To evaluate the performance and methodological quality of sepsis mortality prediction models and identify common predictors associated with sepsis mortality.
Approach:
Key Findings:
84 studies included, reporting 235 prediction models for sepsis mortality, involving approximately 2.7 million patient records and 461,387 deaths.
Pooled AUC of externally validated models was 0.794, indicating moderate discriminative performance.
Common predictors of mortality included age, lactate, albumin, SOFA score, and vasopressor use.
76.19% of studies had a high risk of bias in model development.
Interpretation:
Externally validated prediction models generally demonstrate moderate discriminative performance for predicting sepsis mortality, which has implications for clinical decision-making.
Limitations:
High risk of bias in a substantial proportion of studies.
Limited generalizability due to study design, including a predominance of retrospective cohort studies and small cohort sizes.
Conclusion:
Future studies with larger cohorts and rigorous designs are warranted to improve generalizability and facilitate clinical implementation.
