Predicting the 10-year risk of cataract surgery using machine learning techniques on questionnaire data: findings from the 45 and Up Study.

Abstract

To investigate the feasibility and accuracy of using machine learning (ML) techniques on self-reported questionnaire data to predict the 10-year risk of cataract surgery, and to identify meaningful predictors of cataract surgery in middle-aged and older Australians. Baseline information regarding demographic, socioeconomic, medical history and family history, lifestyle, dietary and self-rated health status were collected as risk factors. Cataract surgery events were confirmed by the Medicare Benefits Schedule Claims dataset. Three ML algorithms (random forests [RF], gradient boosting machine and deep learning) and one traditional regression algorithm (logistic model) were compared on the accuracy of their predictions for the risk of cataract surgery. The performance was assessed using 10-fold cross-validation. The main outcome measures were areas under the receiver operating characteristic curves (AUCs). In total, 207 573 participants, aged 45 years and above without a history of cataract surgery at baseline, were recruited from the 45 and Up Study. The performance of gradient boosting machine (AUC 0.790, 95% CI 0.785 to 0.795), RF (AUC 0.785, 95% CI 0.780 to 0.790) and deep learning (AUC 0.781, 95% CI 0.775 to 61 0.786) were robust and outperformed the traditional logistic regression method (AUC 0.767, 95% CI 0.762 to 0.773, all p<0.05). Age, self-rated eye vision and health insurance were consistently identified as important predictors in all models. The study demonstrated that ML modelling was able to reasonably accurately predict the 10-year risk of cataract surgery based on questionnaire data alone and was marginally superior to the conventional logistic model.