Model-guided placement of decentralized diagnostics for spatial control of the 2026 Bundibugyo Ebola epidemic
- 1Gates Foundation
Abstract
Before 2026, Bundibugyo virus disease had only two recognized outbreaks. By August 2026, delayed recognition and multi-day sample transport in eastern Democratic Republic of the Congo had contributed to a large, spatially dispersed epidemic. We hypothesized that model-guided placement of decentralized diagnostics in frontier health zones could help contain spatial spread by acting as a spatial fence: delayed recognition would be converted into timely polymerase chain reaction (PCR) confirmation, isolation, and contact tracing before imported lineages became established.
We fit a Bayesian spatial branching process to public situation reports through 2 August 2026, with Flowminder movement and a latent-burden anchor. Lineages had overdispersed offspring, PCR access and turnaround, clinical recognition, isolation, tracing, safe burial, and local behaviour change. Because the observed history did not identify future control, we assumed a national response ramp yielding West-Africa-scale status-quo burden and replayed 2,000 posterior trees with 5-100 zone-level PCR laboratories, deployment delays of 0-180 days, and postmortem rapid diagnostic test (RDT) coverage.
The posterior reproduced case, death, latent-burden, and spatial targets. Speed dominated: 35 PCR laboratories averted a median 49% of future deaths if deployed immediately, 38% after 14 days, and 11% after 60 days, with 90% prediction intervals of 30-65%, 23-53%, and 5-20%, respectively. Placement also mattered: with a 30-day delay, model-informed placement averted 18-33% of future deaths with 20-100 laboratories, versus 1-8% under random placement. Postmortem RDTs partly filled gaps left by live-case detection, increasing future deaths averted from 26% to 41% with 20 laboratories, a 14-day delay, and 75% coverage. Decentralized diagnostics can slow spatial spread when they are placed ahead of likely importations and paired with staffing, isolation, tracing, and safe-burial capacity.
Related
- Reported fatality ratios and surveillance data readiness: a test applied to health zones in the 2026 Bundibugyo virus disease epidemic, Democratic Republic of the Congo
- epymorph turns spatial epidemic model comparisons into controlled forecasting experiments
- Choosing the spatial scale of aggregation in epidemic models: compression versus distortion
- VSV-EBOV vaccination induces cross-reactive T-cell responses against Bundibugyo virus in humans
- Ensuring children's inclusion in the Bundibugyo Ebola response: Striving to do better this time
BibTeX
@misc{klein2026modelguided,
title = {{Model-guided placement of decentralized diagnostics for spatial control of the 2026 Bundibugyo Ebola epidemic}},
author = {D. J. Klein and K. H. Grantz and C. O. Buckee and D. B. George and S. F. Dowell},
howpublished = {medRxiv},
year = {2026},
month = sep,
doi = {10.64898/2026.09.29.26364338},
url = {https://www.medrxiv.org/content/10.64898/2026.09.29.26364338v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Model-guided placement of decentralized diagnostics for spatial control of the 2026 Bundibugyo Ebola epidemic
AU - D. J. Klein
AU - K. H. Grantz
AU - C. O. Buckee
AU - D. B. George
AU - S. F. Dowell
PY - 2026
DA - 2026/09/29
PB - medRxiv
DO - 10.64898/2026.09.29.26364338
UR - https://www.medrxiv.org/content/10.64898/2026.09.29.26364338v1
AB - Before 2026, Bundibugyo virus disease had only two recognized outbreaks. By August 2026, delayed recognition and multi-day sample transport in eastern Democratic Republic of the Congo had contributed to a large, spatially dispersed epidemic. We hypothesized that model-guided placement of decentralized diagnostics in frontier health zones could help contain spatial spread by acting as a spatial fence: delayed recognition would be converted into timely polymerase chain reaction (PCR) confirmation, isolation, and contact tracing before imported lineages became established. We fit a Bayesian spatial branching process to public situation reports through 2 August 2026, with Flowminder movement and a latent-burden anchor. Lineages had overdispersed offspring, PCR access and turnaround, clinical recognition, isolation, tracing, safe burial, and local behaviour change. Because the observed history did not identify future control, we assumed a national response ramp yielding West-Africa-scale status-quo burden and replayed 2,000 posterior trees with 5-100 zone-level PCR laboratories, deployment delays of 0-180 days, and postmortem rapid diagnostic test (RDT) coverage. The posterior reproduced case, death, latent-burden, and spatial targets. Speed dominated: 35 PCR laboratories averted a median 49% of future deaths if deployed immediately, 38% after 14 days, and 11% after 60 days, with 90% prediction intervals of 30-65%, 23-53%, and 5-20%, respectively. Placement also mattered: with a 30-day delay, model-informed placement averted 18-33% of future deaths with 20-100 laboratories, versus 1-8% under random placement. Postmortem RDTs partly filled gaps left by live-case detection, increasing future deaths averted from 26% to 41% with 20 laboratories, a 14-day delay, and 75% coverage. Decentralized diagnostics can slow spatial spread when they are placed ahead of likely importations and paired with staffing, isolation, tracing, and safe-burial capacity.
ER -