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.

Our simple estimators of transmissibility can be applied generally to study zoonoses. Nipah virus is primarily clinically characterized by fever and encephalitis and was first discovered in a large outbreak in Malaysia in 1998-1999 [17, 18] . During this outbreak (where sick pigs were believed to be the natural reservoir), Parashar et al. [19] conducted a case-control study of the risk factors for infection. They recruited patients who were hospitalized with encephalitis from January through April 1999. Candidate encephalitis patients whose serum specimen tested positive for Nipah antibody were included as cases. Other cases were detected through targeted investigations; but if we restrict analyses to the subset of patients detected through hospital surveillance, this situation resembles surveillance scenario 1 above, where the probability of being detected is independent of cluster allocation. Here, we can therefore estimate the reproduction number as R = 12G. Probability G can be approximated using the data of Parashar et al. [19] by the proportion of cases who (1) either lived or worked in a farm (G 1 ) or (2) handled a pig, or came within 1 m of a pig and came into contact with pig urine or faeces (G 2 ). Parashar et al. [19] only give estimates for the pooled set of cases (i.e., hospital surveillance+targeted investigations) for which G 1 = 95% and G 2 = 92%; however, it should be straightforward to derive these statistics for the subset of cases detected through hospital surveillance alone. For G = 0.92-0.95, we estimate R = 0.05-0.08. This suggests low levels of human-to-human transmission in this first outbreak. [21] . They classified 60 of the 122 identified Nipah virus cases as reservoir-to-human infections (G = 49%). Using the simple estimator R = 12G, we obtain R = 0.51. This is consistent with more detailed contact tracing data that estimated R = 0.48 [21] . However, both these estimators may be biased upwards as the Bengali situation corresponds to surveillance scenario 2 (i.e., detection of a case may trigger an outbreak investigation), which may lead to the selection bias discussed earlier. In such a context, we recommend running our analysis on the subset of first detected cases in each of the 23 introductions and estimating proportion F. Again, it should be possible to derive F from data collected at the time. Since the case detection rate is unknown, we expect 12F to act as a lower bound for R ( Figure 5 ). From the estimates of G and F, we should therefore be able to derive simple bounds on R: 1{F Æ’RÆ’1{G~0:51.