Reformulating patient stratification for targeting interventions by accounting for severity of downstream outcomes resulting from disease onset: a case study in sepsis.

Summary

Across two ICU cohorts, the estimated effect of sepsis on mortality was only weakly correlated with predicted sepsis onset risk, and high-risk groups overlapped by 53–67% depending on the site. Incorporating the mortality effect into stratification identified a different, older population than risk-only approaches, indicating that risk-only targeting may miss high-impact intervention candidates.

Key Findings

• Sepsis risk and estimated mortality effect were weakly correlated (Spearman 0.35 at U-M; 0.31 at BIDMC).
• High-risk patients overlapped by 66.8% (U-M) and 52.8% (BIDMC) between risk-only versus risk+effect stratification.
• Including mortality effect identified an older population than risk-only stratification.
• Among sepsis cases, mortality occurred in 21.9% (U-M) and 26.3% (BIDMC).

Clinical Implications

Decision support should incorporate both disease risk and estimated impact on outcomes (e.g., mortality) when prioritizing alerts and resources, potentially improving yield and equity of sepsis interventions.

Limitations

• Retrospective observational design with potential unmeasured confounding.
• Effect estimation details and generalizability beyond ICU settings may be limited.

Future Directions

Prospective evaluation of effect-aware targeting on clinical outcomes and cost-effectiveness; integration with causal inference frameworks and fairness metrics in deployment.