Fits a compound-Poisson transmission model (individual reproduction number drawn from a gamma distribution, dispersion parameter k) to contact-tracing data from 8 directly-transmitted human diseases (SARS-1, measles, smallpox, pneumonic plague, etc.), showing infectiousness is highly right-skewed: the great majority of cases transmit to few or no others while a small minority (“superspreaders”) generate large, explosive clusters. Formalizes a rigorous statistical definition of a superspreading event (transmission exceeding the 99th percentile of a Poisson(R0) expectation) and shows individual-level heterogeneity, not average R0, governs whether/where an emerging pathogen’s first visible cluster appears.
relevance_note: the foundational quantitative primary behind “venue-amplification” reasoning — a pathogen’s first visible cluster size and location is driven by whichever contact network a rare superspreading event happens to hit, independent of where true spillover occurred; underlies (but does not itself address) the market-specific ascertainment-bias argument in this case.
Extracted (structured summary)
Empirical overdispersion
O-23 - Individual infectiousness is highly overdispersed across eight directly-transmitted diseases; SARS dispersion k approx 0.16
Empirical offspring distributions from detailed contact-tracing/surveillance datasets for eight directly-transmitted diseases; negative binomial(R0,k) has variance R0(1+R0/k), so smaller k means greater heterogeneity (k→inf Poisson, k=1 geometric).
Link to original
Branching-process consequence
A-41 - With overdispersed infectiousness, chance superspreader placement, not average R0 or spillover location, governs whether and where a pathogen first cluster appears
Reasoning
In the negative-binomial branching process (nu gamma-distributed with mean R0 and dispersion k; number of secondary cases Z ~ NB(R0,k), variance R0(1+R0/k)), the probability q of stochastic extinction after a single introduction rises with heterogeneity. For R0=3, q=0.06 under homogeneity (k→inf), 0.33 for k=1, and 0.76 for k=0.16 (the SARS estimate); as k→0, q→1 even for arbitrarily large R0. The expected number of cases before extinction is nearly independent of k, so with high variation extinction happens ‘rapidly or not at all’. Conditional on non-extinction, low-k outbreaks grow infrequently but explosively - matching SARS in 2003, where many settings saw no epidemic despite unprotected exposure while a few cities suffered explosive outbreaks, explained simply by the presence or absence of high-nu individuals in the early generations. A superspreading event is defined rigorously as any case infecting more than the 99th-percentile Z(n) of a Poisson(R) distribution. Because establishment and the size of the first recognized cluster are dominated by these rare right-tail transmission events, the location of the first visible cluster tracks whichever contact network a superspreader happened to hit (a high-throughput venue), not the point of spillover. This is the quantitative foundation for venue-amplification / ascertainment reasoning: a first-detected cluster’s size and place are set by chance superspreading, not by where the pathogen actually entered humans.
Validity verdict
status: approved; reason_if_not_false: checked. Reconstruction — premises: NB(R0,k) branching process with the stated extinction probabilities, plus the (implicit, granted) premise that superspreading transmission is realized in high-throughput contact settings distinct from the spillover point. Load-bearing step: because establishment and first-cluster size are dominated by rare right-tail transmission events, the setting of the first visible cluster tracks whichever venue a superspreader happened to hit, not the spillover locus. Traced the math directly (branching-process results are elementary and standard Lloyd-Smith 2005; q monotone in heterogeneity, expected pre-extinction size ~k-independent → “rapidly or not at all,” explosive conditional on establishment) → checked, author-blind. Probed the undercutting defeater that early chains are short so the superspreader sits geographically near spillover: this constrains distance but not setting/venue, and the conclusion is about which setting/venue the first cluster centres on — so it does not break the step. Approved as stated.
Link to original