NICE Cardiology Guidance Under the Microscope: A Data-Driven Statistical Critique

Abstract

Correction (Version 2 — 10 July 2026): This corrected version repairs reference and arithmetic errors in the original while confirming its core result. Seven of twelve reference PMIDs were wrong and are corrected; the original E-value (~2.1) was arithmetically incorrect and is corrected to 1.85; and the unverifiable claim that NICE was the “last of six” bodies on a specific named technology appraisal is withdrawn as not independently verifiable. The reproducible core is unchanged and independently regenerated by the companion script: the SGLT2-inhibitor class effect on cardiovascular death or heart-failure hospitalisation is HR 0.79 (I²=8%, Q=7.61), and a new analysis shows no difference between dedicated heart-failure trials and diabetes cardiovascular-outcome trials (interaction p=0.72).

Background. Sodium–glucose co-transporter-2 (SGLT2) inhibitors are among the most thoroughly evidenced advances in cardiovascular pharmacotherapy of the past decade. We ask two questions: (i) how statistically robust and reversal-resistant is the class effect on the composite of cardiovascular death or heart-failure hospitalisation (CV-death/HHF), and (ii) can guideline-adoption timing across national bodies be quantified from the public record?

Methods. Eight pivotal randomised trials reporting a CV-death/HHF composite were pooled on the log-hazard-ratio scale using REML (primary), Paule–Mandel and DerSimonian–Laird τ² estimators, each with a Hartung–Knapp–Sidik–Jonkman (HKSJ) t-interval (q≥1 floor) and, for like-for-like reproduction, a random-effects z-interval. We report Cochran Q, I², a 95% prediction interval, leave-one-out influence, a new dedicated-HF-trial-versus-diabetes-CVOT subgroup interaction test, a seeded Monte-Carlo reversal probability, and E-values. Guideline dates were taken from the primary guideline publications; unverifiable claims are flagged, not asserted.

Results. Pooled HR 0.79 (REML; RE z-CI 0.749–0.827; HKSJ 0.739–0.837), I²=8.0%, Q=7.61 (df=7, p=0.37), τ²≈0. The 95% prediction interval (0.74–0.84) excludes 1. All eight leave-one-out estimates lie in 0.78–0.80 with CIs excluding 1. The class effect does not differ between the four dedicated HF trials (HR 0.781) and the four diabetes CVOTs (HR 0.795): interaction p=0.72. Monte-Carlo P(HR≥1) < 0.001%. E-value 1.85 (point), 1.71 (CI bound).

Conclusion. The SGLT2-inhibitor CV-death/HHF class effect is statistically mature, homogeneous across trial designs, and highly reversal-resistant. The existence of measurable inter-body adoption lags is defensible from public dates; the specific claim that NICE was "last of six" is not independently verifiable from the substrate here and is flagged accordingly.

---