A Causal Inference Approach to Mediation Analysis in Vitreomacular Traction: How Much Does Traction Resolution Mediate Functional Outcomes?

Abstract

Modern mediation analysis techniques supplement the primary intention-to-treat analysis with the aim to shed light onto the treatment mechanism. We investigate to what extent the anatomic marker vitreomacular adhesion resolution (VMAR) mediates vision benefits, comparing ocriplasmin vs. a sham regimen. A causal mediation analysis is applied to randomized trial data including 218 participants with vitreomacular traction. Logistic regression models are used to estimate the total treatment effect (TTE) on binary outcomes. Outcomes, assessed at month 24, included visual acuity improvement (VA-I): ≥2-line increase in VA; visual function questionnaire improvement (VFQ-I): ≥5-point increase in the 25-item visual function questionnaire composite score (VFQ-25cs); visual function improvement (VF-I): defined as either a VA-I or a clinically meaningful improvement in the VFQ-25cs. Quantity of interest is the breakdown of TTE into an indirect (through VMAR) and direct effect to estimate the extent to which the TTE is transmitted through the mediating variable (VMAR) vs. other pathways. Causal effects are expressed as risk differences. Indirect effects for VFQ-I, VA-I, and VF-I are 5.7%, 11.8%, and 5.2%, respectively, representing the increase in the probability of a vision improvement if VMAR status were changed for each participant to the extent that it is affected by ocriplasmin. The direct effects are 8.3%, 12.1%, and 24.1% respectively, capturing the effect of treatment on the probability of a vision improvement if ocriplasmin left each participant's VMAR status unchanged. The relative treatment effect of ocriplasmin on the functional outcome VA-I is to a large extent the result of its effect on VMAR, while an improvement in the patient-reported outcomes VFQ-I or VF-I was only partially mediated by VMAR.