Analysis of environmental impact on health status data

A contribution to subproject INT
 
 

Uwe Schlink, Olf Herbarth, Matthias Richter, Ulrich Müller, Gisela Fritz

Dept. of Human Exposure Research and Epidemiology at the UFZ-Centre for Environmental Research Leipzig-Halle Ltd., POBox2, D-04301 Leipzig, Germany





Summary

The environmental impact on health status is studied by help of cross-correlation analysis and transfer-function modelling. The analysis of those impact, generally, is difficult because of the persistency and autocorrelation of both input and output series. The utilized approach provides a suitable technique for handling autocorrelated time series. As a result, the impact of SO2 - concentration on the health status can be demonstrated.

Aim of the research

A tremendous change has taken place during the last few years in the industrial area of Leipzig-Halle with substantial modifications in the pollution situation. An assessment of the current impact of air pollutants on the health of its population is of interest, as well as the study of the performance of the technique of transfer-function modelling in the present context.

Activities during the year

The association between environmental factors and health status is examined in children aged 3 to 7 years which are assumed to react especially sensitive to their surrounding environment. The health status of 267 children has been investigated using daily diaries. For the elimination of the low-frequency and seasonal parts we used a high-pass filter. This is based on a Kalman filter (Schlink et al., 1997; Ng et al., 1990) which shows an abrupt cut-off in the frequency response characteristic and removes all variations with a period of more than 150 days (high-pass filter HPF-150).

The transfer-function model (1) contains the output () at different points in time as well as the input (, representing one of the environmental variables) at different points in time: , with the white noise process . (1)

Here B is the back-shift operator () and  are polynomials in B. In this model the input X is transformed by the transfer function
F = into the output M according to . (2)

If X is not autocorrelated, the transfer function F can be estimated from the cross-correlation between M and X (Box et al., 1976). Otherwise, a so-called prewhitening must be carried out. This prewhitening procedure is in analogy to the minimization of the partial autocorrelation by flexible smoothing proposed by Schwartz (1998). It is done by filtering both the input and the output series with the ARMA - model of the input series. All the filtering processes (high-pass filter, ARMA filter) guarantee the stationarity of the filtered series M and X utilized for cross-correlation analysis as demanded by the theory of transfer-function modelling.

Principal results

The data series of the following environmental data show an annual cycle: global radiation, maximum daytime temperature, daily averages of vapour pressure, air humidity, concentrations of O3, SO2, NO2 and total suspended particles (TSP). This can also be observed for the data series of prevalences of general complaints (Figure 1), respiratory and skin-related symptom complexes.
 
 


 
 


 
 

Figure 1: Incidence (a), prevalence (b) and convalescence (c) of general complaints during the 759 days studied.
 
 
 
  On the other hand, the data indicate that the incidence of respiratory and general complaints is influenced by abruptly changing meteorological conditions. This will often indicate a passing weather front. There appears to be some relationship between illness duration and global radiation. The latter is collinearly related to temperature, air humidity and ozone concentration (Figure 2). Most of the time, they vary in combination but their effects are exerted independently. During the winter months, a positive relationship is observed between the SO2 concentration and prevalences of respiratory and skin-related symptom complexes (Figure 3).
 
 
 
 

Figure 2: Cross-correlation of daily maximum atmospheric temperature in summer (left) and relative air humidity (right hand side) to the global radiation (dashed lines indicate the 95% confidence interval for zero cross-correlation).
 
 
 
  Main conclusions

The cross-correlation analysis proves to be a valuable tool in the evaluation of panel studies. In order to get reliable results, care must be taken to the pre-processing of the data (prewhitening) and the interpretation of significant correlation, in particular in the case of collinearly varying variables.
 
 












Figure 3: Dependence of the prevalence on SO2 load in winter.
 
 

Aim for the coming year

The problem of collinearity between the environmental variables and infection chains in the health status variables (see Figure 1) makes it difficult to find clear associations between environmental and health data. Future study should focus on possibilities to avoid these problems.
 
 
 
 

References

Box GEP, Jenkins GM (1976) Time series analysis - forecasting and control, Prentice Hall, Englewood Cliffs, New Jersey.

Ng CN, Young PC (1990) Recursive estimation and forecasting of non-stationary time series, Journal of Forecasting, 9, 173-203.

Schlink U, Herbarth O, Tetzlaff G (1997) A component time series model for SO2 data: forecasting, interpretation and modification, Atmospheric Environment 31(9), 1285-1295.

Schwartz J (1998) Principles of analysis of epidemiological time series and confounder control, Advanced seminar "Time series analysis in epidemiology", February 12-13, Neuherberg (Munich), Germany.