Novel machine learning model – The Opioid Risk Tool for Opioid Use Disorder – may be used as a preliminary screening tool for opioid use disorder (OUD) among patients with chronic noncancer pain receiving long-term opioid therapy, according to study results published in Pain Medicine. In addition, employing additional screening tools, such as clinical predictors, can also be valuable in determining patient risk.
Patients with chronic noncancer pain are at an increased risk of developing OUD, which underscores the need for accurate and comprehensive assessment of OUD risk in this population. However, tools for screening this risk are currently limited.
Researchers evaluated the benefits of machine learning models in predicting risk for OUD in patients with chronic noncancer pain.
Participants with chronic noncancer pain were enrolled in the study, and data were compared to analyze the performance of 2 sets of machine learning models. The first model used the Opioid Risk Tool for Opioid Use Disorder alone and the second set incorporated 17 additional clinical predictors.
[I]ntegrating predictive models into clinical practice has the potential to support proactive, tailored strategies for OUD prevention and intervention, improving outcomes for patients with [chronic noncancer pain] on [long-term opioid therapy].
Participants completed self-report forms or responded to interviews that were conducted by a research coordinator or assistant in a private setting.
A total of 1300 participants (59.68% women; mean age, 49.03 years) were included in the analysis. Financial strain was noted to be common, with one-third of the cohort reporting “not able to make ends meet.” Rates of depression and anxiety were 29.3% and 7.23%, respectively.
Researchers observed that while both models were strong predictors of OUD, the Opioid Risk Tool for Opioid Use Disorder demonstrated stronger performance (precision, 0.91; [95% CI, 0.90-0.92]; specificity, 0.96 [95% CI, 0.96-0.97]) compared with the random forest model (precision, 0.90 [95% CI, 0.89-0.91]; specificity, 0.96 [95% CI, 0.96-0.96]) and eXtreme Gradient Boosting model (precision, 0.86 [95% CI, 0.86-0.87]); specificity, 0.94 [95% CI, 0.93-0.94]). The random forest and eXtreme Gradient Boosting models showed improved performance on precision-recall area under the curve and F1 scores.
Notably, the researchers also observed that the features that were the most predictive of OUD in the expanded models were nicotine dependence, marital status, opioid misuse behaviors, and pain interference and catastrophizing.
Limitations of this study include a lack of generalizability because the sample was a homogenous population that included individuals who self-identified as White. Data for the study were collected between 2012 and 2018 and prescribing practices and guidelines have since changed. Moreover, there is potential for underreporting due to the model’s features being based on subjective measures. Further, the study used Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) vs DSM, Fifth Edition (DSM-5) OUD criteria.
“Ultimately, integrating predictive models into clinical practice has the potential to support proactive, tailored strategies for OUD prevention and intervention, improving outcomes for patients with [chronic noncancer pain] on [long-term opioid therapy],” the study authors concluded.
This article originally appeared on Clinical Pain Advisor
Leave a comment