In this study, we have analyzed the household-stratified early infection patterns for pandemic influenza in inner-city Birmingham, UK. We have used modern computationally-intensive statistical methods to fit a realistic model for transmission, and our comprehensive literature search [see Additional file 1] indicated that our approach to modelling the case ascertainment of influenza A(H1N1)pdm09 is novel and provides valuable additional information.
Three key conclusions can be drawn from our work. First, the level of within-household transmission can be estimated directly, despite difficulties in case ascertainment. An estimate of this quantity is important if antivirals are distributed prophylactically to household contacts of cases; if there is little transmission within the household then such a policy is less likely to be effective and vice versa.
Considering our results, we arrive at a 'rule of thumb' for the H1N1pdm09 pandemic that transmissibility lies somewhere between what would be predicted from the HPA definition of ILI (swabbed) and a less specific reporting of ARI symptoms (symptomatic). Our results therefore provide evidence that relying solely on laboratory-confirmed cases is excessively stringent and consistently leads to under-estimation of transmission, as would be expected from serological work [17, 18]. An additional consequence of relying on laboratory confirmation is that given this case definition, the transmission probabilities do not decline swiftly as household size increases, while our full model shows a reduction in transmission probability as household size increases, as expected, with the exception of household size seven (we did not find any direct cause for this anomaly). The question of the relative importance of large households for epidemic spread remains significant, and while pre-pandemic analysis of seasonal influenza suggested decline with size [19, 20], this was not a consistent observation during the pandemic as seen in  and our literature review [see Additional file 1]. Study design may be an important part of variability , and our results show that case ascertainment is also relevant.
As part of estimating the transmission process, we also calculated the probability of a false negative PCR result. Forty-two percent of infected cases are estimated to have had a negative laboratory test, which has significant public health importance, and may have been caused by a combination of a number of factors including: problems encountered with taking the swabs [22, 23]; swabbing individuals who were not in the early stages of their illness ; and potentially swabbing individuals with milder forms of illness. Exploring these factors could also be the focus of future work.
Secondly, our analysis provides additional support for the picture of the recent influenza pandemic as one with highly variable clinical outcomes, including significant numbers of cases who did not meet the HPA's diagnostic criteria, but are likely to have been true cases, and a high variance in the infectiousness of cases, that is there were many cases who were not particularly infectious, while a relatively small minority had an extremely high probability of passing influenza on to their household contacts.
Finally, and most significantly, our approach could be used in future outbreaks as a rapid complement to serological work. Serology provides an important independent test of clinical surveillance methods, but is costly and the correct epidemiological interpretation of an individual's titre is not always clear. Our methods are inexpensive and model the epidemiology of disease transmission directly, giving the potential for an early snapshot of the proportion of cases ascertained.
While we have given certain questions priority in our analysis, as is unavoidable, there are factors that were not captured in our model. We believe that the stratification of cases by age is the most significant omission from our analysis, while other potentially important factors are estimation of between-household transmission and the efficacy of interventions such as encouragement of personal hygiene measures and use of antiviral drugs. In general, inclusion of these additional complexities will lead to stratification of our transmission estimates by age, time to treatment and prophylaxis and so on, in addition to household size, but these may still on average be similar to our unstratified estimates. The expectation from our literature review would be for lower transmission among those given antivirals early and adults than those given antivirals late and children, but the often subtle effects of transmission dynamics mean that this can only be conjectured in the absence of a full analysis.
Ultimately, our ability to extend the model relies on sufficient data being available. Our data are of good quality, but still only contain a finite amount of information. Furthermore, as highlighted there is some missing data on which individuals were managed after 18 June, and therefore treated on the basis of clinical suspicion rather than swabbed. In our review of the literature on household transmission of pandemic influenza [see Additional file 1] we found many studies, involving between them several thousand cases and household contacts, that produced relevant data. Of these, only a small fraction fitted a transmission model to extract generalizable epidemiological conclusions.
The current UK Influenza Preparedness Strategy stresses the need for rapid research early in a pandemic to improve understanding and inform response, and to develop appropriate protocols for such research . We suggest that protocols for collection, sharing and meta-analysis of household data should form part of this preparedness. The data for the studies we found were mostly collected before the end of June 2009. Much of the information collected during these studies, in particular syndromic information and household stratification, was not reported and used at the time. An internationally co-ordinated meta-analysis of household data during July 2009, fitting transmission parameters and adjusting for case definitions so that meaningful comparisons could be made across different demographic and healthcare contexts, could have provided useful information about the pandemic at relatively low cost. In particular, these estimates of disease transmission could be used in a timely fashion to guide changes in public health management strategies, which in England in 2009 were made only in areas where there was evidence of sustained community transmission.