Applying Machine Learning Models to Predict Medication Nonadherence in Crohn’s Disease Maintenance Therapy

• Main Authors: Wang L, Fan R, Zhang C, Hong L, Zhang T, Chen Y, Liu K, Wang Z, Zhong J
• Format: Article
• Language: English
• Published: Dove Medical Press 2020-06-01
• Series: Patient Preference and Adherence
• Subjects: crohn’s disease; azathioprine; medication adherence; maintenance therapy; machine learning; support vector machine; back-propagation neural network.
• Online Access: https://www.dovepress.com/applying-machine-learning-models-to-predict-medication-nonadherence-in-peer-reviewed-article-PPA

Lei Wang,1,* Rong Fan,1,* Chen Zhang,1 Liwen Hong,1 Tianyu Zhang,1 Ying Chen,2 Kai Liu,2 Zhengting Wang,1 Jie Zhong1 1Department of Gastroenterology, Ruijin Hospital, Shanghai Jiaotong University School of Medicine, Shanghai, People’s Republic of China; 2CareLinker Co., Ltd., Shanghai, People’s Republic of China*These authors contributed equally to this workCorrespondence: Zhengting Wang; Jie ZhongDepartment of Gastroenterology, Ruijin Hospital, Shanghai Jiaotong University School of Medicine, 197 Ruijiner Road, Shanghai 200025, People’s Republic of ChinaTel [phone] ext. 600901Email [e-mail]; [e-mail]: Medication adherence is crucial in the management of Crohn’s disease (CD), and yet the adherence remains low. This study aimed to develop machine learning models that can help predict CD patients of nonadherence to azathioprine (AZA), and thus assist caregivers to streamline the intervention process.Methods: This single-centered, cross-sectional study recruited 446 CD patients who have been prescribed AZA between Sep 2005 and Sep 2018. Questionnaires of medication adherence, anxiety and depression, beliefs of medication necessity and concerns, and medication knowledge were provided to patients, while other data were extracted from the electronic medical records. Two machine learning models of back-propagation neural network (BPNN) and support vector machine (SVM) were developed and compared with logistic regression (LR), and assessed by accuracy, recall, precision, F1 score and the area under the receiver operating characteristic curve (AUC).Results: The average classification accuracy and AUC of the three models were 81.6% and 0.896 for LR, 85.9% and 0.912 for BPNN, and 87.7% and 0.930 for SVM, respectively. Multivariate analysis identified four risk factors associated with AZA nonadherence: medication concern belief (OR=3.130, p< 0.001), education (OR=2.199, p< 0.001), anxiety (OR=1.549, p< 0.001) and depression (OR=1.190, p< 0.001), while medication necessity belief (OR=0.004, p< 0.001) and medication knowledge (OR=0.805, p=0.013) were protective factors.Conclusion: We developed three machine learning models and proposed an SVM model with promising accuracy in the prediction of AZA nonadherence in Chinese CD patients. The study also reconfirmed that education, psychologic distress, and medication beliefs and knowledge are correlated to AZA nonadherence.Keywords: Crohn’s disease, azathioprine, medication adherence, maintenance therapy, machine learning, support vector machine, back-propagation neural network