This study, notable in its consideration of a broad range of respiratory pathogens in addition to influenza, demonstrates that household transmission of ILI is most strongly associated with host and demographic factors: female sex and the presence of children within the household (Tables 1 and 2).
The observation that female sex may be associated with increased transmission in the absence of children (RR = 2.33 (0.919, 5.90), p = 0.059) is novel, perhaps suggesting that females are fundamentally more infectious, and not simply more connected to children (in terms of both their susceptibility compared with males if a child introduces infection, and their infectiousness to children if they are the primary household case). Behavioural differences whilst ill may drive such an observation. Alternatively, mechanisms by which influenza pathogenesis is sex dependent have been investigated ; whether or not differences extend to infectiousness and susceptibility is not clear. Barbara et al. have recently identified that the reporting of respiratory symptoms may be linked with risk perception  and hence gender . Clearly, we cannot exclude the possibility of gender difference in the reporting of within household transmission.
The association between transmission and the presence of children within the household is consistent with many other studies [18, 22, 23, 29, 36]. The logistic and Poisson model findings (Tables 1 and 2) are consistent with an increased susceptibility for children. This is further supported by the observed increased SHAP in children compared to adults (0.216 compared to 0.110, a risk-ratio of 1.95 (1.19, 3.19). The SHAP in adults did not differ by whether or not their household contained children, suggesting that other ‘indirect’ effects of children are less likely. As our study design limited the analysis to household events with an adult introducer, we were unable to assess the hypothesis that children may be more infectious than adults.
Our Poisson regression analysis on the number of secondary cases given that the participant was the primary case (Table 2) indicates that isolation of influenza in the introducer of infection to the household is associated with an increase in the number of secondary cases. We explored this finding more deeply using a statistical interaction model. While somewhat limited by sample size, we found that in placebo recipients identification of influenza was not significantly associated with an increase in the number of secondary cases (IRR = 1.69 (0.421, 6.80), p = 0.459), while in vaccine recipients the IRR (relative to identification of any other virus, including ‘none’) was 5.24 (2.17, 12.6), p < 0.001. Note that our previous analysis confirms that vaccination is associated with a reduced probability of influenza virus identification . Additionally, ‘breakthrough’ influenza cases have similar symptoms compared to unvaccinated individuals . We therefore suggest that our finding of increased transmission may be explained by infection with an influenza virus mismatched to the vaccine-strain (known to be in circulation during the year of study ), which furthermore may be relatively antigenically novel and to which household members may be expected to have heightened susceptibility. With no virological samples available from other household members and the small number of vaccinated participants who were infected with influenza we are unable to explore this hypothesis further.
Across all virus types isolated and all household ILI events, and assuming that all secondary cases within the household are directly infected by the introducer, we calculate a serial interval of 6.0 days. Restricting to events in which the participant was the primary case and in which influenza was isolated, we calculate a serial interval of 4.5 days. This simple approach, as taken by others [22, 29], cannot account for two important factors: community importation and infection of household members by other non-introducing members (i.e. tertiary cases). While others have partially accounted for these effects [25, 29, 41], a mechanistically-motivated statistical model is required to fully account for such possibilities, for example as introduced by Cauchemez et al.  who determined a serial interval for influenza of 2.6 days (SD = 1.3) compared to 2.9 days if calculated directly from empirical observations. With just 5 events, application of these more advanced model-based techniques is not justified for our data.
Of primary interest for this sub-analysis focussed on transmission is the complication introduced by the monitoring and assessment of ILI in an individual rather than a household. Ideally, a protocol such as that suggested by Klick et al. would have been employed . The lack of virological assessment of household secondary cases and the broad nature of the question used to establish the secondary case count in each house also contributes to uncertainty with regards to our assignment of temporally associated ILI to within-household transmission. Both of these limitations were an unavoidable consequence of the nesting of the data-collection protocol within a randomized placebo-controlled trial. Furthermore, due to the requested timing for completion of the questionnaire, we cannot exclude the possibility that late onset of secondary (or tertiary etc.) cases may have been missed, particularly if a participant’s experience of symptoms was of short duration. Similarly, because the study protocol and analyses effectively assume that individuals are infectious until the end of their symptoms, any systematic differences (by virus type) in this relationship may influence the results. However the prompt to return the diary upon symptom cessation was in an effort to ensure timely reporting of questionnaire information to minimise recall bias. Conversely, our Poisson model implicitly assumes independence among household members, attributing all household infections to the primary case. More advanced model based methods that account for tertiary (and subsequent) cases and community introduction would be warranted with more complete data sources.
As with all protocols based purely on symptomatic presentation (as opposed to active surveillance for nonclinical signs of infection such as virological or immunological measures [24, 42]), we are unable to account for potential sub-clinical infection routes, with potential impact for our assessment of whether or not transmission did occur, the primary case status of our participants and determination of the size of the susceptible pool within a given household. Conversely, taking a non-simulation approach to analysis, we are unable to discount our estimate for the SHAP due to the effects of community introduction into the household, or account for community introduction and tertiary cases in our estimate for the serial interval .
Our study sample had an over-representation of females (166 of 258 (64.3%) individuals for the 279 captured episodes; 105 of 167 (62.9%) individuals for the 177 primary-participant introductions). Furthermore, it should be noted that the study population were originally volunteers in a randomized controlled trial and as such more likely to represent a group who were more concerned with their health than the general population. Eligibility was restricted to healthy adults without recognized risk factors for severe influenza infection.