal distribution. When the continuous variable was normally distributed, an independent t-test was used. If the normality assumption was violated, a nonparametric Wilcoxon rank sum test was used. To minimize the imbalance in baseline characteristic covariates between the sepsis and non-sepsis groups, we performed 1:1 propensity score matching. Covariates that may have caused interference or biased the association between exposure and outcome of interest including demographic characteristics, comorbidities and severe exacerbations of COPD in the one year prior to the index date (never, 1, or ≥2 times/year) were included in the propensity matching.
The cumulative incidence rates of severe exacerbations were compared between the sepsis and non-sepsis groups before and after propensity score matching. Crude and adjusted hazard ratios (HRs) were calculated using Poisson regression with the non-sepsis cohort selected as the reference group.
The crude incidence rates of other outcomes (mortality, pneumonia and serious pneumonia) were obtained as the total number of events during the follow-up period divided by person-days. The Schoenfeld Residuals test was used to test the proportional hazards assumption. Cox proportional regression models were used with the non-sepsis cohort as the reference group. Crude HRs and adjusted HRs of individual outcomes were also evaluated after adjusting for the propensity score. We also used the Fine-Gray model in our incidence rate and time-to-event model to account for competing risks in the mortality model to derive subdistribution HRs (sub-HR) of secondary outcomes. A two-sided p-value < 0.05 was considered to indicate statistical significance in all analyses.

Baseline Demographic Characteristics
The medical records of 91,197 patients with COPD from 2000 to 2010 were reviewed, of whom 14,238 were enrolled as the sepsis cohort and 76,959 as the non-sepsis cohort. The demographic characteristics of both groups are summarized and shown in Table 1. The sepsis group tended to be older than the non-sepsis group. In addition, there were significant differences in sex, COPD index year, monthly income, medications for hypertension (beta-blockers, angiotensin-convertingenzyme inhibitors (ACEis) and angiotensin receptor blockers (ARBs), medications for COPD and statins between the two groups. In terms of baseline comorbidities, except for diabetes mellitus, the sepsis cohort had higher rates of myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, rheumatologic disease, peptic ulcer disease, hemiplegia or paraplegia, renal disease, moderate or severe liver disease, and tumors. After 1:1 propensity score matching, 8774 patients with COPD were identified in each group to have similar characteristics including age, sex, income, baseline comorbidities and medications ( Figure 1).

Outcomes Analysis
Following matching, the post-index overall cumulative incidence rates of total acute exacerbations were 11.19/person-years in the sepsis group and 6.21/person-years in the non-sepsis group (adjusted HR = 1.38, 95% CI, 1.38-1.40). With regards to overall acute exacerbation events, the sepsis group also had higher risks of severe exacerbations (adjusted HR = 2.05, 95% CI, 2.02-2.08), severe exacerbations requiring hospitalization (adjusted HR = 2.30, 95% CI, 2.24-2.36), and severe exacerbations leading to an emergency room visit (adjusted HR = 1.91, 95% CI, 1.87-1.94). The results before matching were similar to those after matching ( Table 2). Table 2. Cumulative incidence rates and hazard ratios of COPD patients with acute exacerbation, severe acute exacerbation, severe acute exacerbation requiring hospitalization and ED visit before and after propensity score matching in sepsis and non-sepsis cohort. Regarding the effect on secondary outcomes, the sepsis group had higher risks of mortality (incidence rate: 23.75 per 100 person-days vs. 11.34 per 100 person-days, adjusted HR = 2.27, 95% confidence interval (CI), 2.14-2.41), pneumonia (incidence rate: 26.41 per 100 person-days vs. 10.34 per 100 person-days, adjusted HR = 2.70, 95% CI, 2.5-2.91), and serious pneumonia (incidence rate: 5.84 per 100 person-days vs. 1.98 per 100 person-days, adjusted HR = 2.89, 95% CI, 2.5-3.33) after propensity score matching compared with the non-sepsis group. These results were consistent to those before propensity score matching. In competing risk analysis performed by adjusting for mortality due to causes other than the outcomes of interest, sepsis was still significantly associated with pneumonia (sub-HR = 1.86, 95% CI, 1.75-1.98) and serious pneumonia (sub-HR = 2.09, 95% CI, 1.86-2.35) after matching (Table 3). Kaplan-Meier analysis for the cumulative incidence also yielded significant differences and higher risks of mortality (p < 0.001), pneumonia (p < 0.001), and serious pneumonia (p < 0.001) in the sepsis group (Figure 2). Table 3. Incidence rates and risks of pneumonia, serious pneumonia of COPD patients comparing between sepsis and non-sepsis cohort before and after propensity score matching. .

Figure 2.
Kaplan-Meier analysis of the cumulative incidence for (A) pneumon pneumonia and (C) mortality between the sepsis and non-sepsis cohort after matchin sub-HR, subdistribution hazard ratio.

Discussion
This study compared two matched cohorts each with 8774 patients with COPD with or without sepsis, and found that the sepsis group had poorer outcomes due to higher risks of severe exacerbations, mortality, pneumonia, and serious pneumonia. After adjusting for mortality from causes other than the outcomes of interest as a competing risk and propensity score, the results still showed consistent findings. In addition, in subgroup analysis, the negative effect of sepsis on the outcomes of the patients with COPD remained significant.
This study has several strengths. Spirometry is crucial and the most objective diagnostic tool used to diagnose COPD [13]. In order to ensure the accuracy of the diagnosis of COPD, we only enrolled patients who received a pulmonary function test within one year before and after the COPD diagnosis was made. Patients were also excluded if the COPD diagnosis was altered after the pulmonary test. The 2016 Surviving Sepsis Campaign (SSC) guidelines and the published 3rd International Consensus Definitions for sepsis and septic shock (Sepsis-3) revised definitions of sepsis emphasize organ dysfunction related to a dysregulated host response to infection [5,14]. Although the Sequential Organ Failure Assessment (SOFA) or quick SOFA (qSOFA) score is currently the most commonly used scoring system, the most appropriate scoring system to assess organ dysfunction is controversial [14][15][16]. In this study, we defined sepsis according to ICD-9-CM codes including all infections with related acute organ dysfunction as validated in a previous study [17]. These selection criteria were more comprehensive and similar to a modern definition of sepsis. We also performed propensity score matching to minimize the effects of possible confounding variables, and therefore we believe our findings are more representative and generalized.
Sepsis is associated with a high rate of mortality, however the reported rate varies widely from 10-52% depending on the how the data are collected [18][19][20]. Sepsis is also associated with poor long-term outcomes and increased risks of death following discharge, recurrent infections, readmission and worse quality of life. An increased incidence of bacterial infection of both respiratory tract and non-respiratory tract (i.e., urinary tract) origin has been reported in patients with COPD [6]. This increased incidence may be due to the following factors: 1. impaired barrier function of the respiratory tract in COPD resulting in bacterial colonization in the airway; 2. systemic inflammatory characteristic of COPD associated with multiple comorbidities [21]; 3. the usage of corticosteroids and anticholinergic agents [10,[22][23][24]; and 4. possible adverse behavior, especially smoking [25]. However, to the best of our knowledge, the effect of sepsis on COPD subgroups has not been investigated in detail in previous studies.
Although the most common cause of acute exacerbations in patients with COPD appears to be from the respiratory tract infection (viral or bacterial) [13,26], exacerbations do not always coexist with an acute infection. The ecological features of infection, including increased bacterial burden with low community diversity associated with multiple indices of host inflammation such as alveolar neutrophilia and increased alveolar concentration of catecholamines, are consistently lacking in patients with exacerbations of COPD. However, critical illnesses such as sepsis will alter the ecosystem of the body's microbiota due to various pathophysiological (i.e., glucose and electrolyte disturbances, impairment in IgA production, loss of secreted antimicrobial peptides in the mucosal barrier or endogenous opioid production) and therapeutic (sedatives, opiates, and catecholamines) factors, and is clearly involved in the pathogenesis of subsequent frequent exacerbations [27]. Furthermore, sepsis-induced organ dysfunction, such as renal function impairment, central nervous system dysfunction, and muscle wasting, may also contribute to long-term disabilities and increase the risk of mortality in patients with COPD [7,28,29].
Currently, the most commonly used tool to assess COPD is based on the "ABCD" grouping re-defined by the 2017 Global Initiative for Chronic Obstructive Pulmonary Disease (GOLD) guidelines, which also recommends the choice of therapy [13]. The COPD assessment tool takes account of the presence and severity of abnormalities in spirometry, clinical symptoms, exacerbation history and comorbidities, but not the influence of previous sepsis. In light of the fact that the impact of sepsis on patients with COPD is significant, sepsis should not be neglected when making treatment decisions for patients with COPD. Several therapies and approaches have been shown to have beneficial impacts on sepsis, including statins, theophylline, macrolides, and vaccinations, and they may play a role in treatment for COPD [30][31][32][33][34][35]. On the other hand, the agents commonly used to control COPD which may increase the risk of infection including corticosteroids and anticholinergics, should be used more cautiously in select patients [10,22,23]. Additional large and well-designed studies with more in-depth research on the control and prevention of sepsis in patients with COPD are essential and may lead to a more comprehensive COPD treatment strategy.
There are several limitations to this study. First, the diagnosis of COPD was made according to ICD-9 codes, which were mainly registered by clinical physicians. Although the accuracy was confirmed by strict screening criteria in our study, and the reliability of ICD-9 codes has been validated in previous studies [17,36,37], this issue remains a concern. Secondly, COPD is now considered to be a disease with single or combined "phenotypes" that describe the differences between individuals [38,39].
However, the impact of sepsis on different phenotypes of COPD was not available in this study. Third, the risk of mortality from sepsis varies depending on the severity of sepsis and comorbidities, however the association between the severity of sepsis and different groups of COPD (i.e., "ABCD" groups) could not be determined in the present study. Fourth, the exact results of spirometry were not be available in NHIRD. Thus, we could not classify the severity of COPD in this study. However, we used the number of COPD exacerbation episodes as an indicator and matched this data between sepsis and non-sepsis cohort. Fifth, our study revealed a relatively high proportion of patients received oral steroid therapy. Even though it was a real-world condition in Taiwan, this could limit the generalizability of our findings. Finally, similar to other database studies, some possible confounding factors such as smoking, environmental factors, appropriateness of antimicrobial use for pneumonia, Medical Research Council (MRC) Dyspnea scale, radiologic findings of pneumonia, and compliance of orders which may also have affected the outcomes were lacking.

Conclusion
Under Taiwan's NHI program, our investigation showed that sepsis survivors among patients with COPD had a higher risk of severe exacerbations, pneumonia, serious pneumonia, and mortality compared to patients with COPD without sepsis. In light of these important findings, the negative effects of sepsis on patients with COPD should be considered in clinical practice. Furthermore, more efforts should be put into the early detection and prevention of sepsis to improve the prognosis and quality of life in patients with COPD.