Home // International Journal On Advances in Life Sciences, volume 16, numbers 1 and 2, 2024 // View article

Approaches to Improving Medication Adherence Prediction in Chronic Disease Patients

Authors:
Ben Malin
Ejike Nwokoro
Tatiana Kalganova
Joshua Hinton

Keywords: medication adherence; CNN; healthcare; homecare; adherence prediction; length of service

Abstract:
This study aims to identify the impact of a patient’s treatment/ support service duration (LOS) on the ability of a machine learning model to predict their medication adherence. The insight generated from this study can support the adaptation of patient support interventions, based on the evolution of predicted adherence at different treatment or service durations. For adherence prediction, we use medication delivery data, driven by the patient’s prescription, to calculate a patient’s stock level at any given time during their participation in a homecare support service, whilst allowing for medication stockpiling. This data is visualized and inputted into a Convolutional Neural Network (CNN). To evaluate adherence for a range of LOS values, every patient’s first x months on service are extracted, with the final month used as the target variable. To define nonadherence, we use Proportion of Days Covered (PDC) of 100% for this period, where if a patient does not have any medication during this month they will be classed as nonadherent. Using this approach, we found that as LOS changes, there is a variation in both the proportion of the population that are adherent as well as the prediction model’s performance. Across the studied timeframe of 4-12 months LOS, proportion of the study population that are adherent varies between 54.6% and 66.1%, with an Area Under the Curve (AUC) varying from 88.1% to 98.6% for our best performing model. We also found that additional variables linked with adherence such as: communications with the service provider, demographic data, socioeconomic data and diagnosis specific average PDC, improve model performance. The model in this study achieves its highest AUC and adherence prediction accuracy of 98.6% and 92.8% respectively, at an LOS of 9 months. Additional evaluation was performed to identify variation across therapies offered through a Homecare service. The results from this evaluation show diversity across both adherence to these therapies as well as the accuracy to be expected from the adherence predictions. We conclude that this diversity is linked to medication delivery/prescription frequency, volume of medication stock prescribed as well as therapy-specific diagnosis differences.

Pages: 33 to 43

Copyright: Copyright (c) to authors, 2024. Used with permission.

Publication date: June 30, 2024

Published in: journal

ISSN: 1942-2660