OBJECTIVES: Ophthalmology trials are designed to assess clinical primary endpoints such as lesion growth reduction in geographic atrophy (GA). For economic modeling, the emphasis is on quality-of-life impacting outcomes, such as visual acuity (VA), which trials may not have been powered or randomized for evaluating. To illustrate, results of two phase 3 trials of intravitreal pegcetacoplan showed statistically significant reductions in lesion growth but not VA decline; however, these trials did not account for key VA risk factors like macular occupancy (MO). This analysis applied causal inference methods to better reflect the complex relationship between lesion growth and vision loss.

METHODS: The randomized trial sample (n=1211) was subset to those with complete baseline measures (n=888) and stratified by subfoveal vs. extrafoveal (≥250μm from foveal center) lesions. VA was evaluated as change from baseline up to 24 months compared to sham. The adjusted impact of treatment was estimated using inverse probability of treatment weighted regression models, with propensity scores estimated from baseline MO, lesion size, VA, age, sex, geographic region, and multifocal, bilateral, pseudophakic, and better- vs. worse-seeing eye status.

RESULTS: Several VA risk factors differed between treatment arms, with the most substantial imbalance seen in MO. For the subfoveal group, where adjustment is expected to have less impact on VA, the observed and estimated differences in VA loss at 24 months were nearly identical (-1.7 and -1.6 letters for treatment relative to control). However, for the extrafoveal group, where lesion location is critical to VA impact, covariate adjustment increased the estimated treatment effect from an observed +2.2 letters to an adjusted +5.6 letters (p=0.078).

CONCLUSIONS: Adjustment methods, widely accepted in observational research, are now increasingly being applied to secondary outcomes and sub-groups analyses of RCTs. In these phase 3 trials, adjustment for baseline imbalances allowed for estimation of treatment effect on functional measures.