Precision treatment of chalazion based on AI-driven risk stratification: a prospective study of triamcinolone acetonide combined with intense pulsed light therapy.

PURPOSE: To evaluate the rates of chalazion recurrence following intralesional triamcinolone acetonide (TA), intense pulsed light (IPL), or combined TA + IPL therapy and to validate an AI-enhanced risk stratification strategy.

METHODS

This prospective, observational study enrolled 366 patients. To eliminate immortal time bias and circular reasoning, LASSO-Cox regression with 10-fold cross-validation was employed to identify purely baseline predictors: age, Demodex infection, chronicity, and meibomian gland dysfunction (MGD). These baseline predictors were integrated into a clinical Cox proportional hazards model and a survival-based XGBoost model, which were subsequently merged into a hybrid framework. Performance was assessed via time-dependent area under the curve (AUC), net reclassification improvement (NRI), Kaplan-Meier analysis, and decision curve analysis (DCA).

RESULTS

The 3-month complete resolution rate was significantly higher in the combined TA + IPL group (95.2%) compared to the TA-only (90.8%) and IPL-only (79.4%) groups (p < 0.001). During follow-up, 60 recurrence events (16.4%) occurred. Recurrence rates differed significantly (χ2 = 31.73, p < 0.001) as follows: IPL group (38.2%), TA group (15.4%), and TA + IPL group (8.3%). Multivariate Cox analysis revealed that demodex infection, age, chronicity, and MGD were independent risk factors for chalazion recurrence. The hybrid model achieved an AUC of 0.822, outperforming the clinical (0.745) and XGBoost (0.821) (p < 0.05) models. AI integration significantly refined the risk estimates (NRI = 0.365, p = < 0.001). In the high-risk subgroup, recurrence rates reached 83.3-91.7% with monotherapy, whereas combined TA + IPL therapy significantly reduced the recurrence rate to 60.0%. DCA confirmed that the hybrid model consistently yielded the greatest net clinical benefit.

CONCLUSION

A hybrid AI-clinical model significantly improved the prediction of chalazion recurrence. Combined TA + IPL therapy demonstrated superior efficacy, particularly for high-risk patients. This framework provides a robust decision-support tool for precision management and the systematic reduction of chalazion recurrence.