Chemometrics in Occupational Hygiene—how and Why! a Picture Can Tell More than a Thousand Words and Figures!

In this introductory article the author argues for an increased use of a multivariate analytical approach to the complex problems encountered in occupational hygiene. Relations between exposure at the work place and reported health effects arc mostly so complicated and depend on so many factors that methods other than the traditional statistical techniques should be applied. Chemometrics is a field within chemistry where mathematics, statistics and modem computer technology are used to perform multidimensional data analysis. Graphical plots are extensively used to extract the most relevant information from the measurements. The possibility of performing soft modelling through pattern recognition and multifactorial regression analysis will simplify the management of large data sets. A 'metric' philosophy is introduced to describe similarity and dissimilarity among many objects characterized with many variables. This article emphasizes the use of principal component analysis and partial least-squares regression for such purposes. Application of the SIMCA method for classification of objects is also described. These methods arc not dependent upon a priori formulated hypotheses, as in the classical modelling techniques. Instead of being restricted to accepting or rejecting previously formulated hypotheses, these methods may lead to new insights and unperceived features of a complex problem. The application of such exploratory methods may produce new hypotheses and further investigations are necessary to confirm or discard any 'new' chemometric findings. Copyright © 1996 British Occupational Hygiene Society. Published by Elsevier Science Ltd. INTRODUCTION The occupational hygienist is faced with complex problems. As efforts are intensified towards a description and analysis of the total work environment there is a need for a multivariate approach. This is particularly so in studies where occupational health problems are related to the complex hygienic quality at the work place. There are several analytical methods that can handle large, complex data. Some of these methods apply graphical presentations and plots extensively to simplify interpretation. In an earlier paper (Schneider et ai, 1993) a favourite plot of chemometrics appeared in this journal for the first time. In that article Schneider et al. (1993) used a variable loading plot to interpret correlations between the respirable fibre concentrations and various properties of vitreous fibre types, including the content of oil from the production. Such a variable loading plot may be used to find the most relevant or important variables in an investigation. The recognition of variables with low or no importance at all is quite helpful in cases with many variables. In addition the correlations among the variables can be depicted. A variable loading plot is schematically shown in Fig. 1. The variable loadings (XI . . . X6) are given with coordinates in the range + 1 and related to two axes, t} and t^These are calculated so