Determination of Medication Regimen Complexity Trends in Older Adult Patients With Insomnia: A Triangulation of Findings From Artificial Neural Network and Logistic Regression Analyses
Author(s)
Hsiang-Wen Lin, RPh, M.S., Ph.D.1, Chun-Wei Tu, RPh., M.S.2, Daniel Hsiang-Te Tsai, RPh, MSc3, Yu-Chieh Chen, RPh, M.S., Ph.D.4, Chun-Hui Liao, M.D., M.S.5, Chih-Hseuh Lin, M.D., Ph.D.6;
1School of Pharmacy and Graduate Institute, College of Pharmacy, China Medical University; Department of Pharmacy, China Medical University Hospital, Taichung City, Taiwan, 2Department of Pharmacy, China Medical University Hospital, Taichung, Taiwan, 3School of Pharmacy, Institute of Clinical Pharmacy and Pharmaceutical Sciences, College of Medicine, National Cheng Kung University, Tainan, Taiwan, 4Department of Pharmacy, China Medical University Hospital, Taichung City, Taiwan, 5Department of Psychiatry, China Medical University Hospital, Taichung City, Taiwan, 6School of Medicine, College of Medicine, China Medical University; Department of Geriatric Medicine, China Medical University Hospital, Taichung City, Taiwan
1School of Pharmacy and Graduate Institute, College of Pharmacy, China Medical University; Department of Pharmacy, China Medical University Hospital, Taichung City, Taiwan, 2Department of Pharmacy, China Medical University Hospital, Taichung, Taiwan, 3School of Pharmacy, Institute of Clinical Pharmacy and Pharmaceutical Sciences, College of Medicine, National Cheng Kung University, Tainan, Taiwan, 4Department of Pharmacy, China Medical University Hospital, Taichung City, Taiwan, 5Department of Psychiatry, China Medical University Hospital, Taichung City, Taiwan, 6School of Medicine, College of Medicine, China Medical University; Department of Geriatric Medicine, China Medical University Hospital, Taichung City, Taiwan
OBJECTIVES: As the trends of the Medication Regimen Complexity Index (MRCI), encompassing dosage form, frequency, and additional instruction sub-sections, over time remain unclear, this study aimed to determine its optimal cut-off score, assess its one-year trends, and develop MRCI prediction model for older adults with insomnia.
METHODS: Data were collected at baseline and one year later from a prospective, longitudinal study for older adults with insomnia in Taiwan. Five steps were conducted:1) Calculating MRCI total and sub-scores; 2). Identifying the optimal baseline cut-off score; 3). Analyzing factors influencing MRCI trends; 4). Training and testing prediction models using logistic regression (LR) and artificial neural networks (ANN) for complex MRCI after one year; 5). Estimating risk probabilities. The best model was determined based on the Area Under Receiver Operating Characteristic Curve (AUROC), prediction accuracy, and clinical relevance.
RESULTS: Among 132 older adults being followed-up over one-year, the average number of prescribed medications and MRCI total scores at baseline were 8.6 ± 4.4 and 23.6 ± 14.1, respectively. An MRCI total score ≥16 was classified as complex (C), while <16 was non-complex (NC). Over one year, 55% remained complex (C-C), 23% remained non-complex (NC-NC), 7.6% shifted from NC to C, and 14.4% from C to NC. Significant differences were found in 19 baseline factors across the four groups. Both the 10-parameter LR and 12-feature ANN models demonstrated excellent performance (AUROC ~0.90). The LR model, with baseline MRCI complexity as the strongest predictor, was the best model.
CONCLUSIONS: An MRCI total score ≥16 serves as an effective threshold, with over half of older adults maintaining complex medication regimens after one year. Baseline MRCI complexity emerged as the primary predictor in the final 10-parameter logistic regression model. MRCI could be a valuable quality indicator for guiding deprescribing efforts, though further validation in diverse populations is recommended.
METHODS: Data were collected at baseline and one year later from a prospective, longitudinal study for older adults with insomnia in Taiwan. Five steps were conducted:1) Calculating MRCI total and sub-scores; 2). Identifying the optimal baseline cut-off score; 3). Analyzing factors influencing MRCI trends; 4). Training and testing prediction models using logistic regression (LR) and artificial neural networks (ANN) for complex MRCI after one year; 5). Estimating risk probabilities. The best model was determined based on the Area Under Receiver Operating Characteristic Curve (AUROC), prediction accuracy, and clinical relevance.
RESULTS: Among 132 older adults being followed-up over one-year, the average number of prescribed medications and MRCI total scores at baseline were 8.6 ± 4.4 and 23.6 ± 14.1, respectively. An MRCI total score ≥16 was classified as complex (C), while <16 was non-complex (NC). Over one year, 55% remained complex (C-C), 23% remained non-complex (NC-NC), 7.6% shifted from NC to C, and 14.4% from C to NC. Significant differences were found in 19 baseline factors across the four groups. Both the 10-parameter LR and 12-feature ANN models demonstrated excellent performance (AUROC ~0.90). The LR model, with baseline MRCI complexity as the strongest predictor, was the best model.
CONCLUSIONS: An MRCI total score ≥16 serves as an effective threshold, with over half of older adults maintaining complex medication regimens after one year. Baseline MRCI complexity emerged as the primary predictor in the final 10-parameter logistic regression model. MRCI could be a valuable quality indicator for guiding deprescribing efforts, though further validation in diverse populations is recommended.
Conference/Value in Health Info
2025-05, ISPOR 2025, Montréal, Quebec, CA
Value in Health, Volume 28, Issue S1
Code
EPH168
Topic
Epidemiology & Public Health
Disease
No Additional Disease & Conditions/Specialized Treatment Areas, SDC: Geriatrics