Geodemographics profiling of influenza A and B virus infections in community neighborhoods in Japan

Influenza is a highly contagious acute respiratory disease that causes periodic seasonal epidemics and global pandemics, and shows marked seasonality in many countries [1, 2]. The spread of influenza viruses in a community is influenced not only by the type of virus [3, 4], but also by factors such as age [5, 6], immunological conditions of person [7, 8], climate [9, 10], indoor crowding [11, 12], school activity [13‐15], and social contacts [16‐22].

Recently, the geographical spread of seasonal influenza was investigated with the aid of geographic information systems (GIS) [23‐27]. We found that the spread of influenza in Japan showed a particular pattern every year from western-central Japan to northeast Japan [25].

Geodemographics is widely defined as "analyses of people by where they live", and is constructed by linking classified neighborhoods [28] and some indices of interest such as economy, health, crime, or education. One of the commonly used commercial geodemographics tools is Mosaic, which was originally developed in the UK. The Japanese version of this product classifies 0.2 million census districts into 11 Groups and 50 Types by clustering socio-economic and demographic variables.

Until now, no reports have focused on the relationship between the incidence of influenza and characteristics of small neighborhoods in a community. We started to map the incidence of influenza at the census enumeration district level in a local city, Isahaya City in Nagasaki Prefecture, Japan using an influenza patient registration program conducted by the Isahaya Medical Association since 2004. Nearly all pediatric and internal medicine outpatient clinics joined the project.

By combining influenza mapping and commercial geodemographics segmentation (Mosaic Japan) at the small enumeration district level, we aimed to profile neighborhoods where influenza frequently or infrequently occurs. Using these results, we speculated the socio-demographic factors affecting the transmission of influenza in a community.

To the best of our knowledge, this is the first study to correlate influenza occurrence in a local community with geodemographics data. We found that the incidence of influenza A and B in the neighborhood group "Burdened Optimists" (Mosaic Group G) was 10-40% higher than expected in the study area (Table 3). This group consists of parents in their 30s and 40s living with their children (Table 1 and Additional file 2). Supporting this finding, the "Company Town" (Mosaic Type G28) the subcategory of this group, where many families in their 30s to mid 40s live with children aged 0-14 year (Additional file 2), was approximately 100-230% higher than expected (Additional file 1, Table S1). On the contrary, the incidence of influenza A and B in neighborhood groups with an aging society in rural areas where the proportion of elderly citizens was high (Additional file 2), the "Rural Fringe" and "Deeply Rural" groups (Mosaic Groups J and K) was 20-50% lower than that expected, a difference that was statistically significant (Table 3).

This finding was a reflection of the higher incidence in children and lower incidence in the elderly for influenza A and B drawn from age group analysis in the entire area, but the results tended to be similar even after age adjustment. Therefore, it was suggested that the clustering of children in young families was a cause for the higher transmission of influenza. Children in households play a key role in influenza transmission, and we assume that the parents in their 30s and 40s are also relatively susceptible to influenza compared to the elderly due to greater chances of contact with children and a lesser chance of having a history of past infection.

Population density is also another factor that affects influenza transmission in neighborhoods. Influenza incidence tended to be higher in the "Social Housing Tenants" groups that had the highest population density and with many small children, and the incidence was lower in the sparsely populated neighborhood groups with many elderly, the "Rural Fringe" and "Deeply Rural" groups. However, the group with the fourth highest population density and a high proportion of children, "Burdened Optimists" had a higher influenza incidence during our study period. Thus, the crowding of people in neighborhoods with many small children could explain the increased levels of influenza in such neighborhoods, just as a sparse population with an aging society can explain the low incidence of influenza; however, population density is not the only factor explaining these differences. We have to consider factors such as social contacts, influenza susceptibility by age group and other socioeconomic factors that can help interpret our study results.

During our study period, influenza circulated in all four seasons, but influenza B caused community outbreaks only in two seasons. The alternating circulation patterns of influenza A and B are among the more prevalent characteristics of influenza [30].

Our age specific incidence analysis demonstrated that children have higher attack rates during typical seasonal influenza outbreaks than adults and the elderly (Table 2). Among them, in the 5-9 years age group, the incidence of influenza A was highest in all four seasons, and that of influenza B was highest in the 2004/2005 season. However, in the 10-14 years age group, the incidence of influenza B was highest in the 2006/07 season. A previous community based survey showed that the highest attack rates were observed in children aged <10 years for influenza A and in those aged 10-19 years for influenza B[30]. Furthermore, our observed attack rates regarding age specific incidence were consistent with the age specific characteristics of influenza.

Several reasons are responsible for a high attack rate in children. First, children are more susceptible to influenza than adults because they are immunologically naive with a lower likelihood of previous infections [31]. Second, young children shed influenza virus for longer periods and in higher titers than adults during illness [32, 33]. Third, children have frequent social contacts with their schoolmates [20, 21].

Social contact studies suggest that individuals in all age groups tend to mix assortatively; in other words, they mix with people of similar age [16‐22], especially in the case of children and adolescents [17, 20‐22]. Furthermore, these studies show that children mix intimately with their parents, particularly for the 30 to 39 year age group, in which such mixing occurs mainly in their homes [20, 22]. Simulation studies using data on social contact indicated that school-aged children have the highest incidence of infection and play a major role in the further spread of infection during initial phases of epidemics by respiratory dissemination [17, 20]. By using survey-based contact data and mortality data, optimal vaccination is achieved by prioritization of schoolchildren and adults aged 30-39 years [34]. These observations suggest that the virological characteristics of children and their social contacts strongly contribute to influenza transmission in the community.

On the contrary, the "Rural Fringe" and "Deeply Rural" groups, in which the percentages of people in younger age groups were low but those of people in older age groups were high, and the incidence of influenza A and B was a significantly low. Residents in these neighborhoods are mainly engaged in self-employed farming or fishery work. Therefore, infrequent social contact within these neighborhoods, especially the contact of elderly people with virus-carrying children, would result in a relatively low risk for influenza transmission in addition to immunity from past infections [35].

Geodemographics classifies residential areas according to various characteristics, providing geographers with new analytical information to help identify what type of residents live in a neighborhood [28]. These data have been used to study issues related to the social structure and physical environment in small neighborhoods, identified by their zip code or census tract code. In recent years, social marketing principles and techniques have been central to government proposals for improving health and tackling inequalities in health [36]. Geodemographics is used not only in commerce but also in various areas of public heath, such as drug abuse [37], smoking cessation programs [38], Type 2 diabetes [39], primary dental care service [40], and self-rated health [41]. The use of geodemographics profiles offers the possibility of improving our understanding of the probability of the incidence or inequality in them between districts and within communities. The use of this approach enables the health sector to target interventions effectively in some neighborhood groups [36]. In this study, we used a commercially available dataset, Mosaic Japan. A range of geodemographics tools are currently in use, but the ways in which they are constructed are broadly similar. The tools tend to use variables drawn either entirely or in part from the census data. Regarding the Mosaic Japan dataset, a large number of variables were collected from census data and commercial data. Census data including age group, sex, occupational type, working situation, housing type, population density, and other variables were obtained from a commercial database to infer income levels, life styles and consumer behaviors. Many !variables were collected at the household level by census research or consumer survey, and they were aggregated at the census enumeration district level. Segmentations were generated by clustering those multi-variables using a multivariate classification method such as K-means cluster analysis [28]. In the case of Mosaic Japan's geodemographics clusters, all 0.2 million Japanese census tracts were classified into 50 different neighborhood types that were then aggregated into 11 neighborhood groups. One of the reasons why we used a commercial database was that Mosaic Japan contains variables not included in the Japanese Census data, such as income level and life style. These variables can potentially influence the profiles of neighbors, but they are difficult to obtain unless expensive surveys are conducted. The advantage of using existing datasets is especially applicable to decision makers, because of the ease in elucidating some of the information inherent in multivariate classification analysis, and eventually one is able to extrapolate results from small areas to wider regions such as prefectures or to the nationwide level if similar profiles exist. Besides, social and economic structures differ from society to society, and census data collections also differ from that among counties. Consequently, each country tends to have its own geodemographics profiling dataset, but these commercial datasets have a universal method of application. This indicates that the existing datasets not only permit interpolation of the results to other areas in Japan but also have a potential application for comparison with datasets of other countries.

It is common in epidemiological studies to list only adjusted results as in the case of standard mortality rate (SMR) such as cancer to evaluate the disease incidence (or mortality) by census enumeration district, municipality, or prefecture. In those epidemiological studies, age distribution is considered a strong factor that affects disease incidence, and age-adjusted calculation is applied to compare regional differences.

However, many of previous geodemographical studies provided non-adjusted results and did not implement adjustment [37, 39, 42]. In our paper, non-age adjusted results showed that the influenza incidence was high in the segments with young families with children, who had the highest incidence of influenza among age groups, and low where elderly, who had the lowest incidence, dwell, and the age adjusted results demonstrated that the infection rates across generations were still high in the former segments and low in the latter.

Both non-age adjusted and age-adjusted results are valuable for understanding the different effects on the incidence of influenza between the compositional effects of age groups of residents and contextual effects in the community.

Thus, we believe that our findings on influenza may lead to generalized ways of capturing characteristics of influenza circulation in societies. This will particularly be useful for allocation vaccines and anti-influenza drugs to high risk neighborhoods if the number of cases is rapidly growing and the decision maker has to choose target areas with the little delay.

This study has several limitations. Regarding data collection, patient medical consultation seeking behaviors between or among different age groups regarding influenza-like illnesses remained unknown. However, one OECD study showed that the rate for outpatient visits per person in Japan was the highest among all studied countries in 2007 [43]; therefore, non-inclusion of cases because of failure to seek medical attention may be lower than that in other countries. When we compared school absenteeism in elementary and junior high schools in a different season of 2008/09 in Isahaya City, our patient number was twice as high as that for school absenteeism (data not shown). It often happens that the networks of parents and children are strong conduits via which information and decisions are spread. If, for example, one school concludes that it has a concerning number of influenza cases, the children and adults associated with that neighborhood might be on higher alert. They may be quicker to seek medical care and prescription of anti-influenza drugs. This information supported the high consultation rate for influenza-like illness in children, but the other age groups remain uninvestigated. In addition, medical consultation seeking behaviors may be different based on the location of residence. Patients in rural areas may not seek medical service because of difficulties in accessing these services. To our knowledge, no study has been conducted in Japan on the medical consultation rate of patients with influenza-like illnesses in the community. Thus, these problems should be solved by future studies combining the data obtained from social questionnaire surveys and data already in our possession. The influence of selection bias from refusal for registration appeared to be minimal because the Isahaya Medical Association assured that an extremely low number of patients refused to participate in the study; however, the possibility of a larger bias remains after excluding cl!inicallydiagnosed and migrated patients who were referred to medical facilities outside the study area.

In the present study, influenza A had consistent results for higher or lower index values for particular Mosaic Groups and Types over the seasons even after age adjustment, but the results for influenza B were less consistent. One reason is that influenza B has different transmission patterns, affecting different age groups and group sizes, which led to slightly different area profiles compared to those for influenza A. In addition, as our study was based on an ecological analysis, we believe it is difficult to accurately determine all the reasons why influenza frequently or infrequently occurs in a particular neighborhood together with possible small number problems [42].

We believe that understanding the incidence of influenza in neighborhood groups is a valuable basis for community strategies to control influenza and that a simple statistical analysis using geodemographics tool is an effective means to aid the understanding of differences in the incidence of influenza among neighborhood groups. Our results are useful for stake holders in finding areas of priority to allocate vaccines and anti-influenza drugs in the case of a sudden increase in the number of influenza patients in a community. We demonstrated that geodemographics is a potentially powerful method for elucidating the correlation between social aspects in small areas and communicable diseases such as influenza. We aim to continue our study to analyze pandemic influenza in 2009 and other communicable diseases in the future.