Illustrative AI review sample — cardiology

Key Points

1Nationwide Taiwan NHI claims cohort showing lower incident heart failure with NOACs versus warfarin in ≥65-year-old patients with AF and diabetes; target trial emulation strengthens causal inference but residual channelling bias, active-comparator design considerations, and HbA1c confounding remain unaddressed.

Major Issues

Methods: Prevalent NOAC users are included alongside incident initiators; this violates the new-user active-comparator design principle and introduces depletion-of-susceptibles bias, as patients who have already tolerated their anticoagulant for months differ systematically from new initiators.

Methods: HbA1c and left ventricular ejection fraction (LVEF) are not available in claims data and are omitted from the propensity model; since glycaemic control drives both anticoagulant choice (HbA1c instability favours warfarin avoidance) and HF risk, this unmeasured confounding inflates the apparent NOAC benefit.

Results: The 4:1 NOAC-to-warfarin ratio reflects post-2015 prescribing shifts; a secular time-trend analysis is not performed, leaving open the possibility that calendar-time confounding (improving overall cardiovascular care) explains part of the NOAC advantage.

Discussion: No E-value or quantitative bias analysis is provided for the primary HR 0.80; the minimum unmeasured confounding required to shift the lower CI above 1.0 should be reported to help readers gauge robustness.

Risk of heart failure in elderly patients with atrial fibrillation and diabetes taking different oral anticoagulants: a nationwide cohort study

Key Points

Major Issues