Estimating the Potential Cardiovascular Benefit of Lifestyle Risk Factors Using Causal Machine Learning: Insights from UK Biobank

Background Increasingly recommendations for prevention focus on global risk assessment in lower risk individuals where sufficient lifestyle changes may reduce the need for pharmacotherapy; yet these benefits are hard to quantify, undermining adherence. We aimed to derive a tool that personalises the cardiovascular benefit associated with lifestyle risk factors. Methods Baseline exposures were linked to longitudinal outcomes in 371,102 UK Biobank participants without prior CVD. Causal effects were estimated using directed acyclic graphs with inverse propensity weighting (IPW), augmented IPW, generalised linear models, and causal machine learning for smoking (former/never vs current), obesity (BMI <30 vs ≥30 kg/m²), physical activity (≥600 vs <600 MET-minutes/week), and alcohol (less-than-daily vs daily). The outcome was a 10-year composite of cardiovascular death, non-fatal MI, or stroke. Results Median age was 58 years (IQR 50–63), 56% female; 15,752 CVD events occurred over ten years. The largest risk difference was for former/never versus current smoking (absolute risk difference [ARD] 3.2–3.8%), followed by BMI <30 versus ≥30 kg/m² (ARD 1.0–1.1%) and higher versus lower physical activity (ARD 0.2–0.4%); less-than-daily versus daily alcohol carried a small but higher risk. Differences were larger at higher baseline risk and with co-existing risk factors, increasing with pack-year for smoking (β=+0.23% per SD), and were slightly lower at higher BMI (β=−0.04%). Conclusion Among the lifestyle risk factors examined, smoking provided the largest risk difference, increasing with pack-years. Estimates for BMI were smaller and broadly consistent across the obesity range. Findings are accessible via a free tool (lifestyle-cvd.casefinding.com).