Insufficient data for model comparison.
Data: 20 post-ruxolitinib HRV observations, 20 post-ruxolitinib HR observations from Oura Ring Gen 4 (consumer wearable, not clinical-grade).
Linear model: OLS regression on day number since treatment start (2026-03-16). Bootstrap: 1000 resamples for confidence intervals.
Exponential model: y = a · (1 - e-t/τ) + baseline, fit via scipy.optimize.curve_fit with bounded parameters. Model selection by AIC (lower = better).
Three scenarios: Optimistic (75th pct slope), Expected (50th pct), Conservative (25th pct).
Limitations (small N): With only ~20 data points, forecasts carry wide confidence intervals and may shift substantially as more data accumulates. Physiological recovery is non-linear and depends on factors not captured in wearable data (infection status, medication changes, immune reconstitution). These projections are exploratory, not clinical recommendations.