Prediction of sustained opioid use in children and adolescents using machine learning

In recent years, opioid use for chronic pain has surged, alongside related mortality and morbidity.1 Although opioids remain essential for anaesthesia and analgesia, their role in chronic noncancer pain has become increasingly controversial owing to concerns about addiction and limited long-term effectiveness.1, 2, 3 This prompted the US Centers for Disease Control and Prevention to declare opioid overdose prevention as one of the five main public health challenges in the USA in 2014.4

The opioid epidemic is not exclusive to adults. Mounting evidence also points to increasing opioid use in children and adolescents,5 with estimates of prevalence rates ranging from 1% to 15%, peaking among those aged 13–17 yr.6, 7, 8 Importantly, opioid use during brain development can cause long-term effects, including an increased risk of addiction to opioids and other substances.5,6 Despite its potentially hazardous effects, opioid use in children and adolescents is understudied.

Similar opioid use trends have been observed in Israel, in adults and children alike. Studies from the past decade in Israel have demonstrated that opioid use among adults almost doubled between 2009 and 2016, with oxycodone and fentanyl showing the greatest increase.9 This increase has also extended to children and adolescents, especially among potent opioids (oxycodone, fentanyl, morphine, and buprenorphine), in non-central areas, and for noncancer patients.10 Although Israel's opioid-related morbidity and mortality rates remain lower than other Western countries,11 the associated economic burden is still substantial. Opioid users in Israel were found to incur healthcare costs three to nine times higher than non-users, straining resources and highlighting the need for effective prevention and treatment strategies.12

Despite prevention efforts, opioid-related mortality rates continue to increase, partially because of initial overemphasis of prevention efforts on non-medical users.1 Therefore, prevention efforts have shifted to focus on medical opioid users, and on the physicians treating them, as physician over-prescription and misuse of opioids is considered one of the main reasons for the increasing prevalence of opioid use.1,13 Opioid misuse prevention includes primary prevention, involving educating doctors on opioid efficacy and risks; secondary prevention, focusing on early detection of opioid-addicted patients and referral to rehabilitation programmes; and tertiary prevention, centred on harm reduction.1

Opioid misuse is difficult to identify solely from electronic medical record (EMR) data, as its definition relies on motivation and intentions. Thus, sustained opioid use is often targeted as an alternative outcome. Machine learning (ML) algorithms are increasingly being used in medical studies and practice to predict outcomes of interest and provide decision support for physicians, to assist in secondary prevention. Several studies have shown the ability of ML algorithms to predict sustained opioid use, addiction, and overdose among various populations.14, 15, 16 However, to the best of our knowledge, there are no studies of sustained opioid use in the paediatric outpatient population, using solely EMR data.

In this study, we developed a ML classifier to differentiate occasional opioid users from sustained users, among outpatient patients at the ages of 0–19, using EMR data available at or near the time of first prescription fulfilment. Our classifier is also available as an online tool containing visualisations of the model's prediction explanation, to help physicians understand the predicted personalised risk of patients for sustained opioid use.

Read more