Illustrative AI review sample — internal medicine

Key Points

1Longitudinal analysis of 2,140 diabetic patients from a Chinese community health cohort examining HbA1c category and incident hypertension over a median 4-year follow-up; reverse causation risk, HbA1c measurement timing variability, and absent antihypertensive-use adjustment limit causal inference.

Major Issues

Methods: Antihypertensive medication initiation during follow-up is not modelled as a time-varying confounder; patients with poor glycaemic control may be prescribed antihypertensives earlier (protecting against the outcome) or later (masking blood pressure elevation), creating a competing events problem that biases the HbA1c–hypertension association.

Methods: HbA1c is measured at annual clinic visits only; within-person glycaemic variability (coefficient of variation, time-in-range) is not captured; a patient with HbA1c 7.8% at one visit may have substantially different integrated glucose burden compared to a patient with HbA1c 7.8% across all visits.

Methods / Results: Hypertension is ascertained from clinical coding without ambulatory blood pressure measurement (ABPM) validation; coding of hypertension in a community diabetes registry is subject to under-ascertainment bias (undiagnosed hypertension) and coding heterogeneity across participating clinics.

Discussion: The J-curve relationship between HbA1c and cardiovascular outcomes is not explored; the reference category (HbA1c ≤7%) includes patients at risk of hypoglycaemia-related cardiovascular events; a spline analysis would reveal whether the relationship is truly monotonic.

Association of glycemic control with hypertension in patients with diabetes: a population-based longitudinal study

Key Points

Major Issues