Linking of Environmental and Health Indicators by Neural Networks: Case of Breast Cancer Mortality, Slovak Republic

Abstract

This study deals with the analysis of relationship between 67 environmental indicators in geological environment and breast cancer mortality in Slovakia. Primary data comprises a database of 67 environmental indicators for groundwater and soils as a mean value for every Slovak municipality and the data on relative mortality from breast cancer for Slovak municipalities (about 2900) collected for the period of 22 years. The training neural network was used as a mathematical model for data analysis. The top 200 networks have identified 12 environmental indicators (8 for soils: pHKCl, carbonates, Ca, Mg, Na, Co, Sr and 4 for groundwater: Ba, SiO2, Zn, Fe) with high rate of influence. As the most influential environmental indicator was identified pHKCl in soil that should be within the range of acids, definitely under the neutral point. Most of environmental indicators were unambiguously identified as negative (Mg, Ca, Sr, carbonates, Ba, SiO2, Zn and Fe). These indicators should be at the lowest level of their concentration in related geological component. Three environmental indicators, Ce, Na and Co had parabolic function where certain essential concentration levels of these elements are presumed with protective effect on human health. For all influential indicators limit values with respect to the lowest mortality were proposed. The verification of achieved results was performed through regressive model of breast cancer mortality for 12 calculated influential environmental indicators for all municipalities in the Slovak Republic. The revealed difference between real mortality levels and the model values was found for 6 environmental indicators (pHKCl, Mg, Ca, Sr, Ba, Zn). Each executed limit has decreased breast cancer mortality for about 0.79 cases per 100,000 inhabitants.