Individualized Prospective Prediction of Opioid Use Disorder

Abstract

Objective Opioid use disorder (OUD) is a chronic relapsing disorder with a problematic pattern of opioid use, affecting nearly 27 million people worldwide. Machine learning (ML)-based prediction of OUD may lead to early detection and intervention. However, most ML prediction studies were not based on representative data sources and prospective validations, limiting their potential to predict future new cases. In the current study, we aimed to develop and prospectively validate an ML model that could predict individual OUD cases based on representative large-scale health data. Method We present an ensemble machine-learning model trained on a cross-linked Canadian administrative health data set from 2014 to 2018 ( n = 699,164), with validation of model-predicted OUD cases on a hold-out sample from 2014 to 2018 ( n = 174,791) and prospective prediction of OUD cases on a non-overlapping sample from 2019 ( n = 316,039). We used administrative records of OUD diagnosis for each subject based on International Classification of Diseases (ICD) codes. Results With 6409 OUD cases in 2019 (mean [SD], 45.34 [14.28], 3400 males), our model prospectively predicted OUD cases at a high accuracy (balanced accuracy, 86%, sensitivity, 93%; specificity 79%). In accord with prior findings, the top risk factors for OUD in this model were opioid use indicators and a history of other substance use disorders. Conclusion Our study presents an individualized prospective prediction of OUD cases by applying ML to large administrative health datasets. Such prospective predictions based on ML would be essential for potential future clinical applications in the early detection of OUD.

Predictive Models to Assess Risk of Persistent Opioid Use, Opioid Use Disorder, and Overdose

“…Primary outcomes included opioid overdose (17% of studies), 18,32,35,38,39,46,48 fatal opioid overdose (14.6%), 24,32,35,39,50,51 OUD (41.4%), 9,16,17,19,21-23,26-28,31,33,34,41,44,47,55 and persistent opioid use (17%) 25,29,30,42,43,53,54 . OUD and overdose were typically defined using ICD-9 or ICD-10 codes, although one study used DSM-5 to infer diagnosis 55 .…”

“…The 41 included studies developed more than 160 models to predict risk of persistent opioid use, OUD, or opioid overdose. The most common modeling approach was regression modeling, 9,17-20,22,23,25-29,31,32,34,36,38,40-42,44,46,48-53 including logistic regression with LASSO regularization 17,23,25,48 and stepwise logistic regression 27,52 . Machine learning approaches were also commonly used; these included random forest, 18,19,29,34,36-38,44,46-48,53,54 neural network, 19,26,28,29,34,36,44,47,48 gradient boosting machine, 19,45-48,53,55 support vector machine, 29,33,38,44,54 elastic net, 18,47 decision tree, 36,44 Bayesian belief network, 53 ADA Boost, 38,54 XGBoost, 19,38,54 and natural language processing 21,41 .…”

“…External validation uses data that were not used for model development to evaluate the model 15 . Most models were validated using internal validation, 9,17-19,21-23,25,26,28,29,31,32,34-41,45,47-49,51,53-55 with the most common technique being k-fold cross-validation 17,19,23,25,29,35,37,38,40,53,55 . Sixteen studies 17,20,24,27-29,31,37,40,43-46,48,50,51 used external validation, and 12 studies did not specify usage of validation 9,18,20,24,26,27,39,44-48 …”

“…20 Seven studies used data from insurance or managed care databases, including Healthcare Maintenance Organization data, 21 Group Health Cooperative data, 22 Kaiser Permanente, 23,24 the Veterans Affairs Informatics and Computing Infrastructure, 25,26 and Humana Research Database. 27 Five studies [28][29][30][31][32] used regional research databases, and 1 study 33 used an online database. Eleven studies 21,[34][35][36][37][38][39][40][41][42][43] drew data from an EHR.…”

Predicting opioid use disorder before and after the opioid prescribing peak in the United States: A machine learning tool using electronic healthcare records

“…A recent study utilized the Canadian administrative health records billing data to achieve a balanced accuracy score of 86%, and found opioid-related poisoning, sedative hypnotic-related disorders, and polysubstance-related disorders to be predictive of OUD. 12 Their billing data did not include prescription/refill variables. EHR data contains more patient level variables (e.g., pain scores, laboratory results), and may out-perform models developed using medical claims data.…”