A Bayesian joint model that reads tumour scans as survival evidence
Our new paper couples a mechanistic model of tumour growth to a multistate survival model and fits both at once. In a small-cell lung cancer case study, it forecasts the mature progression-free survival curve from nine patients, and converges on overall survival eight months before it read out.
Temporal cross-validation for hierarchical mechanistic survival models
LFO cross-validation was built for univariate time series. Here I work out what it takes to bring it into hierarchical joint longitudinal–survival models: staggered enrolment, dual observation clocks, ELPD decomposition, and a patient-level bootstrap for standard errors.
Sequential decision-making under uncertainty for Effective Altruism funders
A simulation study of planning policies for an EA funder maximising welfare through intervention selection. Naive policies — single-study belief updating — perform substantially worse than probabilistic policies that optimise re-evaluation timing, with the gap widening in sequential settings.