Subject recruitment and selection

Full details of subject recruitment and selection for the primary phase IV vaccine efficacy trial have been described previously [35]. Briefly, across 23 study sites in Australia and New Zealand, 7544 healthy adults aged ≥18 to <65 years were recruited for a placebo-controlled trial of a licensed trivalent influenza vaccine (Fluvax®, CSL Ltd) in 2008 (Clinicaltrials.gov #NCT00562484). Study participants were randomized to receive either placebo or vaccine in a 1:2 ratio prior to the southern hemisphere 2008 influenza season. From an available pool of 5624 participants from the primary study at 12 study sites, we consider the 581 persons (adults) who experienced at least one ILI episode – meeting the case definition of at least one respiratory symptom (cough, sore throat, runny nose or nasal congestion) and at least one systemic symptom (fever (oral temperature ≥ 37.8°C), feverishness, chills or myalgia) [35], and who provided written informed consent for participation in the nested cohort-study which required contribution of a valid biological sample (COPAN^TM dry flocked swab). Samples were tested for a range of respiratory viruses using a combination of multiplexed and uniplexed conventional and real-time polymerase chain reaction (PCR) assays [35]. Of those, 258 were members of multi-occupancy households, allowing investigation of transmission within the household. Using a non-specific and sensitive ILI definition, they reported an episode of ILI on 279 occasions. For each episode they returned a study questionnaire (Additional file 1) detailing respiratory symptoms (see [35] for details), health seeking behaviour (health care provider consultations, hospital admission, time off work), household characteristics (number of adults (≥18 years) and children (<18 years)) and temporally associated symptoms of ILI (if any) in other household members. From herein, we consider the illness episode as the primary unit of analysis.

The primary phase IV vaccine efficacy trial and nested sub-study were approved by Ethics Committees at all study sites: Royal Children’s Hospital Ethics in Human Research Committee (Victoria), Princess Margaret Hospital for Children, Ethics Committee (Western Australia), Redcliff-Caboolture Health Service District (Queensland), Human Research Ethics Committee Network (Tasmania), Children, Youth and Women’s Health Service Research Ethics Committee (South Australia), Bellberry Ltd (South Australia, Queensland), Cairns District Health Service Cairns Base Hospital Ethics Committee (Queensland).

Outcome measures and analysis

Virology results were classified into 5 virus groups [35]: influenza (influenza A, influenza B), coronaviruses (OC43, 229E, NL63, HKU1), picornaviruses, other viruses (parainfluenza viruses (1, 2, 3), adenoviruses, human metapneumovirus (hMPV), bocaviruses, RSV and KI and WU polyomaviruses) or none, where none indicates that no ‘tested-for’ virus was detected in the participant’s sample, as opposed to a missing sample or inconclusive result. Study participants’ vaccination status (as determined by the primary phase IV trial intervention), physical location (i.e. study site), and socio-demographic characteristics were also recorded.

The relevant outcome measure for this sub-analysis was evidence of transmission within households based on experience of symptomatic illness in at least one other member of the study participant’s household. Study participants were asked to complete the diary on the day following cessation of their own symptoms. They recorded the date of onset of symptomatic ILIs in household members from between 14 days prior to the study participant’s illness through to the day of diary completion. Note that the primary case in the household may or may not be the study participant.

For household ILI events in which the participant was the primary case (183 of 279), transmission may or may not have occurred in the household and so it is statistically valid to develop univariate and multivariate explanatory models. For participants who reported recurrent ILI episodes during the study in which the same virus was isolated, we exclude all but the first episode. Events with co-introduction, defined as onset of symptoms in the participant and one or more household members on the same day, were also excluded. Following these exclusions, 177 episodes remained for analysis.

For household ILI events in which the participant was not the primary case (95 of 279), while we do know who the introducer was for these events (via the questionnaire data), other household ILI events initiated by that introducer that did not involve the participant are unobserved. That is, any events in which a child (or for that matter, any other adult member of the household) introduced an infection that did not infect the study participant are not captured by the study protocol. This observation necessitates the exclusion of all household ILI events in which the participant was not the primary case from the analyses.

In the one remaining household ILI event, the status of the participant (primary or not) was unknown, so the episode was excluded from the analysis.

For the 177 episodes in which the participant was the primary case, logistic regression models were used to explore associations between host, demographic or virus variables with any observation of within-household transmission (outcome variable = transmission in household for each recorded ILI episode in a study participant). The secondary household attack proportion (SHAP) was calculated as the proportion of potentially exposed household members (assumed susceptible) experiencing illness, averaged over all recorded episodes. We present descriptive statistics for the SHAP and its variation by virus, participant and demographic variables. Poisson regression models were used to assess the influence of virus, participant and demographic variables on the number of secondary cases within a given household, offset against the number of potentially exposed household members (outcome variable = number of secondary cases in household for each recorded ILI episode in a study participant).

Vaccination status of participants was not included in the primary logistic and Poisson statistical analyses due to its known mitigating effect on the likelihood of influenza acquisition [35]. Investigation of the influence of prior immunisation on influenza transmission in ‘breakthrough cases’ was explored in a secondary analysis by inclusion of a statistical interaction term between vaccination and influenza-identification status.

We make an empirical calculation of the mean time between the onset of symptoms in the primary case and the onset of symptoms in the household contacts (the serial interval), for all household ILI events, events in which the participant was the primary case, and events in which influenza was isolated from the participant’s virological sample.

All statistical analyses were conducted in Stata/IC 11.1.

Factors associated with any transmission within the household

The study participant was the primary case in 183 of the 279 recorded household ILI events. One re-introduction of picornavirus into the household and five co-introductions were recorded, leaving 177 episodes for analysis. Transmission occurred in 39 (22.0%) episodes. Stratifying by presence of child (Figure 1), transmission occurred in 17 of 54 (31.5%) households in which at least one child was present and 22 of 123 (17.9%) households without children, a risk-ratio (RR) of 1.76 (1.02, 3.04), p = 0.045.

Female sex of the participant was associated with increased transmission: 30 of 108 (27.8%) episodes in women compared to 9 of 69 (13.0%) episodes in men, a risk ratio of 2.13 (1.08, 4.21), p = 0.021. The association with female sex remained (at borderline significance) when restricted to households without children (123 episodes): 23.3% transmission compared to 10.0% for males, a risk ratio of 2.33 (0.919, 5.90), p = 0.059. The risk ratio for female sex in households with children (54 episodes) was 1.76 (0.668, 4.66), p = 0.224.

There is marked co-linearity between the variables ‘presence of child in household’, ‘age-category’ and ‘household size’. For example, respondents aged 35 – 44 years had significantly greater odds of having a child in the household than those aged 18 – 24 years (OR 53.2 (13.0, 217)), while no participant aged more than 55 years lived with a child. The relationship between the household size distribution and presence or absence of children is depicted in Figure 2. We retained ‘presence of child in household’ in the final multivariate model for transmission due to its strong predictive role, intuitive appeal, presumed causal role in our observed (univariate) association with age-category, and previous research indicating an association between transmission and children [18, 22, 23, 29, 36]. In the multivariate model, the observed increased risk of transmission with female sex remains (OR = 2.45 (1.01, 5.93), p = 0.047). Presence of children in the household is both the strongest and most statistically significant factor associated with transmission (OR = 2.63 (1.18, 5.88), p = 0.018).

Factors associated with the number of secondary household infections

Within 258 multi-occupancy households, 177 primary-participant introductions gave rise to 54 secondary cases among 391 potentially exposed individuals, a secondary household attack rate (SHAP) of 0.138. Of 102 exposed children, 22 developed ILI (SHAP = 0.216) compared to 32 of 289 adults (SHAP = 0.110), a risk-ratio of 1.95 (1.19, 3.19), p = 0.012. The adult SHAP did not differ by presence of children in the household (0.117 vs. 0.109, RR = 1.07 (0.486, 2.35)).

In households in which the participant was female, 41 secondary infections were reported among 238 exposed household members (SHAP = 0.172), compared with 13 secondary cases among 153 contacts in households in which the participant was male (SHAP = 0.085), a risk-ratio of 2.03 (1.12, 3.66), p = 0.016 (2-sided Fisher’s exact). In households with children, 27 secondary infections were reported among 162 exposed household members (SHAP = 0.167), compared with 27 secondary cases among 229 contacts in households without children (SHAP = 0.118), a risk-ratio of 1.41 (0.863, 2.32), p = 0.182.

In a secondary analysis, we considered the influence of prior vaccination on the reported number of secondary household cases among participants testing positive for influenza compared with all other participants. In a Poisson model for secondary attacks including a statistical interaction between influenza detection (true/false) and vaccination status (placebo/vaccine), the IRR for influenza positive cases in those receiving placebo was 1.69 (0.421, 6.80), p = 0.459. The factor increase (interaction term) for the IRR for vaccinated participants was 3.10 (0.608, 15.8), p = 0.174, yielding a net IRR for vaccinated influenza-positive participants relative to vaccinated influenza-negative participants of 5.24 (2.17, 12.6), p < 0.001).

Serial interval

Under the simplifying assumption that the introducer of infection into the household is responsible for all subsequent infections, we may calculate an empiric serial interval, the time from symptom onset in one individual until symptom onset in another. We first consider infections to be related if symptoms are reported within 14 days following onset in the primary case. Across all virus-type isolations, we calculate a mean serial interval of 6.0 days (SD = 3.6) for all household ILI events (where the study participant was the primary case or otherwise), and 5.1 days (SD = 3.2) for the events in which the participant was the primary case. For the five events in which the primary participant had virologically confirmed influenza and transmission occurred, the mean serial interval was 4.5 days (SD = 1.6). If we limited the maximum serial interval to seven days, the mean was reduced to 4.0 days (SD = 1.7) for all household ILI events and 3.9 days (SD = 1.9) for events in which the participant was the primary case. No gaps of greater than seven days occurred for the five events in which the primary participant had virologically confirmed influenza and transmission occurred.

Sensitivity analysis

In our main analysis, we made two assumptions that we now subject to a sensitivity analysis. Of the 183 events in which our participant was the primary case, 6 were classified as co-introductions as (at least) one other household member recorded symptoms beginning on the same day. As the latent period for respiratory infections may vary from individual to individual, here we exclude a further 4 episodes in which there was a 1 day interval from onset of symptoms in the study participant to onset of symptoms in another household member. The resulting multivariate models (equivalent to Tables 1 and 2) are materially unchanged, with the expected slight reduction in statistical power (data not shown).

A second assumption made was that, for participants who reported multiple ILI episodes during the study period, we only excluded the latter ILI episode where the same respiratory pathogen was isolated on both occasions. However, if we conservatively exclude all ILI episodes except for the first (10 episodes excluded (by virus type: 6 “none”, 1 “picornavirus”, 2 “influenza”, 1 “coronavirus”)), again we find no material change in either the logistic or Poisson analyses (data not shown).