((A-188)) (A-188) Group-Based Trajectory Models to Identify Potential Reasons for Treatment Switch Among Patients with Neovascular Age-Related Macular Degeneration (nAMD)

Background: Visual acuity (VA) trends prior to switching could be used to identify reasons for switching treatment among patients with nAMD in EHR databases where this information is often missing. Group-based trajectory modeling (GBTM) offers a data-driven approach for identifying nAMD patient subgroups and potential reasons for switching to aflibercept (AFL) 8mg.

Objectives: To compare GBTM vs. a clinical algorithm in classifying nAMD patient eyes based on VA trends prior to switching to AFL 8mg.

Methods: Previously treated eyes with nAMD who initiated AFL 8 mg (index date) between 8/18/23-6/30/24 were identified in the Intelligent Research in Sight® (IRIS) registry. Eligible patients were required to meet the following criteria: at least 1 visit ≥6 months prior to index, aged ≥ 50 years, ≥1 VA measure, anti-VEGF use 12 months before index (baseline) and antecedent to index, but no other nAMD therapy on index or baseline diagnoses of DME/RVO. A clinical algorithm was used to create 4 mutually exclusive groups of eyes based on VA trend prior to switch (index VA – prior VA): eyes with poor VA (≤38 letters); declining VA (>5-letters loss); improving VA (>5-letter gain); and stable VA (≤5-letter gain/loss). GBTM was used to identify groups of eyes based on their VA at monthly intervals (± 15 days) during baseline. The final model was selected using Bayesian Information Criterion and an average predicted probability of group membership >70%. The percent overlap between methods was calculated using the Ns classified by the clinical algorithm as the denominator.

Results: Analyses included 25,846 eyes with nAMD switching to AFL 8mg. Majority of patients were White (84%), male (60%) with a mean age of 81 years. The clinical algorithm categorized 4055 (15.7%) as poor VA, 4251 (16.4%) as declining VA, 4463 (17.3%) as improving VA, 12,790 (49.5%) with stable VA, and 241 (1%) with unknown VA trends. GBTM (n=25,800) identified four groups of trajectories of VA prior to switching: 1 included poor and declining VA (n=5578), 2-4 included stable VA with differing intercepts (i) (i: 79, n=6670; i:71, n=8321; i:60, n=5231), no groups had improving VA, and 46 eyes with a single VA measure were excluded. GBTM group 1 had a 95% and 6% overlap with the poor VA group and declining VA groups, respectively. There was a 41%, 34%, and 21% overlap between the stable VA group and GBTM groups 2, 3, and 4, respectively, while a < 50% overlap with the declining and improving VA groups was observed.

Conclusions: Although GBTM was effective in identifying patients with poor VA, the overlap with other clinical groups was < 50%. While GBTM is useful for understanding relationships over time and handling missing data, results should be interpreted with caution as GBTM may not always identify clinically relevant subgroups.