Researcher discusses how machine studying helps establish sufferers in danger ranges for opioid use dysfunction

Ronen Rozenblum, Ph.D., MPH, director of the Unit for Modern Healthcare Apply & Expertise and director of Enterprise Growth of the Heart for Affected person Security Analysis and Apply at Brigham and Ladies’s Hospital, and an assistant professor at Harvard Medical College, is the principal investigator and senior creator of a brand new research revealed in JMIR Medical Informatics, “A Machine Studying Software to Classify Sufferers at Differing Ranges of Threat of Opioid Use Dysfunction: Clinician Based mostly Validation Research.”

On this article, Dr. Rozenblum discusses this analysis.

How would you summarize your research for a lay viewers?

Our research centered on utilizing superior machine studying (ML) to assist clinicians extra precisely establish sufferers prone to growing opioid use dysfunction (OUD). Regardless of strict tips for managing opioids, OUD stays a severe public well being difficulty.

We evaluated an ML software known as MedAware, which alerts clinicians to sufferers who could also be at larger threat of OUD by analyzing affected person data. Our findings confirmed that ML can present clinicians with dependable alerts a few affected person’s stage of threat. This sort of expertise has the potential to considerably improve how physicians and different clinicians assess and deal with OUD, with the objective of offering extra correct, safer, and personalised look after sufferers early of their opioid remedy.

What query have been you investigating?

This scientific validation research investigated how effectively the ML system in comparison with clinicians’ evaluation of a affected person’s threat of OUD. We examined the settlement between the ML software and clinicians’ structured overview of medical data to categorise sufferers receiving opioid remedy into three distinct classes of OUD threat (i.e., not excessive threat, excessive threat, or suspected OUD). We additionally evaluated the explanations for discrepancies between clinicians’ judgments and ML threat evaluation.

What strategies or method did you employ?

Outpatient knowledge of 649,504 Mass Basic Brigham sufferers and a random pattern of 180 sufferers have been used to develop the ML mannequin and the validation research, respectively. We developed an OUD threat classification scheme and knowledge extraction device to validate these alerts. Clinicians independently carried out a scientific and structured overview of medical data and reached a consensus on every affected person’s OUD threat stage, which was then in comparison with the ML software’s threat assignments.

What did you discover?

Our findings revealed that the ML software efficiently categorized sufferers into differing ranges of OUD threat and demonstrated substantial settlement with clinicians’ overview of medical data. The best settlement between the 2 strategies was noticed for sufferers categorized as excessive threat for OUD and suspected OUD. Thus, the outcomes of this research display that this ML software can generate clinically legitimate and helpful alerts for screening sufferers prone to OUD. Moreover, we recognized key themes explaining disagreements between the ML software and clinician opinions.

What are the implications and subsequent steps?

The importance of those findings lies in the truth that solely a restricted variety of research have examined the scientific validity and utility of ML purposes in distinguishing between varied ranges of OUD threat in sufferers.

These outcomes counsel that ML purposes, comparable to MedAware, can considerably improve clinicians’ potential to evaluate sufferers’ threat for OUD early in opioid remedy, selling extra personalised and safer care. This functionality is anticipated to enrich conventional rule-based approaches in alerting physicians and different clinicians about opioid issues of safety.