Initial lesion growth rates and other baseline prognostic factors can improve the design of clinical trials in Geographic Atrophy (GA)

Abstract

Purpose : Adjusting for prognostic variables when analyzing clinical trial data leads to greater power by reducing unexplained variability. Previous studies have identified various prognostic factors at baseline related to subsequent GA lesion growth, with initial lesion growth rates being the most predictive of the various baseline variables.

Methods : The present study evaluated various prognostic factors using data from the SPECTRI (NCT02247531) and CHROMA (NCT02247479) Phase 3 studies (n=975 and 906, respectively), with an objective to quantify the potential impact on trial efficiency of these baseline factors for future trials in GA. A benchmark model was developed using machine learning techniques and considering all baseline variables except initial growth rate. This model was developed using SPECTRI, validated with CHROMA and consists of a simple linear function based on three variables derived from fundus autofluorescence imaging (baseline GA lesion size, lesion location, and lesion contiguity) and one variable derived from normal and low-light visual acuity assessments (low luminance deficit).

Results : The benchmark model explained 20% of variability in subsequent 12-month growth. As a stand-alone prognostic factor, the 6-month initial growth rate explained 37% of variability in subsequent 12-month growth. A model combining both the variables from the benchmark model and the 6-month initial growth rate explained 42% of subsequent growth.

Conclusions : The percent of explained variability can be translated into an effective sample size increase (ESSI), indicating how much larger a trial would have to be when using unadjusted analysis to have the same power as an analysis adjusting for the prognostic factors. Based on the above results, the ESSI of the benchmark model was 30% and the ESSI of the combined 6-month initial growth rate + benchmark model was 70%. The net increase in sample size purely due to the 6-month initial growth rate after accounting for the value of the benchmark model variables was 34%. The authors suggest that patients who may be eligible for future GA clinical trials have regular imaging to potentially avoid prolonged screening, and further speed enrollment in this disease for which there are no current therapies.

This is a 2020 ARVO Annual Meeting abstract.