TrialReviewer
← Back to the library

RECOVERY (dexamethasone)

A

When evidence changes practice

Input · the abstract you paste

In the RECOVERY platform trial, 6,425 hospitalized COVID-19 patients were randomized to dexamethasone 6 mg daily (up to 10 days) or usual care. Dexamethasone reduced 28-day all-cause mortality (rate ratio 0.83), with the benefit concentrated in patients requiring respiratory support — largest in those on invasive mechanical ventilation (NNT 8) — and no benefit, with a possible harm signal, in patients needing no oxygen.

For illustration only. TrialReviewer appraises evidence quality to inform professional judgment; it is not medical advice, and clinical decisions are made by the clinician.

This appraisal is generated by AI (based on clinical-epidemiology methods) to inform professional judgment; it does not replace expert human review.

Study type: open-label RCT (platform trial) · N=6,425 (dexamethasone 2,104 / usual care 4,321) · Primary endpoint 28-day · Evidence level: Full text

Bottom Line
Appraisal confidenceA

In hospitalized COVID-19 patients requiring respiratory support, dexamethasone provides practice-changing evidence: a large, simple platform trial with a hard 28-day mortality endpoint and an NNT of ≈8 in the mechanical-ventilation subgroup. Benefit is also present in oxygen-only patients; the evidence does not support use in patients not requiring oxygen, where the point estimate favors harm.

Reduces 28-day all-cause mortality in respiratory-support COVID-19 (IMV ARR 12.1pp crude, NNT 8; oxygen-only ARR 2.9–4.2pp crude–adjusted, NNT ≈24–34), with the primary outcome graded High, pre-registered endpoints, 99.7% ascertainment, and no material bias to the primary endpoint — grade A.

PICO

P — Population
Hospitalized adults with suspected / confirmed COVID-19 (UK NHS, 176 sites)
I — Intervention
Dexamethasone 6 mg daily, up to 10 days
C — Comparison
Usual care
O — Outcome
28-day all-cause mortality (primary); discharge, progression to IMV (secondary)

Key endpoints

Endpoint / subgroupResultARR · NNT
28-day all-cause mortality (overall)RR 0.83 (0.75–0.93)ARR 2.8pp (crude) · NNT 36
Mechanical ventilation subgroupRR 0.64 (0.51–0.81)ARR 12.1pp (crude) · NNT 8
Oxygen-only subgroupRR 0.82 (0.72–0.94)ARR 2.9–4.2pp (crude–adjusted) · NNT ≈24–34
No respiratory support subgroupRR 1.19 (0.92–1.55)Harm signal (not significant)
28-day all-cause mortality RR 0.83 (95% CI 0.75–0.93)
1.0
The CI excludes the null (1.0) — a statistically significant benefit.

Absolute effect — ARR / NNT / NNH

ARR
2.8pp
overall · IMV subgroup 12.1pp
NNT
36
overall · IMV subgroup NNT 8
NNH
not estimable
Adverse events were sparsely reported and likely under-ascertained in the open-label pragmatic design, so a reliable NNH cannot be estimated.

The full text reports arm-specific rates, so ARR / NNT are genuinely computed — an NNT of 8 in ventilated patients is exceptionally low for a mortality endpoint. Subgroup absolute differences here are crude observed-rate differences; back-derived from the age-adjusted rate ratio, the oxygen-only benefit is somewhat larger (adjusted ARR ≈4.2pp).

Methodology concerns

  1. 1

    Open-label design and adverse-event ascertainment.

    Why it matters：Only 4 serious adverse reactions were reported across 2,104 patients, whereas steroid effects such as hyperglycemia should be commoner [Schulz et al. 2010].

    Impact：The safety profile is incomplete (AEs under-ascertained); it does not affect the primary mortality conclusion.

  2. 2

    The no-respiratory-support subgroup harm signal is not statistically confirmed.

    Why it matters：RR 1.19 (0.92–1.55), CI crosses 1.0.

    Impact：The point estimate favors harm and is biologically coherent (early viral-replication phase), so use in this subgroup is not supported.

  3. 3

    About 8% glucocorticoid crossover in the usual-care arm.

    Why it matters：Contaminating the control biases the observed effect toward the null.

    Impact：The reported benefit is a conservative estimate; the true effect may be larger.

  4. 4

    No physiologic / laboratory / virologic data collected.

    Why it matters：The mechanistic interpretation (immunopathologic-phase hypothesis) is inferential.

    Impact：It limits mechanistic certainty but does not affect the primary clinical conclusion.

  5. 5

    28-day follow-up horizon.

    Why it matters：The pre-specified 6-month outcomes are not in this report.

    Impact：Long-term sequelae of 10-day dexamethasone (secondary infections, adrenal suppression) are uncharacterized.

Benefit – harm tradeoff

Benefit

Mortality falls sharply in respiratory-support patients (IMV NNT 8, oxygen-only NNT ≈24–34), with better discharge, ventilator liberation, and less renal-replacement therapy.

Harm

Glucocorticoid class effects (few reported, likely under-ascertained); the no-oxygen subgroup's point estimate favors harm — the evidence does not support use there.

Net clinical value

Strongly positive in respiratory-support patients — this is exactly “evidence changes practice”; but no-oxygen patients gain nothing and may be harmed, a clear boundary.

Why downgraded (GRADE)

Start: High (RCT)
Primary outcome (28-day mortality): no downgrade
Risk of Bias: open-label doesn't affect the hard mortality endpoint, full ITT, attrition <1%, pre-registered; Inconsistency / Indirectness / Imprecision (CI 0.75–0.93 excludes 1) / publication bias all clean [Balshem et al. 2011]. (The safety profile and the no-oxygen subgroup carry separate limitations — see Concerns.)
Primary outcome (28-day all-cause mortality) · evidence quality: High → appraisal confidence: A

Citations & references

In the full report every threshold judgment carries an inline [Author Year] citation. Sources for this example:

  • Balshem H et al. (2011). GRADE guidelines: 3. Rating the quality of evidence. J Clin Epidemiol; 64(4):401–6.
  • Guyatt GH et al. (2008). GRADE: rating quality of evidence. BMJ; 336(7650):924–6.
  • Chan AW et al. (2004). Empirical evidence for selective reporting of outcomes in RCTs. JAMA; 291(20):2457–65.
  • Rothwell PM (2005). Subgroup analysis in randomised controlled trials. Lancet; 365(9454):176–86.
  • Schulz KF et al. (2010). CONSORT 2010 statement. BMJ; 340:c332.
Start my own appraisal →