It is empirically known that, prior to a national epidemic, small sporadic influenza outbreaks occur widely. Various factors such as social and environmental factors i.e., temperature and humidity [7, 8], host immunity by vaccination [9, 10] and previous infection, human transportation [11–13], and population density  may affect whether certain local outbreaks can successfully trigger a national outbreak. We demonstrated that WSD values decreased to the lowest value before each epidemic peak then increased in inter epidemic period, and this trend was observed in nine of ten influenza seasons. The weighted standard distance method is useful for measuring the compactness of geographic distribution as shown in this study. This method is the classical way to assess distribution pattern: localized or diversified, however, very few application case was found in research study . Also, only a limited study was found in the literature which utilized spatial statistical method to clarify the spreading pattern in infectious diseases [16–18]. WSD values represent spatial distributions of influenza outbreak as numerical data: clustered distribution is expressed as small value; dispersed one is represented as large value. We postulated that the reason for a decrease in WSD values before the peak in influenza activity is due to local outbreak and clustering in a limited region which results in a small standard distance value. Furthermore, when successive cases disperse nationwide, the influenza outbreak peaks, thus resulting in an increasing WSD value. Also in inter epidemic season, sporadic imported cases with travelers and visitors might result in increased WSD value. This study is the first to show a spatial measurement to evaluate compactness of the distributions of influenza outbreak nationwide with WSD method. In the current study, we combined the number of reported influenza cases and spatial indicator calculated by the distribution of influenza cases, and found a significant association between these two factors. This is a novel measurement of nationwide influenza outbreak in Japan which can be applied for influenza epidemiology.
We found that the duration between the lowest WSD week and the peak week, reflected a significantly negative correlation with the proportion of A/H3N2 subtypes, and a significantly positive correlation with the proportion of type B viruses, in early phase of the each corresponding season. These results suggest that higher prevalence of A/H3N2, the predominant epidemic strain, may correlate with a faster diffusion from the lowest WSD week to the peak of a nationwide epidemic, and that B strain correlate with a slower diffusion from the lowest WSD week to the peak of a nationwide epidemic. Other studies employed spatial statistics tools to analyze spreading patterns and to trace infectious agents [17, 18]. Although those studies have shown distribution pattern using spatial tool and they did not find any significant associations between spatial component and magnitude of the epidemic.
The fastest spreading epidemic was observed during the 2002/2003 influenza season with the new antigenic variants of A/H3N2. This was also shown in our previous report through a kriging method . Faster evolution of influenza A/H3N2 may contribute to increased susceptibility among the population, thus accelerating the rate of epidemic spread [8, 19]. In addition, influenza A/H3N2 virus causes the most severe symptoms for human among the three types and subtypes of influenza; A/H3N2, A/H1N1, and B . From a clinical standpoint, patients infected with A/H3N2 virus could shed more virus and spread the infection . As a result, rapid A/H3N2 spreading patterns at the community level may lead to a rapid spread at the national level. However, cause of different spreading patterns among type and subtype is still unclear.
In the 2005/2006 season, the peak epidemic week came atypically before the lowest WSD week. In this season, A/H3N2 was the predominant circulating strain (64.8%) in the early phase of epidemic. Perhaps, very large and fast outbreak occurred in the clustered area in the early phase of epidemic, thus peak epidemic and clustering occurred simultaneously. Further detailed research study is needed to elucidate timing and location for local clustering in the early phase of epidemic.
In inter-epidemic periods, unstable WSD trends were observed. These trends may be the result of fluctuating and sporadic influenza cases in local areas due to imported cases from abroad. Easily changeable WSD values, especially in inter-epidemic period, might not be reliable for analysis. In order to decrease influence by these fluctuating values, we adjusted the data with un-weighted moving average method.
The WSD represents compactness of influenza epidemics except in cases of clustered distributions at opposite sides of the study area. For instance, if a local outbreak was clustered in different areas at the same time (e.g. in far north and far south); this may result to a large WSD value. However, based on the successive ten-year national influenza surveillance data, we found that the WSD value reached a minimum just before the peak epidemic week indicating the presence of certain local clusters prior to each seasonal epidemic. The WSD method has proven valuable for the analysis of influenza outbreaks throughout Japan due to the country’s unique and isolated geography with limited entrance routes. Further studies are needed to apply the result for continental countries as well as the temperate and tropical/subtropical zones, and also needed to find different characteristic of WSD trend in different geographical condition.
In our analysis, we evaluated only compactness by using standard distance method. In Figure 1, circles moved weekly depending on the locations of weighted mean center. Those circles’ movement does not match the result of the previous study . However, the method was different among this study and the previous one. The previous study focused on only the peak week of the epidemic. On the other hand, our study focused distribution of weekly influenza cases over time.
One of the limitations of our study is that we did not take social and environmental factors into consideration. Socio-economic factors might affect influenza spreading pattern . However, we could not obtain enough socio-economic information to analyze this time. We plan to conduct a study to evaluate those impacts of surrounding factors nationwide. However, it may be more appropriate to perform detailed analysis in focused area including social and environmental factors. For example, dispersion pattern may be different with regional areas if early outbreak occurs in major metropolitan areas such as Tokyo and Osaka which have function of transportation hub.
A limitation of using influenza surveillance data in this study is reflected in the lack of conclusive laboratory diagnosis that divides the cases by influenza strain. For more accurate analysis, more research must be done on data that clearly diagnoses the specific viral strain. There are some more limitations in our study. As we can exactly recognize when is the lowest WSD week only after epidemic was over, we cannot predict the timing of starting using the WSD value in the middle of epidemic. Interestingly WSD value was close to 400 km when influenza epidemic peaked, which may be a kind of threshold value. However, even this method cannot predict the timing of peak epidemic because WSD value exceeded 400 km in several times before the peak epidemic. Further study is needed to predict the timing of onset and peak of influenza epidemic nationwide.
Another limitation is from elongate shape of Japan islands. Location of weighted mean center which is calculated by using each location of points (centroid of each prefecture) affects WSD value. Clustered outbreak far from mean center tend to be underestimated resulting as large WSD value, and vice versa. However, we expressed a compactness of nationwide influenza distribution as a whole, not focusing on only local region. Even though WSD values were affected by distance from mean center, WSD values were still meaningful because those trends were clearly generalized in this study and had significant relationship with influenza epidemic trend. If there is better way to evaluate skewed distribution, more exact analysis of spatial pattern of influenza epidemic could be done, further detailed study is crucially needed.
In our analysis, we did not employ data from the 2009/2010 pandemic H1N1 influenza season due to excessive influenza reporting during this time. Analysis of this pandemic season should be performed separate and could give interesting insight on the pandemic spreading patterns compared to seasonal outbreaks. We plan to perform a transmission analysis for pandemic H1N1 influenza virus (A/H1N1pdm) in the near future.