OphthaDT
An LLM-based digital-twin benchmark for ophthalmology that serializes longitudinal patient histories from 3,220 patients across four Phase III clinical trials into structured narratives to forecast best-corrected visual acuity (BCVA) over horizons up to 100 weeks in neovascular AMD and diabetic macular edema.
Composite
54.5
Experimental validation
N/A — retrospective clinical-trial digital-twin benchmark
Stages
Clinical Development (cross-phase)Phase IIIDisease modeling
Modalities
biologic (mAb/ADC/bispecific)
Task types
regression
Size
patients: 3,220
trials: 4
horizon_weeks: 100
splits: {'train': 0, 'val': 0, 'test': 0}
note: 3,220 patients across four Phase III trials (nAMD and DME); BCVA forecasting up to 100 weeks; baselines include Random Forest and XGBoost
trials: 4
horizon_weeks: 100
splits: {'train': 0, 'val': 0, 'test': 0}
note: 3,220 patients across four Phase III trials (nAMD and DME); BCVA forecasting up to 100 weeks; baselines include Random Forest and XGBoost
License
Other — derived from proprietary Phase III trial data (controlled-access upstream)
First release
2026-06-20
Last updated
2026-06-20
Official site
Leaderboard
→ leaderboard
Dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
OphthaDT: Generative Digital Twins for Forecasting Visual Acuity Trajectories in Ophthalmology · Pietro Belligoli, Nikita Makarov, Sayedali Shetab Boushehri, Fabian Schmich, Raul Rodriguez-Esteban, Michael Menden · 2026 · paper · doi:N/A — arXiv preprint 2606.22101 · 0 citations
Flags
none
Experts
Groups
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
4
coverage
2
maintenance
2
adoption
1
quality
4
accessibility
2
industry_relevance
4
Notes
Rare clinical digital-twin benchmark grounded in real Phase III trial data with industry-affiliated authorship (Roche/Genentech collaborators; Menden lab), giving high industry relevance (4) and data quality (4). Honest reporting that linear models remain competitive for stable DME responses while the digital twin helps most as trajectory complexity grows. Coverage narrow (two retinal indications, BCVA endpoint) and accessibility limited by proprietary trial data; adoption 1 as a June 2026 preprint.