Transmissibility of influenza viruses in households

Abstract

Influenza viruses cause a substantial burden of morbidity and mortality worldwide. Household studies have been used to study the epidemiology of influenza for decades. One more recent development is the use of household transmission studies with short-term follow-up rather than following up over an entire season.

Using data from household transmission studies done in Hong Kong, I explored the role of two biological variables on transmission, specifically antibody titers measured by hemagglutination-inhibiting (HAI) or microneutralization (MN) assays and viral shedding measured by reverse-transcription polymerase chain reaction (PCR) test on nose and throat swabs. To explore the relationship between antibody titers and protection, I used individual-based hazard models that can account for the household transmission dynamics. I developed a Bayesian Markov chain Monte Carlo (MCMC) algorithm to impute the missing antibody titers and estimate the model parameters jointly. To explore the relationship between viral shedding and infectivity, I first fitted a log-linear random effects censored regression model to reconstruct the individual viral shedding trajectory to overcome missing observations of viral shedding. Then I extended the abovementioned household transmission model to incorporate with data on viral shedding to compare the consistency between pattern of viral shedding and timing of secondary infection in households. I discovered that an HAI titers of 1:40 against seasonal influenza A(H1N1) and A(H3N2) viruses were associated with 31% (95% CI: 13% to 46%) and 31% (CI: 1% to 53%) protection respectively, which were much lower than 50% protection in other settings. I also demonstrated the association between MN titers and protection. The pattern of viral shedding trajectories was not particularly consistent with the timing of secondary infection and individual viral shedding was at most weakly associated with individual infectivity. To estimate the cumulative risk of infection of 2009 pandemic influenza virus outbreak from a separate community-based cohort study with multiple sera collections, I developed a Bayesian hierarchical model to account for the timing of sera collection and missing of sera collection in the study. I found that using all of the collected sera in the analysis could substantially improve the precision of estimation, compared with a simple analysis that only used the proportion of 4-fold or greater rise in participants as estimates. The association between an HAI titer of 1:40 and protection was substantially less than 50% in households, perhaps due to the intense and frequent contacts between persons in households. Viral load measured by RT-PCR in the nose and throat was as most weakly correlated with individual infectivity in households, and other correlates of infectivity should be examined in future studies. The statistical framework that I developed to account for the timing of sera collection can provide robust estimates of cumulative risk of infection during an epidemic.