Our analysis of data from City of Milwaukee households supports the following conclusions: 1) Early oseltamivir use was associated with approximately a 42% reduction in the odds of at least one secondary case in the household, though this finding was not statistically significant and may be affected by some of the biases described below 2) Children had a statistically significantly higher risk of becoming infected than adults. 3) Household size greater than 3 was associated with an increased risk of at least one secondary case in the household, though this effect was statistically significant only in the univariate analysis and could be impacted by a selection bias. No association between individual risk of a symptomatic secondary infection and household size was found.
Observational studies of household transmission are crucial for rapid evidence-gathering in a newly emerging influenza pandemic, but they are also susceptible to a number of biases intrinsic to observational studies, compounded by the sometimes formidable logistical challenges of gathering accurate data during a local public health response to a pandemic. We shall now present several potential sources of biases and our attempts at addressing them.
A key potential source of bias is a form of selection bias [11
], related to the fact that not all infected households were recorded by the City of Milwaukee Health Department, since not all infected persons came to medical attention and received testing. One might expect that some households would come to medical attention only after having secondary cases in the household, and that this may have occurred several days after the onset of symptoms in the index. Such households would by definition have a secondary case but no early oseltamivir use (since the index case was not initially diagnosed). We have attempted to correct for this form of bias by excluding households for which the index case did not have a case report form, since these would be households in which the secondary cases would be most likely to have triggered the inclusion of the household in the study. As expected, this moved the univariate odds ratio for early oseltamivir use and secondary transmission toward the null value of 1, from 0.28 to 0.47. Additionally this selection bias should correlate with a larger household size. Since household size was included in the multivariate analysis on the household level and was found to be positively associated with having at least one secondary case (see Table ), this should further reduce the bias.
Another potential source of bias is related to the inclusion criteria for data completeness in the primary cohort which have differentially excluded households with different outcomes. For instance, oseltamivir receipt with a missing date or a missing date for symptom onset could cause exclusion from the primary analysis cohort. Although our study is a cohort study, this form of selection bias is analogous to that present in all case-control studies, in which the outcome influences the probability of inclusion in the study; however this alone does not bias the odds ratio. To see that, consider the odds of inclusion into the primary analysis cohort for households with vs. without secondary attack given index oseltamivir, and the corresponding odds given no index oseltamivir. It is a reasonable first approximation that those 2 odds (2 ratios of probabilities) are the same, because the probability that index oseltamivir start date is recorded should not differ depending on whether there was a secondary case in the household. Hence, in this approximation, the odds ratio for inclusion is 1, or, in other words, the odds ratios in the original population and in the group analyzed in this study are the same.
The associations between early oseltamivir usage and reduced risk of secondary attack seen in univariate analyses were attenuated slightly in multivariate analysis. This attenuation reflects a positive correlation between smaller household sizes (which are also protective) and early index oseltamivir usage in the data.
We do not believe that oseltamivir chemoprophylaxis contributed to a significant upward bias in our estimate of the benefit of early oseltamivir index usage. In fact, out of the 24 households with index oseltamivir usage on days 0-1 and no secondary attack, only 2 had some contact chemoprophylaxis.
We found that the time from index onset to the onset of the first secondary case in the household had a mean of 3.32 days and a standard deviation of 2.23 days, broadly consistent with some previous estimates [12
] and slightly longer than some other estimates for the serial interval for H1N1 [14
]. Moreover the serial interval estimated from the first secondary case in this way will be somewhat shorter than the mean time at which a case will cause secondary infections [17
], but this bias is relatively minor as there a rarely more than two cases in a household.
We have chosen a case definition for secondary cases that is slightly different from the Centers for Disease Control and Prevention surveillance definition of influenza-like illness (ILI) as fever plus cough or sore throat. A similar distinction was made in [14
], both of which made estimates of the secondary attack rates for acute respiratory infection similar to our estimate. Influenza A/H1N1 occasionally leads to mild self-limiting illness, and not all the confirmed index cases had ILI. 9 of 55 contacts in the primary analysis cohort who met our definition of secondary cases failed to meet the standard ILI definition, and 3 of these individuals were tested and virologically confirmed.
In our sample, 21% (29/135) of index cases received oseltamivir within the first two days of symptoms. This likely represents an upper bound on the true frequency within the population, since inclusion in our study depends on coming to medical attention. It is therefore likely that little impact of oseltamivir treatment on total transmission of influenza occurred during the spring in Milwaukee.
Oseltamivir is thought to be effective in reducing transmission of seasonal influenza in households and other settings [3
]. Our findings suggest that it is comparably effective in the setting of H1N1pdm infection in households. Given the biases inherent in observational studies conducted in the middle of an emerging pandemic, randomized studies are needed to assess the magnitude of this effect more precisely. The United States has limited supplies of oseltamivir, approximately enough for ¼ of the population to obtain one course. While this is larger than the supply of virtually any developing country, it is less than the stockpile held by some other wealthy countries, and less than most models predict would be necessary to undertake a sustained effort to reduce transmission with antiviral treatment [19
] or chemoprophylaxis. In countries with a larger stockpile, these findings may support the consideration for use of treatment as a transmission-reduction measure, but the findings also suggest that early treatment of symptomatic cases is essential if transmission-reduction is to succeed. Given that the benefits of oseltamivir for the treated patient are best documented (for seasonal influenza) when treatment is given within 48 hours of symptom onset (though later treatment can also be beneficial, [20
]), the results here, combined with considerations of individual benefit of early treatment in reducing disease severity [4
], suggest that additional efforts are needed to ensure timely access to antiviral drugs.