Derivation and Validation of a Novel Severity Scoring System for Pneumonia at Intensive Care Unit Admission

Abstract

Background

Severity stratification scores developed in intensive care units (ICUs) are used in interventional studies to identify the most critically ill. Studies that evaluate accuracy of these scores in ICU patients admitted with pneumonia are lacking. This study aims to determine performance of severity scores as predictors of mortality in critically ill patients admitted with pneumonia.

Methods

Prospective cohort study in a general ICU in Brazil. ICU severity scores (Simplified Acute Physiology Score 3 [SAPS 3] and Sepsis-Related Organ Failure Assessment [qSOFA]), prognostic scores of pneumonia (CURB-65 [confusion, urea, respiratory rate, blood pressure, age] and CRB-65 [confusion, respiratory rate, blood pressure, age]), and clinical and epidemiological variables in the first 6 hours of hospitalization were analyzed.

Results

Two hundred patients were included between 2015 and 2018, with a median age of 81 years (interquartile range, 67–90 years) and female predominance (52%), primarily admitted from the emergency department (65%) with community-acquired pneumonia (CAP, 80.5%). SAPS 3, CURB-65, CRB-65,and qSOFA all exhibited poor performance in predicting mortality. Multivariate regression identified variables independently associated with mortality that were used to develop a novel pneumonia-specific ICU severity score (Pneumonia Shock score) that outperformed SAPS 3, CURB-65, and CRB-65. The Shock score was validated in an external multicenter cohort of critically ill patients admitted with CAP.

Conclusions

We created a parsimonious score that accurately identifies patients with pneumonia at highest risk of ICU death. These findings are critical to accurately stratify patients with severe pneumonia in therapeutic trials that aim to reduce mortality.

Severity scores are inaccurate in those admitted with pneumonia to the intensive care unit (ICU), particularly elderly patients. Clinical trials may misclassify pneumonia severity, leading to conflicting mortality outcomes. The Pneumonia SHOCK score is a simple tool that accurately predicts ICU pneumonia mortality.

(See the Editorial Commentary by Guillamet and Kollef on pages 950–2.)

Pneumonia remains the principal infection leading to admission to intensive care units (ICUs) throughout the world. Moreover, pneumonia persists as a significant cause of sepsis deaths with mortality rates consistently reaching 50% [1–3]. Mortality is highest in developing countries, and in the Brazilian public health system, pneumonia is the second most common ICU admission diagnosis and the third major cause of in-hospital mortality [4].

Severity prediction scores have been refined and new ICU scores developed to create an ideal systematic model that performs well in a diverse, complex, and increasingly aging ICU population. These tools, however, often include variables that are cumbersome to obtain within the first 24 hours after admission [5, 6].

Widely used scores such as the Simplified Acute Physiology Score 3 (SAPS 3) lack sufficient evidence for use in patients with pneumonia admitted to the ICU [5, 7, 8]. Furthermore, severity of pneumonia by scores such as CURB-65 (confusion, urea, respiratory rate, blood pressure, age), CRB-65 (confusion, respiratory rate, blood pressure, age), and the Pneumonia Severity Index have been used in recent trials to stratify patients by severity to evaluate the efficacy of corticosteroids, often with conflicting mortality outcomes [9–13]. Another trial of corticosteroids specifically excluded patients requiring immediate transfer to the ICU, a population that arguably is in dire need of adjuvant therapy for pneumonia [14, 15]. It is unknown whether these mixed results accurately reflect the effect of corticosteroids in those with severe pneumonia and, more broadly, whether pneumonia-specific severity scores are accurate in an ICU population admitted with pneumonia.

Given these research gaps, additional studies are required to determine whether pneumonia-specific scores developed outside the ICU or severity scores specific to the ICU are accurate measures of mortality risk in an increasingly elderly population admitted to the ICU with pneumonia. In this study, we evaluated whether ICU and non-ICU pneumonia severity scores accurately predict mortality in critically ill patients admitted with pneumonia.

METHODS

This was an observational analytical cohort study conducted over a period of 3 years between August 2015 and July 2018 in a general ICU with 22 beds in a tertiary care hospital in Salvador, Bahia, Brazil. Over the study period, 2401 patients were admitted to the ICU, of whom 200 met inclusion criteria with a diagnosis of pneumonia at time of admission. The external validation cohort was derived from the Community-Acquired Pneumonia Organization (CAPO) cohort of patients admitted to the ICU with pneumonia, yielding 362 patients with complete data (among 405 patients total) with documented inspired oxygen percentage or assumed fraction of inspired oxygen (FiO₂) ≥ 30% undergoing mechanical ventilation (Figure 1).

Figure 1.

Patient flowchart of discovery and validation cohort. A total of 2401 patients were initially admitted to the intensive care unit in the discovery cohort between August 2015 to July 2018, of which 200 met inclusion criteria. In the validation cohort, 405 patients were initially included from the Community-Acquired Pneumonia Organization dataset, of which 43 had incomplete data, resulting in 362 patients included in the final analysis. Abbreviations: CAPO, Community-acquired Pneumonia Organization; ICU, intensive care unit.

The primary outcome evaluated was ICU mortality. Data for our cohort were prospectively collected for all those admitted to the ICU with pneumonia from the emergency department, hospital wards, or interhospital transfers. Pneumonia was defined by clinical and radiographic data with infiltrate on chest imaging and compatible clinical syndrome for pneumonia. Nosocomial pneumonia, as defined by the Brazilian Consensus, is pneumonia acquired 48 hours or more after hospital admission, in contrast with community-acquired pneumonia (CAP), which is present within the first 48 hours of admission [16].

Clinical and laboratory data were prospectively collected daily, and end of follow-up was determined by discharge from the ICU. Study variables included age, weight, height, sex, comorbidities, functional capacity, admission diagnosis, origin, length of ICU and hospital stay, and physiological and laboratory data within the first 6 hours of admission. Complications including need for mechanical ventilation, vasopressors, and other supportive therapy in the ICU were noted. Calculated prognostic scores were recorded.

We analyzed the performance of SAPS 3 and the quick Sepsis-Related Organ Failure Assessment (qSOFA) as ICU-specific severity scores, and the pneumonia-specific scores CURB-65 and CRB-65. Other prognostic ICU scores that were evaluated included the Modified Frailty Index (MFI) and Charlson Comorbidity Index (CCI). Data were prospectively recorded in the Epimed Monitor system, which contained all variables of interest for this study. This study was approved by the Research Ethics Committee of Hospital Ana Nery under the number 2.571.265 and Certificate of Presentation for Ethical Appreciation 52892315.1.0000.0045.

Statistical Analysis

Categorical variables were expressed as frequency and percentage, and continuous variables were expressed as median with interquartile range (IQR).

The proportion of categorical variables between groups was compared using the Fisher exact test or χ ² test. The medians of continuous variables were compared using the Mann-Whitney test when analyzing the outcome groups. All tests were 2-tailed and considered statistically significant at P ≤ .05.

To assess for potential confounders, variables that demonstrated possible statistical associations in univariate analysis (P < .1) were transformed from continuous variables into categorical variables whose cutoff values were identified using receiver operating characteristic (ROC) curve analysis for a specificity and sensitivity of 0.80 and the median (25th–75th percentile) value of the variable in nonsurvivors. Clinically important variables that are routinely available in most resource settings, including creatinine and sodium, were also transformed to categorical variables using cutoffs determined by ROC curve analysis. Our stepwise multivariate logistic regression model yielded 8 variables associated with mortality (P < .2) that were included in the composite Pneumonia Shock score (Figure 2): age ≥ 75 years, heart rate ≥ 110 beats per minute, hematocrit ≤ 38%, white blood cell count ≥ 15 × 10³ cells/μL, sodium ≥ 145 mmol/L, FiO₂ ≥ 30%, use of vasopressors, and presence of obtundation by Glasgow Coma Scale < 15.

Figure 2.

Adjusted and unadjusted multivariate regression model for intensive care unit mortality. Univariate analysis yielded unadjusted odds of death. Multivariate regression adjusted for differences in baseline characteristics (variables of P < .1 identified in univariate analysis). *Documented inspired oxygen percentage or assumed fraction of inspired oxygen (FiO2) ≥ 30% undergoing mechanical ventilation. Abbreviations: CI, confidence interval; FiO₂, fraction of inspired oxygen; WBC, white blood cell.

The weight of each variable was determined based on variability in the odds ratio for a confidence interval (CI) of 95%. Given these parameters, age and vasopressor use were weighted 2 points while other variables were given 1 point in the score calculation, with total score values ranging from a minimum of 0 to a maximum of 10. Predictive performance of the Pneumonia Shock score was evaluated by calculating the area under the ROC curve (AUC), with ≥ 0.8 considered most predictive. Performance of the Pneumonia Shock score with the pneumonia and ICU severity scores was compared using a 2-tailed Z test to evaluate the absolute AUC and difference in AUC derived from the empirical ROC curves produced by NCSS statistical software. A Cox proportionate test analysis was performed to determine predictive performance adjusted for differences in baseline characteristics in survivors and nonsurvivors. External validation of the Pneumonia Shock score was performed using data provided by CAPO that was limited to patients with pneumonia admitted to the ICU [17]. The data were analyzed using Microsoft Excel (Office 365), GraphPad Prism version 6.01, and SPSS, version 25.0 software.

RESULTS

The cohort had a median age of 81 (IQR, 67–90) years and a predominance of women (n = 104 [52%]). Patients were primarily admitted from the emergency department (n = 130 [65%]) and the median ICU length of stay was 8 (IQR, 4–16) days. The majority of patients were admitted with a diagnosis of CAP (n = 161 [80.5%]); vital signs at time of admission were notable for a median systolic blood pressure of 130 (IQR, 112–151) mm Hg, mean arterial pressure of 93.5 (IQR, 79–108) mm Hg, heart rate of 91 (IQR, 77–109) beats per minute, respiratory rate of 22 (IQR, 19–25) breaths per minute, and axillary temperature of 36.2°C (IQR, 35.5°C–36.7°C). Additional findings are detailed in Table 1.

Table 1.

Univariate Analysis of Baseline Cohort Characteristics

Characteristic	All Patients (N = 200)	Nonsurvivors (n = 71)	Survivors (n = 129)	P Value
Age, y	81 (67–90)	84 (75–91)	79 (65–89)	.021
Female sex	104 (52.0)	32 (45.1)	72 (55.8)	.183
BMI, kg/m2	23 (20–26.7)	22.2 (19.1–24.4)	23.4 (20.8–27.6)	.014
ICU length of stay, d	8 (4–16)	13 (6–23)	6 (4–13.5)	.001
ICU diagnosis				.001
CAP	161 (80.5)	48 (67.6)	113 (87.6)
Nosocomial pneumonia	39 (19.5)	23 (32.4)	16 (12.4)
Admission source				.409
Emergency department	130 (65.0)	35 (49.3)	95 (73.6)
Ward	22 (11.0)	13 (18.3)	9 (7.0)
Home care	12 (6.0)	6 (8.5)	6 (4.7)
Transfer	36 (18.0)	17 (23.9)	19 (14.7)
Autonomy				.104
Independent	131 (65.5)	42 (59.2)	89 (69.0)
Need for assistance	28 (14.0)	10 (14.1)	18 (14.0)
Restricted/bedridden	41 (20.5)	19 (26.8)	22 (17.1)
Vital signs
Systolic blood pressure, mm Hg	130 (112–151)	127 (108–148)	133 (113–153)	.159
Mean arterial pressure, mm Hg	93.5 (79–108)	73 (61–83)	96 (81.67–110)	.216
Heart rate, beats/min	91 (77–109)	99 (85–114)	87 (74–104)	< .001
Respiratory rate, breaths/min	22 (19–25)	22 (19–26)	22 (19–25)	.573
Temperature, °C	36.2 (35.5–36.7)	36.05 (35.2–36.5)	36.2 (35.8–36.7)	.172
Laboratory results
Leukocytes, × 109/L	12.5 (9.1–17.5)	13.4 (10.2–18.7)	12.2 (8.8–16.1)	.046
Platelets, ×103/μL	217 (168–301)	214 (152–292)	221 (172–304)	.656
Hematocrit, %	34.65 (29.5–38.7)	33.30 (27.4–37.5)	35.50 (30–39.3)	.022
Creatinine, mg/dL	0.80 (0.6–1.3)	0.80 (0.5–1.5)	0.80 (0.6–1.2)	.532
BUN, mg/dL	23.36 (16.4–40.4)	31.78 (19.4–47.8)	21.50 (15–32.7)	< .001
PaO2, mm Hg	89 (71–127)	93 (77–135)	87 (69–115)	.105
FiO2, %	26.5 (25–50)	40 (25–100)	25 (21–50)	< .001
Sodium, mmol/L	140 (136–143)	140 (134–146)	140 (136.75–143)	.411
Outcomes
Confusion	43 (21.5)	26 (36.6)	17 (13.2)	< .001
Mechanical ventilation	57 (28.5)	30 (42.3)	27 (20.9)	.002
Noninvasive ventilation	51 (25.5)	23 (32.4)	28 (21.7)	.127
Use of vasopressors	27 (13.5)	20 (28.2)	7 (5.4)	< .001

Continuous variables are represented as median (interquartile range); values were compared using the Mann-Whitney U test. Qualitative variables are shown as frequency (%) and compared using the Fisher exact test or Pearson χ ² test.

Abbreviations: BMI, body mass index; BUN, blood urea nitrogen; CAP, community-acquired pneumonia; FiO₂, fraction of inspired oxygen; ICU, intensive care unit; PaO₂, partial pressure of oxygen.

ICU supportive care included use of vasopressors in 27 (13.5%) and mechanical ventilation in 57 (28.5%) patients. Other variables evaluated can be found in Table 1. Prognostic ICU scores determined by the CCI, MFI, and Braden scale demonstrated a median value of 1 (IQR, 0–3), 2 (IQR, 1–3), and 12 (IQR, 10–14), respectively. Severity of disease was evaluated by pneumonia and ICU severity scores that included SAPS 3, CURB-65, CRB-65, and qSOFA with a median value of 55 (IQR, 50–62), 2 (IQR, 2–3), 2 (IQR, 1–2), and 1 (IQR, 1–2), respectively (Table 2).

Table 2.

Severity of Illness Scores in Cohort Stratified by Mortality

Severity Score	All Patients (N = 200)	Nonsurvivors (n = 71)	Survivors (n = 129)	P Value
Braden	12 (10–14)	10 (9–12)	13 (11–15.75)	< .001
MFI	2 (1–3)	2 (1–3)	2 (1–3)	.636
CCI	1 (0–3)	2 (1–3)	1 (0–3)	.129
SAPS 3	55 (50–62)	60 (55–71)	53 (47–58)	< .001
qSOFA	1 (1–2)	1 (1–2)	1 (1–2)	.001
CURB-65	2 (2–3)	3 (2–3)	2 (1–3)	< .001
CRB-65	2 (1–2)	2 (1–2)	2 (1–2)	.001

Data are expressed as median (interquartile range). Increased Braden score reflects decreased risk of pressure ulcers. Increases in all other scores are associated with increased frailty, comorbidities, and risk of death, respectively.

Abbreviations: CCI, Charlson comorbidity index; CRB-65, confusion, respiratory rate, blood pressure, age; CURB-65, confusion, urea, respiratory rate, blood pressure, age; MFI, Modified Frailty Index; qSOFA, quick Sequential Organ Failure Assessment; SAPS 3, Simplified Acute Physiology Score 3.

Comparison Between Nonsurvivors and Survivors

Nonsurvivors were significantly older than survivors (median age, 84 [IQR, 75–91] years vs 79 [IQR, 64.5–88.5] years; P = .021). Initial heart rate was significantly higher in nonsurvivors compared with survivors (99 [IQR, 85–114] beats per minute vs 87 [IQR, 74–104] beats per minute; P ≤ .001), whereas respiratory rates did not differ (22 [IQR, 19–26] breaths per minute vs 22 [IQR, 19–25] breaths per minute; P = .573). FiO₂ used was significantly higher in nonsurvivors (40% vs 25%; P < .001). Laboratory results were notable for significant increases in leukocytes (13.4 vs 12.2 10⁹/L; P = .046) and blood urea nitrogen (31.8 vs 21.5 mg/dL; P < .001) in nonsurvivors.

Mental confusion (n = 26 [36.6%] vs 17 [13.2%]; P < .001) was significantly higher among nonsurvivors compared with survivors. CURB-65, CRB-65, qSOFA, SAPS 3, and Braden scores were significantly higher among nonsurvivors than survivors (P < .001), but no significant differences were noted in MFI and CCI scores (P = .636 and P = .129) (Table 2).

Existing Score Performance and the Pneumonia Shock Score

The Pneumonia Shock score performed significantly well in prediction of ICU mortality with an AUC of 0.80 (95% CI, .73–.86) (Figure 3). In those individuals with a score ≤ 2, the mortality rate was 9.3%, whereas those with a score ≥ 3 had a mortality rate > 26% (Supplementary Figure 1).

Figure 3.

Performance of Pneumonia Shock score in the discovery cohort. Receiver operating characteristic curve analysis to determine accuracy of the Pneumonia Shock score in predicting intensive care unit death. Abbreviations: AUC, area under the receiver operating characteristic curve; CI, confidence interval; Sens., sensitivity; Spec., specificity.

Discriminate function of SAPS 3, qSOFA, CURB-65, and CRB-65 was limited, with no ICU or pneumonia severity score reaching an AUC threshold of 0.80 to accurately detect those at highest risk of death admitted to the ICU with pneumonia (AUC = 0.74, 0.64, 0.65, and 0.63, respectively). The Pneumonia Shock score did not differ significantly in performance in those admitted with community-acquired or nosocomial pneumonia. Evaluation of the composite score by ROC analysis identified a cutoff score of 3.5 that was accurate and significantly outperformed all pneumonia severity scores in the discovery cohort (P < .0008) (Figure 4A). To more fully determine differences in model performance beyond AUC comparisons, predictive performance by Cox proportionate test analysis demonstrated the superiority of the Pneumonia Shock score compared to SAPS 3 in prediction of ICU mortality in those admitted with pneumonia (Supplementary Figure 2).

Figure 4.

Comparisons of discriminate function of pneumonia scores, intensive care unit (ICU) scores, and the Pneumonia Shock score in the discovery and validation cohort. Receiver operating characteristic (ROC) curve analysis to determine score accuracy in prediction of ICU death. A, The Pneumonia Shock score outperformed the Simplified Acute Physiology Score 3 (SAPS 3), the quick Sequential Organ Failure Assessment (qSOFA), and CURB-65 (confusion, urea, respiratory rate, blood pressure, age) and CRB-65 (confusion, respiratory rate, blood pressure, age), as well as in prediction of mortality in the discovery cohort. P values refer to comparisons of area under the ROC curve (AUC) of the Pneumonia Shock score with the severity models analyzed in the discovery cohort. B, In the validation cohort, the Shock score performed equally well with comparable discriminate function. The Pneumonia Shock score was significantly superior to all severity models analyzed in the validation cohort, with P values referring to AUC comparisons.

Patient characteristics, vital signs, and mortality rate in the CAPO cohort were similar, albeit patients were younger than those included in our discovery cohort (Supplementary Table 1). Evaluation of pneumonia- and ICU-specific scores demonstrated poor performance similar to findings in our discovery cohort, with the Pneumonia Shock score significantly outperforming CURB-65, CRB-65, and qSOFA as determined by AUC comparisons (P < .006). Furthermore, the Pneumonia Shock score performed well with continued excellent discriminate function in this external cohort with comparable performance to our discovery cohort (P > .80) (Figure 4B).

DISCUSSION

The findings presented here identify the shortcomings of both ICU- and pneumonia-specific severity scores in those patients admitted to the ICU with pneumonia, in both elderly and nonelderly populations. As ICU populations age around the world and are increasingly admitted with pneumonia, early identification and intervention are critical [18–20]. Validation of this score in ICU patients admitted with pneumonia from the CAPO cohort demonstrated that the Pneumonia Shock score also performs well in a younger ICU population distinct from our discovery cohort. Reliance on existing scores may fail to accurately identify those at highest risk of death in the ICU, thereby resulting in delays in early interventions and robust monitoring. Moreover, it is apparent that elderly individuals may be misclassified at a higher risk of death by inaccurate scoring systems. Furthermore, poor score performance may confound clinical trial enrollment that aims to decrease mortality related to severe pneumonia, resulting in conflicting results. Our Pneumonia Shock score is a simple tool that leverages data gathered in routine ICU care within the first 6 hours of admission to determine the mortality risk of those admitted to the ICU with either CAP or nosocomial pneumonia. Importantly, organisms identified in respiratory cultures in our discovery cohort were primarily healthcare-associated gram-negative pathogens, suggesting that the clinical definition of CAP based on duration of hospitalization may be inadequate to distinguish nosocomial from community-acquired pneumonia. In this context, the Pneumonia Shock score was found to perform well in those admitted to the ICU with either nosocomial or community-acquired pneumonias. Prior work that specifically evaluated pneumonia and ICU severity scores, such as the Pneumonia Severity Index, CURB-65, SAPS, and Acute Physiology and Chronic Health Evaluation (APACHE), have demonstrated poor performance in predicting pneumonia mortality in the ICU [21–23]. While other studies have evaluated pneumonia and ICU severity scores, most failed to include patients in the ICU and others focused exclusively on patients with community-acquired pneumonia [24–26]. Furthermore, these studies have either excluded those patients directly admitted to the ICU or focused narrowly on ventilator-associated pneumonia [21, 23, 27]. Other scores designed for use in patients with pneumonia requiring ICU admission, such as PIRO (predisposition, insult, response, and organ dysfunction), include variables not easily obtained within 24 hours after admission, thus compromising routine use for early prediction of mortality and clinical trial enrollment [27, 28]. In contrast to findings in the comorbid diseases (‘D’) and oxygen saturation (‘S’)‐CRB‐65 study, performance of our Pneumonia Shock score did not improve with inclusion of comorbidities including heart failure, neoplasm, chronic renal or hepatic disease, or neurologic dysfunction [29]. With the emergence of simple evidence-based interventions for those with hypoxic respiratory failure and severe pneumonia, including prone positioning, our score may identify those most likely to benefit from early implementation. Our study cautions intensivists’ use of established pneumonia and ICU severity scores as they may inaccurately determine mortality risk, particularly in an elderly ICU population. As the ICU population worldwide comprises an increasingly elderly population, often admitted with pneumonia, accurate tools to predict mortality will be critical to target resources toward those at highest risk of death.

While our study has multiple strengths, including a well-characterized elderly discovery cohort with external validation and comparable mortality rate to other studies, there are certain limitations that should be acknowledged. First, as a single-center study, there may be unknown patient and healthcare provider factors that were not readily apparent in this analysis. Given the robust prospective data collection utilized in this study, and heterogeneous cohort population, it is unlikely that our study participants differ from other populations that may confound the performance of the Pneumonia Shock score. Second, the relatively small but well-characterized study population in our hospital may have specific factors that independently improve score performance. While infection with resistant organisms in our Brazilian cohort may be distinct from other countries, the performance of the Pneumonia Shock score in the CAPO cohort from the United States suggests that this is unlikely to impact score performance. The lack of available data on antibiotic resistance as well as on potential inappropriate use of antibiotics may have impacted the occurrence of shock and need of vasoactive drugs. Finally, delay in appropriate antibiotic therapy was not evaluated and may impact mortality rates in our cohort. While this may be a factor in determining variables associated with mortality, the performance of our score relies on simple, readily available data at the time of admission that accurately identifies those at highest risk of ICU death.

CONCLUSIONS

The Pneumonia Shock score is a novel parsimonious tool designed to aid intensivists and emergency physicians to accurately triage and intervene in those patients admitted to the ICU with pneumonia. The composite score developed here outperformed prior scores analyzed in our cohort, demonstrates excellent discriminate function in a distinct validation cohort, and offers an alternative prognostic tool with robust performance to predict mortality in those with pneumonia.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Author contributions. K. M. A., N. M. F. F., and B. B. A. were responsible for study design, implementation, and manuscript preparation and had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. T. A. C., I. B. B. F., R. C. M., G. P. T., M. L. O., L. P. N., and S. A. contributed substantially to the study design, data analysis, and interpretation, and writing and review of the manuscript. K. F. F. and M. B. A. were responsible for advance statistical analysis, figure generation, and manuscript review and preparation.

Acknowledgments. The authors acknowledge the research groups Grupo de Estudo em Medicina Intensiva, linked to the Núcleo de Ensino e Pesquisa do Hospital da Cidade, and Multinational Organization Network Sponsoring Translational and Epidemiological Research Initiative, linked to Fundação Oswaldo Cruz. We thank the Community-Acquired Pneumonia Organization for providing data from its database.

Disclaimer. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Financial support. The work of B. B. A. was supported by the National Institutes of Health (grant number U01AI115940). K. F. F. received a fellowship from the Programa Nacional de Pós-Doutorado/Coordenação de Aperfeiçoamento de Pessoal de Nível Superior.

Potential conflicts of interest. The authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest.