Skip to main content

Foundational Trials

SONIA

Image: Shutterstock / ChaNaWiT

Publication

  • Title: A Pragmatic Trial of Glucocorticoids for Community-Acquired Pneumonia
  • Acronym: SONIA (Steroids in Pneumonia)
  • Year: 2025
  • Journal published in: New England Journal of Medicine
  • Citation: Lucinde RK, Gathuri H, Mwaniki P, et al. A pragmatic trial of glucocorticoids for community-acquired pneumonia. N Engl J Med. 2025;393(22):2187-2197.

Context & Rationale

  • Background
    • CAP is a major cause of morbidity and mortality, with disproportionately high case-fatality in sub-Saharan Africa despite a younger patient population.
    • Glucocorticoids are biologically plausible adjuncts in CAP, targeting potentially harmful host inflammatory responses that contribute to respiratory failure and shock.
    • Prior RCTs and meta-analyses (largely high-income, often ICU-enriched) suggested possible reductions in mortality and clinical deterioration, but effect estimates varied by severity, steroid class/dose, and concomitant care.
    • In many African settings, ward-based care predominates and ICU access is constrained, creating uncertainty about the applicability of ICU-centric steroid evidence to routine inpatient care.1
  • Research Question/Hypothesis
    • Does early adjunctive low-dose systemic glucocorticoid therapy (prednisolone-equivalent 50 mg once daily for up to 10 days) reduce 30-day all-cause mortality versus standard care alone in adults admitted with clinical CAP to Kenyan public hospitals?
    • Hypothesis: adjunctive glucocorticoids would reduce time-to-death risk over 30 days compared with standard care.
  • Why This Matters
    • If effective, a low-cost, widely available, orally deliverable intervention could have large population-level impact in LMIC inpatient settings with high CAP mortality.
    • If ineffective or harmful, empiric steroid use (especially in settings with overlapping aetiologies such as TB, PCP, heart failure, and undifferentiated sepsis) could be actively discouraged.
    • The trial directly tests effectiveness in real-world Kenyan ward practice, rather than extrapolating from ICU trials and high-resource pathways.

Design & Methods

  • Research Question: In adults admitted with clinical community-acquired pneumonia to Kenyan public hospitals, does adjunctive low-dose systemic glucocorticoid therapy reduce 30-day all-cause mortality compared with standard care alone?
  • Study Type: Pragmatic, open-label, individually randomised, multicentre, parallel-group, investigator-initiated trial across 18 public hospitals in Kenya (general wards; HDU/ICU eligible but uncommon).
  • Population:
    • Setting/timing: Inpatients admitted within the previous 48 hours.
    • Inclusion: Adults ≥18 years; clinician diagnosis of CAP using a clinical definition (≥2 symptoms/signs for <14 days: cough, fever, dyspnoea, haemoptysis, chest pain, or crackles on examination); written consent or deferred consent where too unwell.
    • Key exclusions: Hospital-acquired pneumonia (>48 hours in hospital without symptoms at admission); clinician-determined requirement for glucocorticoids; known/suspected condition requiring glucocorticoids (e.g., COPD, asthma, adrenal insufficiency, PCP); pregnancy/breast feeding; contraindication to glucocorticoids; confirmed COVID-19 (if result known at enrolment) or strong clinician suspicion of COVID-19 prompting steroid use.
  • Intervention:
    • Core strategy: Standard care plus adjunctive low-dose systemic glucocorticoids for up to 10 days (prednisolone-equivalent 50 mg once daily).
    • Formulations/dosing (protocol equivalence set): Betamethasone 5 mg, dexamethasone 6 mg, hydrocortisone 160 mg, methylprednisolone 30 mg, prednisolone 50 mg, prednisone 50 mg once daily (choice driven by randomisation card and medication availability).
    • Route/timing: Oral (including NG) preferred; IV permitted if oral unavailable or clinically necessary with switch to oral when possible; intended early after admission.
    • Implementation feature: Participants randomised to the intervention arm were additionally randomised to a specific glucocorticoid formulation (two formulations later removed due to pill burden; see Controversies for implications).
  • Comparison:
    • Standard care: Clinician-directed treatment according to local guidelines and practice (protocol-referenced empiric beta-lactam + macrolide pathway), plus supportive care (oxygen, fluids, vasopressors, etc. as available).
    • Rescue/co-interventions: Open-label glucocorticoids permitted if subsequently clinically indicated for non-trial reasons (recorded as withdrawals/crossover).
  • Blinding: Unblinded (open-label); no placebo.
  • Statistics: Planned 2180 participants to detect a 5% absolute reduction in 30-day mortality (from 20% to 15%) with 85% power at a two-sided 5% significance level, allowing ~5% loss; primary analysis intention-to-treat using Cox proportional hazards regression with trial site as a stratification variable.2
  • Follow-Up Period: 30 days post-randomisation (phone follow-up at day 14 and day 30 where applicable).

Key Results

This trial was not stopped early. Recruitment completed to the prespecified target sample size (n=2180) with prespecified interim monitoring.

Outcome Adjunctive glucocorticoids + standard care (n=1089) Standard care (n=1091) Effect p value / 95% CI Notes
All-cause death by day 30 (primary) 246/1089 (22.6%) 284/1091 (26.0%) HR 0.84 95% CI 0.73 to 0.97; P=0.02 Time-to-event; trial site stratified in Cox model.
All-cause death by day 7 156/1089 (14.3%) 179/1091 (16.4%) HR 0.86 95% CI 0.73 to 1.00; P=Not reported Secondary time-to-event analysis.
All-cause death by day 14 203/1089 (18.6%) 224/1091 (20.5%) HR 0.87 95% CI 0.74 to 1.01; P=Not reported Secondary time-to-event analysis.
All-cause death by day 21 231/1089 (21.2%) 262/1091 (24.0%) HR 0.84 95% CI 0.73 to 0.98; P=Not reported Secondary time-to-event analysis.
In-hospital death (≤30 days) 196/1089 (18.0%) 223/1091 (20.4%) Not reported Not reported Count/proportion only reported.
Out-of-hospital death (≤30 days) 50/1089 (4.6%) 61/1091 (5.6%) Not reported Not reported Count/proportion only reported.
ICU transfer 2/1089 (0.2%) 3/1091 (0.3%) Not reported Not reported Reflects very low ICU access/transfer in this setting.
Duration of hospital stay Median 5 days (IQR 2–10) Median 5 days (IQR 2–9) Not reported Not reported Secondary descriptive outcome (no effect estimate reported).
≥1 adverse event (patients) 184/1089 (16.9%) 154/1091 (14.1%) Not reported Not reported Adverse event attribution performed in the intervention arm; 62/211 events (29.4%) judged related to glucocorticoids.
Serious adverse event (patients) 44/1089 (4.0%) 48/1091 (4.4%) Not reported Not reported Serious adverse events judged related to glucocorticoids occurred in 5 patients (0.5%).
Hyperglycaemia (events) 35 events 3 events Not reported Not reported Reported as adverse-event counts (not unique patients).
  • Adjunctive glucocorticoids were associated with lower 30-day mortality than standard care (HR 0.84; 95% CI 0.73 to 0.97; P=0.02).
  • Sensitivity analyses were directionally consistent: complete-case analysis HR 0.84 (95% CI 0.73 to 0.96) and modified ITT HR 0.83 (95% CI 0.72 to 0.97).
  • Selected prespecified subgroup estimates (interaction P values not statistically significant): women HR 0.70 (95% CI 0.55 to 0.89) vs men HR 0.94 (95% CI 0.79 to 1.12; P interaction 0.10); oxygen saturation <90% HR 0.92 (95% CI 0.75 to 1.14) vs ≥90% HR 0.80 (95% CI 0.65 to 0.97; P interaction 0.37).

Internal Validity

  • Randomisation and allocation: Site-specific, computer-generated block randomisation with allocation revealed only after enrolment details were entered (tablet-based system with sealed opaque envelopes as back-up), supporting allocation concealment in principle.
  • Dropout/withdrawals and missingness: 194/2180 (8.9%) were withdrawn by day 30 (88/1089 [8.1%] intervention; 106/1091 [9.7%] control); day-30 visit missing for unknown reasons in 98/2180 (4.5%); missingness not MCAR (Little’s test P<0.001) but appeared non-differential by arm, and sensitivity analyses were concordant.
  • Performance/detection bias: Open-label design creates potential for co-intervention and discharge/withdrawal-related bias; primary endpoint (mortality) is objective, partially mitigating detection bias.
  • Protocol adherence: 1063/1089 (97.6%) received trial glucocorticoids; median time from randomisation to first dose 1 day (IQR 1–1); median in-hospital glucocorticoid duration 4 days (IQR 2–8); post-discharge continuation to day 10 was intended but not directly monitored.
  • Baseline comparability: Groups were well balanced (e.g., median age 53 years in both arms; women 46.1% vs 46.5%; oxygen saturation <90%: 37.3% vs 36.9%; HIV positive: 16.3% vs 15.2%).
  • Heterogeneity: Pragmatic delivery across 18 hospitals, variable diagnostic testing (radiography available in ~39%), and multiple glucocorticoid formulations introduce clinical heterogeneity; primary model stratified by site.
  • Dose and timing: Fixed low-dose prednisolone-equivalent strategy with early initiation (median 1 day) supports biological plausibility for effect during the early inflammatory phase; duration shortened in-hospital by discharge and may dilute intended exposure.
  • Separation of the variable of interest: Trial glucocorticoids were delivered to 97.6% of the intervention arm, while post-randomisation non-trial glucocorticoid initiation occurred in 33/1091 (3.0%) of the standard-care arm (withdrawn for clinical steroid indication), indicating modest contamination but substantial overall separation.
  • Adjunctive therapy balance: Antibiotic use was similar (1054/1089 [96.8%] intervention vs 1053/1091 [96.5%] control), with broadly comparable regimen patterns (e.g., beta-lactam + macrolide in ~74–76%); ICU transfer was very uncommon (0.2% vs 0.3%).
  • Outcome assessment/statistical rigour: Primary analysis prespecified (ITT, Cox model); proportional hazards assumption supported (Schoenfeld P=0.8391); concordant sensitivity analyses reduce (but do not eliminate) concern about informative censoring from withdrawals.

Conclusion on Internal Validity: Moderate. Allocation processes and an objective primary endpoint support credibility, but open-label delivery, heterogeneous formulations/exposure duration, and post-randomisation withdrawals with non-MCAR missingness modestly weaken causal certainty.

External Validity

  • Population representativeness: Adults admitted with clinically defined CAP in Kenyan public hospitals, including patients with prevalent comorbidities (e.g., HIV) and variable access to diagnostics, resembles “real-world” ward CAP in many LMIC settings.
  • Setting realism: Very low ICU transfer rates and variable ward resources suggest applicability to settings where escalation pathways are limited and early mortality risk is concentrated on general wards.
  • Applicability across health systems: Generalisability is strongest to sub-Saharan African/LMIC inpatient wards with similar aetiologic mix and constrained critical care; extrapolation to high-income ICU-managed severe CAP (with different pathogen profile, diagnostics, and escalation capacity) is uncertain.
  • Intervention feasibility: Oral once-daily prednisolone-equivalent dosing is scalable where steroids are available; implementation requires capacity for glucose monitoring and for distinguishing CAP from steroid-sensitive/contraindicated diagnoses where possible.

Conclusion on External Validity: High for LMIC ward-based CAP care similar to the Kenyan trial sites; limited for high-resource ICU contexts and for phenotypes where robust diagnostics (e.g., to exclude TB/PCP) substantially alter the treated population.

Strengths & Limitations

  • Strengths:
    • Large, pragmatic, multicentre trial in a high-mortality, understudied setting, directly addressing global generalisability gaps.
    • Objective, patient-centred primary outcome (30-day all-cause mortality) with time-to-event analysis.
    • High delivery of allocated intervention (97.6% received trial steroids), and high antibiotic use in both arms.
    • Prespecified sensitivity analyses and model checks (including proportional hazards testing).
  • Limitations:
    • Open-label design with potential performance bias and withdrawal-related bias (including non-trial steroid initiation).
    • Clinical (not microbiologic) CAP definition and limited radiography availability increase diagnostic heterogeneity and potential inclusion of alternative aetiologies.
    • Heterogeneous glucocorticoid formulations and mid-trial formulation changes complicate attribution to a single agent and dose-exposure construct.
    • Planned 10-day exposure frequently truncated in-hospital (median 4 days); post-discharge adherence was not directly verified.

Interpretation & Why It Matters

  • Clinical signal
    In Kenyan inpatients with clinically defined CAP, adjunctive low-dose glucocorticoids were associated with lower 30-day mortality than standard care alone (HR 0.84; 95% CI 0.73 to 0.97).
  • Implementation reality
    The intervention is pragmatic and low-cost, but requires operational capacity for glucose monitoring (hyperglycaemia events were more frequent in the glucocorticoid arm) and careful clinical reassessment for alternative diagnoses.
  • How this shifts prior assumptions
    SONIA strengthens the case that steroid benefit may extend beyond ICU-enriched severe CAP cohorts, but its effect estimate is embedded in a ward-based LMIC care pathway with very limited escalation, and may not be portable to all settings.

Controversies & Subsequent Evidence

  • Pragmatic diagnosis versus aetiologic specificity: The accompanying editorial highlighted that a clinical CAP definition with limited radiography/microbiology may include heterogeneous syndromes (including conditions where steroids could plausibly help, harm, or be neutral), complicating mechanistic interpretation and cross-setting generalisability.3
  • Multi-formulation exposure and protocol pragmatism: The protocol prespecified multiple equivalent-dose glucocorticoids driven by availability; hydrocortisone and prednisone were removed from the within-intervention randomisation because of pill burden, leaving small samples for those formulations and reducing precision for formulation-level inference.2
  • Relationship to ICU severe-CAP trials: CAPE COD (ICU severe CAP) found lower 28-day mortality with hydrocortisone versus placebo, supporting steroid benefit in a tightly defined high-resource severe-CAP phenotype; SONIA extends the question to ward-based, resource-limited care with substantially higher baseline mortality.4
  • Conflicting platform-trial signal: In REMAP-CAP’s severe CAP corticosteroid domain, a 7-day hydrocortisone course was unlikely to reduce mortality (meeting a futility criterion), underscoring persistent uncertainty about which CAP phenotypes derive net benefit and how context modifies treatment effects.5
  • Meta-analytic synthesis: A recent non-viral CAP meta-analysis (including modern ICU trials) supports an overall mortality reduction signal with adjunctive corticosteroids, but emphasises heterogeneity by severity and trial context; SONIA would be expected to materially influence future pooled estimates because of its size and LMIC setting.6
  • Guidelines remain in flux (and largely pre-SONIA): A 2024 SCCM focused update issued a strong recommendation to use corticosteroids in hospitalised patients with severe CAP, while other national guidance has applied more restrictive criteria (e.g., severe CAP requiring critical care, with exclusions). Formal guideline updates explicitly incorporating SONIA were not identified in targeted searches at the time of writing.78
  • Harms and monitoring: Hyperglycaemia was more frequently reported in the glucocorticoid arm (35 vs 3 events), but serious adverse events were similar overall and serious events judged related to glucocorticoids were uncommon (0.5%); implementation should include pragmatic glucose monitoring capacity and risk stratification.

Summary

  • Large pragmatic Kenyan RCT (SONIA) tested adjunctive low-dose systemic glucocorticoids (prednisolone-equivalent 50 mg daily for up to 10 days) versus standard care in adults admitted with clinical CAP.
  • 30-day all-cause mortality was lower with glucocorticoids: 22.6% (246/1089) vs 26.0% (284/1091); HR 0.84; 95% CI 0.73 to 0.97; P=0.02.
  • ICU transfer was rare (0.2% vs 0.3%), highlighting a ward-based, resource-limited pathway where early inpatient management is pivotal.
  • Adverse events and serious adverse events were broadly similar; hyperglycaemia events were more frequent with glucocorticoids (35 vs 3 events).
  • Internal validity is moderate (open-label, withdrawals, heterogeneous exposure), but effect estimates were supported by concordant sensitivity analyses.

Further Reading

Other Trials

Systematic Review & Meta Analysis

Observational Studies

Guidelines

Notes

  • Most published guideline recommendations identified in targeted searches pre-date SONIA’s December 2025 print publication; future updates may alter strength and scope of recommendations for ward-based LMIC CAP phenotypes.
  • Where effect estimates were not provided in the SONIA manuscript/supplement (e.g., length of stay, in-hospital mortality comparisons), “Not reported” is shown rather than derived calculations.

Overall Takeaway

SONIA is a landmark pragmatic trial because it demonstrates an association between early adjunctive low-dose systemic glucocorticoids and reduced 30-day mortality in ward-managed CAP in a low-resource Kenyan setting. Its scale, mortality endpoint, and real-world implementation features make it highly influential for LMIC inpatient practice, while simultaneously sharpening debates about diagnostic heterogeneity, exposure fidelity, and transportability to ICU-enriched high-resource populations.

Overall Summary

  • Adjunctive low-dose glucocorticoids reduced 30-day mortality in Kenyan inpatients with clinically defined CAP (HR 0.84; 95% CI 0.73 to 0.97).
  • Exposure was pragmatic (oral, availability-driven) with high initiation but shortened in-hospital duration (median 4 days) and unverified post-discharge adherence.
  • Hyperglycaemia was more frequently reported with glucocorticoids; serious adverse events were similar overall.

Bibliography