Using Machine Learning to Predict Medication Therapy Problems among Patients with Chronic Kidney Disease.

Area of Science:

• Pharmacogenomics
• Nephrology
• Artificial Intelligence in Medicine

Background:

• Patients with chronic kidney disease (CKD) face a high burden of comorbidities and medications, increasing their risk of medication therapy problems (MTPs).
• The Kidney Coordinated HeAlth Management Partnership (Kidney CHAMP) trial provided data to develop predictive models for MTP risk in CKD patients within primary care.

Purpose of the Study:

• To develop and validate a machine learning model for predicting medication therapy problems (MTPs) in patients with chronic kidney disease (CKD).
• To identify key clinical factors associated with MTP risk in the CKD population within a primary care setting.

Main Methods:

• Utilized baseline data from 730 patients in the Kidney CHAMP trial, splitting into 80% training and 20% testing sets.
• Evaluated eight candidate machine learning models, selecting the top three (random forest, support vector machines, gradient boosting) for refinement based on AUROC.
• The best-performing random forest model (AUROC 0.72) was identified using SHapley Additive exPlanations to determine predictor importance.

Main Results:

• A significant proportion of CKD patients (77.5%) experienced at least one MTP at baseline.
• The random forest model demonstrated good predictive performance with an AUROC of 0.72, sensitivity of 0.80, and specificity of 0.64.
• Key predictors for MTP risk included diabetes status, hemoglobin A1C, urine albumin-to-creatinine ratio, systolic blood pressure, and age.

Conclusions:

• A machine learning-based risk calculator utilizing routinely available clinical data can effectively identify CKD patients at high risk for MTPs in outpatient primary care.
• This predictive tool supports proactive medication management and intervention for high-risk individuals, potentially improving patient outcomes.