A Machine Learning Approach for Predicting Therapeutic Adherence to Osteoporosis Treatment Ggaliwango Marvin and Md. Golam Rabiul Alam Department of Computer Science and Engineering. Brac University Dhaka, Bangladesh Email: ggaliwango.marvin@g.bracu.ac.bd, rabiul.alam@bracu.ac.bd Abstract—--Osteoporosis is a great disability burden with an expected cost increase of almost 50% by 2025. Due to its long term treatment, 50–70% of the patients withdraw from their osteoporosis medications within the first year of initiation. This necessitates an urgent need for improved osteoporosis and phar- macologic management tools most especially for pregnant women, postmenopausal women and the elderly to ensure therapeutic adherence of the patients during treatment. In this paper, we developed and tested accuracy of Machine Learning Models for predicting therapeutic adherence of patients to enable health professionals to compatibly decide on the therapeutic treatments and approaches for osteoporosis treatment and pharmacologic management of their patients. We were the first to develop and test Machine Learning Models for Predicting Therapeutic Adherence treatments. The ML Model accuracy results are summarized as classical metrics where the ExtraTree Model exhibited the highest accuracy of 100%, 85.0%, 94.5% on the training, testing and overall dataset respectively using Synthetic Minority Over-sampling Technique Support Vector Machine Learning (SMOTE-SVM). Keywords— Artificial Intelligence, Predictive Models, Os- teoporosis Management, SMOTE, SVM, Machine Learning, Pregnancy, Pharmacologic Management, Therapeutic Ad- herence. I. I NTRODUCTION A. Background and Motivation Osteoporosis is one of the huge socioeconomic burdens revealed by demographic patterns with a liability to meeting the financial and social needs yet it is increasing with the growing number of elders [14, 4] who are greatly constraining the Health care systems. This calls for an urgent need to strategically and sustainably control the health care costs of the elderly. Moreover 1 in 3 women over the age of 50 years and 1 in 5 men will experience osteoporotic fractures in their lifetime according to the International Osteoporosis Foundation statistics [19, 7]. Adherence to neridronate therapy in pregnant women due to pregnancy osteoporosis during postpartum or last trimester due to vertebra fractures is also obscure [9] . Clinicians offer pharmacologic and therapeutic treatments to reduce osteoporosis although the adherence to treatment is not satisfactory [21, 3]. The perceptions and experiences of patients during treatment greatly affects their adherence to osteoporosis therapy [16] yet they are extremely unpredictable hence a need for interdisciplinary collaboration to improve long term treatment approaches [18, 6]. This is what motivated us to use Machine Learning Models to predict the adherence of patients in order to develop strategies to improve adherence to medications individually. B. Our Contributions. In this paper, we developed Machine Learning models for predicting therapeutic adherence of patients to osteoporosis treatment and tested them on a real dataset for Drug Per- sistence [12] with 69 features and about 3414 samples. We optimized and tested the accuracy of different ML models and classified the accuracy metrics of the results depending on the training, testing or overall dataset where the ExtraTree Model showed the finest accuracy of 100%, 85.0% and 94.5% with respect to the datasets. The outcomes of the tests prove that the implementation of Machine Learning Predictive Models that use the ExtraTree Classification algorithms with SMOTE- SVM enable health professionals to compatibly decide on the individualized therapeutic treatments and approaches for osteoporosis treatment and pharmacologic management of their patients. The summary of the contributions IS stated below. 1) We proposed and developed Machine learning Mod- els for predicting therapeutic adherence to osteoporosis treatment for physicians and researchers to develop suitable adherence-improving interventions. 2) We optimized the Models with various sampling tech- niques for the imbalanced data on osteoporosis. 3) We evaluated the performance and accuracy of the models with both synthesized and real datasets for Drug Persistence classification. 4) Finally, we recommended the most accurate Machine Learning Models for adoption and deployment for re- searchers, physicians and investors in the therapeutic adherence domain. The rest of the paper is organized as follows: Literature review in section II, Problem and Solution Formulation, Data preparation, Data Processing and Training and Sampling in section III, Section IV represents simulated results and per- formance evaluation of the Models. Finally Conclusion and Recommendations for future work are in section V. II. EXISTING WORKS 2021 IEEE Asia-Pacific Conference on Computer Science and Data Engineering (CSDE) | 978-1-6654-9552-3/21/$31.00 ©2021 IEEE | DOI: 10.1109/CSDE53843.2021.9718416 Authorized licensed use limited to: GGALIWANGO MARVIN. Downloaded on March 03,2022 at 00:45:58 UTC from IEEE Xplore. Restrictions apply.