Developing and Validating Machine Learning-Driven Risk Indices to Predict Patient Dropout During Referral,...

Area of Science:

• Nephrology
• Transplant Surgery
• Health Disparities
• Machine Learning in Healthcare

Background:

• Kidney transplant is the optimal treatment for kidney failure, but access disparities persist.
• Existing national registries do not capture early dropout stages in the transplant process.
• Risk indices were developed to predict dropout at referral, evaluation, and waitlisting.

Purpose of the Study:

• To develop and validate risk indices for predicting early dropout in kidney transplant candidates.
• To identify patient characteristics associated with higher dropout risk at different stages of the transplant process.
• To enable early identification of at-risk patients for targeted interventions.

Main Methods:

• Utilized machine learning (ML) models to predict dropout risk in kidney transplant candidates.
• Included demographic, clinical, and socioeconomic variables from electronic health records and census data.
• Evaluated model performance using Area Under the Receiver Operating Characteristic (AUROC) curves.

Main Results:

• 46% of referred patients missed their first evaluation; 54% of eligible patients were not waitlisted; 31% of waitlisted patients became inactive.
• Higher risk patients were older, obese, socioeconomically disadvantaged, and belonged to minority ethnic groups.
• Stage-specific risk factors included social determinants at referral, comorbidities at evaluation, and digital exclusion at waitlisting.

Conclusions:

• ML models effectively identified kidney transplant dropout risk at multiple stages.
• Early identification of at-risk patients is possible through these models.
• Targeted interventions can potentially reduce disparities and improve transplant access.