|M.Sc Student||Wafa Deabat|
|Subject||Investigation of the Conclusions on the Relationship between|
Air Pollution and Respiratory Diseases in the
Haifa Metropolitan Area
|Department||Department of Industrial Engineering and Management||Supervisor||Ms. Cohen Ayala (Deceased)|
|Full Thesis text - in Hebrew|
The motivation for this research was a problem which was raised following analyses on the short term effect of air pollution on health. These analyses were carried out in several research projects, among them a large scale worldwide project known as APHEA - (Air Pollution and Health Europe Approach). In Israel, a similar research was carried out on data related to the Haifa metropolitan area for the period June 1'st
1996-May 31'st 1999, (Cohen et al., 2002). The current work concerns the data of the Haifa research by Cohen et al.
The underlying model used in these analyses was Poisson regression with the log as the link function. The explanatory part was modeled by both parametric and non-parametric components. The relationship between the dependent variable and each of the pollutants, as well as the humidity, were modeled as a linear function, while the confounders: time and temperature; were modeled by applying a smooth non-parametric component. The model also accounted for the daily and season variations, by including dummy variables and an interaction term between season and humidity. The model was fitted by applying the GAM (Generalized Additive Models) procedure of the software Splus.
In 2002 researchers at the University of Johns Hopkins found out that there were two significant problems in the software Splus regarding the fitting algorithm for GAM. These problems are as follows: 1) The parameters of the default choice in the Backfitting algorithm did not guarantee the convergence of the algorithm. This could have led to biased estimates of the regression coefficients relating the pollutants with the dependent variables. 2) The estimation method applied for the standard errors of the regression coefficients was based on the assumption that the non-parametric component was linear. This assumption resulted in an underestimation of these standard errors.
The current research was aimed to reanalyze the Haifa data by applying correct methods and to compare previous versus new results. Several smoothing methods were used for estimating the non-parametric component in the GAM models. These methods are: Natural Cubic Splines and Cubic Regression Splines.
When comparing the results of the current research with those results of the previous one, it was found that there are differences in the power of the associations between outdoor air pollution and visits to emergency rooms or admission, but not in the direction. In general, most of the relative risks that were received are lower than these of the previous research.