Social Networks and Transmission and Control of Influenza
Social Networks and Transmission and Control of Influenza
We have quantified the marked effect that being infected with ILI has on people's social mixing behavior. People report significantly fewer social encounters when they have influenza symptoms. Not only does the number of encounters fall, but the patterns of contact change; people with ILI have fewer contacts at work/school and during leisure activities, which results in fewer contacts in their own age group. This can be understood as a result of people taking time off from work/school and avoiding social activities when ill.
We have found that patterns of incidence during the early stages of the influenza A/H1N1pdm pandemic in England and Wales suggest that symptomatic persons generated the majority of cases. Therefore, there is scope to focus intervention resources on targeting symptomatic persons, encouraging efficient treatment, and recommending time off from work and social distancing for persons with symptoms. It appears that during the A/H1N1pdm epidemic, symptomatic persons were, in general, sufficiently unwell that they moderated their social behavior; had a substantial fraction of them continued to engage in normal social activities, there would have been a large increase in transmission.
R0 is a common measure of viral fitness. Using this measure, we have demonstrated that there is a considerable fitness cost associated with illness through the disruption of social networks. Changes in behavior, in our sample, were correlated with severity of illness (measured by number of symptoms reported). However, the presence of symptoms is also correlated with higher transmissibility: If, as seems plausible, between one-third and two-thirds of A/H1N1pdm infections were symptomatic, then our calculations suggest that symptomatic infections were 3–12 times as infectious as asymptomatic infections. Therefore, there appears to be a trade-off between infectiousness and opportunities for transmission. Symptomatic persons are more infectious per contact but make fewer contacts than asymptomatic persons.
Accessing symptomatic persons is challenging and, as with any such study, it is possible that the participants in this study were not representative of all people with ILI. However, the fact that patterns of incidence during the early stages of the epidemic were so well predicted by the mixing patterns reported by participants suggests that we succeeded in capturing true epidemiologically relevant behavior. Furthermore, when healthy, the participants had mixing patterns very similar to those found in other social-contact studies, suggesting that the sample was reasonably representative in terms of normal social mixing behavior. We note that the social-contact data used here were collected only in England, while data from other sources have covered all of Great Britain or England and Wales; however, in all cases, the majority of the information came from England, so we are confident that the differences in sampling locations did not affect our conclusions.
The potential limitations of the diary-based approach are well known. Collecting data from young children is difficult, since proxies have to be used and people (of any age) may not record all of their contacts. Electronic methods can reduce these problems, but such methods require very high rates of participation in order to record most contacts and are therefore not suitable for a general population survey. Different biases could appear when participants are unwell, which could have affected our results. For example, diaries could be filled in better when people are ill, since fewer contacts are easier to remember and record, or they could be filled in less well, since illness may make participation more burdensome. We were unable to quantify the magnitude of such effects.
In this analysis, the duration of infectiousness was assumed to be equal for symptomatic and asymptomatic persons. The influences of this assumption have not yet been investigated, since they were beyond the scope of this initial paper, but will be the focus of further work. We assumed that infected persons are behaviorally "asymptomatic" or "symptomatic" throughout their infectious period. Alternatively, it is possible that people with symptoms change their social mixing behavior some time after symptom onset or that people are infectious for some time before symptoms appear. Exploring the impact of these possibilities will be the focus of ongoing work; in the models presented here, in the absence of good data with which to link infectiousness, symptom onset, and behavior change, we have taken the simpler approach.
To our knowledge, this study was the first to assess the individual-level impact of infection on social contact patterns and the consequences of altered social encounters on epidemic transmission. Even though the 2009 pandemic influenza strain was relatively mild, the change in contact patterns induced was large and epidemiologically significant. Epidemic models that fail to take these changes into account are likely to be inaccurate—even more so if disease severity is high. Symptomatic persons appear to have been responsible for most of the infection transmission, and efforts to reduce influenza transmission should target them. Indeed, should this result hold for other strains of influenza, identifying, treating, and isolating symptomatic individuals should be the focus of public health efforts in order to prevent transmission to others in the community.
Discussion
We have quantified the marked effect that being infected with ILI has on people's social mixing behavior. People report significantly fewer social encounters when they have influenza symptoms. Not only does the number of encounters fall, but the patterns of contact change; people with ILI have fewer contacts at work/school and during leisure activities, which results in fewer contacts in their own age group. This can be understood as a result of people taking time off from work/school and avoiding social activities when ill.
We have found that patterns of incidence during the early stages of the influenza A/H1N1pdm pandemic in England and Wales suggest that symptomatic persons generated the majority of cases. Therefore, there is scope to focus intervention resources on targeting symptomatic persons, encouraging efficient treatment, and recommending time off from work and social distancing for persons with symptoms. It appears that during the A/H1N1pdm epidemic, symptomatic persons were, in general, sufficiently unwell that they moderated their social behavior; had a substantial fraction of them continued to engage in normal social activities, there would have been a large increase in transmission.
R0 is a common measure of viral fitness. Using this measure, we have demonstrated that there is a considerable fitness cost associated with illness through the disruption of social networks. Changes in behavior, in our sample, were correlated with severity of illness (measured by number of symptoms reported). However, the presence of symptoms is also correlated with higher transmissibility: If, as seems plausible, between one-third and two-thirds of A/H1N1pdm infections were symptomatic, then our calculations suggest that symptomatic infections were 3–12 times as infectious as asymptomatic infections. Therefore, there appears to be a trade-off between infectiousness and opportunities for transmission. Symptomatic persons are more infectious per contact but make fewer contacts than asymptomatic persons.
Accessing symptomatic persons is challenging and, as with any such study, it is possible that the participants in this study were not representative of all people with ILI. However, the fact that patterns of incidence during the early stages of the epidemic were so well predicted by the mixing patterns reported by participants suggests that we succeeded in capturing true epidemiologically relevant behavior. Furthermore, when healthy, the participants had mixing patterns very similar to those found in other social-contact studies, suggesting that the sample was reasonably representative in terms of normal social mixing behavior. We note that the social-contact data used here were collected only in England, while data from other sources have covered all of Great Britain or England and Wales; however, in all cases, the majority of the information came from England, so we are confident that the differences in sampling locations did not affect our conclusions.
The potential limitations of the diary-based approach are well known. Collecting data from young children is difficult, since proxies have to be used and people (of any age) may not record all of their contacts. Electronic methods can reduce these problems, but such methods require very high rates of participation in order to record most contacts and are therefore not suitable for a general population survey. Different biases could appear when participants are unwell, which could have affected our results. For example, diaries could be filled in better when people are ill, since fewer contacts are easier to remember and record, or they could be filled in less well, since illness may make participation more burdensome. We were unable to quantify the magnitude of such effects.
In this analysis, the duration of infectiousness was assumed to be equal for symptomatic and asymptomatic persons. The influences of this assumption have not yet been investigated, since they were beyond the scope of this initial paper, but will be the focus of further work. We assumed that infected persons are behaviorally "asymptomatic" or "symptomatic" throughout their infectious period. Alternatively, it is possible that people with symptoms change their social mixing behavior some time after symptom onset or that people are infectious for some time before symptoms appear. Exploring the impact of these possibilities will be the focus of ongoing work; in the models presented here, in the absence of good data with which to link infectiousness, symptom onset, and behavior change, we have taken the simpler approach.
To our knowledge, this study was the first to assess the individual-level impact of infection on social contact patterns and the consequences of altered social encounters on epidemic transmission. Even though the 2009 pandemic influenza strain was relatively mild, the change in contact patterns induced was large and epidemiologically significant. Epidemic models that fail to take these changes into account are likely to be inaccurate—even more so if disease severity is high. Symptomatic persons appear to have been responsible for most of the infection transmission, and efforts to reduce influenza transmission should target them. Indeed, should this result hold for other strains of influenza, identifying, treating, and isolating symptomatic individuals should be the focus of public health efforts in order to prevent transmission to others in the community.
Source...