Spatiotemporal characteristics of pandemic influenza
© Skog et al.; licensee BioMed Central Ltd. 2014
Received: 29 November 2013
Accepted: 30 June 2014
Published: 9 July 2014
Skip to main content
© Skog et al.; licensee BioMed Central Ltd. 2014
Received: 29 November 2013
Accepted: 30 June 2014
Published: 9 July 2014
Prediction of timing for the onset and peak of an influenza pandemic is of vital importance for preventive measures. In order to identify common spatiotemporal patterns and climate influences for pandemics in Sweden we have studied the propagation in space and time of A(H1N1)pdm09 (10,000 laboratory verified cases), the Asian Influenza 1957–1958 (275,000 cases of influenza-like illness (ILI), reported by local physicians) and the Russian Influenza 1889–1890 (32,600 ILI cases reported by physicians shortly after the end of the outbreak).
All cases were geocoded and analysed in space and time. Animated video sequences, showing weekly incidence per municipality and its geographically weighted mean (GWM), were created to depict and compare the spread of the pandemics. Daily data from 1957–1958 on temperature and precipitation from 39 weather stations were collected and analysed with the case data to examine possible climatological effects on the influenza dissemination.
The epidemic period lasted 11 weeks for the Russian Influenza, 10 weeks for the Asian Influenza and 9 weeks for the A(H1N1)pdm09. The Russian Influenza arrived in Sweden during the winter and was immediately disseminated, while both the Asian Influenza and the A(H1N1)pdm09 arrived during the spring. They were seeded over the country during the summer, but did not peak until October-November. The weekly GWM of the incidence moved along a line from southwest to northeast for the Russian and Asian Influenza but northeast to southwest for the A(H1N1)pdm09. The local epidemic periods of the Asian Influenza were preceded by falling temperature in all but one of the locations analysed.
The power of spatiotemporal analysis and modeling for pandemic spread was clearly demonstrated. The epidemic period lasted approximately 10 weeks for all pandemics. None of the pandemics had its epidemic period before late autumn. The epidemic period of the Asian Influenza was preceded by falling temperatures. Climate influences on pandemic spread seem important and should be further investigated.
An early understanding of factors allowing for efficient spread of a new influenza virus is vital for the handling of a pandemic. Correct predictions of the onsets of epidemic periods in different areas may allow for optimal distribution of vaccines, antivirals and of social distancing, whereby the consequences of the outbreak may be mitigated. Pre-existing immunity, social behaviour and, most likely, climate factors [1, 2] are major determinants for the spread of pandemic influenza viruses. The propagation of an influenza is also dependent on certain traits of the different viruses and possibly also by interference from other viruses [3, 4], which may be more efficiently spread than the influenza virus under certain climate conditions . As a result the R0-value  may vary for the same influenza virus.
Studies on the spread of viruses among humans and results obtained from guinea pig experiments under different climate conditions [7, 8] may facilitate the understanding of viral traits affecting the dissemination. However, we can also get new knowledge from old pandemics. To better understand factors governing the dissemination of pandemic viruses we have investigated the spread patterns for three past influenza pandemics in Sweden, using spatiotemporal analysis, spatial modeling and visualization. The study is based on reported case data from the Russian Influenza in 1889–90 [9–12], the Asian Influenza in 1957–58  (An animated map of the global spread can be found in Additional file 1) and the A(H1N1) pdm2009-2010 [14–17]. The paper also discusses possible co-variation of influenza dissemination with temperature and precipitation for the years 1957–1958.
In the beginning of 1890, shortly after the outbreak of the Russian Influenza in Sweden, all Swedish physicians were asked by the Swedish Society of Medicine to provide information about the start and the peak of the pandemic and the total number of cases in their region. They were also asked to fill in a questionnaire on the number, sex and age of infected persons in the households they visited. General answers on the epidemic were received from 398 physicians and data on individual patients were available for more than 32,600 persons  starting from the first week of December 1899. The information was compiled into a table, providing detailed information on the development of the influenza at 69 locations.
In 1957, when the Asian Influenza arrived, the Royal Medical Board  asked the district physicians throughout the country to report on clinically diagnosed influenza cases on a weekly basis. The reports on diagnosed cases, provided by 713 out of 720 district physicians during the period of June 1957 to February 1958 , have been stored in the Swedish National Archives . For this study the weekly reports (1200 pages) were scanned into digital documents. From the scanned pages we entered location of the physicians, date and the weekly number of infected cases into an Excel spread sheet. In total 275,000 cases were recorded.
For the period of the Asian Influenza in Sweden we also acquired daily observations of temperature and precipitation from 39 weather stations. The data was made available by the Swedish Meteorological and Hydrological Institute . For the years 1957–58 there are unfortunately no data on humidity available for Sweden.
Approximately 60 infectious diseases are continuously surveyed by the Swedish Institute for Communicable Disease Control  (now Public Health Agency of Sweden ) through statutory notifications according to the Communicable Disease Act. From 15 May, 2009 and onwards all laboratory verified Influenza A(H1N1)pdm2009-2010 (hereafter called A(H1N1)pdm09) cases have been reported in the web-based reporting system SmiNet . Up-to-date census data per municipality are published each year by Statistics Sweden  and historic records on population (for the time of the Russian and the Asian Influenza) have been collected and made available to researchers at the Demographic Data Base  at Umeå University. The present municipality division, established in 2003, was used to enable comparisons over time of the influenza data from the three pandemics.
No ethical approval was required to access the data for the study.
The case tables from the Russian Influenza were checked for inconsistencies and some of the place names were changed to the spelling of today. The data was converted into Excel format, to enable geocoding. In ArcGIS  the tables were spatially joined to the municipalities with reported cases. The weekly number of cases was assigned to each municipality according to the present municipality division.
The geocoding for the A(H1N1)pdm09 is based on the 10077 laboratory verified cases with identifiable postal codes in SmiNet reported between 1 June, 2009 and 8 March, 2010. In SmiNet it is mandatory to report national registration numbers. As the address information in some cases was incomplete, PAR AB , a company specialized in supplying address information for commercial purposes, was engaged to match the incomplete cases with a database called SPAR . The database contains address and other information for all individuals in Sweden with a national registration number. For this study, it was decided to geocode at postal code level only, in order not to violate Swedish laws on personal integrity. Geocoding was performed in ArcGIS, connecting the case data with the centroids of the postal code areas and with their respective municipality.
From the daily observations of temperature and precipitation we calculated the mean temperature and the accumulated precipitation for each week of the duration of the Asian Influenza pandemic in Sweden (i.e. the period for which we have data on infected persons). The meteorological data was combined with the case data in two ways. First we combined the weekly weather observations with the case data from the eight major cities (from Malmö in the South to Luleå in the North representing 24.4% of the population in 1957) where weather stations were located (the rest of the 39 weather stations were located in places with very few inhabitants). Least squares linear regression was used to calculate trend lines for temperature for the weeks prior to the onset of epidemic periods. Weekly z-score values ((observation-mean value)/standard deviation) for temperature, precipitation and number of cases were used to create diagrams to study the impact of temperature and precipitation on the local number of influenza cases. Furthermore, we used the spatial statistics functions in ArcGIS to interpolate (inverse distance) weekly weather data for all locations with diagnosed cases of influenza in order to determine whether there was any correlation between the number of cases and temperature and/or precipitation.
We developed a new method to estimate and compare the duration of the epidemic periods for the three pandemics. First, z-score values for the weekly number of cases for each pandemic were calculated, disregarding whether the number of cases were based on clinical observations or if they were laboratory verified. We then defined the start of the epidemic period as the first week when the slope of the graph for the z-scores exceeded 20, provided that it did so for three consecutive weeks. Using this technique we identified the onset of the pandemic period for all (21) county capitals. The end of the epidemic period was defined as the week when the slope, after a decline in the number of cases to approximately the same level as before the start of the peak, was exceeding −10 for three consecutive weeks. We also created diagrams of the z-score values for the weekly number of cases.
The peaking week for each municipality was identified and added as an attribute to the municipality feature class. Using this data we had new means to statistically study the spatiotemporal characteristics of the pandemics. We created maps for the three pandemics and Moran’s I was used to describe the pattern of peaking weeks. Spatial correlation between latitude and peaking week was also investigated.
where w i is the incidence in municipality i.
The weekly maps were converted into video animations. From the weekly GWM locations for the epidemic period geographic vectors were created and analyzed to establish a statistical mean direction for each of the three moving patterns of the GWM.
Number of cases for the three pandemics
Total number of cases
Cases within 10 weeks epidemic period
Total number of weeks
The Russian influenza outbreak started in the Stockholm region in November 1889. Data are available from the first week of December 1889 (week 8949) when the incidence was already high in the Stockholm region and isolated cases were found in many places along the railroad network from Malmö in the south to Sundsvall in the north. This is also the start of the national epidemic period. The peak was reached in week 9001 (the first week of 1890). In this week there were also 11 cases in the city of Umeå, 100 kilometers north of the railroads at that time. In week 9002 there were 2 cases in Pajala, north of the Arctic Circle near the Finish border. When this northern area peaked in January the influenza was more or less over in the southeastern parts of the country. When our data series ends in week 9009 (last week of February, 1890) the influenza showed high incidence levels only in the northernmost region.
In the last week of June 1957 (week 5726) the first 14 cases of the Asian influenza were diagnosed in the county of Gävleborg, 200 km north of Stockholm. In week 5728 the disease was diagnosed in various places from Kristianstad in the South to Sorsele in the North. After 4 weeks (week 5730) the whole country was affected including Gothenburg in the West and Pajala in the North. The number of diagnosed cases was still very low, in total 61 or 0,002% of all diagnosed cases reported.
In week 5740 (the first week of October) there were 1635 (0.6% of total) diagnoses reported. The major epidemic period started week 5739 and the peak of the influenza was reached during the second week of November (week 5743) with 43,290 diagnosed cases (15.7% of the total). At the end of the year (week 5752) the number of diagnosed cases was 4,271 and the last week of January (week 5804) there were 1,705 cases. The reports for the first week of February 1958 (week 5805) contain only 89 diagnosed cases and thereafter the reporting stopped.The spatial distribution of incidence values for municipalities in week 5743 (the peak of the influenza) showed no global autocorrelation when tested with Moran’s I (Moran’s Index: 0.096895, z-score: 1.612274 and p-value: 0.106902). When the spatial distribution was tested with Ripley’s K-function (99 permutations) for the same week local clustering could be found with a Kernel distance up to 350 km. For weeks outside the epidemic period there was no clustering at all. Moran’s I on peaking weeks for all municipalities showed no clustering and there was no clear correlation between peaking week and latitude (a map is available in Figure 5).
Least squares linear regression analysis of temperature prior to onset of epidemic periods
Adjusted R Square
Temp. drop from
Temp. drop to
The first laboratory confirmed case of the Influenza A(H1N1)pdm09 was reported close to Gothenburg in the first week of June 2009 (week 0923), followed by 8 cases in Stockholm the week thereafter (Before mandatory reporting including full identity was introduced, two cases were reported from the Stockholm area. The first was hospitalized on April 29, 2009). In week 0925 there was one case in Kristianstad in the south of Sweden. In week 0929 there was also a case in Umeå in the north and in week 0932 (the first week of August) the influenza had reached the city of Boden. The influenza was thereby seeded over the whole country without any major outbreaks. In week 0940 a high number of cases was reported in Skellefteå in the north. The epidemic period started in week 0941. During weeks 0941 and 0942 outbreaks were reported from all over the country. The maximum was reached in week 0946. In week 0950 the epidemic period was over and at the end of 2009 only a few new cases were registered.The statistical evaluation of the spatial distribution of incidence values for municipalities showed patterns similar to those found for the Asian Influenza. For week 0947 there was no global autocorrelation when tested with Moran’s I (Moran’s Index: 0.0492795, z-score: 2.767462 and p-value: 0.005649). When testing with Ripley’s K-function local clustering could be found with a Kernel distance up to 350 km for the weeks 0946 to 0949. For weeks outside the epidemic period there was no clustering at all. When applying Moran’s I on peaking weeks the z-score of 1.66 (p-value: 0.14), indicates a random pattern. As for the other two pandemics, there is no clear correlation between peaking week and latitude (a map is available in Figure 5).
Directions of GWM during epidemic period
Compass angle of resulting vector
Directional mean (Clockwise from due North)
Average X (km)
Average Y (km)
Average length of vector (km)
The pandemics studied in this paper have been discussed in numerous articles on epidemiology, but they have never been analyzed in a common spatiotemporal framework. Using spatial analysis we have been able to map the spread in Sweden of these three influenza pandemics, which have been appearing over a time span of 120 years. During this period infrastructure and social behavior have changed dramatically. The population maps in Additional file 7 show the changes in the distribution of population over these 120 years. Despite those changes we have identified some common patterns of spread, in particular regarding the epidemic period, which lasted approximately 10 weeks for all three pandemics. For obvious reasons influenza reporting has never been and can never be complete, and it never provides more than an estimate of the actual spread.
During the 120 years between the pandemics studied different methods of data collection (case reporting) have been used. We have no means to evaluate the accuracy and completeness of the reporting. However, the high and geographically widespread participation in the post-pandemic study of the Russian Influenza indicates high data completeness. A national system for district physicians for reporting infectious diseases including influenza was established in 1911, and the large number of reporting physicians in 1957 indicates completeness. The reporting by district physicians was abolished in 1989. ILI-reporting, which was used after 1989, has not been functioning well. Instead laboratory verified reports have been evaluated and now they are the leading indicator for influenza surveillance in Sweden . The inclusion of laboratory verified influenza A(H1N1)pdm09 among reportable diseases and a well-established web-based reporting system ensured next to complete reporting from the laboratories. A reference PCR-method was further used for diagnosis and quality control was organized. However, sampling may still have varied. The ratio of positive samples to the number of examined specimen ranged from 0.37 to 0.76 in the 17 of 21 regions that delivered denominators. The highest ratios were reported from the northern and western part of the country. The geographic differences are more likely explained by vaccination and climate than sampling. Sweden offered the pandemic vaccine Pandemrix® to the whole population. The vaccine arrived simultaneously with the onset of the epidemic period, too late to have a profound effect on the spread . The vaccine arrived in weekly batches that were evenly distributed over the country. The pandemic peaked first in the north, and therefore vaccinations may have affected mainly the activity in the southern part of the country. A milder climate in the south may also have diminished the spread. Measures that might have had a general influence on the spread of the pandemics were not recommended in either 1898 or 1957.
The use of incidence is a good measure to describe the spread of a single influenza or a pandemic - thus we believe that the data are rather reliable for the individual pandemics. Incidence per se cannot be used to compare different pandemics with different reporting systems. We believe that the method we developed allows for the comparison of time series of different influenza outbreaks, despite the differences in reporting systems and number of reported cases.
The very high proportion of cases within the epidemic period for the Russian Influenza may partly be explained by lack of data prior to and after the epidemic period. The first known cases were reported in the middle of November, and already in December “it had acquired an epidemic character” . The railroad network played an important role in the dissemination of the Russian Influenza. By 20 December, approximately one month after the first case, the number of places with reported cases was almost twice as high for locations with a railway station as for those without a railway station . The municipalities in the north, not connected by the railroad network, were affected much later than the municipalities in the south and middle of Sweden. Apart from the communications, weather conditions by the end of November were probably very favorable for an immediate spread of the Russian Influenza.
Directional trends for Swedish municipalities
Short axis (km)
Long axis (km)
Rotation (degrees clockwise from noon)
Municipalities (no weight factor)
Municipalities with population 1890 as weight factor
Municipalities with population 1957 as weight factor
Municipalities with population 2009 as weight factor
Charland et al.  showed that the timing of the epidemic periods for the annual influenza in cities in the United States was positively correlated to latitude. Chowell et al.  made similar observations for the A(H1N1)pdm09 influenza pandemic in Chile, noting that the southern regions experienced earlier pandemic activity than the northern ones. The spread of the Russian Influenza, first appearing in a nation-wide winter climate, was determined by social contact and structure, with the railway being an efficient means of spread. In todays society with almost daily contact facilities covering the whole country, climate may be of greater importance. The similarities in seeding, time of onset and duration of the main peak between the Asian Influenza and the A(H1N1)pdm09 are surprising in view of the rather dramatic changes in contact pattern that have taken place since 1957. However, already in 1889 it was demonstrated that the arrival of one infected person at a new location was sufficient to create a local epidemic . The increased number of contacts today compared to 1957 may therefore be of limited importance for national spread of a pandemic influenza since the virus was already seeded nation-wide before the onset of peaking activity, both in 1957 and in 2009.
The saying that influenza “is spread with the speed of man” seems very true, but the onset of epidemic spread seems to depend on many factors and climate may be one. Atchison et al.  found a direct relationship between cold weather and rotavirus transmission in Great Britain and The Netherlands. Shaman et al.  calculated vapor pressure (absolute humidity) from relative humidity and temperature data. They showed that the onset of increased wintertime influenza-related mortality in the United States is associated with abnormally low absolute humidity levels during the weeks preceding the onset. When absolute humidity is low, influenza virus survival and transmission is high. Experiments with guinea-pigs have shown that airborne transmission of influenza is facilitated by low temperatures . Chowell et al.  showed similar results in studying the propagation of the A(H1N1)pdm09 in New Zealand where the influenza spread in a climate-dependent pattern. Åman  reported similar behavior of the Spanish influenza.
Our climate analysis points in the same direction; falling temperature is of importance for the onset of epidemic spread of a seeded influenza. The seeding of the influenza in all parts of Sweden over the summer period, followed by epidemic spread first in October for both the Asian Influenza and the A(H1N1)pdm09 indicates that lowering of temperature is vital for the onset of epidemic spread. The level of precipitation does not seem to have affected the spread of the Asian influenza. Whether the decrease in absolute humidity that accompanies temperature fall is the real reason cannot be verified in this study since information on humidity was not registered for the years 1957–58. However, our results clearly support previous studies that point to climate as an important factor behind influenza spread, and further studies and focus on climate factors are warranted for modeling and planning in the face of a new pandemic.
A major strength of this study is that we have been able to calculate incidence per municipality for our analyses. Census data is not always available and historic population data is generally difficult to obtain in many countries. In Sweden census data and historic population data are easily available and that has been of utmost importance to this work. The main weakness, but also the strength of the study is the data on influenza cases. Access to such detailed data from the same country and for three different pandemics is unique, even though there are limitations in all three data sets. Despite the limitations we believe that the data is sufficiently good to allow for the analysis made. In the end, presenting case data as z-scores of incidence per municipality reduces the shortcomings created by differences in quality and measurement methodology. Our findings shed new light on the spread of pandemic viruses and onset of epidemic periods. These findings may be valuable for future analysis and modeling in the face of a pandemic.
Looking at the results of the studied pandemics one can conclude that
Each pandemic had an epidemic period of approximately 10 weeks.
The weekly geographic weighted mean (GWM) of the incidence of the Russian and the Asian Influenza pandemics is mainly moving along a line from southwest to northeast. The weekly GWM for the A(H1N1)pdm09 influenza pandemic moved in the opposite direction, but along a line parallel to the one above.
The epidemic period of the Asian Influenza was preceded by falling temperatures.
The power of spatiotemporal analysis and modeling has been clearly demonstrated and epidemiologists have been given a new tool that can be used to develop new models for preparedness, prediction and analysis.
Geographic weighted mean
Influenza A (H1N1)2009pdm2009-2010
Simon Lindmark entered data from the scanned medical records. Sharon Kühlmann Berenzon collected and prepared case data from the A(H1N1)pdm09 for geocoding.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.