The evaluation of machine-based learning models for predicting acute postoperative pain

Use of machine-based learning models for prediction in medicine may hold promise. Yet, predicting postoperative pain following surgery using this approach remain biased.
Data from this study (n=14,263 patients) showed that the CatBoost machine-based model demonstrated bias towards age, race, area deprivation index (ADI), and insurance type. In contrast, the model demonstrated fairness in terms of sex, language, and health literacy.
Although overall performance in predicting acute postoperative pain was favorable, bias remained for certain attributes. According to the authors, further assessment to warrant the fairness of machine-based learning tools is necessary before advancing to the clinical implementation stage.