Finalist
Student Team 13272 from Makerere University: Africano Byamugisha (Team lead) and Wilbroad Owor
Africa bears the world’s highest per capita AMR mortality but holds the sparsest surveillance data in the Vivli AMR Register. Expanding sentinel site surveillance is expensive, and current scale up is largely ad hoc. No principled framework exists to answer: given what we already know, which country × pathogen × antibiotic cells should be sampled next to maximally reduce uncertainty about African AMR per dollar invested?
Bayesian optimal experimental design (BOED) is a mature statistical framework for selecting experiments that maximise expected information gain (EIG) about quantities of interest. It has been applied in clinical trial design, ecological monitoring, and infectious disease surveillance, but not in AMR surveillance prioritisation.
Scientific question: can BOED, applied to the Vivli AMR Register and constrained by realistic laboratory capacity and investment data from the Global AMR R&D Hub, identify surveillance priorities that would have detected African resistance shifts 1–2 years earlier than historical expansion actually did?
Methodology.
Step 1: a Bayesian hierarchical model of MIC distributions with pathogen × antibiotic × country × year random effects and Gaussian process spatial priors, fit with NumPyro on ATLAS, SIDERO WT, and J&J DREAM.
Step 2: compute expected information gain for each candidate country × pathogen × antibiotic cell via nested Monte Carlo.
Step 3: constrained integer programming (OR Tools) over laboratory capacity budgets derived from the Global AMR R&D Hub Investment Gallery and WHO GLASS enrolment tiers.
Step 4: retrospective validation — truncate ATLAS at 2018 and show BOED prioritised cells align with observed 2019–2023 resistance shifts.
Deliverables: open source BOED AMR library (GitHub, MIT), interactive prioritisation dashboard, a Uganda case study co developed with Makerere clinicians, a two page policy brief for the Uganda Ministry of Health, and a Wellcome Open Research manuscript