Using Routine Surveillance Data to Estimate the Epidemic Potential of Emerging Zoonoses: Application to the Emergence of US Swine Origin Influenza A H3N2v Virus

Cauchemez, Simon; Epperson, Scott; Biggerstaff, Matthew; Swerdlow, David; Finelli, Lyn; Ferguson, Neil M.

For efficient prevention and control, quantitative and rigorous assessment of the risks associated with emerging zoonoses is desirable-in particular the risk that an emerging pathogen evolves to cause sustained human-to-human transmission. One approach to such risk assessment is by monitoring the reproduction number R of zoonoses in humans, with an alarm being raised if R increases or approaches 1 [9] [10] [11] . However, until now, estimating R required detailed outbreak investigations of human clusters [10, 11] and suffered from three important limitations: (1) the resources, access, and expertise needed to conduct investigations is not always available; (2) the proportion of cases that are missed during outbreak investigations may vary by setting and be difficult to assess; (3) even if the study is complete, the data collection process can be affected by a selection bias whereby larger outbreaks are more likely to be detected so that estimates of transmissibility may be biased upward. Consider for example a scenario where R = 0.7, where each case has the same detection probability r = 1%, and assume that once a cluster is detected, detailed outbreak investigation ensures that all cases in the cluster are detected. With an average size of 18.3 and a 21% probability of 1-case cluster, clusters that are detected are substantially larger than normal ones (average size: 3.3; 65% probability of 1-case cluster) ( Figure 1A ). As expected, this selection bias leads to R being overestimated as illustrated for methods that use the distribution of detected cluster sizes ( Figure 1B ) [10] .