OBJECTIVES: Previous economic evaluation of ocriplasmin for treatment of VMT used a de novo model consisting of a decision tree and nested Markov components, simulating the MIVI-TRUST trial and extrapolating long-term disease progression. Dependent on patient characteristics and disease history, patients could transition between six different vision health states (HS). Patient-level data requirements for estimating a 6x6 transition matrix poses a challenge with smaller sample sizes of new trial evidence. Learnings from survival partition models were adopted to investigate a more flexible approach to estimate patient distribution in vision HS (partitioned distribution model).

METHODS: Eight vision HS and eight disease HS were used. Patient vision, for each disease HS, at each time point, is described using mean and standard deviation (SD) following (scaled) beta distribution to capture both lower and upper bounds of the 0-100 Early Treatment Diabetic Retinopathy Study (ETDRS) vision scale. Mean and SD were estimated from OASIS data. Beta distribution was partitioned according to vision HS cut-offs determining the vision HS proportion of patients for each disease HS.

RESULTS: Preliminary results (ETDRS letters read) indicate mean (SD) baseline best-corrected visual acuity (BCVA) for HS1 (VMT with FTMH) and HS2 (VMT without FTMH) are 57.82 (9.81) and 66.00 (8.29), respectively. For 24-month disease HS distributions using linear regression models and OASIS observed data, mean BCVA (SD) were: HS3 (unresolved FTMH) 57.69 (9.60), HS4 (surgically resolved FTMH) 68.65 (11.26), HS5 (Non-surgically resolved FTMH) 70.00 (8.23), HS6 (unresolved VMT) 67.96 (10.39), HS7 (surgically resolved VMT) 68.95 (13.24), and HS8 (non-surgically resolved VMT) 75.82 (9.00).

CONCLUSIONS: Compared to previous modelling techniques, this approach offers a simpler, more clinically intuitive methodology to simulate patient vision. Patient vision over time requires only three parameters (mean, SD and change in mean over time) as opposed to a large and granular transition probability matrix.