Quantifying transmission of emerging zoonoses: Using mathematical models to maximize the value of surveillance data

Monique R. Ambrose; Adam J. Kucharski; Pierre Formenty; Jean-Jacques Muyembe-Tamfum; Anne W. Rimoin; James O. Lloyd-Smith

In the monkeypox analysis, the size of the administrative unit used as the broader contact 356 zone has a strong effect on the resulting parameter estimates (Fig 5A) . When larger 357 administrative units are assumed to represent the broader contact zone, a given pair of cases is 358 more likely to belong to the same broader contact zone, giving the model more opportunities to 359 infer inter-locality human-to-human transmission events and resulting in larger estimated 360 reproductive number R and a smaller spillover rate λ z . Mean values of the posterior distribution 361 of R range from 0.29 when transmission is assumed to occur only within localities to 0.52 when 362 transmission is assumed to occur among all localities in the country (Table 1) Both the choice of broader contact zone and the assumed total number of localities have a 407 large impact on estimates of R and λ z ( Fig 5B) . As noted above, models assuming smaller 408 broader contact zones allow fewer opportunities for human-to-human transmissions to be 409 inferred, and these models estimate substantially lower R values and correspondingly higher 410 spillover rates. In contrast, assuming that a smaller fraction of surveilled localities were observed 411 . CC-BY 4.0 International license is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It District-level broader contact zone preferred in model comparisons. To assess which broader 427 contact zone assumption is most appropriate for the monkeypox system, we used the deviance 428 information criterion (DIC) to perform model comparisons for the corrected-denominator 429 method as well as for each assumption about the number of surveilled localities. For the 430 corrected-denominator method, the district-level model had the best DIC score, followed by the 431 region and country-level models ( Table 1) . The locality-level model received a much larger DIC 432 value, indicating that the data strongly support models that allow transmission between localities. 433 . CC-BY 4.0 International license is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. It . https://doi.org/10.1101/677021 doi: bioRxiv preprint