Smoking data from BRFSS surveys, 2000 - 2008 Robert Delongchamp, MPH, PhD Professor of Epidemiology,...

45
Smoking data from BRFSS surveys, 2000 - 2008 Robert Delongchamp, MPH, PhD Robert Delongchamp, MPH, PhD Professor of Epidemiology, CoPH Professor of Epidemiology, CoPH Consultant, ADH Consultant, ADH Tobacco Master Settlement Tobacco Master Settlement 9/21/2010 1 Health Research, Policy and Health Promotion Conference

Transcript of Smoking data from BRFSS surveys, 2000 - 2008 Robert Delongchamp, MPH, PhD Professor of Epidemiology,...

  • Slide 1
  • Smoking data from BRFSS surveys, 2000 - 2008 Robert Delongchamp, MPH, PhD Professor of Epidemiology, CoPH Consultant, ADH Tobacco Master Settlement Tobacco Master Settlement 9/21/20101Health Research, Policy and Health Promotion Conference
  • Slide 2
  • To construct credible estimates of smoking prevalence for specific sexes, ages, locations and times from data collected in BRFSS surveys That is, I want the 2008 smoking prevalence among 40 year old males living in Mount Ida! 9/21/20102Health Research, Policy and Health Promotion Conference
  • Slide 3
  • Outline Behavioral Risk Factor Surveillance System Direct estimates of smoking prevalence Simulated smoking prevalence in a cohort State trends in smoking prevalence Geographically weighted regression (GWR) Regional trends in smoking prevalence 39/21/2010Health Research, Policy and Health Promotion Conference
  • Slide 4
  • Behavioral Risk Factor Surveillance System www.cdc.gov/brfss/about.htm 9/21/20104Health Research, Policy and Health Promotion Conference
  • Slide 5
  • BRFSS Annual telephone survey (nation-wide) probability sample of Arkansans with landlines self-reporting of several risk factors selected the surveys from years, 2000 through 2008 9/21/20105Health Research, Policy and Health Promotion Conference
  • Slide 6
  • Relevant BRFSS Data Annual telephone survey Smoking Questions respondents assigned to 3 categories (_smoke3) Current smoker (> 100 cigarettes & smoke: every day or some days) Former smoker (> 100 cigarettes & smoke: not at all) Never smoked (< 100 cigarettes) doesnt deal with smokeless tobacco, pipes & cigars 9/21/20106Health Research, Policy and Health Promotion Conference
  • Slide 7
  • Relevant BRFSS Data Annual telephone survey Smoking Questions Demographic Information age: selected 35 through 84 sex 9/21/20107Health Research, Policy and Health Promotion Conference
  • Slide 8
  • Relevant BRFSS Data Annual telephone survey Smoking Questions Demographic Information zip code convert to a region a.k.a. Zip Code Tabulation Area (ZCTA) deleted respondents w/o ZCTA 9/21/20108Health Research, Policy and Health Promotion Conference
  • Slide 9
  • Records 30,457 from Arkansas 47 from Mount Ida 18 From Huttig 9/21/20109Health Research, Policy and Health Promotion Conference
  • Slide 10
  • Estimates of smoking prevalence based on the survey design Some can be downloaded from the web site 9/21/201010Health Research, Policy and Health Promotion Conference
  • Slide 11
  • Age-sex-year specific estimates direct estimates (design-based approach) 180 parameters (5 x 9 x 2 x 2) 5 age groups: 35 44, , 75 84 9 years: 2000 to 2008 Both sexes trinomial response (never, current, former) 9/21/201011Health Research, Policy and Health Promotion Conference
  • Slide 12
  • Arkansas estimates direct estimates 180 parameters / 270 estimates large CIs small numbers of respondents Mt Ida? 47 respondents 9/21/201012Health Research, Policy and Health Promotion Conference
  • Slide 13
  • Working toward regional estimates Major problem: sample size limits the precision of direct estimates 180 parameters are too many for precise estimates of Arkansas rates let alone Mt. Ida Need to reduce number of parameters or increase sample sizes Approach: reduce parameters by modeling the main trends in these data 9/21/201013Health Research, Policy and Health Promotion Conference
  • Slide 14
  • No Interaction!!! Patterns with age and sex in smoking prevalence are stable across the years. Can collate the information about age and sex patterns across years. What are the patterns? Type 3 Analysis of Effects EffectDF Wald Chi-SquarePr > ChiSq year1615.790.4675 agegrp81271.92
  • Annual Change Brown regions had increasing prevalence of current smokers (APC > 100) Prevalence of current smokers declined in the remaining regions; highest declines in shades of blue. 9/21/201040
  • Slide 41
  • Prevalence of Current Smokers This shows the decline with age as well as the decline with year. Thus, Mena has fewer current smokers among the cohort which was 40 years old in 2000.
  • Slide 42
  • Conclusions Are ZCTA estimates credible? data-based established methodologies population weighted survey sample (weights respondents) multinomial logistic regression (trinomial responses) geographically weighted regression (weights locations) statistical properties essentially moving averages of state-wide estimates state-wide estimates are good 9/21/201042Health Research, Policy and Health Promotion Conference
  • Slide 43
  • Conclusions Are ZCTA estimates credible? GWR, as defined herein, always gives estimates shrinkage estimate (to state-wide estimates) even where direct estimation is unreliable even at locations w/o data (e.g. White River Refuge or Oklahoma) 9/21/201043Health Research, Policy and Health Promotion Conference
  • Slide 44
  • Nothins free direct estimation is unbiased but may have large variance model-based estimates achieve more precision but may have large bias only as accurate as the model is valid 9/21/201044Health Research, Policy and Health Promotion Conference
  • Slide 45
  • 9/21/201045Health Research, Policy and Health Promotion Conference