Machine learning models demonstrated comparable discriminative performance to traditional logistic regression for predicting DR screening adherence, while offering distinct profiles in calibration and clinical net benefit, with random forest achieving the highest AUC (0.774) and net benefit across most threshold probabilities.
The random forest model achieved the highest AUC but showed no statistically significant difference from logistic regression or decision tree models.
The decision tree model exhibited the best calibration among the three models.
Decision curve analysis indicated that the random forest model provided the highest net benefit across most threshold probabilities.
History of ocular disease was identified as the strongest positive predictor of DR screening adherence.
Higher HbA1c, lower self-efficacy, higher perceived severity, absence of exercise therapy, and smoking were associated with poorer DR screening adherence.
DR screening adherence was suboptimal in the study sample, with the majority of participants classified as poor adherers.
Integration of PMT-based psychological constructs with clinical and behavioral factors provided a multidimensional framework for predicting DR screening behavior.
This research suggests that among older adults with diabetes living in the community, predicting who will regularly attend diabetic retinopathy (eye) screening is possible using both traditional statistical methods and newer machine learning approaches—and that neither type of model is clearly superior. The study recruited over 1,000 adults from community health centers in China and found that only about 16% had good screening adherence. When researchers compared a traditional logistic regression model against two machine learning models (decision tree and random forest), all three performed similarly in their ability to distinguish good from poor adherers, with no statistically significant differences in accuracy. However, the random forest model appeared most useful in clinical decision-making scenarios, while the decision tree had the best calibration (its predicted probabilities matched real-world outcomes most closely). The study also identified key factors linked to poorer screening attendance: having higher blood sugar levels (HbA1c), not exercising, smoking, lower confidence in managing one's health (self-efficacy), and paradoxically, a stronger belief that diabetic retinopathy is a serious condition (perceived severity). The single strongest predictor of better adherence was having a prior history of eye disease, which tripled the likelihood of attending screening. Notably, this study went beyond typical clinical or demographic factors by including psychological measures drawn from Protection Motivation Theory, such as self-efficacy and perceived severity, highlighting that mental and motivational factors play a meaningful role in whether patients follow through with screening. This research suggests that combining psychological, behavioral, and clinical information together—rather than relying on any single type of factor—offers a more complete picture of who is at risk for missing eye screenings. This could help healthcare providers in community settings identify patients who need additional support or targeted outreach to improve early detection of diabetic retinopathy, which is a leading but preventable cause of blindness.
Wang Y, Xia L, Geng G, Ge C, Qian X, Shen H. (2026). Machine learning versus traditional regression models for predicting diabetic retinopathy screening adherence among community-dwelling older adults with diabetes.. Frontiers in endocrinology. https://doi.org/10.3389/fendo.2026.1909594