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

Abstract

Existing predictive models of opioid use disorder (OUD) may change as the rate of opioid prescribing decreases. Using Veterans Administration’s EHR data, we developed machine-learning predictive models of new OUD diagnoses and ranked the importance of patient features based on their ability to predict a new OUD diagnosis in 2000–2012 and 2013–2021. Using patient characteristics, the three separate machine learning techniques were comparable in predicting OUD, achieving an accuracy of >80%. Using the random forest classifier, opioid prescription features such as early refills and length of prescription consistently ranked among the top five factors that predict new OUD. Younger age was positively associated with new OUD, and older age inversely associated with new OUD. Age stratification revealed prior substance abuse and alcohol dependency as more impactful in predicting OUD for younger patients. There was no significant difference in the set of factors associated with new OUD in 2000–2012 compared to 2013–2021. Characteristics of opioid prescriptions are the most impactful variables that predict new OUD both before and after the peak in opioid prescribing rates. Predictive models should be tailored to age groups. Further research is warranted to determine if machine learning models perform better when tailored to other patient subgroups.

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 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 . Other modeling approaches included Cox proportional hazards 24,35,39 and Poisson modeling 30 …”

“…Other measures of model performance included screening metrics such as positive predictive value (PPV), sensitivity and specificity, 21,23,31,39,46 the Akaike information criterion, 9 and Euclidean loss 33 . Gradient boosting models 19,45-48,53,55 and neural network models 19,26,28,29,34,36,44,47,48 tended to perform well in the context of their own study—often outperforming other models made with the same dataset—with high c statistics up to 0.959, to predict outcomes of persistent opioid use, 53 OUD, 19,26,28,29,34,36,44,47,55 and opioid overdose 45-48 . The Cox proportional hazards model by Glanz et al, 24 the Bayesian belief network by Anderson et al, 53 and the decision tree by Gao et al 44 performed poorly in the context of their own study, all with c statistics of 0.73, to predict outcomes of opioid overdose, persistent opioid use, and OUD, respectively.…”

“…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 …”

Socioeconomic risk factors for long‐term opioid use: A national registry‐linkage study

“…According to the Stanford-Lancet Commission, opioids show little to no effectiveness in the treatment of chronic pain and may instead worsen pain perception and quality of life overall (Humphreys et al, 2022). Long-term opioid use is also associated with increased risk of opioid use disorder (Banks et al, 2023;Edlund et al, 2014). Excess prescribing of strong opioids such as oxycodone has been identified as one of the main causes of the ongoing North American opioid epidemic, and countries such as Brazil, Spain and the Netherlands show signs of a spreading global opioid epidemic (Bedene et al, 2019;Humphreys et al, 2022;Serra-Pujadas et al, 2021).…”