Document rBX7r2E6bkGjQyQd1NQ38Q3Kr

ENVIRONMENTAL FACTORS THAT EXPLAIN VARIATION IN SERUM DIOXIN CONCENTRATIONS IN A COMMUNITY IN MICHIGAN, USA David H. Garabrant, MD, MPH The University of Michigan Dioxin Exposure Study The University of Michigan, Ann Arbor, MI, USA School of Public Health Institute for Social Research College of Engineering Center for Statistical Consultation & Research August 21, 2006 Authors School of Public Health Alfred Franzblau, MD Lynn Zwica, MS Kristine Knutson, MPH Elizabeth Hedgeman, MS, MPH Qixuan Chen, MS Shih-Yuan Lee, MS Center for Statistical Consulting and Research Brenda W. Gillespie, PhD Camelia Sima, MS Scott Swan, MS Danielle Gwinn College of Engineering Peter Adriaens, PhD, PE Avery Demond, PhD, PE Tim Towey, MS Shu-Chi Chang, PhD Institute for Social Research James Lepkowski, PhD, MPH Barbara Lohr Ward, MBA Kathy Ladronka Kristen Olson, MS Jennifer Sinibaldi Financial support for the study comes from The Dow Chemical Company through an unrestricted grant to the University of Michigan. University of Michigan Dioxin Exposure Study Slide 2 August 21, 2006 Study Design We measured the WHO 29 PCDD, PCDF, and PCB congeners in serum, house dust and soil. Today we are reporting the results for the TEQ and 7 specific dioxins that are the major contributors to the TEQ in people's blood in our study. Congener 2,3,7,8-TCDD 1,2,3,7,8-PnCDD 1,2,3,6,7,8-HxCDD 2,3,4,7,8-PnCDF PCB 126 PCB 118 PCB 156 1998 WHO TEF Value 1.0 1.0 0.1 0.5 0.1 0.0001 0.0005 University of Michigan Dioxin Exposure Study Slide 3 August 21, 2006 Study Design We studied people who live in five geographic areas: The Floodplain of the Tittabawassee River The Near Floodplain The Midland Plume Other Midland/Saginaw For comparison, Jackson/Calhoun Counties Midland/Saginaw: Floodplain, Near Floodplain, Midland Plume and Other Midland/Saginaw combined into one geographic area. University of Michigan Dioxin Exposure Study Slide 4 August 21, 2006 Study Design University of Michigan Dioxin Exposure Study Slide 5 August 21, 2006 Number of study participants University of Michigan Dioxin Exposure Study Slide 6 August 21, 2006 Modeling Strategy for Blood Dioxin Predictors Backwards selection from multiple imputed data sets Identify potential explanatory factors for consideration in further models Linear regression models Log10(blood) = + 1(factor1) + ... + n(factorn) + error University of Michigan Dioxin Exposure Study Slide 7 August 21, 2006 Modeling Strategy for Blood Dioxin Predictors Backwards selection from multiple imputed data sets. Rest of the predictors 946 participants, 190 potential predictors. Force age, age2, BMI, breast feeding, pk-yrs, region, dust, soil_hp1 Soil_max, soil_fp1 Backward selection on Section A&H: Model 1a Backward selection on Section B: Model 1b Backward selection on Section C: Model 1c Backward selection on Section D: Model 1d Backward selection on Section F&G: Backward Model 1f selection on Section E: Model 1e Insignificant variables Backward selection on all the predictors identified in the previous models: Combined Model 1 Force all predictors in the combined model 1 into the backward selected models, and see if any other variables enter: Combined model 2 University of Michigan Dioxin Exposure Study Slide 8 August 21, 2006 Modeling Strategy for Blood Dioxin Predictors Final model from imputation 1: Combined model 2a Final model from imputation 2: Combined model 2b Final model from imputation 3: Combined model 2c Variable is selected for inclusion if it is significant in > 3 models, then refit all 5 models with this list of variables Backwards selection from variables selected in previous step. Each step is based on 5 imputed data sets. Final model from imputation 4: Combined model 2d Final model from imputation 5: Combined model 2e University of Michigan Dioxin Exposure Study Slide 9 August 21, 2006 Issues about models Collinearity: many of the foods (and some other variables) were collinear. Variables were removed from consideration when the variance inflation factor was 10 or greater. Because some parameter estimates were unstable, we chose to estimate the contribution of variables in blocks (food, work, property use, etc.). The contribution of a block of variables to a model is assessed by the difference in adjusted R2 between the full model and the reduced model (without the block of variables). Choice of functional form of the variables Both linear and categorical forms were modeled for ordinal variables. We chose categorical forms to avoid assuming any specific functional form for predictors. University of Michigan Dioxin Exposure Study Slide 10 August 21, 2006 Results: Health/Demographic predictors of serum dioxin concentration Variables in blue are forced into models. Parameter estimates in pink are positive associations (p<0.05) Parameter estimates in green are negative associations. (p<0.05) Age and BMI are positively associated with most congeners. There are important interaction terms between age*BMI, age*sex, and BMI*sex. University of Michigan Dioxin Exposure Study Slide 11 August 21, 2006 Interaction between age, BMI and sex Among females, TEQ 60 rises less as BMI increases 50 Male BMI 20 Male BMI 40 Among males, Female BMI 20 Female BMI 40 TEQ rises more 40 as BMI increases 30 TEQ 20 10 0 20 30 40 50 60 AGE University of Michigan Dioxin Exposure Study Slide 12 70 August 21, 2006 Results: Region, soil, and house dust predictors of serum dioxin concentration Region: People who live in the Floodplain have higher levels of TCDD, 2,3,4,7,8-PeCDF, and 1,2,3,7,8PeCDD Near Floodplain have higher levels of TEQ, TCDD, 2,3,4,7,8-PeCDF, 1,2,3,7,8-PeCDD, and PCB-126 Other Midland/Saginaw have higher levels of TCDD and 1,2,3,7,8-PeCDD Midland Plume have higher levels of TCDD than do people who live in Jackson/Calhoun. University of Michigan Dioxin Exposure Study Slide 13 August 21, 2006 Results: Region, soil, and house dust predictors of serum dioxin concentration Soil and House dust: Living on property with a maximal soil concentration of 1,000 parts per trillion TEQ of dioxins was associated with higher levels in blood of 0.7 parts per trillion (2%) for the TEQ. 4% of the properties tested had a soil TEQ at or above 1,000 parts per trillion (among all soil samples on the property). Living on property with house perimeter top 1 inch soil containing 1,000 parts per trillion of PCB-118 was associated with higher levels in blood of 18 parts per trillion (less than 1%) for PCB-118. Living on property with house perimeter top 1 inch soil containing 40 parts per trillion of PCB-126 was associated with higher levels in blood of 0.9 parts per trillion (5%) for PCB-126. House dust was not related to serum dioxin concentrations with the exception of a weak association for PCB-118. University of Michigan Dioxin Exposure Study Slide 14 August 21, 2006 Results: Milk and eggs predictors of serum dioxin concentration Eggs and Milk: Positive association for eggs raised in other areas (not in the floodplain of the Tittabawassee River, Saginaw River, or Saginaw Bay) and most congeners. Positive association for store bought milk and PCB-126 University of Michigan Dioxin Exposure Study Slide 15 August 21, 2006 Results: fruits and vegetable predictors of serum dioxin concentration Fruits and Vegetables: Generally negative associations for fruits, vegetables, and root vegetables, whether raised in the contaminated areas or raised elsewhere A few positive associations for store bought fruits, vegetables, and root vegetables. University of Michigan Dioxin Exposure Study Slide 16 August 21, 2006 Results: Meat, game meat, and hunting predictors of serum dioxin concentration Meat, Game Meat, and Hunting: Findings suggest that meat (bplv= beef, pork, lamb, veal), poultry (ctdg=chicken, turkey, duck, goose), game meat (sr=squirrel, rabbit), and hunting are associated with higher dioxin levels in blood. Game meat and hunting in the contaminated areas not clearly different than in other areas. University of Michigan Dioxin Exposure Study Slide 17 August 21, 2006 Results: Fish consumption and fishing predictors of serum dioxin concentration Fish and Fishing: Eating fish in general, eating fish from the contaminated areas, and fishing in Saginaw River and Saginaw Bay are associated with higher dioxin levels in blood. University of Michigan Dioxin Exposure Study Slide 18 August 21, 2006 Results: Fish consumption and fishing predictors of serum dioxin concentration People who ate fish from the Tittabawassee River, Saginaw River, and Saginaw Bay between 1980 and the present have higher levels of some dioxins in their blood than people who did not eat fish from these areas. For every one year of consumption the increase is: 0.23 parts per trillion (0.9%) for the TEQ 0.03 parts per trillion (2%) for TCDD 0.05 parts per trillion (1.1%) for 1,2,3,7,8 PentaCDD 0.34 parts per trillion (0.9%) for 1,2,3,6,7,8 HexaCDD No apparent effect on the other specific dioxins University of Michigan Dioxin Exposure Study Slide 19 August 1251, 2006 Conclusions University of Michigan Dioxin Exposure Study Slide 20 August 1251, 2006 Conclusions University of Michigan Dioxin Exposure Study Slide 21 August 1251, 2006 Results: Explained variation in serum dioxin concentration The regression model explains 78% of the variation in serum TEQ. 51% of the variation in serum TEQ is explained by Health/Demographic variables: age, sex, BMI, smoking, breast feeding. Region, soil contamination, and house dust contamination explain only small fractions of the variation in TEQ or any specific congener. University of Michigan Dioxin Exposure Study Slide 22 August 21, 2006 Conclusions Age, sex, BMI, and demographic factors account for ~50% of the variation in the blood levels of dioxins (TEQ) among people. These are the most important factors related to levels in people's blood. Eating fish and game (especially from the contaminated areas), doing water-related activities and certain occupations combined to account for 1-6% of the variation in blood levels of dioxins among people. Living on contaminated soil, living in Midland/Saginaw, and contaminated household dust accounted for about 0.2-1.0% of the variation in the blood levels of dioxins among people. University of Michigan Dioxin Exposure Study Slide 23 August 1251, 2006 END