We describe the combined analysis of 7 household infection studies of pH1N1 influenza conducted in early 2009 to determine the serial interval and the temporal distribution of secondary infections within households. The mean serial interval was 2.9 days. Only 18% of transmission events were estimated to take place >2 days after the onset of clinical symptoms among those ill with pH1N1 virus, and only 5% >3 days after onset. To our knowledge, this is the largest analysis of its type to date and the first to estimate the temporal distribution of secondary infections within households.
The gold standard data for the estimation of the serial interval distribution of a disease are linked pairs of dates of the onset of clinical symptoms in infectors and infectees. Although it cannot typically be proven that within-household pairs are in fact infector-infectee (particularly within communities affected by ongoing disease transmission), rigorous analysis of clinically apparent serial intervals–in which unrealistically long serial intervals are excluded and external introductions of infection are allowed for–was implemented to infer the underlying distribution of true serial intervals. Our approach may lead to slight overestimation of the true serial interval because it did not correct for tertiary cases. However, such bias would be expected to be relatively small, given that studies of pH1N1 infections have shown low secondary attack rates in households, implying relatively few tertiary infections [20
]. An indication of the level of bias can be derived by comparing the estimates obtained from study 1 in the current analyses with those derived from an analysis that corrected for tertiary transmission [20
]. In the latter analysis, the mean serial interval was estimated to be 2.6 days, whereas our methods, applied only to study 1, yielded an estimate of 2.8 days. We therefore conclude that our estimates of a mean clinically apparent serial interval of 2.9 days is consistent with estimates of 2.5–2.7 days obtained from analyses that corrected for tertiary transmission [20
From the point of view of providing information relevant to the recommended duration of case isolation, our estimates are conservative, in that not correcting for tertiary transmission means we slightly overestimate the proportion of within-household transmission occurring after a particular number of days after symptom onset. However, depending on the occupation of an infected individual, the risks of onward transmission associated with return to work after a particular number of days after symptom onset may be much greater than those implied by the household data---for example, in the case of an individual caring for immunocompromised patients.
The distributions of within-household transmissions (following the onset of clinical symptoms), implied by these estimated serial interval distributions, consistently indicate that only approximately 15% to 20% of within-household transmission happens >2 days after symptoms appear. This supports the view that requiring isolation of those with ILI for a full week after the onset of clinical symptoms may not be necessary to reduce substantially the risks of transmission from clinically affected individuals to communities and workplaces. However, reducing transmission substantially at a population level will not be, of course, the only consideration, because there will be a desire to reduce the risk of transmission from individuals working in high-risk settings to virtually zero. Furthermore, if a future influenza pandemic were associated with increased severity, the balance between preventing onward transmission and the need to limit societal disruption might be such that isolation guidelines would be more restrictive. Interpretation of the results should be cautious because of the absence of information on the periods of isolation undertaken by the index individuals under study; it is possible that within-household transmission was somewhat reduced following the return of the index individual to his or her usual activities (eg, work, school, and shopping).
This study demonstrates the value of using detailed epidemiological data obtained through household studies to determine the serial interval and the temporal distribution of secondary infections within households. By analyzing multiple studies simultaneously, we were able to increase precision and to show the consistency of the observed data. To our knowledge, this is the largest analysis to date of data from linked pairs of dates of the onset of ILI or ARI clinical symptoms in clinically apparent infectors and infectees. Our methods provide the estimated proportion of within-household transmission risk that remains following a particular number of days after symptoms, which, when combined with information on severity of the virus, informs policymakers of the impact of specified periods of isolation. Future methodological research priorities for this type of analysis include developing approximate methods for correcting for tertiary transmission in the absence of household-level data.
The pH1N1 Household Investigations Working Group
The pH1N1 Household Investigations Working Group: Adebowale Awosika-Olumo, Crystal Beasley, Achuyt Bhattarai, Matthew Biggerstaff, Dianna Blau, Karen Chronister, Fatimah Dawood, Ryan P. Fagan, Anne Marie France, Gale Galland, Michele Ginsberg, Matt Gladden, James Hadler, Jenifer Jaeger, Annie Kao, Paula Kriner, Hyewon Lee-Han, Karla Lopez, Michael Lynch, Azarnoush Maroufi, Christine Mattson, Ma'ria E. Moll, Stephen Ostroff, Byron Oujesky, Diane Rexin, Margarita Rios, Carol Roach, Trudi Shim, Kirstin Short, and David Sugerman.