Predicting Kidney Transplant Recipient Cohorts' 30-Day Rehospitalization Using Clinical Notes and Electronic Health...

Area of Science:

• Nephrology
• Medical Informatics
• Data Science

Background:

• Rehospitalization after kidney transplant presents significant costs and adverse outcomes.
• Few studies have explored the predictive value of clinical notes from electronic medical records (EMR) for rehospitalization.
• Predictive models are crucial for identifying high-risk kidney transplant patients.

Purpose of the Study:

• To evaluate whether incorporating clinical notes data enhances the prediction of 30-day rehospitalization (30DR) in kidney transplant recipients.
• To compare the predictive performance of models using structured EMR data versus those including unstructured clinical notes.

Main Methods:

• Retrospective observational study of adult kidney transplant recipients (2005-2015).
• Utilized natural language processing (NLP) on 8 types of clinical notes and structured EMR data.
• Employed unsupervised machine learning for predictive modeling, assessing accuracy with ROC and PRC curves, and 5-fold cross-validation.

Main Results:

• 30.7% of 2060 kidney transplant recipients experienced 30-day rehospitalization.
• Models incorporating clinical notes did not substantially improve prediction accuracy over structured data alone (ROC 0.6821).
• Models using both structured data and progress notes achieved the highest performance (ROC 0.6902).

Conclusions:

• Integrating clinical notes into risk prediction models did not significantly enhance the accuracy for 30-day rehospitalization in kidney transplant patients.
• Future research should focus on multi-institutional data pooling to increase sample size and mitigate model overfitting.