Background Patients with chronic kidney disease (CKD) are at risk of medication therapy problems (MTPs) due to high comorbidity and medication burden. Using data from the Kidney Coordinated HeAlth Management Partnership (K-CHAMP) trial, we used machine learning to identify predictors of MTPs in patients with CKD in the primary care setting.
Methods We used data from patients enrolled in intervention arm of K-Champ trial which tested a population health management strategy, including medication management, for improving CKD care. The dataset was divided into 80% training and 20% testing subsets. The area under the ROC curve (AUROC) served as the metric to assess the classifier’s efficacy in distinguishing between patients with and without MTP (Medication Therapy Problems). Initially, eight models were assessed, and the top three models, Random Forest (RF), Support Vector Machines (SVM), and Gradient Boosting (GB), were selected. These three models underwent further refinement, and the best-performing model was chosen based on the highest AUROC using the testing set, while considering the Bias/Variance trade-off.
Results: Among 730 patients who received medication review at baseline, 566 (77.5%) had at least 1 MTP at baseline. Their mean age 74, 55% females, 92% Whites, 64% with diabetes, 5.8 mean number of meds at baseline. AUROC for the 3 models are shown in Fig 1 for both the training and testing data sets. RF had best performance on the testing set with highest AUROC (0.72), a sensitivity of 0.80 and specificity of 0.64. The five most influential variables, ranked in descending order of importance for predicting individuals with MTP, were diabetes status (yes/no), A1C levels, UACR levels, age, and the count of comorbidities.
Conclusion: Random forest provided the highest performance in predicting MTP for patients with advanced CKD. Future work will focus on developing a user-friendly online tool aimed at identifying patients who may benefit most from pharmacist-led medication management.