Ai-Based Clinical Stratification Using Tear Film Biomarkers In Ded To Identify Sub-Clinical Disease

• Dry eye disease (DED) poses a substantial burden with diverse clinical presentations, necessitating precise patient classification for optimal treatment and monitoring. Tear biomarkers associated with DED prompted our use of an AI-based classifier strategy to identify individuals at risk of DED or associated ocular surface inflammation

Tertiary Eye care Hospital, India 

We measured eight biomarkers in tears from 640 subjects using multiplex ELISA and classified them into control and DED groups based on clinical parameters. An AI model employing a decision tree classifier (DTC) determined the biomarker levels, with MMP-9 emerging as the primary discriminator. The cohort was subdivided into four groups based on MMP-9 ranges: Controls, Subclinical Inflammation-1, Subclinical Inflammation-2, and DED. A Random Forest (RF) AI model was applied to classify the cohort, both with and without clinical parameters.

The DTC AI model exhibited an AUC of 0.76, correctly predicting 68% of Group-1 and 71% of Group-2 eyes. The RF AI model (MMP-9-based grouping) achieved an AUC of 0.79, with 94.8% sensitivity and 88.6% specificity. Incorporating clinical parameters slightly improved accuracy to 0.81, enhancing specificity towards DED eyes to 78%.

Our AI model, using MMP-9 as a key discriminator, identified controls and subclinical cases at risk of DED. Biomarker-based subgrouping correlated well with clinical parameters, providing valuable insights for subject stratification in a clinical setting. This study highlights the efficacy of AI in classifying DED patients based on tear biomarkers, especially in identifying subclinical cases not apparent through traditional assessments.