Jeroen van Meijgaard – UCLA School of Public Health April 14, 2011

1. Population Health Forecasting – Exploring utilization of simulation modeling in Health Impact Assessment Jeroen van Meijgaard – UCLA School of Public Health April 14, 2011

2. HEALTH FORECASTING AT UCLA • Health Forecasting is • a sister project of Health Impact Assessment, both based at the UCLA School of Public Health • a collaborative effort between UCLA, Los Angeles County Department of Public Health, California Department of Public Health • conceived and principally led by Dr Fielding • supported by a mix of foundation and federal grants, supporting a small staff of 2-4 researchers • Funding from • The California Endowment • The Robert Johnson Foundation • UniHealth Foundation (local Los Angeles foundation supporting hospitals) • National Institutes of Health • Target audience • Local Health Departments • Foundations • Legislators and legislative analysts • Advocacy groups

3. THE RELATION BETWEEN HEALTH IMPACT ASSESSMENT AND HEALTH FORECASTING AT UCLA Policy and Program Alternatives Behaviors and Exposures Population Health Outcomes HIA Examine impact of a particular policy or program on exposures and subsequent health outcomes in static population Health Forecasting Examine impact of exposures on outcomes in dynamic population (over time) Short-Medium Term (2-5 Years) Long Term (10+ Years)

4. COMBINING SCOPE OF HIA AND HEALTH FORECASTING REMAINS A CHALLENGE Narrowly targeted programs Relevant geographic challenges, e.g. Built Environment Interactions between individuals and environment Need for specifically defined exposures/risks Ad hoc/ Tailored HIA Uniformly Applicable Model Small/Local Populations Large/Regional Populations Broadly applicable policies May assume uniformity across regions Exposures may be averaged

5. ENABLING DECISION MAKERS TO MAKE MORE INFORMED DECISIONS USING HEALTH FORECASTING The model aims to allow decision makers in health related fields to answers questions at various levels of detail – primarily addressing chronic disease conditions How will mortality rates in the state of California (or any county) change over time? What is the incidence or prevalence of disease X in different counties in California, and how is this expected to change in the next 10 years? How much of the differences in disease incidence rates and other key health outcomes across ethnic and geographic segments can be attributed to known factors? 10 years from now, what will be the effect of a public health intervention Y on the health outcomes for different ethnic and racial groups in Ventura County and Los Angeles County?

6. DEVELOPING THE MODEL AND DISSEMINATING THE RESULTS First we determined feasibility and built a prototype model; disseminating the results has required the development of additional tools Disseminate information (e.g. briefs) and educate and train stakeholders through workshops, presentations and mailings Web-based interface to provide public health practitioners and advocates intuitive access to results from the Health Forecasting model Synthesis of evidence-based research intoa comprehensive Health Forecasting Model

7. INTUITIVE INTERFACE – ENABLING STAKEHOLDERS TO USE MODEL RESULTS FOR LOCAL POPULATIONS The full model will be maintained at UCLA by project team – users can request scenarios to be simulated. A user friendly interface that uses static model output to enable users to perform analysis on a local communities or counties. Users may input community specific demographic information, and the interface provides tables and graphics based on modeling results. • The website is a primary means of wide distribution of tools, results, and analyses • Baseline forecasts • Technical documentation • Simplified version of the model that can be used by local health officers, their staffs and other stakeholders.

8. APPLICATIONS OF THE FORECASTING MODEL • Evaluate research questions about the association between sets of variables that can not be observed directly through surveys, e.g. estimates of life time expenditures associated with levels of physical activity and weight, • Inform debate on important policy issues in public health through issue briefs, • Support community advocacy to strengthen local communities and efforts to improve population health – intuitive access via web-based interface (www.health-forecasting.org), and • Provide analysis on the long term impact of proposed policies and programs.

9. Comprehensive modelling to assess health impacts of proposed policies The impact of U.S. smoking trends on cause specific mortality

10. EVALUATE LONG TERM IMPACT OF POLICIES ON MORTALITY, BY CAUSE OF DEATH Q. What is the impact on smoking reductions in the U.S. on cause of death mortality • Smoking prevalence has reduced significantly; and is still declining • How to evaluate current and future benefits of reductions already attained • What additional gains can be made • How is the mix of causes of death expected to change • How important is it to assess this in a comprehensive framework? A. Use comprehensive lifecourse model with competing causes to assess benefits • Model lifecourse of individuals in U.S. population with cohort specific initiation and cessation rates • Alternative trends are used to estimate to total mortality benefit • End point is comprised of each of the 20 leading causes of death in the United States, plus the residual to comprise all cause mortality

11. COMPONENTS OF THE SIMULATION MODEL Smoking initiation and cessation Smoking Prevalence Mortality Outcomes Leading causes of death Independent competing process Cohort specific rates by age, gender and period Keep track of time since last smoked • Relative risks are calculated for smokers and former smokers based on time since last smoked • Different causes of death are assumed to be independent for now • Tracking of smoking prevalence follows naturally from lifecourse model and is determined by initiation, cessation and excess mortality • Initiation assumptions based on surveys at young age; assume limited initiation at later age • Cessation assumptions based on tracking surveys and sequential cohort surveys

12. LESS THAN 50% OF GAINS IN MORTALITY REDUCTION FROM CHANGES IN SMOKING INITIATION AND CESSATION HAS BEEN REALIZED • 2004 Rates • Smoking Prevalence: Female – 18.9%; Male – 23.1% • Life Expectancy: Female – 80.2; Male – 75.1 • If initiation and cessation rates had stayed at 1950s levels through 2004: • Smoking Prevalence: Female – 32.9%; Male – 47.9% • Life Expectancy: Female – 79.6 (-0.6); Male 73.9 (-1.2) • If current initiation and cessation rates continue to prevail for new cohort: • Smoking Prevalence: Female – 14.4%; Male – 16.4% • Life Expectancy: Female – 81.1 (+0.9); Male – 76.4 (+1.3) Reflects 10-15% of total gain in LE since 1960

13. RESULTS – CAUSES OF DEATH FOR COHORTS WITH VARIOUS INITIATION AND CESSATION RATES

14. DISCUSSION • Lifetime simulation model can be used to estimate impacts over long time horizons • Changes in initiation and cessation of smoking have already yielded substantial gains in life expectancy and will continue to extend life for new cohorts relative to older cohorts provided rates stay the same • Simulation models are flexible in analyzing various cohorts and time horizons, and thus showing when and to whom benefits accrue. • Modelling competing causes suggests shifts in burden of disease • Multiple causes of death in a single framework provides an opportunity to better estimate burden of disease and associated expenditures • Limiting outcomes to one or a few causes of death may overstate benefits of changes in health behaviours even if risks of death are independent as non-modelled causes of death may be associated with behaviour (study in progress)

15. Evaluating policies with uncertain impact estimates Using sensitivity analysis to evaluate to potential impact of retail food environment changes on health and economic outcomes

16. EVALUATE POTENTIAL IMPACT OF ACCESS TO HEALTHIER FOOD Q. Impact of policy/program to improve access to health food in U.S. neighborhoods uncertain • Recent literature suggests type of food stores in neighborhoods impact eating patterns and obesity in local populations • Controlling for selection and confounders remains a challenge • Few available studies to estimate impact, with relatively large standard errors, leaves potential benefit in question • Controlled studies are expensive; however investment in such studies may yield valuable information to better evaluate proposed policies A. Conduct sensitivity analysis with existing information • Model upstream determinants linked to health outcomes and economic outcomes • Use multivariate sensitivity analysis to evaluate potential range of benefits

17. SIMULATION MODEL Retail Food Environment Health Behavior Health /Economic Outcomes Life Expectancy $$ Supermarkets vs Convenience Stores Obesity Prevalence Mortality Medical Expenditures • Mortality associated with obesity, controlled for confounders • Medical expenditures caused by obesity related conditions • Relation between obesity prevalence and BMI distribution • Tracking of obesity over the lifecourse • # of supermarkets • # of convenience stores • Effect size of store type on obesity prevalence

18. SOURCE DATA FOR IMPACT OF RETAIL FOOD ENVIRONMENT ON OBESITY (Morland et al, 2006)

19. SENSITIVITY Scenarios • Reference scenarios: Use prevalence of convenience stores and supermarkets by census tracts • Supermarket scenario: Assume that supermarkets would be available where currently no supermarket is available; this would create areas with both a supermarket and a convenience store • Convenience Store scenario: Assume that convenience stores could be transformed to offer supermarket type services. Sensitivity • Although users of the health forecasting model have rarely requested sensitivity analyses of the results, in this specific case it is useful because there is much uncertainty on a number of variables • Uncertainty on the parameters may be incorporated by multivariate sampling on the parameters domain; here sensitivity is incorporated for impact estimates, relative risks for mortality and association with expenditures.

20. RESULTS SUMMARY

21. DISCUSSION Model can be extended to incorporate policy alternatives • In this analysis policy alternatives were incorporated in the simulation model by allowing users to identify the prevalence of different food environments • Extensions could include more complex dynamics of food purchasing behavior and changes in the retail food environment Multivariate sensitivity analysis is an important component, not only to assess confidence, but also potential • Impacts on outcomes of interests were not considered statistically significant, making it difficult to push for policy/program changes • Yet, the potential of impact may be large: at the 95th percentile the return to a single new supermarket was estimated at $1.3 million annually (reduced medical expenditures and value of gained life) • This uncertainty coupled with potential warrants investments in additional studies

22. Backup

23. WHERE ALONG THE LIFECOURSE ARE GAINS MADE Source: CA-BRFS 1984-2000

24. WHERE ALONG THE LIFECOURSE ARE GAINS MADE? Source: CA-BRFS 1984-2000

25. MAPPING OF THE CHANGE IN OBESITY PREVALENCE INTO A BMI DISTRIBUTIONAL SHIFT Source: CA-BRFS 1984-2000

26. EXAMPLE – MODELLING THE IMPACT OF OBESITY ON MEDICAL EXPENDITURES Overweight and Obesity in California • Observations • BMI levels have increased steadily since the early 1980s • Increases are seen among all groups but are most pronounced among younger people and Latinos • Individual BMI levels are highly correlated over time • BMI and Physical Activity are negatively correlated • Model Implementation • Individual BMI levels are determined by gender, ethnicity, age, previous BMI and Physical Activity • BMI impacts mortality though a relative risk function derived from the literature. RR of BMI on mortality decreases as age increases and are gender specific • BMI trends in the model with three scenarios • Decline to 1984 levels by 2025 • Stable at 2005 levels • Continued increase through 2025

27. EXAMPLE – MODELLING THE IMPACT OF OBESITY ON MEDICAL EXPENDITURES (CONTINUED) Medical Expenditures associated with Obesity and Physical Activity • Direct Personal Medical Expenditures associated with Obesity and Physical Activity are estimated using NHIS data linked with data from the Medical Expenditure Panel Survey 1998-2005 • Medical expenditures are significantly higher for Obese people (BMI>30) among the under 65 population, and significantly higher for Overweight and Obese people (BMI>25) among the over 65 population. • Medical expenditures are significantly lower for people over 65 with recommended levels of Physical Activity (>16 METhrs/wk) • The simulation model allows researchers to analyze expenditures as BMI and PA levels change for each individual from year to year, thus enabling analysis of lifetime medical expenditures