Notes
Methodology notes, replication memos, and short pieces on the mechanics of modelling human physiology. Citation-heavy by design.
The Ebola drugs work. The timing cliff starts at 48 hours.
The PALM trial established that mAb114 and REGN-EB3 reduce mortality in Ebola. It did not establish how outcomes depend on when treatment starts. A coupled 22-state model, calibrated against all six PALM arms, says the answer is steeper than the field assumes — and the implication is logistical, not pharmacological.
The asterisk on the anti-amyloid Alzheimer drugs
Lecanemab and donanemab reduce cognitive decline ~25–33% over 18 months. They also shrink the brain — by about half a percent of whole-brain volume in the same window. The pseudoatrophy and the real-neurotoxicity hypotheses make identical 18-month predictions. The data needed to distinguish them does not exist yet.
Foresight is a remarkable model. It is not a patient digital twin.
Why a generative transformer over electronic health records cannot answer the counterfactual question every clinician actually wants to ask — and what the engineering definition of digital twin requires instead.
The mechanism is real. The window is three days.
Hydroxychloroquine, ivermectin, and ribavirin have plausible antiviral activity against hantavirus. A calibrated model of hantavirus cardiopulmonary syndrome says they would cut mortality substantially — if you could give them before the cascade fires. The clinical fact is that you cannot, and the reason is diagnostic, not pharmacological.