Presentation Information
[2Yin-A-02]Structural Causal Graph Analysis of Pharmacological Interventions in Heart Failure Prognosis— A Monte Carlo–Augmented Observational Study —
〇Masato Shimizu1, Makoto Suzuki1, Tetsuo Sasano2 (1. Yokohama Minami Kyosai Hospital, 2. Institute of Science Tokyo)
Keywords:
Structural Causal Graph,Atrial Fibrillation,Prognosis
Background:
Heart failure with mildly reduced ejection fraction (HFmrEF; EF 40–50%) is often treated according to reduced EF guidelines, but the causal effects of pharmacological therapies on prognosis remain unclear.
Methods:
We retrospectively analyzed 90 consecutive patients hospitalized for acute decompensated HFmrEF (median age 83 years; 47 men). Discharge medications and clinical variables were included, and the primary endpoint was major adverse cardiac events (MACE) within 1 year. The dataset was expanded to 7,000 cases via Monte Carlo simulation. A structural causal graph was constructed using a score-based causal model, and average treatment effects (ATEs) were estimated with backdoor adjustment. For drugs with negative ATEs, counterfactual intervention simulations were performed using DoWhy.
Results:
MACE occurred in 23 patients (26%). The causal graph identified six direct and eight indirect causal factors. Beta blockers showed direct effects, while diuretics and SGLT2 inhibitors showed indirect effects. Intervention simulations estimated prognostic improvements of 5.7% for beta blockers and 1.1% for SGLT2 inhibitors.
Conclusion:
Structural causal graph analysis with Monte Carlo simulation enables causal evaluation of treatment effects from observational medical data and provides novel insights into heart failure prognosis.
Heart failure with mildly reduced ejection fraction (HFmrEF; EF 40–50%) is often treated according to reduced EF guidelines, but the causal effects of pharmacological therapies on prognosis remain unclear.
Methods:
We retrospectively analyzed 90 consecutive patients hospitalized for acute decompensated HFmrEF (median age 83 years; 47 men). Discharge medications and clinical variables were included, and the primary endpoint was major adverse cardiac events (MACE) within 1 year. The dataset was expanded to 7,000 cases via Monte Carlo simulation. A structural causal graph was constructed using a score-based causal model, and average treatment effects (ATEs) were estimated with backdoor adjustment. For drugs with negative ATEs, counterfactual intervention simulations were performed using DoWhy.
Results:
MACE occurred in 23 patients (26%). The causal graph identified six direct and eight indirect causal factors. Beta blockers showed direct effects, while diuretics and SGLT2 inhibitors showed indirect effects. Intervention simulations estimated prognostic improvements of 5.7% for beta blockers and 1.1% for SGLT2 inhibitors.
Conclusion:
Structural causal graph analysis with Monte Carlo simulation enables causal evaluation of treatment effects from observational medical data and provides novel insights into heart failure prognosis.