Influence of baseline characteristics on subjective improvement of dry eye after intense pulsed light therapy

Abstract

Purpose: To identify baseline clinical signs and symptoms associated with response to intense pulsed light (IPL) combined with meibomian gland expression in dry eye disease (DED), and to develop machine learning (ML) models for individualized outcome prediction. Methods: This retrospective study analyzed 100 eyes from 100 DED patients (aged 58.6 ± 13.4 years) treated with IPL and meibomian gland expression. Baseline parameters assessed with the Antares system included meibomian gland loss (MGL), tear meniscus height (TMH), non-invasive tear break-up time (NIBUT), conjunc tival hyperemia, and Ocular Surface Disease Index (OSDI). Patients were stratified by change in OSDI after treatment (ΔOSDI): Class 1 (no improvement), Class 2 (mild improvement), and Class 3 (clear improvement). Several ML models were trained to predict ΔOSDI from baseline parameters. Results: IPL significantly improved both symptoms and signs. OSDI decreased from 44.65 ± 18.3 to 28.47 ± 19.3 (p < 0.001), NIBUT increased from 4.5 ± 3.2 to 7.5 ± 6.5 s (p < 0.001), and TMH and conjunctival hyperemia also improved (p < 0.001), while MGL and BCVA remained stable. Greater improvement was observed in pa tients with higher baseline OSDI (p = 0.001). The XGBoost algorithm achieved the highest predictive perfor mance (AUC-ROC = 0.77), with OSDI and NIBUT as the strongest predictors based on SHAP analysis. Conclusions: IPL combined with meibomian gland expression improves symptoms and signs in DED, particularly in patients with more severe baseline symptoms. Baseline OSDI and NIBUT were the strongest predictors of response. ML models demonstrated moderate accuracy, supporting their potential role in personalized DED treatment strategies.