ACCORD
BWhen an intuitive intervention fails
The ACCORD trial randomized 10,251 patients with type 2 diabetes at high cardiovascular risk to intensive glucose lowering (HbA1c <6.0%) or standard therapy. Contrary to expectation, the intensive arm had higher all-cause mortality (HR 1.22, 95% CI 1.01–1.46) and the trial's glucose-lowering arm was terminated early. The primary composite of major cardiovascular events was not significantly reduced.
This appraisal is generated by AI (based on clinical-epidemiology methods) to inform professional judgment; it does not replace expert human review.
Study type: RCT (early-terminated) · N=10,251 · Follow-up 3.5 y · Evidence level: Abstract
Intensive glucose control targeting HbA1c <6.0% did not reduce cardiovascular events; only nonfatal MI favored intensive therapy, while mortality and safety moved toward harm. The evidence does not support tighter glucose targets in this population.
Abstract-only input caps the rating at B; evidence quality Moderate; the mortality CI is fragile (lower bound 1.01); absolute event rates are missing. Full-text review may adjust this.
PICO
Key endpoints
| Endpoint | Result | Direction |
|---|---|---|
| Primary composite (MACE) | Not significantly reduced | No clear benefit |
| All-cause mortality | HR 1.22 (1.01–1.46) | Favors standard |
| Nonfatal MI | Lower with intensive | Favors intensive |
| Severe hypoglycemia | ~3× excess | Favors standard |
Absolute effect — / /
abstract reports no arm-specific event rates
cannot be back-derived from a relative measure alone
no baseline hypoglycemia rate
The abstract reports relative measures () only, without arm-specific event rates. TrialReviewer attempts the recompute and flags what it cannot derive as a data gap — rather than glossing over it.
Methodology concerns
- 1
The trial was terminated early for harm.
Why it matters：Early stopping for harm limits long-term precision and makes the exact magnitude uncertain, but it does not erase the safety signal [Bassler et al. 2010].
Impact：The direction is clinically important; the exact long-term benefit–harm magnitude is less certain.
- 2
The composite endpoint masks divergent component effects — fewer MIs but more deaths.
Why it matters：The components move in opposite directions, so the composite alone is misleading.
Impact：A neutral composite should not be read as neutral clinical value when mortality and severe hypoglycemia move toward harm.
- 3
Arm-specific absolute event rates are missing.
Why it matters：The abstract reports only the HR, not event rates [Nuovo et al. 2002].
Impact：ARR / NNT cannot be reconstructed and the relative HR is hard to read clinically without absolute context.
- 4
No baseline rate for severe hypoglycemia.
Why it matters：The “~3×” is relative, with no absolute denominator.
Impact：NNH cannot be computed — yet this is the most decision-relevant safety endpoint.
- 5
Causal inference is limited at abstract level.
Why it matters：There is no mediation analysis for hypoglycemia → mortality.
Impact：“Intensive lowering causes death” is a hypothesis, not an established causal chain.
Benefit – harm tradeoff
Nonfatal MI reduced (the only component favoring intensive therapy).
All-cause and CV mortality increased; ~3× severe hypoglycemia (a plausible mediator of the excess deaths).
Negative in this phenotype — the MI benefit is offset and outweighed by mortality and hypoglycemia harm.
Why downgraded ()
Citations & references
In the full report every threshold judgment carries an inline [Author Year] citation. Sources for this example:
- Bassler D et al. (2010). Stopping randomized trials early for benefit. JAMA; 303(12):1180–7.
- Guyatt GH et al. (2008). GRADE: rating quality of evidence. BMJ; 336(7650):924–6.
- Nuovo J et al. (2002). Reporting NNT and ARR in RCTs. JAMA; 287(21):2813–4.
- Walsh M et al. (2014). RCT results are frequently fragile: a Fragility Index. J Clin Epidemiol; 67(6):622–8.
- Schulz KF & Grimes DA. (2002). Sample size slippages in RCTs. Lancet; 359(9308):781–5.