Efficacy of higher-dose versus lower-dose corticosteroids in community-acquired pneumonia: a systematic review and network meta-analysis

Abstract

Background

Adjunctive corticosteroids improve outcomes in hospitalized patients with community-acquired pneumonia (CAP), but whether higher-dose regimens provide additional benefit over lower-dose regimens remains uncertain.

Methods

We searched PubMed, Embase, and Cochrane databases for randomized controlled trials (RCTs) up to January 25, 2026. Interventions were standardized to protocol-assigned dexamethasone-equivalent doses: high (≥ 7.5 mg/d) and low (< 7.5 mg/d). A frequentist network meta-analysis was performed. The primary outcome was short-term all-cause mortality. Confidence in network meta-analysis estimates was assessed using the Confidence in Network Meta-Analysis (CINeMA) framework.

Results

32 RCTs involving 9,746 participants were included. Compared with placebo or usual care, higher-dose corticosteroids were associated with lower short-term mortality (Risk Ratio [RR], 0.83; 95% Confidence Interval [CI], 0.74–0.92), as were lower-dose corticosteroids (RR, 0.84; 95% CI, 0.75–0.95). The indirect comparison showed no clear difference between higher- and lower-dose regimens (RR, 0.98; 95% CI, 0.83–1.16). In severe patients with CAP, the corresponding indirect estimate was RR 1.01 (95% CI, 0.77–1.33). Most secondary active-dose comparisons were imprecise and showed no consistent advantage of either strategy.

Conclusion

Compared with placebo or usual care, lower-dose corticosteroids (moderate confidence) and higher-dose corticosteroids (low confidence) were both associated with lower short-term mortality. The indirect higher-versus-lower dose comparison showed no clear mortality difference between dose categories; confidence in this comparison was low. Future research should prioritize large, head-to-head RCTs designed to evaluate whether selected inflammatory phenotypes benefit from higher-dose corticosteroids.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13054-026-06185-5.

Introduction

Community-acquired pneumonia (CAP) remains a leading cause of infectious disease–related mortality globally [1, 2]. Following landmark trials such as CAPE COD [3], adjunctive corticosteroids have emerged as a beneficial therapy for selected patients with severe CAP and have been incorporated into several updated clinical guidelines [4–8]. However, the focus of clinical inquiry has now shifted from whether corticosteroids are effective to a more granular challenge: identifying the optimal dosing intensity. This unresolved question leads to significant practice variation at the bedside [9].

Considerable heterogeneity persists in corticosteroid regimens across clinical practice, with daily doses varying across studies and guideline recommendations [6, 9, 10]. While prior meta-analyses have explored dose-response relationships through subgroup analyses [11–13] or continuous meta-regression [14, 15], these approaches provide crucial foundational insights; however, complementing them with a network framework can further mitigate cross-study confounding and the ecological fallacy. Traditional techniques compare isolated clusters of trials rather than providing relative effect estimates between specific dose categories [16–19]. Consequently, the existence of a “therapeutic window”—where anti-inflammatory benefits are maximized without incurring prohibitive risks of secondary infections [20] or severe hyperglycemia [21]—remains poorly defined.

Network meta-analysis offers a robust methodological framework to address this gap by integrating direct and indirect evidence through a common comparator. Unlike continuous meta-regression, this approach allows for the simultaneous comparison of distinct dosing tiers, providing clinically relevant indirect insights to complement continuous models and support individualized decision-making.

In this systematic review and network meta-analysis, we standardized diverse regimens into a unified dexamethasone-equivalent scale to evaluate the comparative efficacy and safety of higher-dose versus lower-dose corticosteroid strategies in hospitalized patients with CAP.

Methods

Registration and reporting

This systematic review and network meta-analysis was prospectively registered in International Prospective Register of Systematic Reviews (PROSPERO) (CRD420261297943) and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement and the PRISMA extension for network meta-analyses [22, 23]. Additional methodological details are provided in eMethods 1.

Information sources and search strategy

We searched PubMed, Embase, and the Cochrane Library from database inception through January 25, 2026, without language restrictions. The electronic database search was intentionally focused on community-acquired pneumonia, corticosteroids, and randomized controlled trials (RCTs). To improve completeness, we additionally performed citation searching of eligible trials and relevant prior systematic reviews to identify potentially missed studies. The full search strategy is provided in eMethods 2.

Eligibility criteria

We included RCTs enrolling hospitalized patients (aged ≥ 18 years) with CAP that compared systemic corticosteroids with placebo or usual care and reported extractable data for all-cause mortality or at least a prespecified secondary outcome, explicitly excluding trials conducted in ambulatory or outpatient settings. Trials enrolling broader critically ill populations, including sepsis, septic shock, or acute respiratory distress syndrome, were also eligible if CAP-specific subgroup data were reported separately or were otherwise extractable, such that the analyzed population remained restricted to patients with CAP [24]. This approach was intended to capture patients with CAP who progressed to severe systemic manifestations while preserving CAP as the primary infectious syndrome under study. Studies primarily evaluating viral pneumonias, including COVID-19 and influenza, were excluded to maintain etiologic homogeneity. Additional eligibility and exclusion criteria are provided in eMethods 3.

Selection process

Two independent reviewers screened titles and abstracts for eligibility in duplicate, and any citation deemed potentially relevant by either reviewer was advanced to full-text review. Full texts of these articles were subsequently retrieved and assessed independently and in duplicate against the inclusion criteria. Disagreements were resolved through discussion or adjudication by a third reviewer.

Outcomes

The primary outcome was short-term all-cause mortality, defined as the latest reported mortality assessment within 60 days after randomization. Long-term all-cause mortality was defined as the latest reported mortality assessment beyond 60 days. Secondary outcomes also included invasive mechanical ventilation, duration of mechanical ventilation, ventilator-free days, intensive care unit length of stay, hospital length of stay, time to clinical stability, gastrointestinal bleeding, hyperglycemia, neuropsychiatric adverse events, secondary infections and serious adverse events. Detailed outcome definitions are provided in eMethods 4.

Data collection process

Two reviewers independently extracted data using a standardized data extraction form. Collected information included: study characteristics (author, year, country), patient demographics (age, sex), disease severity scores [Pneumonia severity index (PSI) or Confusion, Urea, Respiratory rate, Blood pressure, age ≥ 65 years (CURB-65)], intervention details (type of corticosteroid, daily dose, duration), and outcomes of interest.

Dose standardization

Corticosteroid regimens were standardized to dexamethasone-equivalent daily doses using prespecified conversion factors [24–26]. Dexamethasone was selected as the reference metric because its nearly pure glucocorticoid profile allows for the isolated measurement of anti-inflammatory intensity without the confounding mineralocorticoid (fluid-retaining) effects of hydrocortisone, and it aligns with recent benchmark trials in critical respiratory care. Interventions were classified as higher dose (≥ 7.5 mg/d), lower dose (< 7.5 mg/d), or control (placebo or usual care). The 7.5 mg/d threshold was selected before the primary analysis because it approximates 200 mg/d of hydrocortisone—a commonly used stress-dose regimen recommended in sepsis and septic shock guidance [27] (accessible via standard online conversion tools such as https://www.mdcalc.com/calc/2040/steroid-conversion-calculator) and separated the major dose clusters observed among included trials, avoiding the network sparsity that would result from further stratifying nodes by specific corticosteroid types. It is important to note that this node classification is strictly based on the protocol-assigned initial or target treatment strategy, rather than the actual cumulative dose received by individual patients. Because this cutoff remains partly pragmatic rather than biologically definitive, we examined alternative thresholds in sensitivity analyses. For trials utilizing hydrocortisone combined with fludrocortisone, dose conversions were based exclusively on the systemic glucocorticoid component; fludrocortisone was excluded from dexamethasone-equivalent calculations due to its predominant mineralocorticoid activity, though its potential therapeutic contribution was considered during data interpretation. Detailed dose conversion procedures and trial-level classifications are provided in eMethods 5 and eTable 1.

Study risk of bias assessment

The risk of bias for each included RCT was assessed using the Cochrane Risk of Bias 2 (RoB 2) tool, evaluating domains such as the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result [28].

Effect measures

For binary outcomes, we calculated Risk Ratios (RRs) with 95% Confidence Intervals (CIs). For continuous outcomes, we calculated Mean Differences (MDs) with 95% CIs.

Synthesis methods

We performed frequentist random-effects network meta-analyses using restricted maximum likelihood estimation in R version 4.5.1 with the netmeta package [29]. Treatments were ranked using P-scores. For cluster-randomized trials, sample sizes and event counts were adjusted for clustering before inclusion in the meta-analysis. The design effect was calculated as [1+(m-1)× ICC], where m denotes the mean cluster size and ICC the intra cluster correlation coefficient. A continuity correction of 0.5 was applied to single-zero studies, whereas double-zero studies were excluded from relative effect estimation. Transitivity was evaluated by comparing publication year, trial size, age, disease spectrum and severity, intensive care unit (ICU) enrollment, baseline mechanical ventilation, shock, and blinding across nodes. Heterogeneity was assessed using Cochran’s Q, I² statistics, and Baujat plots. Because the star-shaped network lacked closed loops, formal inconsistency testing (e.g., node-splitting) was unfeasible; therefore, high-vs-low dose comparisons relied solely on indirect evidence. Hartung-Knapp-Sidik-Jonkman methods were used in sensitivity analyses of direct pairwise comparisons [30]. Additional details are provided in eMethods 6.

Subgroup analysis

To address the potential dilution effect of mixing severity levels, we pre-specified subgroup analyses stratified by disease severity (Severe vs. Non-Severe CAP) using independent subgroup networks instead of network meta-regression. Severity was defined by original trial criteria (e.g., ICU admission, PSI class IV-V). Given the reliance on trial-level aggregate data and resulting imprecision, these subgroup findings are considered hypothesis-generating rather than definitive evidence of effect modification.

Sensitivity analysis

Prespecified sensitivity analyses used fixed-effect models, retained only trials at low risk of bias, restricted analyses to trials with at least 100 participants, excluded trials published before 2000, performed leave-one-out analyses, and varied the dose threshold. Post hoc analyses were restricted to trials explicitly enrolling hospitalized patients with CAP, broader critical-care trials reporting CAP-specific subgroup data, and strictly double-blind placebo-controlled trials.

Exploratory analyses

Exploratory dose-response meta-regression and trial sequential analysis were performed after examination of the threshold analyses. These analyses were not prespecified in PROSPERO and were interpreted as supportive and hypothesis-generating. Additional details are provided in eMethods 7.

Reporting bias assessment

Small-study effects and reporting bias were assessed at the direct-comparison level. For eligible binary direct comparisons with at least 10 informative studies, Harbord’s test was used as the sole formal regression-based test for funnel-plot asymmetry. Continuous outcomes and binary comparisons with fewer than 10 informative studies were not formally tested and were assessed qualitatively within the CINeMA reporting-bias domain. Contour-enhanced funnel plots were generated for eligible comparisons, with I² and τ² reported in the legends.

Confidence in the evidence

We assessed confidence in each network estimate using the Confidence in Network Meta-Analysis (CINeMA) framework, considering within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and incoherence [31, 32]. Because the star-shaped network contained no closed loops, statistical incoherence could not be assessed. This domain was recorded as not assessable using a transparent manual judgment, while concerns regarding transitivity, imprecision, and reporting bias were evaluated within their corresponding CINeMA domains. Additional details are provided in eMethods 8.

Results

Study selection

The database search identified 12,833 records (PubMed, n = 2,299; Embase, n = 9,546; Cochrane, n = 988). After 1,169 duplicate records were removed, 11,664 records underwent title and abstract screening. Of these, 80 reports were sought for retrieval and assessed for eligibility in full text, of which 60 were excluded. Citation searching identified 51 additional records; all 51 reports were assessed for eligibility in full text, and 39 were excluded. Overall, 32 RCTs were included (Fig. 1) [3, 33–63]. A trial-level classification of mortality time points is provided in eTable 2.

Fig. 1.

Flow Diagram of Study Selection. Flow diagram showing identification, screening, eligibility assessment, and inclusion of RCTs in the systematic review and network meta-analysis. ALT TEXT: PRISMA flow diagram of study selection. Database and citation searches identified records that were screened and excluded in stages, resulting in 32 RCTs included in the network meta-analysis.

Study characteristics

Table 1 summarizes the characteristics of the 32 included RCTs. The trials enrolled hospitalized patients with CAP ranging from hospitalized non-severe disease to severe illness requiring intensive care. In addition, some trials enrolled broader critically ill populations but contributed CAP-specific subgroup data, including patients with CAP complicated by septic shock or acute respiratory distress syndrome. Corticosteroid regimens included hydrocortisone, methylprednisolone, prednisone or prednisolone, dexamethasone, and, in selected trials, hydrocortisone plus fludrocortisone. Standardized daily corticosteroid exposure ranged from 3.00 to 26.24 mg/d in dexamethasone-equivalent dose. Study-level extracted data for continuous and binary outcomes are provided in eTable 3 and eTable 4, and the distribution of dexamethasone-equivalent doses by year and sample size in eFigure 1.

Table 1.

Characteristics of included RCTs

Study	Country	Study design	Target population	Severity	Total N	Male (%)	Age, mean	Intervention drug	Daily dose*
Angus 2025	Multi-national	Multicenter, open-label RCT	Severe CAP	Severe [all ICU]	455	61	62	Hydrocortisone	7.50
Annane 2006	France	Multicenter, double-blind RCT	Septic shock patients with or without early ARDS	Severe [all ICU]	177	32	61	Hydrocortisone	7.50
Blum 2015	Switzerland	Multicenter, double-blind RCT	CAP	Non-severe [PSI I– III > 50%]	785	62	74	Prednisone	7.50
Confalonieri 2005	Italy	Multicenter, double-blind preliminary RCT	Severe CAP	Severe [majority ICU patients]	46	70	64	Hydrocortisone	10.07
Dequin 2023	France	Multicenter, double-blind RCT	Severe CAP	Severe [PSI scores IV–V > 50%]	795	70	67	Hydrocortisone	5.15
El-Ghamrawy 2006	Saudi Arabia	Single-center, RCT (Blinding NR)	Severe CAP	Severe [majority ICU patients]	34	62	62	Hydrocortisone	10.07
Fernandez-serran 2011	Spain	Single-center, prospective double-blind RCT	Hospitalized CAP	Severe [fine scores IV–V > 50%]	45	67	63	Methylprednisolone	12.91
Gordon 2016	United Kingdom	Multicenter, factorial (2 × 2), double-blind RCT	Septic shock	Severe [all ICU]	104	62	65	Hydrocortisone	4.60
Heming 2024	France	Multicenter, double-blind RCT	CAP-related septic shock	Severe [all ICU]	562	70	65	Hydrocortisone + Fludrocortisone	7.50
Keh 2016	Germany	Multicenter, double-blind RCT	Severe Sepsis	Severe [all ICU]	51	76	70	Hydrocortisone	4.60
Li 2016	China	Single-center, prospective RCT (Blinding NR)	Severe CAP complicated with septic shock	Severe [majority ICU patients]	58	53	61	Methylprednisolone	14.99
Lloyd 2019	Australia	Multicenter, stepped-wedge, cluster, double-blind RCT	Hospitalized CAP	Non-severe [50% of patients with CORB scores < 2; all GIM patients]	816	57	76	Prednisone	7.50
Lucinde 2025	Kenya	Multicenter, pragmatic, open-label, RCT	CAP	Non-severe [Non-ICU; Low-resource setting]	2180	54	53	Mixed Glucocorticoids	6.36
Marik 1993	South Africa	Single-center, double-blind RCT	CAP	Severe [mean Apache II score 13, all ICU patients]	30	NR	36	Hydrocortisone	26.24
McHardy and Schonell 1972	United Kingdom	Single-center, RCT (Blinding NR)	Pneumonia	Non-severe [defined by trials, most patients classified as mild-moderate]	126	45	60	Prednisolone	3.00
Meduri 1998	USA	Multicenter, double-blind RCT	Unresolving ARDS	Severe [all ICU]	24	38	48	Methylprednisolone	16.09
Meduri 2007	USA	Multicenter, double-blind RCT	Early severe ARDS	Severe [all ICU]	91	52	51	Methylprednisolone	8.84
Meduri 2022	USA	Multicenter, double-blind RCT	Severe CAP	Severe [PSI scores IV–V > 50%]	584	96	69	Methylprednisolone	4.76
Meijvis 2011	Netherlands	Multicenter, double-blind RCT	CAP	Non-severe [PSI I– III > 50%]	304	57	64	Dexamethasone	5.00
Mikami 2007	Japan	Single-center, open label, prospective RCT	Hospitalized CAP	Non-severe [PSI I– III > 50%]	31	74	72	Prednisolone	6.00
Nafae 2013	Egypt	Single-center, single-blind RCT	CAP	Severe [based on baseline vitals indi- cating mean CORB score > 2]	80	56	49	Hydrocortisone	10.07
Rezk 2013	Egypt	Single-center, open-label RCT	Early ARDS	Severe [all ICU]	27	85	45	Prednisolone	9.31
Sabry 2011	Egypt	Multicenter, double-blind RCT	CAP	Severe [majority ICU patients]	80	73	62	Methylprednisolone	12.32
Snijders 2010	Netherlands	Single-center, double-blind RCT	CAP	Mixed (Severe & Non-Severe) [Subgroup data extracted separately]	213	58	62	Prednisone	6.00
Steinberg 2006	USA	Multicenter, double-blind RCT	Persistent ARDS	Severe [all ICU]	180	49	49	Methylprednisolone	23.11
Tongyoo 2016	Thailand	Single-center, double-blind RCT	Early sepsis-associated ARDS	Severe [all ICU]	197	51	64	Hydrocortisone	7.50
Torres 2015	Spain	Multicenter, double-blind RCT	Severe CAP and high inflammatory response	Severe [PSI scores IV–V > 50%]	120	61	65	Methylprednisolone	13.12
Venkatesh 2018	Multi-national	Multicenter, double-blind RCT	Septic shock	Severe [all ICU]	840	61	62	Hydrocortisone	7.50
Villar 2020	Spain	Multicenter, open-label, RCT	ARDS	Severe [all ICU]	147	67	54	Dexamethasone	14.99
Wafy 2021	Egypt	Single-center, RCT (Blinding NR)	Severe CAP	Severe [100% ICU; IDSA/ATS 2007 criteria]	50	46	57	Prednisone	6.00
Wagner 1956	United States	Multicenter, Quasi-RCT (with a Randomized Subgroup) (Blinding NR)	Pneumococcal pneumonia	Non-severe [as defined by authors]	113	67	NR	Hydrocortisone	4.12
Wittermans 2021	Netherlands	Multicenter, double-blinded RCT	CAP	Mixed (Severe & Non-Severe) [Subgroup data extracted separately]	401	59	68	Dexamethasone	6.00

Baseline characteristics and trial design features of included RCTs. Daily corticosteroid exposure was standardized to dexamethasone-equivalent dose. For context, 7.5 mg of dexamethasone is approximately equivalent to 200 mg of hydrocortisone or 40 mg of methylprednisolone. * total dexamethasone-equivalent dose per day (mg). Severity classification reflects the trial’s target population and/or enrollment criteria as reported by trial authors. ARDS, acute respiratory distress syndrome; ATS, American Thoracic Society; CAP, community-acquired pneumonia; CORB, confusion, oxygenation, respiratory rate, and blood pressure; GIM, general internal medicine; ICU, intensive care unit; IDSA, Infectious Diseases Society of America; NR, not reported; PSI, pneumonia severity index; RCT, randomized controlled trial. ALT TEXT: Table 1 summarizing characteristics of 32 RCTs, including country, study design, target population, disease severity, sample size, sex, mean age, corticosteroid agent, and dexamethasone-equivalent daily dose. Trials range from non-severe to ICU-treated severe CAP

Risk of bias

Risk of bias was assessed using the Cochrane Risk of Bias 2 tool (Fig. 2). Overall, 14 trials were judged to have low risk of bias, 11 had some concerns, and 7 were judged to have high risk. Judgments were most consistently low for missing outcome data, whereas concerns were more frequent for the randomization process and deviations from intended interventions. Detailed domain-level rationales for individual trials are provided in eTable 5.

Fig. 2.

Risk of Bias Assessment of Included RCTs. Risk of bias was assessed using the Cochrane Risk of Bias 2 (RoB 2) tool. The upper panel displays judgments for individual studies across 5 domains. The lower panel summarizes the overall proportion of studies with low risk of bias (green), some concerns (yellow), or high risk of bias (red).ALT TEXT: Cochrane Risk of Bias 2 assessment across five domains for 32 RCTs. Most studies were judged low risk or some concerns, with fewer studies rated high risk overall.

Primary outcome: short term all-cause mortality

All 32 RCTs contributed data for short-term all-cause mortality. Heterogeneity was low (I² = 17.6%; τ² = 0; Q test P = .20). Compared with placebo, higher-dose corticosteroids (RR, 0.83; 95% CI, 0.74–0.92; low confidence) and lower-dose corticosteroids (RR, 0.84; 95% CI, 0.75–0.95; moderate confidence) were associated with lower short-term all-cause mortality. Applying these estimates to a pooled control-arm risk of 200 deaths per 1,000 patients yielded approximately 34 fewer deaths per 1,000 with higher-dose corticosteroids (95% CI, 16–52 fewer) and 32 fewer per 1,000 with lower-dose corticosteroids (95% CI, 10–50 fewer). The uncertainty intervals overlapped substantially, and the 2-per-1,000 numerical difference should not be interpreted as evidence favoring either dose category. The indirect higher-versus-lower dose comparison showed no clear mortality difference between dose categories (RR, 0.98; 95% CI, 0.83–1.16; low confidence) (Fig. 3 and eTable 6).

Fig. 3.

Network Geometry and Relative Effects for Short-term All-cause Mortality. Network plots depict available direct comparisons for the overall population, severe community-acquired pneumonia (CAP), and non-severe CAP. Node size is proportional to the number of participants randomized to each strategy; line thickness is proportional to the number of trials informing each direct comparison; numbers on lines indicate the number of contributing studies. The forest plot summarizes relative effects as risk ratios (RRs) with 95% CIs for high-dose vs. placebo, low-dose vs. placebo, and high-dose vs. low-dose comparisons across severity strata. The network features a star-shaped geometry, meaning all active-vs-active comparisons are strictly indirect. The sum of studies contributing to the severe and non-severe networks exceeds the total number of trials in the overall network because some trials contributed independent subgroup data to both strata.Abbreviations: CI, confidence interval; RR, risk ratio. ALT TEXT: Treatment network and forest plots for short-term all-cause mortality in overall, severe, and non-severe community-acquired pneumonia. Both dose groups reduced mortality versus placebo, with no clear difference between higher-dose and lower-dose corticosteroids.

Secondary outcomes

For higher-dose versus lower-dose corticosteroids, there was no clear difference in long-term mortality (RR, 1.18; 95% CI, 0.80–1.74; moderate confidence), invasive mechanical ventilation (RR, 0.85; 95% CI, 0.47–1.52; low confidence), ventilator-free days (MD, 2.35 days; 95% CI, − 2.76 to 7.47; low confidence), duration of mechanical ventilation (MD, − 1.57 days; 95% CI, − 5.18 to 2.04; low confidence), ICU length of stay (MD, − 1.30 days; 95% CI, − 2.98 to 0.38; moderate confidence), or hospital length of stay (MD, − 2.00 days; 95% CI, − 4.72 to 0.72; very low confidence). Heterogeneity was substantial for several continuous outcomes. Time to clinical stability was informed by only two trials and was not assigned a CINeMA rating.

Safety outcomes also showed no clear difference between higher- and lower-dose regimens: gastrointestinal bleeding (RR, 1.30; 95% CI, 0.45–3.77; moderate confidence), hyperglycemia (RR, 0.56; 95% CI, 0.20–1.60; moderate confidence), neuropsychiatric adverse events (RR, 0.95; 95% CI, 0.44–2.02; low confidence), secondary infections (RR, 1.11; 95% CI, 0.79–1.55; moderate confidence), and serious adverse events (RR, 0.83; 95% CI, 0.44–1.55; low confidence). Comparisons versus placebo are shown in Fig. 4, and P-score rankings across outcomes are presented in eFigure 2.

Fig. 4.

Summary of Relative Effects and Certainty of Evidence Across Outcomes. Heatmap summarizing comparative effects of corticosteroid dose strategies across outcomes and severity strata. Binary outcomes are reported as risk ratios (RRs) and continuous outcomes as mean differences (MDs), each with 95% CIs. Shading indicates certainty of evidence assessed using the CINeMA framework. Green indicates high certainty; light green, moderate certainty; yellow, low certainty; and red, very low certainty. Abbreviations: CI, confidence interval; CINeMA, Confidence in Network Meta-Analysis; ICU, intensive care unit; MD, mean difference; RR, risk ratio.ALT TEXT: Heatmap of relative effects and certainty across efficacy and safety outcomes by disease severity. Both corticosteroid dose strategies improved mortality versus placebo, while most outcomes showed no clear advantage of higher-dose over lower-dose treatment

Transitivity

Higher-dose trials were somewhat older and smaller than lower-dose trials (median publication year, 2013 vs. 2016; median sample size, 106 vs. 170 participants). Age and blinding were broadly comparable, but higher-dose trials more often required ICU admission (75.0% vs. 41.7%) and reported greater baseline use of mechanical ventilation (median, 96.0% vs. 33.0%) and shock (median, 48.4% vs. 10.5%). CAP-specific baseline characteristics were unavailable for some broader critical-care trials. These differences limit the transitivity assumption (eTable 7, eTable 8, and eTable 9).

Subgroup analysis

The severe-CAP network included 25 trials and the non-severe network 9 trials; two trials contributed separate data to both strata [55, 63]. In severe CAP, higher-dose corticosteroids were associated with lower mortality versus control (RR, 0.79; 95% CI, 0.70–0.91; low confidence), whereas the estimate for lower-dose corticosteroids was less precise (RR, 0.79; 95% CI, 0.62–1.00; moderate confidence). The indirect higher-dose versus lower-dose estimate showed no clear difference (RR, 1.01; 95% CI, 0.77–1.33; low confidence). In non-severe CAP, estimates were RR 1.10 (95% CI, 0.77–1.58; low confidence) for higher dose versus control, RR 0.87 (95% CI, 0.75–1.00; moderate confidence) for lower dose versus control, and RR 1.27 (95% CI, 0.86–1.88; moderate confidence) for higher versus lower dose. These subgroup networks were not formal tests of effect modification.

Sensitivity analysis

Across prespecified sensitivity analyses, findings were generally consistent with the primary analysis. In the overall population, the pattern of lower short-term all-cause mortality for both high-dose and low-dose corticosteroids vs. placebo, with no significant difference between high-dose and low-dose regimens, remained unchanged under fixed-effect modeling, restriction to trials with lower risk of bias, restriction to trials with sample sizes of at least 100, and exclusion of pre-2000 trials. Full results are provided in eTable 10, eTable 11 and eTable 12. Leave-one-out analyses did not materially change the indirect higher-dose versus lower-dose estimates (eFigure 3).

In trials that explicitly enrolled patients with CAP, low-dose corticosteroids remained associated with lower mortality vs. placebo (RR, 0.81 [95% CI, 0.68–0.96]), whereas the estimate for high-dose corticosteroids was imprecise (RR, 0.88 [95% CI, 0.69–1.14]). In complementary analyses restricted to trials enrolling broader critically ill populations with CAP-specific subgroup data, high-dose corticosteroids were associated with lower mortality vs. placebo (RR, 0.81 [95% CI, 0.72–0.91]), whereas the estimate for low-dose regimens was imprecise (RR, 1.08 [95% CI, 0.63–1.86]). In both settings, indirect comparisons between high-dose and low-dose regimens showed no clear mortality difference but remained imprecise. Furthermore, in a sensitivity analysis excluding all open-label trials (retaining 21 strictly double-blind, placebo-controlled trials), the indirect comparison between high-dose and low-dose corticosteroids remained generally consistent with the primary analysis, showing no clear mortality difference in the overall population (RR 1.08; 95% CI, 0.82–1.42) or the severe CAP subgroup (RR 1.05; 95% CI, 0.78–1.40).

In threshold sensitivity analyses, findings varied between the 8.25 mg/d and 9.0 mg/d thresholds. For the 8.25 mg/d threshold (13 high-dose vs. 19 low-dose trials), the indirect comparison yielded a lower mortality estimate for the higher-dose node in the overall population (RR, 0.74; 95% CI, 0.57–0.97; P = .032), but not in the severe CAP subgroup (RR, 0.74; 95% CI, 0.55–1.01; P = .056). For the 9.0 mg/d threshold (12 high-dose vs. 20 low-dose trials), no clear mortality difference was found in either the overall population (RR, 0.77; 95% CI, 0.57–1.04; P = .087) or severe CAP (RR, 0.77; 95% CI, 0.56–1.07; P = .126). High- versus low-dose comparisons could not be estimated in patients with non-severe CAP at either threshold because no trials were classified into the high-dose group (eFigure 3). In Hartung-Knapp-Sidik-Jonkman sensitivity analyses of direct pairwise comparisons, selected outcomes were sensitive to the statistical approach, but these differences did not identify a stable dose threshold and did not alter the interpretation of the indirect comparison between high-dose and low-dose regimens (eTable 13).

Dose-response analysis

Dose–response meta-regression specifications and estimates are presented in eTable 14, with corresponding plots in eFigure 4. In the primary linear meta-regression, there was no linear dose–response association with short-term mortality in the overall population (β = -0.004, SE = 0.012, P = .757), severe CAP (β= -0.002, SE = 0.013, P = .904), or non-severe CAP (β = 0.118, SE = 0.114, P = .301). Exploratory restricted cubic spline models did not identify evidence of a statistically significant nonlinear association in the overall population (P = .452) or severe subgroup (P = .173), whereas models were not estimable in non-severe CAP due to dose clustering. Accordingly, the linear model was retained as the primary dose–response analysis.

Heterogeneity

Between-study heterogeneity for short-term all-cause mortality was low across the overall population and severity subgroups (overall: I² = 17.6%; severe CAP: I² = 28.4%; non-severe CAP: I² = 0%). Baujat analyses indicated that heterogeneity in the overall and severe populations was driven by a small subset of trials, whereas individual trial contributions were minimal in the non-severe subgroup (eTable 15 and eFigure 5). In contrast, substantial heterogeneity was observed for several secondary outcomes, including ventilator-free days (I² = 88.3%), hospital length of stay (I² = 87.6%), duration of mechanical ventilation (I² = 75.7%), serious adverse events (I² = 61.2%), and long-term all-cause mortality (I² = 52.0%).

Trial sequential analysis

Trial sequential analysis showed that, at the primary dose threshold and lower-threshold sensitivity analysis, the required information size was reached for both dose groups in the overall population, although this was not consistent across severity subgroups. At higher dose thresholds, the required information size was generally reached for the low-dose group but not for the high-dose group. Detailed trial sequential analysis parameters and monitoring boundary plots are provided in eTable 16 and eFigure 6.

Small-study effects and reporting bias

For short-term all-cause mortality, Harbord’s test suggested funnel-plot asymmetry for higher-dose corticosteroids versus placebo or usual care in the overall population (P = .013; 20 studies) and in the severe-CAP subgroup (P = .013; 18 studies), but not for lower-dose corticosteroids versus placebo or usual care in the overall population (P = .815; 12 studies). No statistical evidence of asymmetry was detected for eligible secondary-outcome comparisons, and other direct comparisons were not formally tested because of sparse data or because Harbord’s test was not applicable. Full eligibility and Harbord-test results are provided in eTable 17, and contour-enhanced funnel plots are shown in eFigure 7.

Discussion

In this network meta-analysis of 32 RCTs, both higher- and lower-dose corticosteroids were associated with lower short-term mortality than placebo or usual care. The indirect higher-dose versus lower-dose estimate showed no clear mortality difference and was rated as low confidence in the CINeMA assessment. Secondary active-dose comparisons also showed no consistent advantage of either strategy. Because no trial directly compared the two dose categories, these findings do not establish equivalence or an optimal dose threshold.

Comparison with prior evidence

Our findings provide an updated, clinically oriented synthesis regarding the long-standing uncertainty over corticosteroid intensity in CAP [10]. Previous meta-analyses established the overall effects of corticosteroids in CAP and described continuous dose-response trends [14, 15]. Our analysis complements this work by standardizing heterogeneous regimens to dexamethasone-equivalent daily doses, comparing clinically recognizable dose strategies, incorporating recent large trials, and reporting CINeMA confidence with each estimate. The contribution is therefore an updated, clinically oriented synthesis rather than a replacement for continuous dose-response analyses. The absence of direct higher-dose versus lower-dose trials remains the principal evidence gap.

Importantly, the included trials differed substantially in design, enrollment criteria, timing of corticosteroid initiation, respiratory and cardiovascular compromise, glucose management, exclusion of viral pneumonia, and the specific type and dose of corticosteroids used. Therefore, the pooled estimates should not be interpreted as implying complete clinical equivalence across trial populations.

Potential effect modification by clinical phenotype

Exploratory subgroup analyses suggest that dose-response patterns might be phenotype-dependent. Restricting to “pure CAP” RCTs—excluding patients with acute respiratory distress syndrome (ARDS) or septic shock—attenuated the mortality association for higher-dose regimens. In contrast, cohorts with systemic hyperinflammation (e.g., ARDS or sepsis subgroups in REMAP-CAP [33] and DEXA-ARDS [60]) were associated with lower mortality in higher-dose arms. These exploratory findings raise the hypothesis that treatment response may differ by systemic inflammatory phenotype, ARDS, or septic shock status, but they should not be used as a current dose-selection rule. This hypothesis requires confirmation in prospective head-to-head or patient-level analyses.

Threshold fragility and biological continuity

Our study identified statistical fragility in the primary outcome at the 7.5 mg/d threshold. The directional reversal favoring higher doses observed when slightly increasing this cutoff is primarily driven by the dense distribution of data from large trials at this specific margin, rather than a true biological shift. Furthermore, restricted cubic spline models did not identify evidence of a statistically significant nonlinear inflection point across the evaluated dose range. However, dose clustering and uncertainty in the spline estimates preclude excluding a clinically relevant nonlinear relationship. This indicates that corticosteroid intensity should be approached as a continuous clinical spectrum rather than a strict binary cutoff in practice, challenging rigid dose-specific recommendations and advising against arbitrary dose escalation.

Safety considerations

Although no clear dose-related differences were observed in hyperglycemia, secondary infections, gastrointestinal bleeding, neuropsychiatric events, or serious adverse events, most estimates were imprecise and derived from indirect comparisons. In a broader meta-analysis of randomized trials involving patients with CAP, sepsis, or ARDS, systemic corticosteroids increased the risk of hyperglycemia and probably hypernatremia, while having little or no effect on gastrointestinal bleeding or secondary infections [64]. A prespecified CAP trial subgroup further suggested that prednisone-associated glycemic dysregulation did not translate into worse short-term clinical outcomes among patients with diabetes or hyperglycemia on admission [65]. Nevertheless, these studies primarily compared corticosteroids with control rather than higher- with lower-dose regimens and therefore cannot establish dose-related safety equivalence.

Limitations

Strengths of this study include an updated synthesis of 32 RCTs, prespecified dose standardization and threshold analyses, formal assessment of transitivity, multiple robustness analyses, and outcome-specific CINeMA ratings. Several limitations are important. First, the star-shaped network provided no direct higher-dose versus lower-dose evidence and precluded statistical assessment of incoherence. Baseline severity imbalances and incomplete CAP-specific covariate data further limit transitivity. Second, dose categories reflect protocol-assigned daily regimens rather than actual delivered cumulative exposure; they therefore may not capture treatment discontinuation, crossover, or rescue corticosteroids. Dexamethasone-equivalent conversion also cannot capture differences in mineralocorticoid activity, half-life, or tissue effects. Third, mortality time points and secondary-outcome definitions varied across trials, and several continuous outcomes were highly heterogeneous. Finally, aggregate data limited assessment of patient-level effect modification, and some sensitivity, dose-response, and trial sequential analyses were post hoc or exploratory. Small-study effects remain an additional source of uncertainty.

Clinical Implications

Indirect comparisons showed no clear efficacy advantage of higher-dose over lower-dose regimens. However, because these estimates were based on low-certainty indirect evidence and potentially limited transitivity, they do not establish an optimal dose threshold or support a definitive preference for either dosing strategy.

The indirect higher-versus-lower dose comparison showed no clear mortality difference between dose categories (RR, 0.98; 95% CI, 0.83–1.16), with low confidence. This finding should not be read as proof that the two strategies are equivalent; rather, the dose-to-dose comparison remains limited by indirectness, possible transitivity concerns, small-study effects, and protocol-assigned rather than cumulative delivered dose categories. Dose selection should remain individualized according to illness severity, inflammatory phenotype, comorbidities, and risk of metabolic or infectious complications. This interpretation is consistent with the updated American Thoracic Society guideline, which recommends against routine systemic corticosteroids in patients with non-severe CAP but conditionally suggests their use in severe CAP, emphasizing appropriate patient selection rather than uniform treatment across all hospitalized patients [8]. Adequately powered head-to-head randomized trials are needed before recommending a single default dosing strategy.

Future directions

Resolving dosing uncertainty will require prospective, head-to-head trials. Adaptive platform trials such as PANDA-SCAP [66] can help address time-varying confounding inherent to corticosteroid therapy. Such trials are expected to refine phenotype-directed strategies and optimize the therapeutic window in real-world settings. Individual patient data meta-analyses may be particularly valuable because treatment effects are likely modified by baseline inflammatory phenotype. Given the low certainty of current indirect evidence, the critical care community urgently requires large, head-to-head RCTs rather than additional observational studies to establish the optimal corticosteroid dose. Prior patient-level work and randomized trials utilizing CRP-based inclusion thresholds [67] highlight the critical need to evaluate whether inflammatory biomarkers can reliably identify patients in whom the benefits of corticosteroids outweigh metabolic and infectious risks. A recent individual patient data analysis suggested that the mortality benefit of corticosteroids may be concentrated among patients with baseline CRP concentrations greater than 204 mg/L, whereas no statistically significant benefit was observed below this model-derived threshold. This cutoff requires prospective validation before being used for treatment or dose selection [67]. Therefore, future head-to-head RCTs should specifically target patients with CAP and hyperinflammation (e.g., CRP > 200 mg/L) and non-viral etiologies. To adequately power such direct dose-to-dose comparisons—assuming a 4% absolute risk reduction from an 18% baseline mortality, while accounting for heterogeneity—we estimate these trials must recruit 2,000 to 3,000 participants. This scale is essential to definitively determine whether higher-dose regimens offer true survival benefits or merely amplify toxicities like severe hyperglycemia. Ultimately, integrating such biomarker-driven phenotyping into clinical pathways will be essential for achieving personalized, precision corticosteroid therapy in severe CAP.

Conclusion

In hospitalized patients with CAP, lower-dose corticosteroids (moderate confidence) and higher-dose corticosteroids (low confidence) were both associated with lower short-term mortality compared with placebo or usual care. The indirect higher-versus-lower dose comparison showed no clear mortality difference between dose categories (RR, 0.98; 95% CI, 0.83–1.16); confidence in this comparison was low. Given the limitations of indirect comparisons and protocol-assigned dose classification, dose escalation should not be applied as a default without further evidence. Future research should prioritize large, head-to-head RCTs designed to evaluate whether selected inflammatory phenotypes benefit from higher-dose corticosteroids.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ARDS

Acute respiratory distress syndrome

ATS

American Thoracic Society

CAP

Community-acquired pneumonia

CI

Confidence interval

CINeMA

Confidence in Network Meta-Analysis

CORB

Confusion, oxygenation, respiratory rate, and blood pressure

COVID-19

Coronavirus disease 2019

CRP

C-reactive protein

CURB-65

Confusion, Urea, Respiratory rate, Blood pressure, age ≥ 65 years

GIM

General internal medicine

ICU

Intensive care unit

IDSA

Infectious Diseases Society of America

MD

Mean difference

NMA

Network meta-analysis

NR

Not reported

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

PROSPERO

International Prospective Register of Systematic Reviews

PSI

Pneumonia severity index

RCT

Randomized controlled trial

RoB 2

Risk of Bias 2 tool

RR

Risk ratio

TSA

Trial sequential analysis

Author contributions

Conceptualization: Yu Zhang; Data curation: Yikang Ouyang, Jiancheng Lai, Peng Wang, Xiang Yuan, Hongbo Wang, Ziyan Xiang, Yangchun Xiao, Jialing He; Formal analysis: Yikang Ouyang, Jiancheng Lai, Peng Wang; Methodology: Yikang Ouyang, Jiancheng Lai, Peng Wang, Yu Zhang, Lu Jia; Supervision: Fang Fang, Yu Zhang, Jun Wan, Lu Jia; Writing – original draft: Yikang Ouyang, Jiancheng Lai; Writing – review & editing: Peng Wang, Xiang Yuan, Fang Fang, Yu Zhang, Lu Jia. All authors read and approved the final manuscript.

Funding

This work was supported by the Provincial Training Program for Innovation and Entrepreneurship of College Students of Chengdu University (Project Title: An Affordable Solution for Precise Puncture of Primary Intracranial Hemorrhage Based on 3D Slicer and Augmented Reality (AR) Technology, Project No. S202611079027X).

Data availability

This is a meta-analysis whose data is drawn exclusively from publicly available scientific articles.

Declarations

Ethical approval and consent to participate

Not applicable. This study is a systematic review and network meta-analysis of published aggregate data and did not involve new recruitment of human participants, human tissue, or individual patient data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.