Dynamic readmission prediction using routine postoperative laboratory results after radical cystectomy

Area of Science:

• Urology
• Health Informatics
• Medical Data Science

Background:

• Complications and readmissions after radical cystectomy are frequent challenges.
• Existing risk stratification methods may not fully leverage dynamic patient data.
• Electronic health records offer a rich source of real-time physiological information.

Purpose of the Study:

• To evaluate if integrating electronic health record (EHR) data enhances risk stratification and readmission prediction post-radical cystectomy.
• To determine the predictive value of common postoperative laboratory results for 30-day readmissions.

Main Methods:

• Utilized an institutional EHR database for demographic, clinical, and laboratory data from radical cystectomy patients.
• Employed support vector machine learning to analyze the trajectory of postoperative laboratory values.
• Compared predictive models with and without laboratory results for 30-day readmission prediction.

Main Results:

• 26% of 996 patients experienced 30-day readmission.
• Postoperative white blood cell count, urea nitrogen, bicarbonate, and creatinine levels differentiated readmitted patients.
• Incorporating laboratory results substantially improved the accuracy of readmission prediction models.

Conclusions:

• Postoperative laboratory values possess discriminatory power for identifying patients at high risk of readmission after radical cystectomy.
• Dynamic physiological data from EHRs can enable more accurate identification and targeting of at-risk patients.
• This proof-of-concept study warrants further investigation into these predictive techniques.