Enhancing Precision in a Non-Inferiority Trial: The Application of Hierarchical Generalized Linear Mixed Models for Correlated Ophthalmic Data

OBJECTIVES: Non-inferiority studies are critical for demonstrating that new treatments are at least as effective as established standards within a pre-specified margin. In glaucoma management, where control of intraocular pressure (IOP) is essential, accurate quantifying of treatment effects is imperative. Hierarchical Generalized Linear Mixed Models (GLMMs) play a pivotal role in these studies, effectively handling complex and correlated data structures like repeated measurements from both eyes of the same subject. This methodological approach is crucial for maintaining statistical integrity by accounting for correlations within the data.

METHODS: The study, now completed, involved 267 participants, assessing the non-inferiority of home-based self-tonometry against traditional inpatient IOP monitoring using a hierarchical GLMM. This final regression model, chosen from a pool of 50 different tested models, incorporated fixed effects and random effects, with the best model achieving an Akaike Information Criterion (AIC) of 252.5 (vs. our basic model with an AIC = 599.49), indicating a robust fit. Model refinement was guided by Residual Plots, R², Likelihood-Ratio Tests and AIC.

RESULTS: The hierarchical GLMM provided a robust framework for analysis, confirming the statistical non-inferiority (p ≤ 0.001) of self-tonometry in detecting critical IOP fluctuations when compared to the traditional method. The model effectively isolated influential factors such as patient age and intervention duration, demonstrating its capability in handling the inherent data correlations.

CONCLUSIONS: Employing a hierarchical GLMM in this non-inferiority study allowed for a nuanced analysis of complex, nested data that is typical in clinical settings, particularly when measurements involve both eyes. This approach not only supported the main hypothesis but also enhanced the understanding of how patient-specific factors impact the treatment outcome. The findings validate the use of hierarchical GLMMs as a robust tool for analyzing complex clinical trial data, providing a methodological framework applicable to similar future studies.