Development of a Multicenter Ward-Based AKI Prediction Model.

Area of Science:

• Nephrology
• Medical Informatics
• Predictive Analytics

Background:

• Acute Kidney Injury (AKI) poses a significant risk to hospitalized patients, often detected only after serum creatinine levels rise.
• Early identification of at-risk patients is crucial for preemptive interventions to mitigate AKI severity.

Purpose of the Study:

• To develop and validate an AKI risk prediction algorithm using electronic health record (EHR) data for patients on general hospital wards.
• To enable early detection of AKI before significant changes in serum creatinine.

Main Methods:

• A discrete time survival model was employed, utilizing demographics, vital signs, and routine laboratory data from EHRs.
• The algorithm, named Electronic Signal to Prevent AKI, was developed on 60% of the cohort and prospectively validated on the remaining 40%.
• Area Under the Receiver Operating Characteristic curves (AUC) were calculated for AKI prediction within 24 hours.

Main Results:

• The study included 202,961 patients, with 8.6% developing AKI and 0.6% progressing to stage 3.
• The final model achieved AUCs of 0.74 for stage 1 AKI and 0.83 for stage 3 AKI in the validation cohort.
• The algorithm identified patients at risk a median of 42 hours before stage 1 AKI and 35 hours before stage 3 AKI.

Conclusions:

• Electronic health record data can effectively stratify AKI risk with high accuracy.
• Real-time implementation of the Electronic Signal to Prevent AKI tool can facilitate early interventions, potentially improving patient outcomes and reducing healthcare costs.