Results
The general information we used, such as symptoms, severity of illness, and a preliminary analysis of the number of contacts (but not the age distribution of contacts), had been presented previously. It was observed that when healthy, people reported an average of 14.9 (95% confidence interval: 11.2, 20.1) contacts each day and, when ill, they reported 3.8 (95% confidence interval: 3.2, 4.6) contacts daily.
Here, we observed that ill people altered whom they contacted (Figure 1). As expected, a large proportion of encounters reported by participants when they were healthy were with people of a similar age as themselves, with particularly high numbers of contacts occurring among schoolchildren and young adults. Indeed, the contact patterns of participants when healthy were very similar to those observed in the POLYMOD Study (see Web Table 2 and Web Figures 1 and 2). In contrast, the strong like-with-like pattern of social mixing was not evident in encounters reported when participants were unwell. We note that the contact patterns displayed in Figure 1 may be asymmetric, since, for example, people contacted by ill participants may have been either ill or healthy.
(Enlarge Image)
Figure 1.
Daily numbers of social contacts made between age groups during the 2009 A/H1N1pdm influenza epidemic, England, 2009–2010. Contact patterns were reported by study participants who completed a contact diary on 2 occasions, once when they were asymptomatic (left; n = 140) and once when they were symptomatic (right; n = 140). Colors represent the mean number of encounters reported by each participant with contacts in each age group. There was less like-with-like mixing reported by symptomatic participants.
That illness results in a particular reduction in the number of contacts between people of similar ages can be understood when considering the social setting of encounters (Figure 2, Table 1; also see Web Appendix 3). When ill, participants reported having far fewer contacts in the work/school and leisure settings—both settings where like-with-like mixing between age groups would be expected to take place. In contrast, illness has little discernible effect on the number of contacts made at home. Although it is tempting to assume that home contacts provide an appropriate proxy for contacts made when ill, we did not find this to be the case: Comparing the reported contacts made at home with the reported contacts made during illness, we found an R0 ratio significantly different from 1 (R0 ratio = 1.42, 95% confidence interval: 1.01, 2.53). Turning to the duration of encounters, we see in Figure 2 that in most settings there was a tendency for participants to report briefer encounters when they were ill.
(Enlarge Image)
Figure 2.
Numbers and durations of encounters in different social settings during the 2009 A/H1N1pdm influenza epidemic for asymptomatic and symptomatic study participants in the paired data set (n = 140), England, 2009–2010. The blue points show the average number of encounters in each setting (right-hand axis), and the colors show the distribution of encounter durations (left-hand axis). For each pair of bars, the left-hand bar represents contacts made when asymptomatic, and the right-hand bar represents contacts made when symptomatic. In every location except home, a drop in the average number of contacts can be observed. NA, not applicable.
The altered patterns of social encounters when people are symptomatic result in large reductions in the basic reproduction number, R0. When calculated using the symptomatic contact matrix, R0 is only one-quarter of the value it takes when calculated using the asymptomatic contact matrix (Table 2). These changes also affect the expected distribution of infection during the early stages of an outbreak. As Figure 3 shows, the contact patterns of healthy participants would result in a concentration of incidence among children of primary-school age. However, the contact patterns of ill participants resulted in a greater concentration of incidence in young adults. This effect emerged because children reported far fewer contacts with other children when symptomatic. Because like-with-like mixing was less evident in ill persons, mixing among adults and between adults and children played a greater role in transmission.
(Enlarge Image)
Figure 3.
Age distribution of cases in the early stages of the 2009 A/H1N1pdm influenza epidemic as predicted by asymptomatic contact patterns (left; n = 140) and symptomatic contact patterns (right; n = 140) in the paired data set, assuming full susceptibility in the population, England, 2009–2010. The indicated 95% confidence intervals (T-shaped bars) were obtained using a nonparametric bootstrap. Because of the skewness of the underlying distributions, the mean values are not in the middle of the intervals.
Under the assumption that persons with an asymptomatic infection do not change their social contact behavior, we can use our data to estimate the fraction of cases infected by asymptomatic persons, by comparing the age distribution of incidence from general practitioners' consultation data (points in Figure 4 (both panels)) with that predicted using the ill and healthy contact data (lines in Figure 4 (right panel)). The left panel of Figure 4 shows this estimation method for different φ values (assuming qR = 1): If φ = 0, then only symptomatic persons transmit infection, and the age distribution of cases during the exponential phase of the epidemic would be given by the solid line. If φ = 1, then only asymptomatic persons transmit infection, and the age distribution of cases would be given by the dashed line. We find (see Figure 4 (right panel) and Table 3) that the observed distribution of ILI cases during the early stage of the epidemic is consistent with the majority of transmission's being due to persons with overt symptoms of ILI; we found that approximately two-thirds of the infections (Table 3, "all contacts") were caused by symptomatic persons. The best fit to the data was found when using patterns of mixing derived from all reported contacts; restricting encounters to those involving some physical contact or those lasting longer than an hour resulted in a poorer fit to the incidence data (Table 3).
(Enlarge Image)
Figure 4.
Theory (left) and fit (right) of a model to the observed age distribution from general practitioners' consultation data from England and Wales for the early phase of the 2009 A/H1N1pdm influenza pandemic (black dots), England, 2009–2010. The left-hand panel shows different φ values (assuming qR = 1, for illustration) leading to different age distributions. The right-hand panel shows the fit obtained using the best-fitting value of
from all capped contacts (the gray area surrounding it indicates the 95% confidence interval, bootstrapping for
), from skin-to-skin contacts, and from long-duration contacts (>1 hour).
Using the contact matrices describing all contacts, the best-fitting value for
was 5.86. It has been suggested that approximately two-thirds of infections involved any symptoms of ILI, which leads to qR ≈ 2.93, implying that persons with symptomatic infections are about 3 times as infectious per social contact as persons with asymptomatic infections. If it were the case that one-third of infections were symptomatic (e.g., considering only infection that causes fever ), then qR would be estimated to be 11.72. If half of the infections were symptomatic, then qR would be estimated to be 5.86.