• No se han encontrado resultados

The aim o f this study is to quantify the role o f environmental, climatic and agricultural factors in the temporal variation o f ague (or malaria) deaths in English and Welsh counties during 1840-1910. The starting period of 1840 was selected because at this time ague had been very w ell distinguished from other acute fevers with which it had been previously classified (se e Introduction) and we can be confident that the majority o f ague deaths used in the database were really due to malaria. The study terminates in 1910 as this was the last year that deaths from ague were reported at county level. Agricultural data were collected at the county level while climate data were derived from low resolution gridded databases or meteorological station measurements.

The specific objectives o f the study are:

1. To map the geographical distribution of ague deaths at county and statistical district level and compare this to various historical records of ague (or malaria)

2. To test whether inter-annual variability in ague death rates within English and Welsh counties was significantly affected by inter-annual changes in cattle and pig density, crop acreage, wetland acreage, temperature and precipitation.

3. To quantify the impact o f changes in these parameters on the inter-annual variability of ague (by model simulations).

This study w ill be the first to quantitatively investigate the link between climate, agriculture and the decrease of English malaria. By doing this, we will be able to test the common theories about why malaria disappeared from this country and show in a ‘virtual experiment’, using predictive modelling, how the extinction o f the disease may have been delayed or advanced if one or several of the explanatory factors had remained unchanged or been altered in a different way.

Methods

System o f recording deaths

During the study period from 1840 to 1910, any death was recorded first at the parish level by the local physician and noted in the bills of mortality which stated the cause of death as well as the patient’s name, age and address. All parish deaths would then be reported to the corresponding statistical district. A statistical district was a large parish (or town) which accumulated and reported mortality and population data from the surrounding smaller parishes. The data from the statistical districts would ultimately be assembled at county and country level by the Registrar General who was responsible for producing an annual report o f mortality statistics at statistical district, county and country level.

Malaria ("ague ") and other deaths

Measures o f ague deaths and deaths from all causes (excluding accidental and violent deaths, suicides and stillbirths) at county and statistical district level from 1840 to 1910 were collected primarily from the Annual Report of the Registrar General published by His Majesty’s stationery offices. On 5 occasions, ague deaths were not recorded for London and were instead collected from the Weekly Notifications o f Infectious Diseases.

Data from 43 counties (as defined geographically in 1840) and 316 statistical districts were available for the whole period. Over the 70 year period, 2 counties were separated into smaller regions (North and South Wales both separated into 6 new counties in 1903) and all ague or total deaths for these were pooled to comply with the original (1840) county borders. At different times deaths were reported either for the whole o f Yorkshire or separately from the three counties North, East and West Riding. For continuity, all data from Yorkshire were collected only for the whole county and not separately for North, East and West Riding.

As shown in Appendix 2.1, deaths in the Annual Report of the Registrar General were classified under headings and sub-headings. No deaths were shown as *-*. In 1840, there were 102 causes o f death registered (excluding accidental, violent deaths, suicides and stillbirths), in 1880 there were 141 and in 1900 there were 157 causes of death (Appendix 2.2) .

During 1840 to 1870, ague was classified as a subcategory under the heading ‘zymotic diseases' together with 15 other diseases (Appendix 2.1, 2.3 and 2.4). In 1871 ‘zymotic

diseases’ was split into four sub-headings: enthetic, dietic, parasitic and miasmatic. Ague and IS other diseases were classified as ‘miasmatic diseases’. In 1881, ague and remittent fever were still classified as ‘zymotic diseases’ but were now listed under their own sub­

heading ‘malarial diseases’ which contained only the two. In 1901, ague was renamed malaria and listed as a separate disease. The database used in this study contains measures o f deaths from ague only as these are the most likely to represent true malaria deaths. data were provided by the Malaria Reference Laboratory at the London School o f Hygiene

& Tropical Medicine.

Table 2.2. List o f all fever deaths reported in mortality statistic sources during 1840 to 1910.

Ague

Annual population figures for each county up to 1910, interpolated from 10-year population census data - were taken from the Annual Report o f the Registrar General.

Total county land cover (i.e. size) and water acreage (inland water, including marshes,

rivers, lakes and brackish water areas) for each county in each year were collected from the cattle densities were calculated by smoothing two 5-year values into 10 annual values.

Climate data

Climate data for each year during the 70 year period were extrapolated from two databases:

(1) from the 0.5° Gridded 1901- 1995 CRU climate dataset and (2) from historical meteorological station data, kindly provided by D. Lister at the University o f East Anglia Climatic Research Unit. Although meteorological station data do extend beyond 1901, it was decided to use the CRU gridded dataset for simplicity and better coverage. Climate data from the two sources did not differ significantly - e.g. in 1905, the mean average annual temperature measured from London meteorological stations was 9.18°C compared to 9.14°C calculated from the gridded data.

The CRU 0.5° Gridded 1901- 1995 climate dataset from the University o f East Anglia consists of monthly measures o f mean temperature, precipitation, cloud cover, diurnal temperature range, vapour pressure, ground frost frequency and wet day frequency. This climate surface has been constructed by New et al. (2000) by direct interpolation from station observations and estimation of synthetic data from predictive relationships between variables. The database was validated by cross-validation and intercomparison.

The historical meteorological station data comprises measurements from 104 stations throughout the UK, some o f which supply recorded data from the mid 17,h century. Climate variables for 1840 to 1900 obtained from this dataset consisted o f mean monthly (Appendix 2.5) which had been scanned into the Geographical Information System

ArcView 3.1. For all counties, climate data up to 1900 were then derived by calculating the average of all meteorological stations within the county borders. Between 1901 and 1910, climate data for each county was calculated as the average o f all 0.5° grid cells lying inside the 1840 county boundaries. In order to do this, the 0.5° grid map from the climate dataset was overlaid on the 1843 map and all grid cells (with their ID number) with the majority of the cell lying inside the county boundaries were identified visually. In a separate

A full database listing ague deaths, all cause mortality and explanatory variables for each county in each year during 1840-1910 was assembled in Excel ’97. This database was exported into STATA 7.0 for analysis. A summary o f all explanatory variables, their availability and conversion is presented in Table 2.3 below.

Table 2 3 . Explanatory variables and their availability and final conversion for use in the models.

Variable Availability Conversion

Population Annual N/A

County size (area) Annual N/A

Inland water acreage Annual Annual percentage o f county area Crop acreage 5-year Annual percentage o f cou n ty area

Cattle 5-year Annual density per 100 acres

The centre (i.e. name) o f all statistical districts was georeferenced using historical atlases or, alternatively (when required), the Worldwide Directory o f Cities and Towns (2002).

The m ap o f English county boundaries which forms the baseline o f all maps shown in this paper w as originally a 1983 boundary map kindly provided by C. Grundy (copyright County Boundaries Crown and ED LINE). This was edited within the Geographical Information System (GIS) ArcView 3.1. to correspond with the 1843 county boundaries.

Manipulation and display of geographical data was performed in ArcView 3.1.

Descriptive and univariate analyses

The temporal pattern in ague deaths and death rates, all-cause mortality and all explanatory variables during the 70 years was plotted in Excel ’97. Descriptive statistics (mean, minimum and maximum) and frequency distribution of all explanatory variables were generated using the statistical package in Excel ’97.

Univariate analyses (binomial regression) were performed in STATA 7.0 to (1) determine the relationship between ague death rates and each potential explanatory variable in turn and (2) test for non-linearity in the effect of explanatory variables on ague death rates. All univariate analyses were undertaken by adding the explanatory variable of interest to the null model described below. explanatory variables as described above were included in the analysis.

Because the model was designed to test if explanatory variables affected the inter-annual - not inter-county - variation in ague deaths, county was added as a categorical explanatory variable. This is a conservative approach for testing the significance o f the factors of interest, as a certain proportion of the inter-county variation which is being removed is also likely to be due to variation in climate, crops and livestock. The variables which affected the inter-county variation were tested later using a ‘spatial’ approach (see below).

For data with a strong temporal trend (as that observed for ague deaths) it is impossible to distinguish actual causal correlations from coincidental correlations which show a similar trend during the study period. Coincidental correlations are factors which, logically, are unrelated to ague (e.g. the number of cars or telephones in England) but, due to a strong temporal trend, may have been significant if added to the model. Any interaction role for

these ‘coincidental’ associations was excluded by taking out the overall (average) temporal trend from the data - i.e. by including a ‘calendar year’ variable in the model. The variable was tested for linearity and, because it showed a quadratic relationship with ague deaths, was included in the model both as a linear and a quadratic term.

To test for possible temporal autocorrelation, the residuals from previous years (the difference between observed number o f deaths and the number predicted by the model including only the temporal trend and the categorical county variable, in a particular year) were also included as explanatory variables (Brumback et al. 2000). The residuals were added to the model in a forward step-wise manner starting with the residual from one year previously, then the residual from two years previously and so on until the addition o f a residual was no longer significant. If for instance only the residuals from 1 and 2 years previously are significant in the model (p < 0.05), it is concluded that ague death rate in a specific year will depend on the death rates from the two previous years.

The null model was defined as the model containing only the categorical county variable, the temporal trend and the temporal autocorrelation and no explanatory variables.

A minimal adequate model was constructed by adding all explanatory variables and then using a backwards stepwise logistic regression in which the least significant variables were removed one at a time until all remaining variables were significant (p <

0.05).

The fit o f the minimal adequate model was assessed by (1) inspecting the scatter plot of residuals and (2) comparing the deviance of the minimal adequate model to that of the null model. By investigating individual variables in comparison to the minimal adequate model it was possible to determine the importance o f each explanatory variable. Because o f the specially defined null model, the fit o f the model refers to percentage residual variation explained instead o f percentage variation explained.

For comparison, a temporal model for all-cause mortality (excluding ague, accidental and violent deaths, suicide and stillbirths) was constructed containing the same explanatory variables which were significant in the temporal ague model.

For ague deaths, spatial model was also created to explain the inter-county variation in ague mortality. As before, ague death rates were used as the binomial outcome but this time year instead of county was added as a categorical variable to determine the factors which influenced the difference in ague between counties. Spatial autocorrelation was not accounted for in this model. In an ideal situation, this problem should be addressed,

however including spatial autocorrelation in models like these requires sophisticated techniques which are beyond the scope of this study. Excluding spatial correlation may have led to under-estimation of confidence intervals and p-values rather than bias in the estimates of effects and thus the results may have been somewhat different if this had been accounted for. The minimal adequate model was constructed in the backwards stepwise manner and the model fit and importance of explanatory variables all assessed as outlined previously.

All models were scaled for overdispersion using the Pearson chi-squared statistics divided by the degrees of freedom as the scale parameter.

Prediction scenarios

Using the temporal minimal adequate model developed for ague deaths, the number of ague deaths during 1840-1910 was predicted for four different scenarios: (1) Observed historical conditions (2) No change in wetlands and cattle densities (3) No change in wetlands and (4) No change in cattle densities. T he ‘no change’ scenarios for cattle and wetlands use the initial (i.e. 1840) values for cattle densities or inland water acreage throughout the 70 years, thereby keeping either (or both) variable constant. For all prediction scenarios, the residuals were not revised (which would have generated a positive feedback in the absence of any dampening regulatory factor in the model). These regulatory factors (e.g. herd immunity) were not available for inclusion in the model and thus all predictions represent the expected increase or decrease in ague deaths for any county due to a change in the particular parameter value in a single year.

The number o f deaths (n) for each county in each year was predicted using the following back-transformation:

P ~

( Equation 2.1)

where A is the logit number of malaria cases - i.e. the result from the multivariate logistic regression model:

logit number o f cases = c + tiX| + t2X2 + 13X3 + ....t„xn

c, ti, t2, t3 and tn are constants, c is a county-specific constant (i.e. there was a different model for each country). For the spatial predictions, c was a year-specific constant.

X |, X2, X3and x„ are explanatory variables

Results

The frequency distributions o f all explanatory variables are shown in Appendix 2.6. All variables were normally distributed. Their mean, maximum and minimum values were identified and are shown below.

Table 2.4. Mean, minimum and maximum for explanatory variables

Explanatory variable Arithmeti Maximum average temperature ( ‘ C) 16.2 9.59

(South Wales, 1840) (mean, min and max for each variable selected from 3053 data points; i.e . 71 year recordings in 43 counties).

Temporal trends - deaths and demography

A total o f 8209 ague deaths were observed during the study period.

The number of ague deaths and the ague death rate decreased gradually over the 70 years (Figs. 2.3 and 2.4). There were distinctive ‘epidemics’ in 1848 and 1859 where the total number o f deaths reached 228 and 367 respectively. As the only war during the study period, the Crimean War, finished in 1856, none of these epidemics is likely to have been caused by an influx of infected soldiers from the tropics.

Documento similar