Predictive Models for Identifying Adult Patients at High Risk of Developing Opioid-Related Harms: a Systematic Review

Opioids are the most frequently prescribed medications for managing moderate-to-severe pain and are associated with significant potential for harm. Several models have been developed to predict opioid-related harms (ORHs). This study aimed to describe and evaluate the methodological quality of predictive models for identifying patients at high risk of ORHs.

Using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline, we reviewed published studies on developing or validating models for predicting ORHs, identified through a literature search of Scopus, PubMed, Embase, and Google Scholar. The quality of studies was assessed using the Prediction model Risk Of Bias ASsessment Tool (PROBAST). The models were assessed by area under the curve (AUC) or c-statistic, sensitivity, specificity, accuracy, and positive or negative predictive value. The study protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO; CRD42024540456).

We included 36 studies involving participants aged 18 years or older. The frequently modeled ORHs were opioid use disorder (12 studies), opioid overdose (8 studies), opioid-induced respiratory depression (6 studies), and adverse drug events (4 studies). In total, 16 studies (44.4%) developed and validated tools. Most studies measured predictive ability using AUC (31, 86.1%), and some only reported sensitivity (14, 38.9%), specificity (11, 30.6%), or accuracy (4, 11.1%). Of the 31 studies that reported AUC values, 29 (93.5%) had moderate-to-high predictive ability (AUC > 0.70). History of opioid use (66.7%), age (58.3%), comorbidities (41.7%), sex (41.7%), and drug abuse and psychiatric problems (36.1%) were typical factors used in developing models.

The included predictive models showed moderate-to-high discriminative ability for screening patients at risk of ORHs. However, future studies should refine and validate them in various settings before considering the translation into clinical practice.