Heterogeneous blood pressure treatment effects on cognitive decline in type 2 diabetes: A machine learning analysis of a randomized clinical trial

Abstract Aim We aimed to identify the characteristics of patients with diabetes who can derive cognitive benefits from intensive blood pressure (BP) treatment using machine learning methods. Materials and Methods Using data from the Action to Control Cardiovascular Risk in Diabetes Memory in Diabetes (ACCORD‐MIND) study, 1349 patients with type 2 diabetes who underwent BP treatment (intensive treatment targeting a systolic BP <120 mmHg vs. standard treatment targeting <140 mmHg) were included in the machine learning analysis. Seventy‐nine variables correlated with diabetes and cognitive function were used to build the causal forest and causal tree models for identifying heterogeneous BP treatment effects on cognitive decline. Results Our analyses identified four variables including urinary albumin‐to‐creatinine ratio (UACR, mg/g), Framingham 10‐year cardiovascular risk score (FRS, %), triglycerides (TG, mmol/L) and diabetes duration, that categorized the participants into five subgroups with different risk benefits for cognitive decline from BP treatments. Subgroup 1 (UACR ≥65 mg/g) had an absolute risk reduction (ARR) of 15.36% (95% CI, 5.01%–25.46%) from intensive versus standard BP treatment (hazard ratio [HR], 0.36; 95% CI, 0.18–0.73). Subgroup 2 (UACR <65 mg/g, FRS ≥26%, TG <2.3 mmol/L and diabetes duration ≥9 years) had an ARR of 14.74% (95% CI, 4.56%–24.59%) from intensive versus standard BP treatment (HR, 0.34; 95% CI, 0.15–0.77). No significant benefits were found for other subgroups. Conclusions Patients with type 2 diabetes with high UACR, or with low UACR and low TG, but high predicted cardiovascular risk and long diabetes duration were likely to derive cognitive benefits from intensive BP treatment.