# Digitized trial data

Source: Douglas RS, Kahaly GJ, et al. "Teprotumumab for the Treatment of
Active Thyroid Eye Disease" (OPTIC trial). N Engl J Med. 2020;382:341-352.
NCT03298867. Phase 3, randomized, double-masked, placebo-controlled,
teprotumumab 10 mg/kg then 20 mg/kg q3w x8 infusions vs. placebo, N=41
(teprotumumab) / N=42 (placebo).

**Extraction method:** the source PDF's text layer includes the numeric data
labels printed directly on Figure 2's panels (these are exact reported
values, e.g. "−3.32", "78", not pixel/axis-position estimates) plus Table 1
(baseline characteristics) and Figure 1B (summary table). This is precise
tabular extraction, not visual chart digitization.

## Files

- `optic_trial_baseline.csv` — Table 1 baseline characteristics.
- `optic_trial_proptosis_change.csv` — Figure 2B: least-squares mean change
  from baseline in proptosis (mm) at weeks 0/6/12/18/24, both arms. This is
  the primary calibration target (`scripts/calibrate_mechanism_core.R`) —
  the only continuous, visit-by-visit endpoint reported that maps directly
  to a model output (`Proptosis_mm`).
- `optic_trial_responder_rates.csv` — Figure 2A/C/D/E: % of patients meeting
  each binary response threshold at each visit (proptosis response, CAS 0-1,
  overall response, diplopia response). **Not used for direct curve-fitting**
  — these are population responder *rates* against a threshold, not a
  continuous mean score, so they don't map onto `CAS_score()`/
  `diplopia_grade()` without a virtual-population/threshold layer this model
  doesn't have yet (see model-spec.md Section 8, "not yet done"). Used only
  as a qualitative timing sanity-check.
- `optic_trial_goqol.csv` — Figure 2F GO-QOL changes. **Low confidence on
  visit ordering** — the source PDF's text extraction did not cleanly
  preserve which numeric label belongs to which visit for this panel; the
  values are collected but not verified against the original figure image.
  Not used anywhere in the current model (no GO-QOL output exists).

## Known gap

No continuous mean-CAS-by-visit table is reported in this paper (only the
binary "% with CAS 0 or 1" responder rate). The mechanism core's CAS output
is therefore calibrated only indirectly (shared fibroblast/effector
kinetics fit to proptosis) — see model-spec.md Section 8 for the honest
accounting of what is and isn't validated.