• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Aug 19, 2003; 100(17): 9997–10001.
Published online Aug 11, 2003. doi:  10.1073/pnas.1233687100
PMCID: PMC188345
Medical Sciences

Malaria in Britain: Past, present, and future


There has been much recent speculation that global warming may allow the reestablishment of malaria transmission in previously endemic areas such as Europe and the United States. In this report we analyze temporal trends in malaria in Britain between 1840 and 1910, to assess the potential for reemergence of the disease. Our results demonstrate that at least 20% of the drop-off in malaria was due to increasing cattle population and decreasing acreages of marsh wetlands. Although both rainfall and average temperature were associated with year-to-year variability in death rates, there was no evidence for any association with the long-term malaria trend. Model simulations for future scenarios in Britain suggest that the change in temperature projected to occur by 2050 is likely to cause a proportional increase in local malaria transmission of 8–14%. The current risk is negligible, as >52,000 imported cases since 1953 have not led to any secondary cases. The projected increase in proportional risk is clearly insufficient to lead to the reestablishment of endemicity.

The British Chief Medical Officer's recent report asserted that “by 2050 the climate of the U[nited] K[ingdom] may be such that indigenous malaria could become re-established” (1). This prediction was based on model simulations, derived from only a subset of the possible links between climate and malaria (2, 3). To be confident in future predictions, we first need to understand the past by quantifying the factors that governed the disappearance of malaria from Britain. In this article, we describe statistical modeling of the determinants of temporal trends in malaria deaths within English counties and Wales from 1840–1910.

From historical records, we know that a malarious illness referred to as “the ague” or “intermittent fever” caused high levels of mortality in the British marshlands and fens from the 15th to the 19th century (4, 5). Robust evidence that the illness was malaria emerged in the early 19th century, when the increasing use of quinine and advances in fever diagnosis and pathology created a distinct separation from other acute fevers. Definitions of ague in 19th-century medical textbooks uniquely indicate malaria as they invariably refer to noncontagious transmission, distinctive cold, hot, and sweating stages, tertian onset of symptoms, cycling relapses, anemia, splenomegaly or “ague cake,” and susceptibility to quinine (6, 7). The remarkable virulence of this disease in England, given that the pathogen responsible was presumably Plasmodium vivax, has never been explained satisfactorily (5). The situation in Britain was not unique. In Holland at the end of the 19th century, equally high death rates were reported from intermittent fevers, believed to be caused by P. vivax (8). Although currently responsible for 80 million annual cases of malaria worldwide, P. vivax is not now a lethal parasite (9). One possible explanation for the high malaria mortality rates observed in 19th-century Europe is the high likelihood of coinfections with pathogens associated with poor sanitation.

Physicians throughout Britain reported a decline in ague cases from the early 1800s (10). Many hypotheses have since been proposed to explain this gradual disappearance (8). Marsh drainage could have eliminated many breeding sites of the main local vector, Anopheles atroparvus, in the brackish waters of coastal marshes, river deltas, and fens (11). A. atroparvus feeds mainly on livestock but will take human blood when available. Hence, increasing livestock densities may have diverted biting from humans toward cattle, pigs, or horses. Improved housing, better access to health care and medication, and improved nutrition, sanitation, and hygiene all may have reduced transmission and/or mortality rates. Although these theories are widely accepted, the relative importance of the different factors has been discussed only rarely and has never been quantified.

In this article, we analyze the role of climate, agricultural factors, and land cover in the temporal variation in ague (malaria) death rates in Britain during the 19th century. We selected 1840 as a starting point, because at this time acute fever diagnosis was sufficiently accurate that the great majority of reported ague deaths were probably due to malaria. To provide further evidence of the accuracy of these diagnoses, we also investigate the factors associated with spatial variability in ague death rates (expected to be similar to the determinants of temporal variability) and with temporal variability in other causes of death (expected to be different). The study terminates in 1910, the last year that ague deaths were reported at county level.


Annual ague death counts from 43 counties (including North and South Wales; see Fig. 2) over the 71-year period were collected from historical records of the Registrar General. Explanatory data comprised population size, annual minimum, mean, and maximum average temperature, total precipitation, pig and cattle density per 100 acres, and percentage coverage by crops and inland water (wetlands) for each county in each year (Table 1). Data on water acreage per county (inland water including marshes, rivers, lakes, and brackish water areas) were collected annually from the Report of the Registrar General. Crop acreage per county (all crops, bare fallow, and grasses) and the number of pigs and cattle at county level were recorded at 5-year intervals from the Ministry of Agriculture, Fisheries, and Food. Annual population measures were obtained from the Report of the Registrar General (interpolated from censuses undertaken every 10 years). Water and crop acreage was converted to percentage of total county size (from the Report of the Registrar General), and pig and cattle numbers were converted to animal density per 100 acres (2.5 km2) of land. Annual crop acreage and pig and cattle densities for intervening years were calculated by linear regression between the observed time points. Climate measurements for each county and year were based on monthly average data on minimum, maximum, and average temperature and total precipitation, which were taken from historical meteorological data from 104 stations (up to 1901) and from the 0.5° gridded 1901–1995 Climatic Research Unit (CRU) climate data set (12) for later years. Although station data extend beyond 1901, it was decided to use the CRU gridded data set for simplicity and better coverage. Data from the two sources did not differ significantly; for example, in 1905, the mean average temperature measured from London meteorological stations was 9.18°C, compared with a calculated value of 9.14°C from the gridded data. Estimates for years up to 1901 were derived by calculating the average of all georeferenced meteorological stations within the county borders (as shown on an 1843 map of Britain). Between 1901 and 1910, measurements represent the average of all 0.5° grid cells where the majority of the cell lay inside each county. Temperatures were converted to average annual minima, maxima, and averages, and precipitation was converted to annual totals. Explanatory and outcome data were collected from 43 counties (including North and South Wales) over the 71-year period.

Fig. 2.
Total ague death rates in British counties from 1840 to 1910 are shown.
Table 1.
Demographic and agricultural explanatory variables: Britain totals at the start, middle, and end of the study period

A database listing ague deaths, all-cause mortality, and explanatory variables for each county in each year during 1840–1910 was assembled and used for the generation of spatial and temporal multiple logistic models in STATA 7 (Stata, College Station, TX), by using county population sizes each year (interpolated from 10-year census data) as denominators.

Statistical Methods

Using multiple logistic regression, we investigated the role of these explanatory variables in both the temporal and spatial variation in ague death rates. The temporal model was designed to investigate the interannual variation in county ague deaths without drawing on information from differences in average rates between counties. To do this, we included a categorical county indicator variable (i.e., stratifying by county). National rates show a strong decrease over time, but it is impossible to distinguish in this single series how much of this decrease is due to explanatory variables that show similar temporal trends (e.g., wetland coverage) or to factors not included in the analysis. We answer these questions instead by opting for the conservative approach of assessing how well the explanatory variables explain temporal deviations from the national downward trend in county-specific rates. This was done by including the national trend in ague rates in the model, i.e., by incorporating calendar year as a continuous variable. The year variable showed a significant nonlinear relationship with ague death rates, which was captured by including both linear and quadratic terms. Ague rates were found to have a significant temporal autocorrelation of order one (one year's rate was predicted by the previous year's rate). This autocorrelation was incorporated into the model by inclusion of the lagged residual (i.e., the difference between observed ague deaths and those predicted by the null model containing the categorical county indicator, the temporal trend, and the temporal autocorrelation; ref. 13).

The spatial model was designed to investigate the intercounty variation in ague deaths without drawing information from variation in national rates between years, by including a calendar year categorical indicator variable (i.e., stratifying by year). To avoid complexity, we did not control for spatial autocorrelation (i.e., nearby counties having similar rates). It is unlikely that this method biased estimates of effects, but it may have underestimated the width of confidence intervals. The confidence intervals given for deaths predicted by all our models reflect uncertainty in parameter estimates but not in the model-building process.

For both the temporal and the spatial modeling scenario, redundant variables were excluded by using a backward stepwise approach, achieving minimal adequate models in which all remaining variables were significant (P < 0.05) (14). However, for comparability, the temporal model for all-cause mortality included those variables identified as significant by the stepwise process for ague death rates. For comparative purposes, we also investigated how the same explanatory variables were related to temporal patterns in all-cause mortality (i.e., all deaths excluding ague, accidental and violent deaths, suicide, and stillbirths).

All models were scaled for overdispersion by using the Pearson χ2 statistics divided by the degrees of freedom as the scale parameter (15).

Finally, we used our model for temporal variability in ague to predict the number of ague deaths in Britain during 1840–1910 in “virtual experiments,” in which we measured the change in the expected number of deaths had the climate been warmer or had one or more of the significant nonclimatic factors remained at 1840 levels. We ran the model six times to test the impact of six possible sets of circumstances, as described in Table 3. These simulations were carried out by altering the values of one or more of the four explanatory variables for each county and each year. The predicted number of cases in each county and each year was then estimated by using the ague death rates fitted by the temporal ague model (incorporating the underlying national trend and temporal autocorrelation), combined with the appropriate population size. These numbers were then totaled for each simulation (Table 3).

Table 3.
Predicted ague deaths for seven scenarios

Results and Discussion

From 1840 to 1910, a total of 8,209 ague deaths was reported. These deaths declined steadily over time, with distinct epidemics in 1848 and 1859 (Fig. 1). The highest rates were reported from Kent, Essex, and Cambridgeshire (Fig. 2), consistent with historical observations of high ague mortality in coastal and marshy areas (10, 16, 17) where the principal mosquito vector, A. atroparvus, is still present (18).

Fig. 1.
Observed and predicted ague deaths in Britain from 1840 to 1910 are shown.

Temporal variability in ague death rates was significantly associated with changes in inland water coverage, mean temperature, and total precipitation (all positive) and cattle density (negative; see Table 2). Omitting the least malarious counties (i.e., analyzing only counties with malaria death rates >60 or 30 per 100,000 inhabitants during the 70 years) had no significant impact on the coefficient values for the four explanatory variables. Hence, the full analysis (Table 2) did not significantly underestimate their effects by including possibly nonmalarious counties. The departures from the overall temporal trend, including the two epidemics of 1848 and 1859, are well predicted by the model (Fig. 1), although the peaks of the predicted values are dampened because of the inclusion of the previous year's residuals. This result suggests that the epidemics were partly due to above average temperatures and/or precipitation (Fig. 3).

Fig. 3.
Annual average temperature and precipitation in Britain from 1840 to 1910 are shown.
Table 2.
Temporal and spatial models for variation in ague rates

Spatial variation in ague was also significantly associated with inland water, cattle density, mean temperature, and total precipitation (Table 2) with directions the same as those of the temporal model. In contrast, the four explanatory variables for ague death rates had opposite (inland water, cattle density, and precipitation) or no effect (average temperature) on temporal variation in all-cause mortality (Table 2), indicating that the variation in ague was not merely reflecting factors which affected general mortality.

Using the model for temporal variation in ague, we simulated malaria incidence during 1840–1910 for various scenarios (Table 3). If cattle densities or inland water coverage had remained at 1840 levels, we predicted that the number of ague deaths would have been 8.7% and 10.8% greater, respectively (Table 3), from 1840 to 1910. Had cattle densities and inland water coverage both remained unchanged, they would have been 19.5% greater. These results conclusively demonstrate the important, largely independent, and roughly equivalent roles of the changes in cattle densities and wetland coverage in the disappearance of malaria from Britain. Our estimates of the impact of both factors are potentially conservative for three reasons. First, much of the variation in ague was already removed by the year trend. When we excluded the trend from the model, a 4-fold increase in the apparent effect of cattle density was detected, but no significant change in the effect of inland water coverage (data not shown).

Second, local environmental factors cannot explain variation due to imported cases. These were not distinguished in reports until 1919. Imported cases typically increase immediately after overseas wars in malarious zones, as shown by the steady decrease in the number of imported cases reported from 1919 until the early 1920s. However, our study period was relatively peaceful except for the Crimean War (1853–1856), which did not coincide with either of the epidemics. The mean annual number of imported cases reported in the first “stable” decade (i.e., 1926–1935) was 464, which is still relatively high, suggesting that at least some of the malaria deaths notified before 1911 were imported.

Third, although our models incorporate the effect of the previous year's residuals, we did not revise these values for the simulations (which would have generated a positive feedback, in the absence of any dampening regulatory factor in the model). Hence, we did not account for the cumulative effect of any consistent simulated change in risk over time. Spatial differences, unlike one-year temporal differences, should reflect the cumulative effect of a consistent difference in risk. However, none of the coefficient values for the spatial ague models were significantly greater than those for the temporal models, indicating that the cumulative effect is marginal.

Average temperatures in Britain from 1840 to 1910 varied annually across a 2°C range. Our simulations (with similar caveats) indicate that a 1°C increase or decrease was responsible for an increase in malaria deaths of 8.3% or a decrease of 6.5%, respectively (Table 3). This contradicts Reiter's claim (19) that climate was relatively unimportant, based on his observation that a drastic drop in average annual temperatures during the 16th to the 18th centuries had no marked effect on the mentions of ague in historical literature or total mortality in the Kent and Essex marshes. In contrast, Lindsay and Joyce (20) argued that a period of successive cold summers in the 1800s (observed from records of the Central England Temperature Series, a monthly composite of station data) coincided with a decline in infant mortality rate (a proxy measure of ague death rates) and may have helped push malaria toward extinction. Our interpretation, based on analysis of concurrent climate, agricultural, and malaria data, differs from both of the above. We show a significant effect of both temperature and precipitation, explaining the malaria epidemics in the “unusually hot summers” of 1848 and 1859 (10, 16, 2123). However, outside these epidemics, the long-term trend during the 19th century was surprisingly consistent, with no noticeable increase in the rate of decline in the 1860s, and was probably driven by nonclimatic factors.

The most commonly used climate change models predict a rise between 1°C and 2.5°C in average United Kingdom temperatures by the 2050s (24). Assuming that the proportional impact of temperature will be the same in the 21st century as it was in the 19th and 20th centuries, our simulations indicate that global warming could increase the risk of local malaria transmission by 8–15%. However, between 1940 and 1981, 20,000 km2 of wet ground was drained in Britain; in East Anglia alone, 90% of fens have been lost since 1934 [M. Millett (Wetlands Advisory Service), personal communication]. Thus, socioeconomic and agricultural changes have created a country in which R0 (the number of secondary cases arising from each primary case in a susceptible population) is currently minuscule, as demonstrated by the absence of any secondary malaria cases in this country since 1953 (25), despite a cumulative reported total of 52,907 imported cases. The national health system ensures that imported malaria infections are detected and effectively treated and that gametocytes are cleared from the blood in less than a week (D. Warhurst, personal communication). This trend should continue in the future, unless malaria treatment becomes ineffective because of, for example, drug resistance. In Britain, a 15% rise in risk might have been important in the 19th century, but such a rise is now highly unlikely to lead to the reestablishment of indigenous malaria.


We thank Phil Rogers and David Lister for provision of climate data; Matt Millett for information on United Kingdom wetlands; and David Bradley, Paul Coleman, Jon Cox, Chris Curtis, Bo Drasar, Vanessa Harding, and David Warhurst for advice on historical malaria and comments on the manuscript. D.H.C.-L. is supported by the Wellcome Trust, and K.G.K. is supported by the Løvens Kemiske Fabrik, Roblon Foundation, Obel Foundation, and Novo Nordisk Foundation. K.G.K., D.H.C.-L., B.A., and C.R.D. are members of the Medical Research Council and Natural Environment Research Councils (MRC-NERC) Cooperative Group on Climate Change, Ozone Depletion, and Human Health formed by the London School of Hygiene and Tropical Medicine and the University of East Anglia.


1. Department of Health (2002) Getting Ahead of the Curve: A Strategy for Combating Infectious Diseases (Including Other Aspects of Health Protection) (Dept. of Health, London).
2. Rogers, D. J., Randolph, S. E., Lindsay, S. W. & Thomas, C. J. (2001) in Health Effects of Climate Change in the UK, ed. Maynard, R. (Dept. of Health, London), pp. 101–138.
3. Rogers, D. J., Randolph, S. E., Snow, R. W. & Hay, S. I. (2002) Nature 415, 710–715. [PMC free article] [PubMed]
4. Dobson, M. (1980) J. Hist. Geogr. 6, 357–389. [PubMed]
5. Dobson, M. J. (1997) Contours of Death and Disease in Early Modern England (Cambridge Univ. Press, Cambridge, U.K.).
6. Taylor, F. (1894) A Manual of the Practice of Medicine (J. & A. Churchill, London).
7. Tanner, T. H. (1865) A Manual of the Practice of Medicine (Henry Renshaw, London).
8. Bruce-Chwatt, L. J. & de Zulueta, J. (1980) The Rise and Fall of Malaria in Europe: A Historico-Epidemiological Study (Oxford Univ. Press, New York).
9. Mendis, K., Sina, B. J., Marchesini, P. & Carter, R. (2001) Am. J. Trop. Med. Hyg. 64, S97–S106.
10. Whitley, G. (1863) Report by Dr. George Whitley as to the Quantity of Ague and other Malarious Diseases now Prevailing in the Principal Marsh Districts of England, Sixth report of the medical officer of the privy council (Dept. of Health, London).
11. MacArthur, W. (1951) Br. Med. Bull. 8, 76–79.
12. New, M., Hulme, M. & Jones, P. (2000) J. Clim. 13, 2217–2238.
13. Brumbach, B., Ryan, L. M., Schwarts, J. D., Neas, L. M., Stech, P. C. & Burge, H. A. (2000) J. Am. Stat. Assoc. 95, 16–27.
14. Crawley, M. J. (1993) GLIM for Ecologists (Blackwell Scientific, Oxford).
15. McCullagh, P. & Nelder, J. A. (1983) Generalized Linear Models (Chapman & Hall, London).
16. Creighton, C. (1891) A History of Epidemics in Britain (Frank Cass & Co., London).
17. Nuttall, G. H. F., Cobbett, L. & Strangeways-Pigg, T. (1901) J. Hyg. 1, 4–44. [PMC free article] [PubMed]
18. Kuhn, K. G., Campbell-Lendrum, D. H. & Davies, C. R. (2002) J. Med. Entomol. 39, 621–630. [PubMed]
19. Reiter, P. (2000) Emerg. Infect. Dis. 6, 1–11. [PMC free article] [PubMed]
20. Lindsay, S. W. & Joyce, A. (2000) Global Change Hum. Health 1, 184–187.
21. Hirsch, A. (1883) Handbook of Geographical and Historical Pathology (New Sydenham Soc., London), Vol. 1.
22. MacDonald, A. (1920) J. R. Army Med. Corps 35, 99–119.
23. Smith, W. D. L. (1956) Lancet 1, 433–436. [PubMed]
24. Hulme, M. & Jenkins, G. J. (1998) Climate Change Scenarios for the United Kingdom: Summary Report (Climatic Research Unit, Norwich, U.K.).
25. Crockett, G. S. & Simpson, K. (1953) Br. Med. J. 2, 1141. [PMC free article] [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences
PubReader format: click here to try


Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...


Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...