Corticosteroids in Patients Hospitalized With Community-Acquired Pneumonia: Systematic Review and Individual Patient Data Metaanalysis

Abstract

Background

Our aim was to evaluate the benefits and harms of adjunctive corticosteroids in adults hospitalized with community-acquired pneumonia (CAP) using individual patient data from randomized, placebo-controlled trials and to explore subgroup differences.

Methods

We systematically searched Medline, Embase, Cochrane Central, and trial registers (all through July 2017). Data from 1506 individual patients in 6 trials were analyzed using uniform outcome definitions. We investigated prespecified effect modifiers using multivariable hierarchical regression, adjusting for pneumonia severity, age, and clustering effects.

Results

Within 30 days of randomization, 37 of 748 patients (5.0%) assigned to corticosteroids and 45 of 758 patients (5.9%) assigned to placebo died (adjusted odds ratio [aOR], 0.75; 95% confidence interval [CI], .46 to 1.21; P &#x003D; .24). Time to clinical stability and length of hospital stay were reduced by approximately 1 day with corticosteroids (–1.03 days; 95% CI, –1.62 to –.43; P &#x003D; .001 and –1.15 days; 95% CI, –1.75 to –.55; P < .001, respectively). More patients with corticosteroids had hyperglycemia (160 [22.1%] vs 88 [12.0%]; aOR, 2.15; 95% CI, 1.60 to 2.90; P < .001) and CAP-related rehospitalization (33 [5.0%] vs 18 [2.7%]; aOR, 1.85; 95% CI, 1.03 to 3.32; P &#x003D; .04). We did not find significant effect modification by CAP severity or degree of inflammation.

Conclusions

Adjunct corticosteroids for patients hospitalized with CAP reduce time to clinical stability and length of hospital stay by approximately 1 day without a significant effect on overall mortality but with an increased risk for CAP–related rehospitalization and hyperglycemia.

community-acquired pneumonia, corticosteroids, randomized clinical trials, individual patient data metaanalysis

Worldwide, lower respiratory tract infections are the second most common cause of death and the leading cause of death due to infectious diseases [1]. Despite widespread availability of antibiotics, 30-day mortality remains high [2, 3]. Furthermore, hospitalization rates for community-acquired pneumonia (CAP) are increasing, mainly due to rising life expectancy [2]. In Europe, pneumonia costs health funders approximately US$ 11.8 billion annually [4].

An excessive inflammatory cytokine response could be a major contributor to the high mortality rate in CAP [5]. Corticosteroids naturally downregulate proinflammatory cytokine production. Systemic corticoid therapy attenuates the inflammatory response [6] and, by doing so, may decrease the frequency of acute respiratory distress syndrome (ARDS), reduce the length of illness and of hospital stay, and reduce mortality.

Several randomized clinical trials (RCTs) have investigated adjunctive corticosteroid treatment in patients with CAP, and a recent metaanalysis with aggregate data suggested a reduction in the need for mechanical ventilation and ARDS, a decrease in time to clinical stability and in duration of hospitalization, an increased risk of treated hyperglycemia, and, for the subgroup of trials that investigated severe CAP, a reduction in 30-day mortality of 9% with adjunctive corticosteroids [7]. Trials that include only patients with severe CAP all showed beneficial effects [8, 9], while others did not [10], suggesting that particular patient subgroups may benefit more from corticosteroid therapy than others. Subgroup analyses within individual trials are limited by low statistical power, and subgroup analysis in metaanalyses with aggregate data is compromised by ecological bias.

Individual patient data metaanalysis (IPDMA) can overcome these problems and has further advantages including standardized definitions and analyses across studies and adjustment for variations in individual patient prognosis at baseline, and thus allows for more powerful investigations of subgroup effects [11]. In this systematic review and IPDMA, we investigated the association of adjunctive therapy with corticosteroids and patient-important outcomes among adults with CAP. Also, we determined whether effects differ across predefined subgroups.

METHODS

The study protocol for this metaanalysis of IPD is provided in the Supplementary Materials 2.

Eligibility Criteria, Information Sources, and Study Selection

We conducted a systematic search to identify RCTs that compared placebo with low-dose oral or intravenous corticosteroid therapy (≤1 mg/kg/day of methylprednisolone or an equivalent corticosteroid) as adjunctive therapy in CAP. Studies with pseudorandomization or with treatment combinations that did not allow investigation of an independent corticosteroid effect were excluded. Patients had to be aged ≥16 years with a diagnosis of CAP. We replicated the search strategy of a Cochrane review with similar inclusion criteria, which identified studies up to December 2010 [12]. We electronically searched Medline, Embase, and the Cochrane Controlled Trials Registry from 1 January 2010 to 12 July 2017 using medical subject headings based on the terms “pneumonia” and “corticosteroid” (see Supplementary Materials 3 for detailed search strategies). No language restrictions were set.

In addition, we checked trial registries (ClinicalTrials.gov and the World Health Organization International Clinical Trials Registry Platform Search Portal (apps.who.int/trialsearch/) for ongoing or unpublished eligible trials, manually searched the reference lists of included studies and existing systematic reviews as well as all articles citing the included studies on Google Scholar, and hand searched abstracts of meetings related to infectious diseases and pulmonary diseases in 2015 and 2016 (American Thoracic Society, European Congress of Clinical Microbiology and Infectious Diseases, the American Association of Respiratory Care, the American College of Chest Physicians, and the Interscience Conference on Antimicrobial Agents and Chemotherapy).

Data Collection

Working in teams of 2, authors (M. B., S. M. C. S., R. A. C. S) independently assessed trial eligibility based on titles, abstracts, information in registers, and full texts; disagreements were resolved by discussion. We requested protocols and original databases from investigators of all eligible trials. For 3 trials, we were unable to obtain IPD despite several email requests and phone calls with investigators [13–15]. Data from the remaining trials were checked against reported results, and queries were resolved with corresponding principal investigators or trial statisticians. Due to standardized outcome definitions and data analyses, some of the results may differ slightly from those published in original articles.

Risk of Bias and Quality Assessments

We systematically assessed the risk of bias in included trials with respect to allocation concealment, patients lost to follow-up, and early stopping for benefit [16]. We used the Grading of Recommendations Assessment, Development, and Evaluation system to rate the certainty of evidence (also known as quality of the evidence) [17]. Certainty of evidence takes into consideration the study design (here, RCTs); risk of bias, precision, consistency, and directness of the evidence; and the possibility of publication bias.

Outcomes and Subgroups

The primary outcome was 30-day mortality. Secondary outcomes were admission to the intensive care unit (ICU), length of hospital stay (LOS), early (<72 hours) and late (>72 hours) treatment failure, CAP-related readmission within 30 days after discharge, duration of intravenous antibiotic treatment (switch to oral in case of substantial clinical improvement), side-effects of corticosteroid treatment (rate of hyperglycemia, nosocomial infections, gastrointestinal hemorrhage, severe neuropsychiatric symptoms), and development of empyema. Since definitions of some outcomes differed in the published trials, we predefined all variables across trials (Supplementary Materials 4). If variables were not prospectively collected, an attempt was made to collect them retrospectively. Prespecified subgroup analyses were performed for the following 5 patient baseline characteristics with the following hypotheses: patients with more severe CAP benefit more from corticosteroids (low-risk group pneumonia severity index [PSI] classes I–III vs high-risk group PSI classes IV–V); patients immediately admitted to the ICU benefit more from corticosteroids than those admitted to normal wards; patients with higher than median C-reactive protein levels benefit more from corticosteroids than others; patients with ≥2 systemic inflammatory response syndrome (SIRS) criteria benefit more from corticosteroids than those with 0–1 SIRS criteria; and patients with antibiotic treatment before hospital admission benefit more from corticosteroids than those without antibiotic pretreatment.

Furthermore, we performed the following 5 exploratory subgroup analyses: patients aged <70 years vs >70 years, patients with vs without chronic obstructive pulmonary disease (COPD), patients with vs without positive blood cultures, longer (≥6 days) vs shorter (<6 days) duration of corticosteroid therapy, and patients without identified microorganism vs patients with identified Streptococcus pneumoniae vs patients with other identified microorganism.

Statistical Analyses

The primary analyses followed the intention-to-treat principle, with all patients analyzed according to the arm to which they were randomized. Results from all trials were pooled using multilevel models, with patient-level variables as fixed effects and individual trials as a random effect to account for within- and between-trial variability. We calculated odds ratios (ORs) and 95% confidence intervals (CIs) for binary data and mean differences between groups with 95% CIs for continuous data. All analyses were adjusted for important prognostic factors (eg, age, pneumonia severity) through inclusion of PSI score as a continuous variable and fixed effect.

To examine prespecified baseline characteristics as potential effect modifiers, we added each of these baseline variables in turn to the statistical model together with the corresponding interaction term with treatment group, both as fixed effects.

The multivariable fractional polynomials interaction (MFPI) approach was used to investigate the potential treatment-modifying effects of the continuous variables PSI score, age, and C-reactive protein level at randomization on time to death within 30 days and LOS [18, 19]. We used MFPI with FP1 functions as the most complex allowable function and created treatment effect functions to visualize potential interactions.

We performed 4 sensitivity analyses. First, we conducted an analysis on the per-protocol population (only patients who fully complied with the respective trial protocol). Second, we excluded the RCT by Confalonieri et al. because it was stopped early for benefit and almost certainly overestimated treatment effect [16]. Third, instead of including trials as a random effect in the statistical model, we included a categorical “trial” variable as an additional fixed effect to account for the fact that there were only 6 trials in total. We used Stata, version 13.2 (College Station, Texas), for all analyses, and set P < .05 as the level of statistical significance.

RESULTS

Study Selection

We identified 14 eligible trials (Figure 1). Five trials were still ongoing (Supplementary Materials 5). For 3 trials (including 197 participants and 27 deaths), the investigators did not provide IPD (Supplementary Materials 6) [13–15], leaving 6 trials for inclusion in this IPDMA. After post-randomization exclusion of patients who did not meet the eligibility criteria of the included trials, 748 patients were randomized to corticosteroids and 758 patients to placebo. This represents 88.4% (1506/1703) of all randomized patients in completed eligible trials.

Figure 1.

Study flow chart. Abbreviations: RCT, randomized clinical trial; WHO, World Health Organization. *Chen Y, Li K, Pu H, Wu T. Corticosteroids for pneumonia. Cochrane Database Syst Rev 2011:CD007720.

Study and Patient Characteristics

Included trials were all investigator-initiated and conducted in Europe between 2000 and 2014 (Table 1). All patients were adults with a median age of 70 years (interquartile range, 56–81 years), 59% were men, 52% had severe pneumonia (PSI class IV–V), S. pneumoniae was identified in 22%, and 10% were admitted directly to an ICU. Pooled baseline characteristics were similar between groups (Table 2). Patients received corticosteroid treatment—hydrocortisone [8], prednisone [20], prednisolone [10], methylprednisolone [9, 21], dexamethasone [22]—for 4 [22] to 9 days [21]. Most trials excluded patients at high risk for adverse events from corticosteroids, including those with immunosuppression, gastrointestinal bleeding within 3 months, and pregnancy. Five of the 5 trials were low risk of bias; 1 was stopped early for benefit [8] (Supplementary Materials 7).

Table 1.

Characteristics of Included Randomized, Placebo-Controlled Trials

Trial, Reference and year	Country	Patients, no.	Study Period	Median Age (Interquartile Range), y	Men, no. (%)	Patient Population	Corticosteroid, Dose, Route, and Duration	Follow-up for Mortality, months
Confalonieri, 2005	Italy	46	2000–2003	67.5 (51–76)	32 (70)	Patients with severe CAP according to 1993 ATS severity criteria	Hydrocortisone 200 mg IV bolus followed by 10 mg/h IV for 7 d	3
Snijders, 2010	The Netherlands	204	2005–2008	64.5 (71–80)	118 (58)	Adults hospitalized with CAP	Prednisolone 40 mg IV or orally for 7 d	84
Meijvis, 2011	The Netherlands	302	2007–2010	66.5 (51–79)	169 (56)	Adults with CAP but without need for intensive care	Dexamethasone 5 mg IV daily for 4 d	72
Fernandez-Serrano, 2011	Spain	52	2000–2002	62.5 (47–68.5)	16 (31)	Adults up to age 75 years with severe CAP (consolidation of ≥2 lobes and PO2/ FIO2 <300)	Methylprednisolone 200 mg IV bolus followed by tapering infusion (3.3–0.8 mg/h IV) over 9 d	1
Blum, 2015	Switzerland	785	2009–2014	73 (61–83)	487 (62)	Adults hospitalised with CAP	Prednisone 50 mg oral daily for 7 d	6
Torres, 2015	Spain	120	2004–2012	69.5 (63–81)	74 (62)	Adults with severe CAP according to ATS or pneumonia severity index criteria and C-reactive protein >150 mg/L	Methylprednisolone 0.5 mg/kg IV twice daily for 5 d	1

Abbreviations: ATS, american thoracic society; CAP, community-acquired pneumonia; IV, intravenous.

Table 2.

Baseline Characteristics of the Pooled 1506 Patients from 6 Randomized Clinical Trials on Adjunctive Corticosteroid Therapy in Community-Acquired Pneumonia

	Corticosteroid	Placebo
Characteristic	(N = 748)	(N = 758)
Age (y), median (IQR)	71 (56–81)	70 (55–81)
Female sex, no. (%)	318 (42.5)	294 (38.8)
Current smoker, no. (%)	199 (27.8)	189 (26.4)
Antibiotic treatment before hospitalization, no. (%)	227 (30.5)	233 (30.9)
Immediate admission to intensive care unit, no. (%)	70 (9.4)	77 (10.2)
Clinical parameters
Days with symptoms, median (IQR)	4 (2–7)	4 (2–7)
Temperature (°C), median (IQR)	37.8 (37.1–38.6)	37.9 (37.1–38.6)
Systolic blood pressure (mmHg), median (IQR)	126 (112–141)	126 (112–141)
Heart rate (beats per min), median (IQR)	90 (78–105)	90 (77–106)
Respiratory rate (breaths per min)	24 (19–30)	24 (19–30)
SaO2 (%), median (IQR)	94 (92–96)	95 (92–97)
Confusion, no. (%)	70 (9.5)	70 (9.5)
Laboratory values
C-reactive protein (mg/L), median (IQR)	191 (93–290)	183 (84–292)
White-blood-cell count (109 cells/L), median (IQR)	12.7 (9.3–16.7)	12.6 (9.0–16.7)
Glucose (mmol/L), median (IQR)	6.9 (5.9–8.4)	6.8 (5.9–8.2)
Positive blood cultures (bacterial community-acquired pneumonia), no. (%)	89 (12.8)	87 (12.5)
PSI scorea
class I, ,no. (%)	83 (11.1)	79 (10.4)
class II, no. (%)	132 (17.7)	135 (17.8)
class III, no. (%)	127 (17.0)	172 (22.7)
class IV, no. (%)	285 (38.1)	250 (33.0)
class V, no. (%)	121 (16.2)	122 (16.1)
Total PSI score (points), median (IQR)	95 (68–120)	90 (66–114)
Identified pathogen
Streptococcus pneumonia	171 (22.9)	154 (20.3)
Other pathogenb	146 (19.5)	184 (24.3)
No pathogen found	431 (57.6)	420 (55.4)
Number of systemic inflammatory response syndrome criteria
0, no. (%)	32 (4.5)	19 (2.6)
1, no. (%)	137 (19.1)	151 (20.4)
2, no. (%)	246 (34.2)	239 (33.2)
3, no. (%)	213 (29.6)	218 (30.2)
4, no. (%)	91 (12.7)	94 (13.0)
Comorbidities
Diabetes mellitus (any type), no. (%)	123 (16.4)	130 (17.2)
Insulin treatment, no. (%)	57 (7.7)	46 (6.2)
Chronic obstructive pulmonary disease, no. (%)	125 (16.7)	112 (14.8)
Chronic heart failure, no. (%)	134 (17.9)	126 (16.7)
Cerebrovascular disease, no. (%)	86 (11.5)	78 (10.3)
Renal disease, no. (%)	153 (20.5)	139 (18.4)
Active neoplastic disease, no. (%)	46 (6.2)	49 (6.5)
Liver disease, no. (%)	22 (3.0)	15 (2.0)
Coinfections at admission, no. (%)	48 (7.1)	50 (7.3)

Abbreviations: IQR, interquartile range; PSI, pneumonia severity index.

^aPSI is a clinical prediction rule to calculate the probability of morbidity and mortality among patients with community-acquired pneumonia. PSI risk class I: aged ≤50 years and no risk factors; II: <70; III: 71–90; IV: 91–130; V: >130 points.

^bOf 330 patients with other identified pathogens, there were, for example, 45 (13.6%) with Legionella species, 31 (9.4%) with influenza virus A or B, 29 (8.8%) with Haemophilus influenzae, 29 (8.8%) with Coxiella burnetii, 25 (7.6%) with Mycoplasma pneumoniae, and 18 (5.5%) with Chlamydophilia species.

Primary Outcome: 30-Day All-Cause Mortality

Overall, there were 37 (5.0%) deaths within 30 days in the corticosteroid groups and 45 (5.9%) deaths in placebo groups (aOR, 0.75; 95% CI, .46 to 1.21; P = .24; moderate certainty; Table 3). We found significant heterogeneity in treatment effects for mortality across included trials (P = .046 by likelihood ratio test). Most of this heterogeneity could be explained by the large mortality reduction with corticosteroids in the trial by Confalonieri et al. [8]. Exclusion of Confalonieri et al. in a sensitivity analysis confirmed no mortality difference between groups (5.1% vs 5.2%; aOR, 0.88; 95% CI, .53 to 1.45; P = .61; Supplementary Materials 8). Other sensitivity analyses using the per-protocol population (Supplementary Materials 9) yielded similar results.

Table 3.

Primary and Secondary Outcomes at 30 days After Randomization Using Random Intercepts for Included Trials

Outcome	Corticosteroid (n = 748)	Placebo (n = 758)	Intention-to-Treat Regression analysis, OR or Coefficient (95% Confidence Interval), P Value
Primary
All-cause mortality, no. (%)	37 (5.0)	45 (5.9)	OR 0.75 (0.46 to 1.21), P = .24
Secondary
Secondary intensive care unit admission, no. (%)a	38 (5.6)	43 (6.3)	OR 0.74 (0.45 to 1.21), P = .23
Length of hospital stay, days	7.0 (5.0–11.0)	8.0 (5.0–12.0)	–1.15 days (–1.75 to –0.55), P < .001
Time to clinical stability, daysb	3.0 (2.0–5.4)	4.0 (2.5–7.0)	–1.03 days (–1.62 to –0.43), P = .001
Intravenous antibiotic treatment, daysc	4.0 (3.0–6.0)	5.0 (3.0–7.0)	–0.62 days (–1.07 to –0.16), P = .01
Early (≤72 hours) treatment failure, no. (%)d	40 (5.7)	45 (6.4)	OR 0.84 (0.53 to 1.34), P = .47
Late (>72 hours) treatment failure, no. (%)d	67 (9.5)	66 (9.3)	OR 0.97 (0.67 to 1.40), P = .86
Community-acquired pneumonia–related rehospitalization, no. (%)e	33 (5.0)	18 (2.7)	OR 1.85 (1.03 to 3.32), P = .04
Nosocomial infections, no. (%)	33 (4.4)	25 (3.3)	OR 1.31 (0.77 to 2.24), P = .32
Hyperglycaemia requiring insulin, no. (%)f	160 (22.1)	88 (12.0)	OR 2.15 (1.60 to 2.90), P < .001
Empyema/complicated parapneumonic effusion, no. (%)	12 (1.6)	14 (1.9)	OR 0.90 (0.41 to 1.96), P = .79
Gastrointestinal bleeding, no. (%)	5 (0.7)	5 (0.7)	OR 0.95 (0.27 to 3.33), P = .93
Neuropsychiatric complications, no. (%)	6 (0.8)	2 (0.3)	OR 2.98 (0.60 to 14.9), P = .18

Data are median (interquartile range) or n (%) unless otherwise stated.

Abbreviation: OR, odds ratio.

^aPatients immediately admitted to an intensive care unit were excluded from this analysis (ie, analysis based on 1359 patients).

^bData on time to clinical stability were not available from Meijvis et al. and Confalonieri et al. [8] (ie, analysis based on 1158 patients).

^cData on duration of intravenous antibiotic therapy were not available from Confalonieri et al. [8] (ie, analysis based on 1460 patients).

^dData on early or on late treatment failure were not available from Fernandez-Serrano et al. and from Confalonieri et al. [8, 21] (ie, analysis based on 1411 patients).

^eData on community-acquired pneumonia–related rehospitalizations within 30 days of discharge were not available from Torres et al. [9] (ie, analysis based on 1386 patients).

^fData on hyperglycemia requiring insulin treatment were not available from Confalonieri et al. [8] (ie, analysis based on 1460 patients).

Secondary Outcomes

Time to clinical stability and length of hospital stay were on average 1 day shorter in patients with corticosteroids compared to patients on placebo (adjusted difference, –1.03 days; 95% CI, –1.62 to –.43 days; P = .001; high certainty and –1.15 days; 95% CI, –1.75 to –.55 days; P < .001; high certainty, respectively; Table 3). Patients in the corticosteroid groups had a shorter duration of intravenous antibiotic treatment than those in the placebo groups (adjusted mean difference, –0.62 days; 95% CI, –1.07 to –.16 days; P = .01). Rates of secondary ICU admission, early treatment failure, and late treatment failure were similar in both treatment groups. Patients in corticosteroid groups had a higher incidence of hyperglycemia that required insulin treatment (22.1% vs 12.0%; aOR, 2.15; 95% CI, 1.60 to 2.90; P < .001; high certainty) and a higher incidence of CAP-related rehospitalizations within 30 days of discharge (5.0% vs 2.7%; aOR, 1.85; 95% CI, 1.03 to 3.32; P = .04; moderate certainty). Of note, in the largest and most recently performed trial, this rise in glucose level was only transient in all but 4 patients who were all pretreated with oral antidiabetics (2 of whom had admission hemoglobin A1c levels above 8%) [20]. Other CAP and corticosteroid-related adverse events (nosocomial infections, empyema, gastrointestinal hemorrhage, neuropsychiatric complications) were rare, with similar incidence in corticosteroid and placebo groups. Results were similar in all sensitivity analyses for all secondary outcomes.

Subgroup Analyses

We found no statistically significant effect modification for 30-day all-cause mortality in any of the prespecified interaction analyses, although there was a consistent trend toward larger benefit from corticosteroid treatment for patients who suffered from more severe pneumonia (Table 4). Sensitivity analyses without Confalonieri et al. (Supplementary Materials 10) support these findings. Results from subgroup analyses on LOS, CAP-related rehospitalizations, and hyperglycemia that required insulin treatment are presented in Supplementary Materials 11–13. Sensitivity analyses, including trials as a fixed effect in the statistical model, confirmed reported results (Supplementary Materials 14 and 15).

Table 4.

Subgroup Analyses for the Primary Outcome All-Cause Mortality at 30 days After Randomization Using Random Intercepts for Included Trials

Subgroup Variable	Corticosteroid (n = 748)	Placebo (n = 758)	Logistic Regression, Odds Ratio (95% Confidence Interval), P Value	P for Heterogeneity (Interaction)a
Pneumonia severity indexb				.35
class I–III, no. (%)	3 (0.9)	2 (0.5)	1.70 (0.28–10.3), P = .56
class IV and V, no. (%)	34 (8.4)	43 (11.6)	0.70 (0.44–1.13), P = .14
Initial intensive care admission				.20
No, no. (%)	33 (4.9)	33 (4.9)	0.85 (0.50–1.45), P = .54
Yes, no. (%)	4 (5.7)	12 (15.6)	0.34 (0.10–1.12), P = .08
Initial median C-reactive protein				.72
< 188 mg/L, no. (%)	21 (5.8)	24 (6.4)	0.84 (0.44–1.61), P = .61
≥ 188 mg/L, no. (%)	16 (4.3)	20 (5.5)	0.68 (0.32–1.43), P = .31
Systemic inflammatory response system criteria				.08
0–1, no. (%)	11 (6.4)	7 (4.1)	1.70 (0.60–4.79), P = .31
2 or more, no. (%)	22 (3.9)	33 (5.9)	0.59 (0.33–1.06), P = .08
Antibiotic treatment before hospital admission				.23
No, no. (%)	25 (4.8)	26 (5.0)	0.93 (0.51–1.70), P = .83
Yes, no. (%)	12 (5.3)	18 (7.7)	0.48 (0.21–1.10), P = .08

^aLogistic regression model including an interaction term of the respective subgroup variable with treatment group.

^bThe pneumonia severity index (PSI) is a clinical prediction rule to calculate the probability of morbidity and mortality among patients with community acquired pneumonia. PSI risk class I: aged ≤50 years and no risk factors; II: < 70; III: 71–90; IV: 91–130; V: >130 points.

MFPI analyses suggested greater reductions in LOS for patients with more severe pneumonia (higher PSI score; Figure 2). We did not find evidence for an effect modification of adjunct corticosteroids on mortality by PSI score. However, the interpretability of this MFPI analysis was limited by the small number of deaths toward the lower (PSI score <110) and higher (PSI score >170) ends of the PSI range (Supplementary Materials 16). There was no evidence for an effect modification by age or C-reactive protein level at randomization on LOS or time to death within 30 days.

Figure 2.

Multifractional polynomial interaction for pneumonia severity index score. Abbreviations: CI, confidence interval; LOS, length of stay; PSI, pneumonia severity index.

Exploratory subgroup analyses suggested that CAP patients with identified microorganism other than S. pneumoniae might benefit more from corticosteroid treatment than CAP patients without identified microorganism (Supplementary Materials 17). We found no evidence for effect modification through age, history of COPD, blood culture positivity, duration of corticosteroid treatment, and identification of S. pneumoniae.

DISCUSSION

Our metaanalysis of IPD from 6 randomized, placebo-controlled trials that evaluated corticosteroid treatment, including 1506 patients hospitalized for CAP, showed no statistically significant mortality reduction (moderate certainty; Table 5). We confirm that corticosteroid treatment reduces the time to clinical stability and LOS by 1 day (high certainty); duration of intravenous antibiotic therapy was reduced by 0.6 days, on average. However, corticosteroid treatment resulted in a higher incidence of CAP-related readmissions (number needed to harm [NNH], 45; 95% CI, 18 to 1235; moderate certainty) and an increase in hyperglycemia treated with insulin (NNH, 9; 95% CI 6 to 17; high certainty). This rise in glucose level, however, was transient in general [20]. Other adverse events compatible with corticosteroid treatment were rare and similar in both groups. We found no conclusive evidence for a subgroup effect by PSI class, immediate ICU admission, C-reactive protein level, SIRS criteria, or antibiotic therapy before admission on mortality or other outcome parameters. The exploratory MFPI analysis on LOS suggested a stronger effect of corticosteroid therapy in patients with higher PSI score. This effect was not driven by age.

Table 5.

Summary of Findings

Outcome Timeframe	Study Results and Measurements	Absolute Effect Estimates		Certainty in Effect Estimates (Quality of Evidence)	Summary
		Placebo	Adjunctive Corticosteroid Therapy
Mortality 30 days	OR, 0.75 (95% CI, .46 to 1.21) Based on data from 1506 patients in 6 studies Follow-up 30 days	59 per 1000	45 per 1000	Moderatea	Corticosteroids probably have little or no effect on 30-day mortality
		Difference: 14 fewer per 1000 (95% CI, 31 fewer to 12 more)
Secondary intensive care unit admission More than 30 days	OR, 0.74 (95% CI, .45 to 1.21) Based on data from 1359 patients in 6 studies Follow-up 30 days	63 per 1000	47 per 1000	Moderatea	Corticosteroids probably have little or no effect on secondary intensive care unit admissions
		Difference: 16 fewer per 1000 (95% CI, 34 fewer to 12 more)
CAP-related rehospitalization More than 30 days after discharge	OR, 1.85 (95% CI, 1.03 to 3.32) Based on data from 1386 patients in 5 studies Follow-up 30 days	27 per 1000	49 per 1000	Moderatea	Corticosteroids probably increase CAP-related rehospitalizations
		Difference: 22 more per 1000 (95% CI, 1 more to 57 more)
Hyperglycemia requiring insulin More than 30 days	OR, 2.15 (95% CI, 1.6 to 2.9) Based on data from 1460 patients in 5 studies Follow-up 30 days	120 per 1000	227 per 1000	High	Corticosteroids increase hyperglycemia treated with insulin
		Difference: 107 more per 1000 (95% CI, 59 more to 163 more)
Nosocomial infections 30 days	OR, 1.31 (95% CI, .77 to 2.24) Based on data 1506 patients in 6 trials Follow-up 30 days	33 per 1000	43 per 1000	Moderatea	Corticosteroids probably have little or no effect on nosocomial infections
		Difference: 10 more per 1000 (95% CI, 7 fewer to 38 more)
Length of hospital stay	Scale: days Based on data from 1506 patients in 6 studies	8.0 (Median)	7.0 (Median)	High	Corticosteroids reduce length of hospital stay
		Difference: MD 1.15 fewer (95% CI, 1.75 fewer to .21 fewer)
Time to clinical stability	Scale: days Based on data from 1158 patients in 4 studies	4.0 (Median)	3.0 (Median)	High	Corticosteroids reduce time to clinical stability
		Difference: MD 1.03 fewer (95% CI, 1.62 fewer to .43 fewer)

Abbreviations: CAP, community-acquired pneumonia; CI, confidence interval; MD, mean difference; OR, odds ratio.

^aOutcomes with moderate certainty were rated down for imprecision and small number of events.

Strengths of our study include an explicit study protocol, standardized outcome definitions and post-randomization exclusions across included trials, and analyses based on the intention-to-treat as well as the per-protocol population. To minimize overfitting and data-driven associations, we tested a limited number of prespecified variables for potential effect modification [23]. Further interaction analyses were exploratory with hypothesis-generating interpretation. We allowed for clustering effects of the pooled data by using random intercepts for included trials and conducted various sensitivity analyses applying alternate statistical approaches. We followed current recommendations for subgroup analyses in IPDMA [24] and used the MFPI approach to investigate potential continuous effect modifiers, thereby overcoming limitations of aggregate-data metaanalyses [7] and arbitrary cutoffs for dichotomization of continuous variables. Most included trials met high methodological quality standards with minimal losses to follow-up until day 30 after randomization.

Our study has several limitations. First, investigators of 3 RCTs did not provide IPD: 2 trials including 80 patients each and a follow-up until ICU discharge [14, 15] and 1 trial including 34 patients with in-hospital follow-up [13]. These 3 trials happened to report a significantly larger reduction in mortality with corticosteroids than the 6 trials that provided IPD (aggregate data metaanalyses in Supplementary Materials 18 and 19). Overall, we included 90% of randomized patients in eligible trials. Second, many of the subgroup analyses had limited statistical power. Third, exploratory analyses in this IPDMA suggested a reduction in all-cause mortality in the corticosteroid group if the identified CAP-causing pathogen was different from S. pneumoniae. However, the number of identified pathogens was too low to explore effects in patients with other specific pathogens. In addition, this subgroup finding has low credibility, in part, because the direction of effect is opposite from what we expected a priori. Fourth, we did not specify “all-cause re-admissions at 30 days” as an outcome, because this outcome was not available for almost half of the included patients. Instead, we specified “CAP-related re-admissions” with a standardized definition across trials, which was available for all trials except for et al [9]. Finally, since only European studies were included in the IPDMA, it is unknown whether our results can be generalized to other parts of the world such as low- and middle-income countries where most deaths from CAP in absolute terms occur. In those countries, CAP aetiology and availability of and care in hospitals differ from hospitals of included trials, which could influence the effect of adjunctive corticosteroids. In Western countries such as the United States and Canada, however, current empirical treatment of CAP is quite similar [25, 26].

Recent metaanalyses suggested a significant reduction in all-cause mortality in a subgroup of trials with severe CAP [7, 27, 28], a finding that our IPDMA could not confirm although the effect estimate is similar. This discrepancy is possible due to the exclusion of 3 trials from our IPDMA because IPD were not available, leading to decreased statistical power. However, we note that at least 2 of these 3 trials had high risk of bias [13, 14]. In addition, the finding of an increased risk for CAP-related readmission with corticosteroids in our IPDMA had not been reported by any other previous metaanalysis. Reasons for this discrepancy are likely that definitions of patient populations and readmissions (CAP-related, all-cause) varied considerably across trials included in previous metaanalyses, leading to less stringent analyses compared to ours; other metaanalyses did not exclusively focus on placebo-controlled trials; and other metaanalyses included all or some of the 3 trials for which investigators did not share their IPD [13–15, 29]. In contrast to the metaanalysis by Siemieniuk et al. [7], we did not specify “development of ARDS” or “need for mechanical ventilation” as outcomes in our IPDMA, because in more than a third of included patients none of these outcomes were captured [10, 22, 30]. We examined the related outcome “secondary admission to the ICU,” which was documented for all included trials and should include patients with new development of ARDS and those in need of mechanical ventilation. Our results regarding shorter LOS, faster time to clinical stability, and increased risk of hyperglycemia echo the conclusions of several recent metaanalyses [7, 27, 31].

Adjunctive corticosteroids reduce the time to clinical stability from approximately 4 days to 3 days and in LOS from 8 days to 7 days. Given that the total cost of CAP is more than US$ 11 billion per year in Europe, much of which is related to in-hospital care, a reduction in LOS by 12.5% would have considerable economic impact. This advantage of adjunct corticosteroid therapy, however, may need to be balanced against a possible increase in CAP-related readmissions (NNH of 45). Only an economic analysis can determine whether adjunctive corticosteroids in CAP are cost-effective. Larger RCTs or RCTs in specific subgroups based on pneumonia severity, pathogens, or biomarkers [32] are warranted.

CONCLUSIONS

Adjunctive treatment with corticosteroids in patients hospitalized with CAP reduced time to clinical stability and LOS by approximately 1 day without a significant effect on mortality. Corticosteroids may increase the risk for CAP-related rehospitalization and hyperglycemia.

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. M. B.: participated in study design, data acquisition, analysis of data, interpretation of analysis results, manuscript drafting, and manuscript editing for intellectual content. S. M. S.: participated in study design, data acquisition, interpretation of analysis results, manuscript drafting, and manuscript editing for intellectual content. W. B., W. J. B., and M. C. C.: participated in study conception, interpretation of analysis results, and manuscript editing for intellectual content. B. K.: participated in data acquisition, analysis of data, and manuscript editing for intellectual content. D.S., A.T., S. F. S., U. M., A. G., C. A. B., M. C., and R. A. S.: participated in data acquisition and manuscript editing for intellectual content.

Acknowledgments. We thank Ruud Duijkers for acquiring information from the study by Snijders et al. We thank Willi Sauerbrei for his advice and support regarding multifractional polynomial interaction analysis.

Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Author notes

M. B., S. M. C. S., W. J. W. B., and M. C. C. contributed equally to this manuscript

All authors belonging to the study groups are listed in Supplementary Materials 1.