Derivation and validation of predictive indices for cardiac readmission after coronary and valvular surgery - A...

Area of Science:

• Cardiovascular Surgery
• Health Informatics
• Predictive Analytics

Background:

• Cardiac surgery patients face significant risks of readmission.
• Accurate prediction of readmission is crucial for resource allocation and patient management.
• Existing models often lack generalizability or utilize limited data sources.

Purpose of the Study:

• To develop and validate predictive models for one-year cardiac readmission risk.
• To utilize routinely collected electronic medical record data for model derivation.
• To assess model performance after isolated coronary artery bypass grafting (CABG), aortic valve replacement (AVR), and combined CABG+AVR.

Main Methods:

• Retrospective cohort study using Canadian clinical registries and administrative databases.
• Development of Fine and Gray subdistribution hazard models within a competing-risk framework.
• External validation of models in an independent patient cohort.

Main Results:

• Models demonstrated good predictive performance: c-statistics ranged from 0.66 to 0.74.
• Post-CABG model: c-statistic 0.73 (derivation), 0.70 (validation).
• Post-AVR model: c-statistic 0.74 (derivation), 0.73 (validation).
• Post-CABG+AVR model: c-statistic 0.70 (derivation), 0.66 (validation).

Conclusions:

• Predicting one-year cardiac readmission after CABG, AVR, and combined procedures is feasible using multidimensional data.
• The derived models show improved discrimination compared to existing registry-based models.
• These models offer a parsimonious yet effective tool for risk stratification.