221-OR: A Machine Learning Approach Identifies Modulators of Congestive Heart Failure Hospitalization Prevention among Patients with Type 2 Diabetes: A Revisit to the ACCORD Trial

Objective: To examine patient characteristics that modulate the efficacy of risk factor reduction in preventing hospitalization for congestive heart failure (CHF) in diabetes management.

Methods: The Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial tested the cardioprotective effect of achieving A1c< 6.0%, systolic blood pressure (SBP) <120 mmHg, and intensified lipid control by adding fenofibrate among individuals with type 2 diabetes, using a 2x2 factorial design. We analyzed 8,992 participants from the ACCORD glucose trial, 4,178 from the SBP sub-trial, and 4,178 from the lipids sub-trial separately. We applied the casual tree (CT) with cross-validation to each trial to identify baseline characteristics that clustered participants into subgroups with different levels of CHF risk reduction by the corresponding intervention.

Results: Individuals with baseline Alanine transaminase (ALT) at the lowest quartile (<18 mg/dl) benefited the most from SBP control (absolute risk reduction (ARR) in CHF: -1.7%, 95% CI: -3.1% to -0.4%). Individuals with ALT >18 mg/dl and low-density lipoprotein > 110 mg/dl who were receiving ACE inhibitor at baseline also benefited substantially from SBP control (ARR: -1.6%, 95% CI: -2.8% to -0.3%). Optimizing lipid therapy with fenofibrate lowered CHF risk by -1.3% ~ -2.4% in individuals with normal baseline creatine phosphokinase (CPK) (< 200 mg/dl) and A1c < 8%. However, individuals with abnormal CPK at baseline were not associated with changes in CHF risk from fenofibrate. For glucose control, individuals with diastolic blood pressure (DBP) above 69 mmHg and very lower density lipoprotein below 25 mg/dl had an ARR of -0.8% (95% CI -1.6% to 0.0%), and individuals with DBP below 69 mm/Hg had no benefit.

Conclusion: This novel approach demonstrates that ML can identify patients with characteristics that are more or less likely to reduce CHF risk from intensive therapy.