Development of a model for the clinical prediction of contact lens discomfort

To evaluate which ocular and contact lens (CL)-related dimensions, and their interactions, contribute most to CL discomfort (CLD), and to determine a set of clinical parameters for CLD assessment.

Cross-sectional study of consecutive CL wearers was performed. Variables were grouped into six dimensions (personal factors, CL characteristics, symptoms, cornea and conjunctiva, tear film, and lid margin and Meibomian glands). Within each dimension composite indices (0-100) were generated with a principal component analysis (mixed data). A structural equation model (SEM) was used to define the latent construct "CLD", followed by unsupervised hierarchical agglomerative clustering. Cluster membership was subsequently predicted using a Least Absolute Shrinkage and Selection Operator (LASSO)-regularized logistic regression model with an 80/20 train-validation split. A simplified model intended for clinical use was developed.

One hundred and fifty CL wearers (97 female; 34.4 ± 12.6 years) were included. SEM indicated a multidimensional structure in which the CL and symptom dimensions contributed most to latent CLD. Clustering yielded two distinct phenotypes that differed (p ≤ 0.043) in age, CL type and replacement schedule, daily wearing time, symptom severity, and several ocular surface parameters. A simplified predictor model (daily wearing time, CL replacement, Contact Lens Dry Eye Questionnaire-8, NIBUT, and tarsal hyperemia) predicted membership in the CLD phenotype with 85% accuracy and area under the curve of 0.956 in the validation set.

Modeling CLD as a latent, multidimensional construct integrating subjective and objective measures allows identification of clinically meaningful phenotypes and supports the development of an easy-to-use classifier with excellent discrimination.