Non-Invasive Ventilation for Community-Acquired Pneumonia: Outcomes and Predictors of Failure from an ICU Cohort

Background and Objectives: The use of non-invasive ventilation (NIV) for community-acquired pneumonia (CAP) remains controversial. NIV failure in the setting of acute hypoxemic respiratory failure is associated with increased mortality, highlighting the need for careful patient selection. Methods and Methods: This is a retrospective observational cohort study. We included 140 patients with severe CAP, treated with either NIV or invasive mechanical ventilation (IMV) as their primary oxygenation strategy. Results: The median PaO2/FiO2 ratio and SOFA score upon ICU admission were 151 mmHg and 6, respectively. We managed 76% of patients with NIV initially and report an NIV success rate of 59%. Overall, the 28-day mortality was 25%, whilst for patients with NIV success, the mortality was significantly lower at 13%. In the univariate analysis, NIV failure was associated with the SOFA score (OR 1.33), the HACOR score (OR 1.14) and the presence of septic shock (OR 3.99). The SOFA score has an AUC of 0.75 for NIV failure upon ICU admission, whilst HACOR has an AUC of 0.76 after 2 h of NIV. Conclusions: Our results suggest that a SOFA ≤ 4 and an HACOR ≤ 5 are reasonable thresholds to identify patients with severe CAP likely to benefit from NIV.


Introduction 1.Background
Community-acquired pneumonia (CAP) is a leading cause of hospitalization and mortality globally [1].The serious complications of CAP include sepsis and acute respiratory failure (ARF), both of which may require intensive care unit (ICU) admission.Over the past decade, the use of non-invasive ventilation (NIV) as a respiratory support strategy for CAP with associated primary hypoxemic respiratory failure has increased, [2][3][4] despite a lack of strong evidence on its efficacy.Although NIV is accepted as the first-line respiratory support for patients with hypercapnic respiratory failure or acute heart failure, its use for CAP remains controversial, and guidelines generally do not support its routine use [5,6].However, more recently, the use of helmet continuous positive airway pressure (CPAP) and NIV has been shown to be effective in patients with hypoxemic respiratory failure in improving outcomes [7,8].
In patients with pneumonia, NIV appears to improve oxygenation and, in those who respond, may reduce the requirement for invasive mechanical ventilation (IMV) and mortality [9,10].However, the success of NIV in the context of ARF associated with severe CAP ranges from 20 to 76%, and selecting patients who will respond to NIV is a challenge [11].The failure of NIV appears to be associated with the type of pneumonia, the disease severity, physiological derangement, the presence of other organ failures and worsening oxygenation [12][13][14][15][16][17][18][19].Furthermore, NIV failure and the delayed initiation of IMV may increase mortality and complications associated with intubation [18,20,21].The need for careful patient selection and early identification of those at risk of NIV failure is therefore clear.In response to these findings, the Heart Rate, Acidosis, Consciousness, Oxygenation and Respiratory Rate (HACOR) score was developed for patients with hypoxemic ARF [22].Although the HACOR score appears to predict NIV failure in unselected hypoxemic respiratory failure [23,24] and more recently in COVID-19 [25][26][27], its predictiveness for NIV failure has yet to be studied exclusively in patients with CAP.

Aims and Objectives
We aimed to investigate the outcomes of NIV use and predictors of NIV failure in patients with ARF secondary to severe CAP.Our primary objective was to report the outcomes of NIV use in this context.Our secondary objectives were to identify whether the HACOR score, Sequential Organ Failure Assessment (SOFA) score, or Ratio of Oxygen Saturation index (ROX index) predict NIV failure for severe CAP cases in an ICU setting.

Study Design and Setting
In this single-center retrospective cohort study, we included consecutive adults with CAP admitted to our ICU prior to the COVID-19 pandemic.University Hospital Southampton is a large tertiary hospital in the south of England serving 1.9 million people.The study data were collected for the period between 1st February 2016 and 30th April 2017.University Hospital Southampton NHS Foundation Trust (RHM CRI 0370) sponsored this study, and ethical approval was obtained from the NHS Health Research Authority (IRAS 232922).The study is compliant with local ethical standards, and no identifiable patient data are presented here.This manuscript complies with STROBE guidelines [28].

Inclusion and Exclusion Criteria
We identified eligible patients by searching our electronic patient records (MetaVision CIS, iMDsoft, Tel Aviv, Israel) by diagnosis upon ICU admission.Our inclusion criteria were (1) a diagnosis of CAP, (2) ARF requiring respiratory support on an ICU and (3) a PaO 2 /FiO 2 (P/F) ratio ≤ 300 mmHg.We excluded patients who received high-flow nasal oxygen (HFNO) as their sole first respiratory support; however, we included patients who received intermittent HFNO to facilitate breaks during NIV.

Data Collection
Anonymized patient data were retrieved from our electronic patient records.The data collected included demographic information, co-morbidities (described using Charlson's Comorbidity Index, CCI) [29] and laboratory values upon ICU admission.We categorized co-morbidities as ischemic heart disease (IHD), congestive cardiac failure (CCF), chronic obstructive pulmonary disease (COPD), other chronic respiratory diseases, cerebrovascular disease, diabetes mellitus, chronic kidney disease (3b or worse), active cancer (solid organ or hematological) or immunocompromise (active cancer or immunosuppressive medication).Acute Physiology and Chronic Health Evaluation 2 (APACHE II) and SOFA scores were calculated at various timepoints [30,31].Our exposure variable was the type of respiratory support used upon ICU admission (NIV vs. IMV), which was entirely dependent on clinician choice.NIV included both continuous positive airway pressure (CPAP) and bilevel positive airway pressure (BIPAP).We also categorized patients who received NIV first as either NIV success (defined as discharged alive from the ICU) or NIV failure (defined as a requirement for IMV or death).For patients who received NIV first, we calculated the original HACOR score [22], the updated HACOR score (uHACOR) [24] and the ROX index [32] upon ICU admission and after 2 h of NIV.We defined sepsis and septic shock according to the Sepsis-3 consensus definitions [33].The primary outcome reported is 28-day mortality from ICU admission.Our secondary outcome measures are the rate of NIV failure, the requirement for organ support, ICU mortality, ICU days and hospital days.

Statistical Analysis
Our data are reported using conventional descriptive statistics, with categorical data presented as numbers (percentage).We used the Kolmogorov-Smirnov test to assess continuous data for normality, and as our dataset was generally non-normally distributed, we presented continuous variables as medians (inter-quartile range; IQR).Comparisons were made between survivors and non-survivors at 28 days, between patients who received NIV and IMV as their first respiratory support and between NIV success and failure.The Mann-Whitney U test was used to compare continuous variables, and Fisher's exact test was used for proportions between groups.We used a univariate logistic regression to investigate the relationships between variables on the dichotomous outcomes of NIV failure and 28-day mortality.Kaplan-Meier survival curves are also used to describe 28-day mortality.The ability of variables to predict NIV failure and 28-day mortality was investigated using a receiver operating characteristic (ROC) curve analysis.We used SPSS v28 (IBM Corp., Armonk, NY, USA) for our analysis.A p-value of <0.05 was taken to be statistically significant.

NIV Outcomes
We compared outcomes between patients with NIV success and failure (Table 3 the 106 patients who received NIV as their first respiratory support, 63 (59%) survive ICU discharge (i.e., NIV success).The remaining 43 patients (41%) failed NIV and e required IMV as their second respiratory support (n = 36, 34%) or were not eligibl IMV and died subsequently (n = 7, 7%).In patients with NIV success, the prevalenc diabetes mellitus was greater (33% vs. 14%, p = 0.040), whilst the prevalence of septic sh was lower (8% vs. 26%, p = 0.025).However, there were no other differences in pa demographics, comorbidities or the prevalence of ARDS, CPE, or de novo ARF.

Predictors of NIV Failure
We compared laboratory tests and prognostic scores at various timepoints between NIV success and failure (Table 3).After 2 h of NIV, the HACOR score and ROX index data were available for 78% (n = 83) and 84% (n = 89) of patients, respectively.Although non-significant, the NIV failure group had a lower P/F ratio and ROX index.

Discussion
In this single-center retrospective cohort study, we report our use of NIV for patients with ARF secondary to severe CAP.Despite the lack of guidelines, the use of NIV for this purpose is now relatively common [2][3][4], and the most recent literature suggests that NIV may reduce the requirement for IMV and mortality [7][8][9][10].However, NIV failure is associated with increased mortality [18,20], and it remains unclear which patients are most likely to benefit.We found that the SOFA and HACOR scores, but not the ROX index, can be used to accurately predict NIV failure (Table 4).These findings are broadly consistent with previous studies on prognostic scores in patients with hypoxaemic ARF [14,22,24].Although the study was retrospective and single-centered, the findings show that NIV may benefit patients with severe CAP.
The common aetiology of ARF in this cohort was severe CAP, with a median P/F ratio upon ICU admission of 151 mmHg, suggesting moderate hypoxemic respiratory failure.Our inclusion criteria were based upon the diagnosis of CAP, as opposed to hypoxemic ARF more broadly, as we believed that aetiology-specific results would be of more relevance to clinical decision making.The median SOFA score upon ICU admission was six, which suggests that most patients had additional non-pulmonary organ dysfunction secondary to sepsis.The disease severity in this cohort is broadly comparable to other previous similar studies [12,13,18].
In our cohort, the overall 28-day mortality was 25%.We found that non-survivors were older, more comorbid and had higher WCCs and APACHE II scores upon ICU admission.In total, 76% of the patients received NIV as the first mode of respiratory support, and the rest (24%) received IMV first.Overall, those who received IMV first had higher SOFA scores and WCCs upon ICU admission, which may suggest that this decision was based upon the presence of another organ dysfunction.The prevalence of de novo ARF was also higher in those who received immediate IMV, whilst patients who received NIV first reported a higher prevalence of non-COPD respiratory disease.These findings may be explained by clinicians' reluctance to start IMV in patients with significant chronic lung disease, but as all patients have a diagnosis of CAP, they are of unclear significance.We report an NIV success rate of 59%, inclusive of patients who had NIV as a ceiling of therapy and were therefore not eligible for IMV, which is similar to previously reported results [11].Overall, 45% of patients were successfully managed with NIV alone, and for these patients, the 28-day mortality and length of ICU stay were significantly better.Furthermore, unlike previous studies [18,20], we also found that there was no increased mortality when IMV was initiated following NIV failure.
Our data suggest that the SOFA score may predict NIV failure in patients with severe CAP.The use of the SOFA score for this purpose has previously been described [14], although this study included patients with non-pulmonary sepsis.Nevertheless, the authors found that the SOFA score had an OR for NIV failure of 1.24, which is consistent with our finding of 1.33 (Table 4).Furthermore, the SOFA score upon ICU admission has an AUC of 0.75 in our ROC analysis, and this improves to 0.82 after 24 h.Overall, a SOFA score ≤ 4 upon ICU admission has a sensitivity and specificity for predicting NIV success of 61% and 72%, respectively.These results suggest that most patients who benefit from NIV may be identified early using the SOFA score upon admission.
The HACOR score was developed to predict NIV failure in hypoxaemic ARF [22].In patients with pneumonia, the HACOR score has been variously reported to have an AUC for NIV failure of 0.88 to 0.93 after 1 h of NIV [22,23].However, we found that the HACOR score had an AUC of 0.66 for NIV failure upon ICU admission, which improved to 0.71 after 2 h of NIV (Table 4).In our analysis, a HACOR score ≤ 5 after 2 h of NIV had a sensitivity and specificity of 53% and 85%, respectively.We also evaluated the predictive ability of the recently updated HACOR score [24], which incorporates the SOFA score as well as binary variables, including the presence of immunosuppression, septic shock, acute respiratory distress syndrome (ARDS) or cardiogenic pulmonary oedema (CPE).We found that the updated HACOR score had a better AUC for NIV failure of 0.72 upon ICU admission and 0.76 after 2 h.However, a major disadvantage of the updated HACOR score is that a chest radiograph is required, the interpretation of which introduces bias and limits the ability to perform serial measurements.
Our results have several limitations.As a single-center retrospective study, our sample size was limited, and this, in conjunction with significant collinearity between variables, meant we chose not to perform multivariate analysis.Furthermore, all treatment decisions were made by the clinical team, and our analysis is unlikely to account for all factors considered by them at the time.For example, it is possible that some patients were started on NIV as a bridging therapy whilst preparations for IMV were ongoing.We have also been unable to report any data on NIV tidal volume, which has previously been associated with NIV failure [34].Our results should also be interpreted in the context of how ICU care is utilized in the United Kingdom.Our practice includes discussions with patients regarding appropriate levels of therapy and ceilings of care that incorporate their wishes, pre-morbid functional status and frailty and the ability to recover from critical illness following invasive mechanical ventilation.In addition, HACOR data after 2 h of NIV were only available for approximately 80% of patients.This was largely because arterial blood gases were not repeated, which we hypothesize is more likely to have occurred in patients who were improving clinically.We were also unable to perform a multivariate analysis due to the small sample size.Nevertheless, our study suggests that NIV can be used as an initial respiratory support intervention for hypoxemic patients with severe community-acquired pneumonia.While NIV failure outcomes were comparable to those of IMV, NIV success had significantly better ICU outcomes.This suggests that using NIV to optimize oxygenation in severe pneumonia may be beneficial.However, careful patient selection is required, with consideration to ensure that the NIV is only administered in appropriate areas where at-risk patients can be offered immediate access to IMV without undue delay.

Conclusions
NIV may be used as initial respiratory support for patients with community-acquired pneumonia and hypoxemic respiratory failure.The NIV failure rate in this setting was 40.6%.Successful NIV was associated with much better outcomes when compared to immediate mechanical ventilation or patients who had NIV failure.Our results suggest that SOFA and HACOR scores can be used to identify patients who are likely to benefit from NIV early in their ICU admission.A SOFA score ≤ 4 upon ICU admission and a HACOR score ≤ 5 after 2 h of NIV are both predictive of NIV success.

Figure 1 .
Figure 1.Flow diagram of eligible, included and excluded patients by respiratory support received and corresponding 28-day mortality.

Figure 1 .
Figure 1.Flow diagram of eligible, included and excluded patients by respiratory support received and corresponding 28-day mortality.

Table 1 .
Patient characteristics, laboratory tests, prognostic scores and outcomes according to 28-day mortality.

Table 1 .
Patient characteristics, laboratory tests, prognostic scores and outcomes according to 28-day mortality.

Table 2 .
Patient characteristics, laboratory tests, prognostic scores and outcomes according to first respiratory support.

Table 3 .
Prognostic variables at longitudinal timepoints and outcomes according to success or failure.

Table 3 .
Prognostic variables at longitudinal timepoints and outcomes according to NIV success or failure.

Table 4 .
Prognostic variables with associated odds ratios and AUCs for NIV failure.

Table 4 .
Prognostic variables with associated odds ratios and AUCs for NIV failure.