JMIR Publications

ABSTRACT

Background:

Medical disputes are a global public health issue receiving increasing attention. However, studies investigating the relationship between hospital legal construction and medical disputes are scarce. A multicenter model incorporating machine learning (ML) techniques to provide individualized prediction of medical disputes would be beneficial for medical workers.

Objective:

To screen potential risk factors related to medical disputes from the perspective of hospital legal construction, and further utilize ML techniques to build models for predicting the risk of medical disputes.

Methods:

This study enrolled 38053 medical workers from 130 tertiary hospitals in Hunan province, China. All enrolled healthcare workers were randomly split into a training cohort (n=34286, 90%) and an internal validation cohort (n=3767, 10%). Medical workers from 87 tertiary hospitals in Beijing were served as an external validation cohort (n=26285). Six machine learning (ML) models, including the logistic regression (LR), decision tree (DT), random forest (RF), support vector machine (SVM), gradient boosting decision tree (GBDT), and deep neural network (DNN), were introduced in the study. Twelve metrics were employed for performance evaluation, including discrimination and calibration. In addition, a scoring system was developed to select the optimal model. The Shapley Additive exPlanations (SHAP) was adopted to generate the importance coefficients for characteristics. To promote the clinical practice of our proposed optimal model, reclassification of patients was operated, and an online application for medical dispute prediction was conducted and can be easy-to-use for the public.

Results:

Medical disputes occurred in 46.06% (17527/38053) of medical workers in Hunan province, China. Among all the 26 clinical characteristics, multivariate analysis demonstrated that 18 characteristics were significantly associated with medical disputes, and those characteristics were included for ML model development. Among the ML techniques, gradient boosting decision tree (GBDT) was regarded as the optimal model, which occupied the lowest Brier score (0.205), the highest area under the receiver operating characteristic curve (AUROC) (0.738, 95% confident interval [CI]: 0.722-0.754), and the biggest discrimination slope (0.172) and Youden index (1.355), in comparison to other ML models. In addition, it embraced the highest metrics score (63 points), followed by deep neural network (46 points) and random forest (45 points), in the internal validation set. In the external validation set, the performance of GBDT is still comparable with the second biggest metrics score (52 points). The high-risk group was above twice odds of experiencing medical dispute than the low-risk one. The availability of our optimal model was embedded into an online application and it could be accessed at https://yiminer-medical-disputes-main-onczu6.streamlitapp.com/.

Conclusions:

This study establishes a prediction model to stratify medical workers at different risks of encountering medical disputes. Comparing with five ML models, the GBDT has achieved the optimal comprehensive performance which is used for online application construction. Our proposed model can be a useful web tool to identify medical workers with a high risk of medical disputes, and preventive strategies should be more emphasized in the high-risk group.