Findings from a new in silico study from Roche researchers suggests that combining home optical coherence tomography (OCT) with pharmacokinetic/pharmacodynamic (PK/PD) modeling could make early ophthalmology clinical trials substantially more efficient and reduce sample size requirements by 20% to 40% while maintaining statistical power.
Published in Translational Vision Science & Technology, the study by Jacques Hermes, PhD, and Bernhard Steiert, PhD, evaluated whether frequent home-based imaging could complement advanced modeling techniques to overcome the variability and extended timelines that are common in retinal disease trials.
They found that traditional biweekly, on-site monitoring required 41 to 54 patients per arm to detect the effect. Home OCT with PK/PD modeling, however, achieved the same power with only 33 to 35 patients per arm, which represents a 20% to 40% reduction in required sample size. Both simulation settings assumed comparable residual noise (σ=5.5 µm), and simulated central subfield thickness (CST) dynamics reproduced a realistic spread of treatment responses that are seen in neovascular age-related macular degeneration (nAMD) populations. Drs. Hermes and Steiert also confirmed parameter identifiability and model robustness using profile likelihood analysis, multistart optimization, and prediction-corrected visual predictive checks.
The researchers developed a population PK/PD model that described CST dynamics following anti–VEGF therapy in nAMD. The model was calibrated using data from a previously published home OCT study of 15 participants (29 eyes) who performed near-daily scans using a Notal Vision device. Participants averaged 5.7 ± 0.9 scans per week, and 93% of scans were eligible for automated analysis. After excluding eyes with mild disease or insufficient data, 20 eyes (17 under active anti-VEGF treatment) contributed 1,628 CST measurements. The study modeled responses to aflibercept and ranibizumab based on established half-lives (9 days and 6.5 days, respectively). Drs. Hermes and Steiert then applied Monte Carlo simulations and bootstrapping to quantify variability and estimate the number of participants needed per treatment arm to detect a mean CST reduction of approximately 50 µm vs an active comparator.
The findings indicate that integrating high-frequency, patient-operated OCT data with PK/PD modeling could streamline early ophthalmology trials by improving endpoint precision and reducing recruitment needs. While the model was based on a small observational dataset, the authors proposed validating this approach prospectively in interventional trials that incorporate home OCT alongside standard in-clinic monitoring.
“In such a design,” they described, “the observed treatment effect could first be evaluated using the conventional sparse dataset. The same effect could then be tested using the densely sampled data from home OCT, comparing the statistical power and sample size requirements retrospectively. This would allow for benchmarking the in silico-predicted sample size reductions against actual clinical outcomes. Moreover, such a study could directly assess the practical impact of variability in home OCT adherence and data quality on trial endpoints and modeling performance.” They further noted that future work should test the method across broader populations and assess adherence, image quality, and disease diversity in real-world settings because “factors such as health literacy, comfort with technology, and clinical diversity may influence the generalizability of the findings” in their “relatively small and demographically uniform sample.”
Full disclosure statements are available in the published study. RP