D’Souza, Rennie M.*; Becker, Niels G.*; Hall, Gillian*; Moodie, Keith B. A.†
Foodborne illness is a significant public health issue in most countries, including Australia where an estimated 4 million cases occur annually.1 Although most foodborne illness is a mild gastroenteritis, some episodes are more debilitating, resulting in hospitalization and even death. Foodborne disease also has a substantial economic impact as a result of the cost of treatment, lost productivity, and trade implications.2
Pathogens responsible for human disease can originate in animal stock on the farm, from contamination during food processing or transport, or during commercial or home preparation of food. There is evidence that the growth and dissemination of the responsible microorganisms can be influenced by the weather,3–6 and some foodborne illnesses show seasonality.7–11 Salmonellosis is most commonly reported in summer, whereas campylobacteriosis peaks in spring. Reasons for the seasonal patterns could vary, because the various pathogens have different characteristics and methods of replication. Some foodborne organisms can survive in the general environment and replicate outside an animal host when conditions are optimal; such organisms include Salmonella, which replicates in manure and foods at temperatures between 7 to 30°C.12
There is growing evidence of global warming.13,14 The Intergovernmental Panel on Climate Change predicts that the Earth’s average surface temperature will rise by 1.4°C to 5.8°C by the year 2100.15 Increased temperatures of 1 to 6°C are expected in Australia by 2070.16 The possible health effects of the anticipated climate changes are now being studied, including the potential effect on food production and foodborne disease.13,17,18 In particular, a few studies have quantified a possible relation between temperature and the incidence of diarrhea19,20 or temperature and food poisoning.21
The objective of this study was to determine whether there is an association between temperature and foodborne infection. We restricted our analysis to salmonellosis because characteristics of this pathogen are well documented and reported salmonellosis is common.
Salmonellosis notifications for each state were obtained for the period 1991 to 2001 from the National Notifiable Disease Surveillance System (NNDSS). All states report cases of salmonellosis to the NNDSS on a regular basis by postal code. It is estimated that in Australia, 5% to 10% of people with salmonellosis infections are reported to the surveillance system.10 There can be variation in reporting practices across states and over time.
We used data from the metropolitan area of each of the 5 capital cities: Perth, Adelaide, Melbourne, Sydney, and Brisbane. The population for each city was extracted from the 1996 census data to adjust for population size. Daily notifications for each city were aggregated to give the total number of cases for each month over the 11-year period. Queensland Department of Natural Resources and Mines provided the spatial interpolated daily temperature data for each location, which was estimated by fitting a surface through the temperatures observed in Australia on that day. The daily average temperatures for 1991 to 2001 were aggregated to give mean monthly temperatures for each city.22 Mean temperature seems to be an appropriate summary measure of temperature for this study, because temperature in these 5 cities are nearly always in the range where replication of Salmonella in the environment takes place.
To examine patterns over the observation period, both the mean monthly temperature and the monthly number of salmonellosis notifications were plotted over the 11-year period for each city. To display seasonal patterns, we aggregated the 11 years of data by month, and both the average temperature and the numbers of notifications were plotted against month of the year for each city.
Model for the Chance Fluctuations
The main objective of the analysis was to see whether temperatures in the recent past predict some of the variation over time in the number of salmonellosis notifications. Data for the 5 cities were analyzed separately, as well as jointly, to explore associations that were common across cities. The Poisson regression model is the tool most often used for analyses of count data. In this instance, the variation in the data was greater than variation expected under the Poisson model, even when we included all available predictor variables. To allow for this overdispersion, we used negative binomial regression for our model fitting.23 The Poisson regression model is an extreme member of this richer class of models. However, even with this model, there was more variation than the model could explain, as a result of the occasional major outbreaks of salmonellosis infection attributable to a single source. To prevent occasional extreme values from overwhelming the analysis, we included an indicator variable for each outbreak month as a predictor in the model. An outbreak month was defined as one with a monthly salmonellosis notification greater than 2 standard deviations above the 11-year mean monthly count. This definition identified 7 outbreak months in Perth, 6 in Brisbane, Sydney and Melbourne, and 5 in Adelaide. Of these 30 outbreaks, 15 occurred in summer, 13 in autumn, and 2 in spring. Fewer than 5% of the months in the observation period were outbreak months. As is common with Poisson and negative binomial regression, we used a log-linear model for the mean salmonellosis count, as described.
Long-Term Trends Over Time
Gradual changes over time can occur as a result of nontemperature-related factors such as changes in the population size, changes in the surveillance system and level of underreporting, and changes in food production, storage, and consumption practices. To allow for such changes, we included a smooth long-term time trend in the expression for the logarithm of the mean salmonellosis count. A cubic polynomial was used because it can represent a wide range of trends, and a linear model did not adequately describe the trend. We did not explore the possibility that some of the long-term trend is the result of longer-term change in climatic temperature such as El Niño effects. The form of the model containing only the long-term trend and indicator variables for outbreak months is described as follows:
Equation (Uncited)Image Tools
where time = month, which takes values 1, 2, …, 132 and outbreak = categorical variable for outbreak month. The population size term is accommodated by declaring the offset to be log (population size).
The reasons for the long-term trends are not known so that the future direction of trend is not predictable. For this reason, we did not extrapolate beyond the boundary of the observation period.
Seasonal Effects and Temperature Effects
Two levels of detail were used to explain seasonal variation. One level consisted of a categorical variable for spring, summer, autumn, and winter. To account for seasonal variation in greater detail, we declared a categorical variable for each month of the year. Our own data and the study by Bentham et al.21 suggested a relationship between mean monthly temperature and notifications of food poisoning in the following month.
Accordingly, we added temperature of the previous month to the model, giving
Equation (Uncited)Image Tools
where temp1 = mean temperature of previous month and season = categorical variable for quarter or month.
Further Predictors of Salmonellosis Notifications
To permit more elaborate descriptions of the effect of temperature on salmonellosis notifications in each city, we included the mean temperature of the current month (temp0) and added quadratic and cubic terms for each of temp0 and temp1. Cubic polynomials of the temperature variables are able to explain elaborate relationships, including shapes that curve up and down for different parts of the temperature range. To determine whether humidity had any influence on salmonellosis notifications, over and above the effect of temperature, we included mean humidity of the previous month in the model along with mean temperature of the previous month.
Temperature and Salmonellosis Notifications for Each City
Seasonal variation is observed for salmonellosis counts and temperature in each city (Fig. 1). The large spikes correspond to outbreaks of salmonellosis. The 1996 population size, latitude, salmonellosis notifications, and temperature range over the 11-year observation period for the 5 cities are shown in Table 1.
Figure 2 shows the monthly patterns averaged across all 11 years for each city. Lowest counts occur in winter (July) then increase as temperature rises from August. The peak mean temperature occurs in January or February for all cities. The peak of salmonellosis notifications occurs 1 month after the temperature peak in all cities except Sydney, where high rates persist over a 3-month period. Visual inspection of the notification and temperature data in the 5 Australian cities suggest that an association between salmonellosis notifications and temperature in the previous month is likely, which was also found in the United Kingdom.21
To display the trend patterns, the predictions from model 1 are plotted against the logarithm of the observed monthly counts for each city (Fig. 3). The logarithmic scale on the vertical axis is convenient for displaying counts for outbreak months on the same graph. Although notifications of salmonellosis increased overall during the study period for all cities, they decreased somewhat in the last few years in Melbourne, Sydney, and Brisbane.
Effect of Temperature of the Previous Month
Seasonal effects are clearly visible in Figures 1 and 2 for all cities. However, when season and the temperature terms were added to the model simultaneously, season did not provide a substantial improvement in fit to the count data in any of the cities. This is true for both levels (monthly and quarterly) of the season factor. In other words, mean monthly temperature explains the seasonal variation.
When model 2 is fitted to the salmonellosis count data, with season omitted, a strong association is observed between mean monthly temperature of the previous month and the number of salmonellosis notifications in the current month for all cities (Table 2). The estimated effect of temperature during the previous month was similar across all cities except stronger in Brisbane (Table 2).
To see the effect of increased temperature on salmonellosis notifications, we used the fitted model 2 with season omitted to predict salmonellosis counts over a range of monthly mean temperatures. For the different long-term trends not to interfere with a comparison across cities, we fixed the time variable at a value that reflects the recent past. Thus, predicted values can be thought of as predictions of current counts. In doing so, we avoid the most recent times because they can be influenced by edge effects in the fitted (log-linear) trend curves. We calculated for different temperatures the number of cases predicted by the fitted model 2 as of December 2000 (Fig. 4). In all 5 cities, there is an increase in the number of cases as temperature increases. On the scale of actual counts (Fig. 4), all the graphs are curved upward slightly, more so for Brisbane. The 5 curves are linear for the logarithm of counts.
Figure 4 can be used to deduce the increase in the mean number of notified salmonellosis cases as we go from winter months to summer months, and to estimate the increase in monthly counts that global warming might induce. Brisbane shows a 62% increase in the mean number of cases for a 5°C increase in mean temperature. The corresponding increase is 31% for Sydney, 29% for Melbourne, 27% for Adelaide, and 23% for Perth. For a temperature increase of 1°C, the increase in the proportion of salmonellosis cases is 10% in Brisbane, 6% in Sydney, 5% in Melbourne, and Adelaide and 4% in Perth (Table 2).
Inspection of the graphs in Figure 2 suggests that the mean temperature of the current month is not as good a predictor of the number of notifications in a month as the mean temperature of the previous month. Fitting model 2 to the temperature data in the current month in place of the previous month confirmed this. In the absence of temperature information from the previous month, the current month’s temperature is a good predictor and is consistent over the cities. However, adding the current month’s temperature to model 2 did not substantially improve the fit. Quadratic and cubic terms of temperature in either month provided no further improvement in fit. Similarly, mean humidity added no useful information to the models.
Reported salmonellosis has both a seasonal pattern and a latitudinal gradient along the eastern states in Australia, suggesting that climatic temperature could be a factor predictive of incidence.10 Replication studies have also shown that growth of Salmonella in the environment is affected by temperature, suggesting a biologically plausible pathway for a relationship between climatic temperature and salmonellosis.12
We found a positive association between mean temperature of the previous month and the number of salmonellosis notifications in the current month. A study in the United Kingdom21 found a similar association between temperature and “food poisoning.” Our finding was consistent over 5 major cities of Australia, located large distances apart and ranging from 27° to 37° latitude. The increase in salmonellosis notifications per degree increase in mean temperature (previous month) is similar in 4 of the cities and higher in the fifth (Brisbane). The strength of the association, the consistency of the findings in 5 cities, and a plausible biologic pathway lends support to the hypothesis that higher mean ambient temperature is a cause of higher salmonellosis notifications. The stronger association with temperature in Brisbane could be because Brisbane is more tropical. A study of the effect of temperature on salmonellosis notifications is warranted for cities with similar climatic conditions as Brisbane.
The increasing long-term trend for salmonellosis notifications observed over the 11 years 1991 to 2001 in all cities is probably largely as a result of factors other than temperature. Possibilities include improved reporting, increasing population, and a possible effect of El Niño on foodborne disease occurring in Eastern Australia in 1996–1997. Salmonellosis notifications increased at first, then stabilized in recent years. A decline in notifications in the last 3 years has been reported in the United Kingdom and United States.9,24 The reason for this decline is unclear.
Nature of the Temperature Effect
The change in temperature with season is accompanied by changes in rainfall and humidity, as well as other seasonal changes in human behaviors and animal and plant cycles. The influence of these correlated seasonal changes are difficult to separate from the changes in temperature. Variation in the data explained by temperature of the previous month is likely to include both direct and indirect temperature effects on food production and consumption behaviors induced by weather conditions. The practice of reporting cases to the NNDSS is unlikely to be seasonally dependent and is therefore unlikely to result in spurious associations with temperature.
The lag of 1 month in temperature effect suggests that if the effect is directly causal, temperature might be more influential earlier in the production process, 1 month or so before illness, rather than at the food preparation stage. The association between salmonellosis notifications and temperature of the previous month is unlikely to be the result of reporting delay, because the data are recorded as time of onset of illness (although some inaccuracies in recording are possible).
In the U.K. study,21 the association between temperature and notifications of foodborne disease was conducted on deseasonalized data to control for the effects of seasonality. The merit of this approach is that it establishes an association between temperature and foodborne disease over and above any seasonal components. However, this approach is conservative in that temperature is likely to be responsible for much of the seasonal changes in behavior. In our analysis, temperature has been allowed to take over the entire role of season.
Outbreak Control in the Model
It was necessary to control the effect of large outbreaks on the analysis by including an indicator variable for each “outbreak month.” This, in effect, removes “outbreak months” from the analysis. We need to consider how this might influence our conclusions. Fifteen of the outbreak months were in summer, 13 in autumn, and 2 in spring. Their removal is likely to reduce the number of cases in months in which the previous month was warmer than average. Indeed, the average temperatures over the months before an “outbreak month” were higher than the overall average monthly temperature for each city. As a result, we are likely to have underestimated the association between temperature of the previous month and salmonellosis notifications by excluding the data on the outbreak months. The relationship between temperature and point source outbreaks warrants further investigation.
The relationship between climatic temperature and salmonellosis has implications for both the present and the future. Periods of hot temperatures can be a warning to public health authorities that an increase in certain foodborne diseases could be imminent. Preventive public health action could be instituted by warning and educating producers of food, managers of commercial food outlets, and the public about the added risks with hot weather, and the importance of hygiene, correct storage, and temperature control during food production and preparation. If global climate change occurs as anticipated, foodborne diseases could be affected. However, it is not clear that a relationship between short-term fluctuations in temperature and disease rates implies an increase in disease rate when average temperature increases gradually over a long time period. Finally, the relationship between temperature and salmonellosis notifications cannot be extrapolated to other foodborne diseases, and these need to be studied separately.
We thank the Communicable Diseases Network Australia, National Notifiable Diseases Surveillance System for providing the data for notifications of salmonellosis and campylobacteriosis for the years 1991 to 2001 and the Queensland Department of Natural Resources and Mines (http://www.nrm.qld.gov.au/silo/datadrill/index.html) for providing interpolated temperature data for the same years. We also thank Rhonda Owen and Ivan Hanigan from the National Centre of Epidemiology and Population Health for preparing the graphics for this article. This study is part of a 5-country collaborative study coordinated by Sari Kovats and Sally Edwards from the Epidemiology Unit of the London School of Hygiene and Tropical Medicine.
1.Food Safety Standards Costs and Benefits
. Canberra: Austalian New Zealand Food Authority; 1999.
2.Department of Agriculture and Forestry. Australian Food Statistics 2002
. Canberra: Commonwealth of Australia; 2002.
3.Centers for Disease Control and Prevention. Outbreaks of Escherichia coli
O157:H7 Infection and Campylobacter among attendees of the Washington County Fair–New York 1999. MMWR Morb Mortal Wkly Rep
4.Rose JB, Daeschner S, Easterling DR, et al. Climate and waterborne disease outbreaks. J Am Water Works Assoc
5.Curriero FC, Patz JA, Rose JB, et al. The association between extreme precipitation and waterborne disease outbreaks in the United States, 1948–1994. Am J Public Health
6.Rose JB, Epstein PR, Lipp EK, et al. Climate variability and change in the United States: potential impacts on water- and foodborne diseases caused by microbiologic agents. Environ Health Perspect
. 2001;109(suppl 2):211–221.
7.Wallace DJ, Van Gilder T, Shallow S, et al. Incidence of foodborne illnesses reported by the foodborne diseases active surveillance network (FoodNet)—1997. FoodNet Working Group. J Food Prot
8.Crown Public Health Disease Report Salmonellosis. 2001, Crown Public Health: Christchurch.
9.Trends in selected gastrointestinal infections, 2001. Canada Communicable Disease Report.
10.Hall G, D’Souza RM, Kirk MD. Foodborne disease in the new millennium: Out of the frying pan and into the fire? Med J Aust
11.McMichael AJ, Haines A, Slooff R, et al., eds. Change and Human Health
. [Task Force: McMichael AJ, Ando M, Slooff RR, Carcavallo R, Epstein PR, Haines A, Jendritzky G, Kalkstein LS, Odongo RA, Patz J, Piver WTR.] Geneva: World Health Organization, World Meteorological Organization, United Nations Environmental Program; 1996.
12.Hocking A, Arnalod G, Jenson I, et al., eds. Foodborne Microorganisms of Public Health Significance
. Sydney: Australian Institute of Food Science and Technology, Food Microbiology Group: NSW Branch; 1997.
13.McMichael A, Haines A. Global climate change: the potential effects on health. BMJ
14.Woodward A, Hales S. Climate Change: Potential Effects on Human Health in New Zealand
. Report prepared for the Ministry for the Environment as part of the NZ Climate Change Programme, Wellington; 2001;1–22.
15.Albritton D, Allen MR, Baede APM, et al. Technical summary. In: Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change
. Cambridge: Cambridge University Press; 2001.
16.Commonwealth Scientific and Industrial Research Organisation (CSIRO). Climate Change Projections for Australia
. Melbourne Climate Impact Group, CSIRO Division of Atmospheric Research; 2001.
17.Patz J, Engelberg E, Last J. The effects of changing weather on public health. Annu Rev Public Health
18.McMichael AJ, Woodruff RE. Climate Change and Human Health: What Do We Know
[Editorial]? Med J Aust
19.Checkley W, Epstein L, Gilman R, et al. Effects of El Niño and ambient temperature on hospital admissions for diarrhoeal diseases in Peruvian children. Lancet
20.Singh R, Hales S, de Wet N, et al. The influence of climate variation and change on diarrheal disease in the Pacific Islands. Environ Health Perspect
21.Bentham G, Langford I. Climate change and the incidence of food poisoning in England and Wales. J Biometeorol
23.Stata Statistical Software,
release 7.0. College Station, TX: Stata Corp; 2001.
24.Centers for Disease Control and Prevention. Preliminary FoodNet Data on the incidence of foodborne illnesses—selected sites, United States, 2001. MMWR Morb Mortal Wkly Rep
© 2004 Lippincott Williams & Wilkins, Inc.