Advanced

Sambanden mellan inandningsbara och fina partiklar i luften och insjuknande i stroke - poissonregressionsmodeller

Hillström, Jenny and Nsabimana, Joselyne (2009)
Department of Statistics
Abstract
The purpose of this Master’s thesis is to model the short-term association between the number of strokes recorded per day and an increased daily mean level of particulate matter, PM. A short-term association means that the occurrence of stroke is related to the content of PM in the air the same day and up to seven days prior to the stroke (lag 0 to lag 7). Previous studies show that there is a significant association between PM and the incidence of stroke. Poisson regression models are used to explain the relationship in this study.
Three different types of data from Scania are used: daily records of stroke occurrences, daily mean levels of PM2.5 and PM10 and the daily mean temperature. First Scania is divided into eight zones and two... (More)
The purpose of this Master’s thesis is to model the short-term association between the number of strokes recorded per day and an increased daily mean level of particulate matter, PM. A short-term association means that the occurrence of stroke is related to the content of PM in the air the same day and up to seven days prior to the stroke (lag 0 to lag 7). Previous studies show that there is a significant association between PM and the incidence of stroke. Poisson regression models are used to explain the relationship in this study.
Three different types of data from Scania are used: daily records of stroke occurrences, daily mean levels of PM2.5 and PM10 and the daily mean temperature. First Scania is divided into eight zones and two zones are chosen for further analyses. That is zone 3, which includes Helsingborg, Höganäs and Landskrona and zone 6, which includes Malmö. According to previous studies, only patients over 65 years old are included and the data is divided into cold and warm seasons. The cold season ranges from October to April and the warm season from May to September. This study is delimited to study the association at lag 0, 1, 5, 7 accor-ding to previous studies results and furthermore the correlations between the lags that follow each other are very high for the temperature.
Two main procedures, Proc GENMOD and Proc GAM in SAS software are used to estimate the models. For zone 3, one model for each season which includes PM10 and temperature are estimated. While for zone 6, three models for each season are estimated. The first two include each of the particles one by one and the temperature and the third includes all of the three independent variables. Proc GENMOD is first used to estimate the models, but the residuals show that the models are inadequate. Therefore further analyses are done in Proc GAM to discover nonlinear terms that need to be included in Proc GENMOD in order to improve the models. A scatter plot smoother in Proc GAM is also used to help visualise the nonlinear relationship.
To sum, a significant positive relationship is only shown during the cold season 2004-2005 in zone 6. The final estimated models in Proc GENMOD show significant relationship at lag 0, lag 1 and a quadratic term at lag 1 for both PM2.5 and PM10. Lag 0 for PM2.5 and PM10 show a negative relationship to stroke. Lag 1 gives a much greater impact than lag 0, therefore the summed impact on stroke is positive. An increase with 1 μg/m3 for PM2.5 at lag 0, lag 1 and (lag 1)2 implies an estimated increase in stroke occurrence with 33.9 %. Respective value for PM10 is an increase with 12.3 %. However the values estimated in this study cannot literally be compared to the ones estimated in previous studies. Since fewer independent variables and all cases of ischemic stroke are investigated, instead of fatal stroke in previous studies. A model with PM2.5, PM10 and temperature shows that only PM2.5 is significant and it gives the same model as the model with PM2.5 and temperature.
Results from residual analyses show that the residuals are not normally distributed and that the variance is not constant and therefore are the estimated models inadequate. This indicates that the models estimated in this study cannot be used to estimate future stroke occurrences. (Less)
Please use this url to cite or link to this publication:
@misc{1392937,
  abstract     = {The purpose of this Master’s thesis is to model the short-term association between the number of strokes recorded per day and an increased daily mean level of particulate matter, PM. A short-term association means that the occurrence of stroke is related to the content of PM in the air the same day and up to seven days prior to the stroke (lag 0 to lag 7). Previous studies show that there is a significant association between PM and the incidence of stroke. Poisson regression models are used to explain the relationship in this study.
Three different types of data from Scania are used: daily records of stroke occurrences, daily mean levels of PM2.5 and PM10 and the daily mean temperature. First Scania is divided into eight zones and two zones are chosen for further analyses. That is zone 3, which includes Helsingborg, Höganäs and Landskrona and zone 6, which includes Malmö. According to previous studies, only patients over 65 years old are included and the data is divided into cold and warm seasons. The cold season ranges from October to April and the warm season from May to September. This study is delimited to study the association at lag 0, 1, 5, 7 accor-ding to previous studies results and furthermore the correlations between the lags that follow each other are very high for the temperature.
Two main procedures, Proc GENMOD and Proc GAM in SAS software are used to estimate the models. For zone 3, one model for each season which includes PM10 and temperature are estimated. While for zone 6, three models for each season are estimated. The first two include each of the particles one by one and the temperature and the third includes all of the three independent variables. Proc GENMOD is first used to estimate the models, but the residuals show that the models are inadequate. Therefore further analyses are done in Proc GAM to discover nonlinear terms that need to be included in Proc GENMOD in order to improve the models. A scatter plot smoother in Proc GAM is also used to help visualise the nonlinear relationship.
To sum, a significant positive relationship is only shown during the cold season 2004-2005 in zone 6. The final estimated models in Proc GENMOD show significant relationship at lag 0, lag 1 and a quadratic term at lag 1 for both PM2.5 and PM10. Lag 0 for PM2.5 and PM10 show a negative relationship to stroke. Lag 1 gives a much greater impact than lag 0, therefore the summed impact on stroke is positive. An increase with 1 μg/m3 for PM2.5 at lag 0, lag 1 and (lag 1)2 implies an estimated increase in stroke occurrence with 33.9 %. Respective value for PM10 is an increase with 12.3 %. However the values estimated in this study cannot literally be compared to the ones estimated in previous studies. Since fewer independent variables and all cases of ischemic stroke are investigated, instead of fatal stroke in previous studies. A model with PM2.5, PM10 and temperature shows that only PM2.5 is significant and it gives the same model as the model with PM2.5 and temperature.
Results from residual analyses show that the residuals are not normally distributed and that the variance is not constant and therefore are the estimated models inadequate. This indicates that the models estimated in this study cannot be used to estimate future stroke occurrences.},
  author       = {Hillström, Jenny and Nsabimana, Joselyne},
  keyword      = {stroke,partiklar,Poissonregression,generaliserade linjära modeller,generaliserade additiva modeller,Statistics, operations research, programming, actuarial mathematics,Statistik, operationsanalys, programmering, aktuariematematik},
  language     = {swe},
  note         = {Student Paper},
  title        = {Sambanden mellan inandningsbara och fina partiklar i luften och insjuknande i stroke - poissonregressionsmodeller},
  year         = {2009},
}