DETERMINING CAUSALITY IN OBESITY

Transcription

1 DETERMINING CAUSALITY IN OBESITY Jessie P. Buckley, PhD, MPH Assistant Professor Department of Environmental Health and Engineering Johns Hopkins University Bloomberg School of Public Health March 6, 2017

2 Obesity is a heterogeneous, complex disease 2 Yajnik & Yudkin. The Y- Y paradox. Lancet. 2004; 363: 163 Ebbeling et al. Lancet. 2002; 360,

3 Cause 3 Energy in > Energy out 3

4 Going beyond the big two using observa\onal data 4 Risk factor idenificaion Molecular approaches Systems science Causal Inference 4

5 Risk factor idenificaion: From cell to society

6 Gesta\onal smoke exposure and childhood obesity 6 Reference Exposure Relative risk (95% CI) Behl et al. EHP

7 Lep\n signaling pathway 7 Han et al. The Lancet

8 Molecular epidemiology 8 Tools Ø Biomarkers of response Ø Epigene\cs Ø OMICS (e.g., metabolomics) Molecular pathways for environmental obesogens Opportuni\es Ø Biologic pathways Ø Phenotyping Ø Innova\ve study designs Ø Inform poten\al interven\ons More work is needed to determine the public health relevance of associa\ons with molecular markers of obesity 8

9 Societal policies and processes influencing the popula\on prevalence of obesity 9 Kumanyika et al. Circula2on. 2008;118:

10 Systems science 10 Tools Ø System dynamics Ø Agent- based modeling Ø Discrete event simula\on A systems model of childhood obesity Opportuni\es Ø Mul\level influences Ø Dynamic rela\onships Ø Evaluate interven\ons Exis\ng studies integrate across limited set of domains, few applica\ons to childhood obesity Cockrell Skinner & Foster J Obesity

11 Example: dynamics of childhood obesity from community to epigene\cs 11 Geisinger Health System Ø 1288 communi\es in 37 coun\es of Pennsylvania Ø Over 163,000 children aged 3-18 Condi\onal random forest analysis of 44 spa\ally correlated predictors of childhood obesity Will assess obesogenic environments in rela\on to methyla\on of obesity- related genes in a subset Nau et al. Health & Place 2015; Dunstan et al. Clinical Epigene2cs in revision 11

12 Causal inference: IntervenIon framework for decision making

13 in recent decades, our discipline s robust interest in iden\fying causes has come at the expense of a more rigorous engagement with the intent for us to intervene. 13

14 Causal inference approaches 14 Tools Ø Poten\al outcomes framework Ø Causal diagrams Ø G methods Opportuni\es Ø Evaluate interven\ons Ø Leverage longitudinal data Ø Interpretable effect es\mate Ø Direct policy relevance Ø Cost- benefit analyses Few applica\ons of G methods in environmental epidemiology Comparison of risk factor and intervenion approaches Sta\s\cal method Risk factor Regression IntervenIon G methods Goal E\ologic insight Public health impact Sample ques\on What is the dose- response func\on for the associa\on of the exposure with obesity? How many cases of obesity could be prevented by a specific strategy to reduce the exposure? Robins Math Modeling 1986; Taubman et al. Int J Epidemiol

15 Example: interven\ons on childhood obesity 15 Nianogo et al. Pediatric Obesity

16 Types of exposure reduc\on interven\ons 16 Cap: What is the expected change in Y if we implemented an exposure limit on X? ShiR: What is the expected change in Y if we reduced exposure to X for all individuals? Ban: What is the expected change in Y if we eliminated exposure to X? 16

17 Example: evalua\ng exposure limits Cumula\ve lung cancer mortality in the Colorado Plateau Uranium Miners cohort between 1950 and 2005 es\mated under various occupa\onal exposure guidelines for radon 17 Interven\on Risk difference Lung cancer deaths avoided 2 Working- level months (- 4.4, - 2.8) Working- level months (- 5.3, - 3.7) Working- level months (- 6.1, - 4.3) 216 Edwards et al. Epidemiology

18 Mul\causality in an interven\on framework 18 Two poten\al interven\ons to reduce NO x exposure 1) Mandate three- way cataly\c converters à Reduce exposure to NO x 2) Ins\tute a conges\on charge to limit driving trips in urban areas à Reduce exposure to NO x as well as other traffic- related air pollutants and road traffic noise à Increase physical ac2vity 18

19 Bayesian g- formula 19 Extension to incorporate model stabiliza\on techniques developed for mixtures applica\ons Advantages Ø Useful with highly correlated and/or sparse data Ø Incorporate toxicologic knowledge using informa\ve priors Ø Es\mate effect of realis\c interven\ons in the presence of mul\causality and correlated exposures Keil et al. Stat Methods Med Res 19 In press

20 THANK YOU Jessie P. Buckley, PhD, MPH Assistant Professor Department of Environmental Health and Engineering Johns Hopkins University Bloomberg School of Public Health March 6, 2017