Using Routinely Collected Electronic Health Record Data to Predict Readmission and Target Care Coordination

Area of Science:

• Nephrology
• Health Informatics
• Predictive Analytics

Background:

• Patients with chronic renal failure (CRF) face high 30-day hospital readmission rates.
• Routinely collected electronic health record (EHR) data offers potential for predicting CRF readmissions.
• Current generic risk tools may not capture the specific needs of CRF populations.

Purpose of the Study:

• To compare the predictive accuracy of EHR-derived variables versus manually extracted data for CRF readmissions.
• To assess the utility of routinely collected EHR data for stratifying readmission risk in CRF patients.
• To evaluate the potential of EHR data for targeted interventions to reduce readmissions.

Main Methods:

• Multivariate logistic regression analysis was applied to one year of admission data from an academic medical center.
• Three routinely collected EHR variables (creatinine, B-type natriuretic peptide, length of stay) were categorized.
• Comparison of predictive performance was made against paper-based methods using the C-statistic (AUC).

Main Results:

• Categorizing specific EHR variables improved readmission prediction by 30% compared to paper-based methods (AUC).
• Marginal effects analysis yielded patient-specific risk scores ranging from 0% to 44.3%.
• EHR-based prediction demonstrated superior accuracy for CRF readmission risk.

Conclusions:

• Routinely collected EHR data is an effective and efficient strategy for stratifying readmission risk in CRF patients.
• Utilizing EHR data allows for more precise identification of high-risk individuals.
• This approach supports targeted care interventions and may reduce overall hospital readmissions.