$ Does%Place%Matter%for%Policy?%%% The$Effect$of$Local$Characteristics$on$Intervention$Priorities$ $in$the$rethink$health$dynamics$model$ Work&in&Progress Prepared by Jack Homer, PhD on behalf of the RTH Dynamics Team August 2013 % Research points to the importance of place as it affects population health and longevity (Mohney 2013; Wenger 2012), as well as health care utilization and costs (Fisher et al. 2009; Wennberg 2010). Population health differences by geography are closely related to health disparities that exist among different socioeconomic, racial, and ethnic groups, underlining the importance of social determinants of health (Braveman et al. 2010; Braveman et al. 2011; Marmot et al. 2008; Smedley et al. 2002). Cost differences by geography, in contrast, reflect not only health disparities but also the local habits and practice styles of providers (Gawande 2009). Although all may agree that these place-based differences exist, the policy implications are not obvious. For example, should a city or county with a large fraction of disadvantaged people have different priorities for improving its health system than one with a smaller fraction, even though both have the same aims of improving health and reducing costs? This question seems similar to the long-standing debate about whether it is better, from a public health standpoint, to intervene in a targeted way with high-risk subpopulations or to address the broader population (Frohlich and Potvin 2008; Goldman et al. 1989; Mechanic and Tanner 2007; Rose 1985). If the population-wide approach were best, then perhaps all places should adopt the same broad policies. If, on the other hand, some degree of targeting were warranted, then perhaps places with larger high-risk subpopulations should put a higher priority
Work&in&Progress DoesPlaceMatterforPolicy?&&Page2 on such targeting than places with smaller high-risk subpopulations. Computer simulations in the areas of hypertension and cardiovascular disease have in fact suggested that a mix of targeted and untargeted interventions may in general be best (Ahern et al. 2008; Homer et al. 2013). This is a suggestive finding, but it does not address the question of place directly. To do so, a computer simulation model may be calibrated to represent different regions or localities, and the results compared. For example, the Center for Disease Control and Prevention s PRISM model of cardiovascular risk has been calibrated to represent the US overall, as well as several localities within the US (Homer et al. 2010). This model includes several dozen intervention levers that may be clustered under the four categories of Clinical Services, Behavioral Support, Health Promotion & Access, and Taxes & Regulation. An analysis was done comparing the impacts of each of these four clusters on deaths and costs over a simulated period of 30 years. This analysis was done for a calibration of the model representing the US overall, and also for a calibration representing a county with a large black population and a higher prevalence of poverty. Although the relative magnitudes of simulated impact were somewhat different for the US compared with the county, the rankings of the four clusters were identical. For example, in both cases, the top-ranked cluster for reducing deaths in the short term was Clinical Services, while the top-ranked cluster for reducing deaths in the long term and also reducing costs in both the short and long term was Taxes & Regulation (Hirsch et al. 2012). The ReThink Health Dynamics (RTH) simulation model (Homer, August 2013) can similarly be calibrated to represent different localities and tested to determine to what extent place matters for policy. This paper reports the results of a first analysis along these lines, specifically addressing the question: Does place affect the simulated outcome rankings of the
Work&in&Progress DoesPlaceMatterforPolicy?&&Page3 RTH model s 21 health care and health risk interventions, when considering these interventions individually rather than in combination? Calibrating%the%Model%for%Three%Places% The current version of the model (Version 2c; see the Reference Guide, Homer, May 2013) has so far been calibrated to three places: Anytown, with per-capita metrics exactly matching those of the nation overall; Atlanta, Georgia; and Morris County, New Jersey. The model was first calibrated to Anytown, based on a wide variety of US data and studies, including some surveys available only at the national level. These national data allowed us to understand how health risks, health status, and health care utilization differ by population age, insurance status, and income level. Calibration to Atlanta and Morris started with gathering as much data as was available locally. When local data did not exist for a particular metric, we estimated synthetically by applying national coefficients to the age, insurance, and income distribution of the local population. Model input parameters were then adjusted until the model was able to reproduce all available data covering the 2000 to 2010 period data on changes in population size, births, deaths, uninsurance, and disadvantage, as well as data (available for Morris but less so for Atlanta) on the utilization of physicians, hospitals, nursing facilities, and home health care. Table 1 shows a comparison of the three places along multiple population health and health care metrics as estimated for 2010. The hard data for Atlanta (as opposed to the estimates inferred through synthetic estimation, mostly in the area of health care) indicate that, compared with Anytown, Atlanta has fewer seniors and has a lower prevalence of behavioral and environmental health risks. It has the same level of disadvantage but more uninsurance, a
Work&in&Progress DoesPlaceMatterforPolicy?&&Page4 somewhat higher age-standardized death rate, and greater health disparity as reflected by the disadvantaged fraction of deaths. The data for Morris indicate that, compared with Anytown, it is much less disadvantaged, has less uninsurance, and has a lower prevalence of behavioral and environmental health risks. It has significantly less severe chronic physical illness but more physicians per capita and fewer ER visits per capita. For extended care, patients are more likely to use home health care rather than a nursing facility, whereas the reverse is true in Anytown. Morris has a significantly lower agestandardized death rate and, in line with its low disadvantaged population, a rather low disadvantaged fraction of deaths. Taking all of these facts are taken into account, the RTH model (run through its thousands of equations) estimates that the average age-standardized health care cost per capita is about 2% lower in Atlanta than in Anytown, and about 4% lower in Morris than in Anytown. 1 [Table 1 about here] Ranking%the%Intervention%Options%in%the%Three%Places% Following model calibration, base run projections for the period 2010-2040 were created for the Anytown, Atlanta, and Morris models. These projections require future assumptions for the model s 19 specified input trends. The exogenous trends are described in detail in the Reference Guide, including baseline assumptions made for the Anytown model. These trend inputs were set the same or similarly for the Atlanta and Morris models as for Anytown. For example, in all 1 Morris County is known to have health care costs below those of New Jersey as a whole. The Dartmouth Atlas indicates that Medicare costs in Morris are the lowest of all Hospital Referral Regions in the state. Using national MEPS cost data broken out by population segment and applying to the local populations, a synthetic estimation done by the author suggests that New Jersey should be expected to have health care costs greater than those of the US overall, whereas Morris and Atlanta should be expected to have health care costs below those of the US overall. (Spreadsheet and related e-mails available from Dr. Homer upon request.)
Work&in&Progress DoesPlaceMatterforPolicy?&&Page5 three models the base run assumes, for simplicity, no future step changes in insurance eligibility, such as might occur with implementation of the federal Affordabable Care Act. Starting with these base run projections for the three places, we next tested individually each of the intervention options in the RTH model that seek to improve health care or reduce risks to health (see Homer, May 2013, and Homer, August 2013.) The Coordinate Care option was tested both without and with updating of guidelines through regular technology assessment; this brings the total number of options to 21. Each of these options was tested, starting in 2012, to its maximum extent addressing the full population rather than targeting particular subgroups, and with a large innovation fund sufficient to cover all required program spending through 2040. For each of the three models, we ranked the 21 options (from 1 to 21) along five cumulative outcome metrics: healthcare costs per capita (age and price adjusted), death rate (age adjusted), disadvantaged deaths as a fraction of all deaths (a measure of health inequity), value of employee productivity (based on average employment income per population segment reduced by productivity losses from absenteeism and presenteeism), and program spending. These rankings are presented in Table 2. All intervention impacts are accumulated from 2012, when all interventions are assumed to begin, to 2040. For the first four metrics healthcare costs, death, inequity, and productivity the top 5 interventions (ranks 1-5) are highlighted with green cell shading. For the fifth metric, program spending, the five most expensive interventions (ranks 17-21) are highlighted with pink cell shading. [Table 2 about here]
Work&in&Progress DoesPlaceMatterforPolicy?&&Page6 There is much similarity in the Top 5 rankings across the three places, but also a few interesting differences that may be traced to the place-based differences seen in Table 1. With regard to reducing healthcare costs, among the top interventions for all three locations are Coordinate Care (with or without updating), Healthier Behaviors, and Generic Drugs. This leaves one additional top intervention, which for Anytown is Environmental Hazards (effective in a place high in hazards), for Atlanta is Family Pathways (effective in a place high in disadvantage), and for Morris is Malpractice Reform (effective, relatively speaking, in a place not high in either hazards or disadvantage, but, where like everywhere in the US, health care costs are elevated in part by defensive medicine.) Note that Pre-visit Consultation is also a highly-rated intervention for reducing costs in all three places, ranked #6 or #7. With regard to reducing death rate, among the top interventions for all three locations are Healthier Behaviors, Self-Care, and Hospital Infections. This leaves two additional top interventions, which for Anytown are Environmental Hazards and Crime Reduction (effective in a place high in both); for Atlanta are Enviromental Hazards and Family Pathways (effective in a place high in disadvantage and moderate in hazards); and for Morris are Preventive/Chronic Care and Mental Illness Care (effective in a place not high in disadvantage or hazards, but with a high level of insurance coverage allowing the community to benefit from two interventions that can improve the quality and regularity of routine care for chronic physical illness.) With regard to reducing health inequity and improving economic productivity, among the top interventions for all three locations are Healthier Behaviors, Family Pathways, Student Pathways, Mental Illness Care, and Self-Care (which in Atlanta s case is ranked #6). These five are also the most expensive of the 21 interventions in terms of program spending.
Work&in&Progress DoesPlaceMatterforPolicy?&&Page7 Discussion% The three places tested here differ significantly in terms of several population health and health care characteristics in 2010, including levels of disadvantage, uninsurance, aging, behavioral and environmental risk, severe chronic illness, and physician capacity. However, testing of the RTH model indicates that the rank order of individual interventions, in terms of their beneficial impacts on costs, deaths, inequity, and economic productivity, is only somewhat affected by these place-based characteristics. The model suggests that in all three places one should ideally, subject to funding constraints, pursue the interventions of Healthier Behaviors, Self-Care, Care Coordination, Generic Drugs, Family and Student Pathways to Advantage, Mental Illness Care, and perhaps also Pre-visit Consultation. These are the interventions that rise to the top for all three places, when considering the outcome metrics in Table 2. Beyond these place-independent priorities, the results here suggest that certain places, depending on their mix of characteristics, might also make a high priority of Environmental Hazards or Crime Reduction or Malpractice Reform or Preventive/Chronic Care. These results have certain limitations suggesting the need for further analysis. The analysis here considers the 21 health care and health risk interventions individually rather than in combination, and also assumes no funding constraint. It also assumes the existing fee-forservice payment scheme, and does not consider possible interaction of the 21 interventions with movement toward contingent global payments. It also assumes baseline exogenous trends, and does not consider, for example, the possible interaction of the 21 interventions with expansions in insurance eligibility or other changes due to the Affordable Care Act.
Work&in&Progress DoesPlaceMatterforPolicy?&&Page8 The analysis here also considers only the three places to which the current version of the RTH model has been applied. Other places will have their own unique mix of characteristics, and perhaps their optimal intervention priorities will turn out to be somewhat different as well. Some places looking to use RTH for health system planning may thus find value in developing their own customized calibration rather than using one of the existing calibrations as a proxy. References% Ahern J, Jones MR, Bakshis E, Galea S. Revisiting Rose: comparing the benefits and costs of population-wide and targeted interventions. The Milbank Quarterly 2008; 86(4):581-600. Braveman PA, Cubbin C, Egerter S, Williams DR, Pamuk E. Socioeconomic disparities in health in the United States: what the patterns tell us. Am J Public Health 2010; 100:S186- S196. Braveman PA, Egerter S, Williams DR. The social determinants of health: coming of age. Annual Review of Public Health 2011; 32:381-398. Fisher ES, Bynum JP, Skinner JS. Slowing the growth of health care costs lessons from regional variation. New Engl J Med 2009; 360(9):849-852. Frohlich KL, Potvin L. The inequality paradox: the population approach and vulnerable populations. Am J Public Health 2008; 98(2):216-221. Gawande A. The cost conundrum: what a Texas town can teach us about health care. The New Yorker, June 1, 2009. Goldman L, Weinstein MC, Williams LW. Relative impact of targeted versus population-wide cholesterol interventions on the incidence of coronary heart disease: projections of the Coronary Heart Disease Policy Model. Circulation 1989; 80:254-260. Hirsch G, Homer J, Wile K, Trogdon JG, Orenstein D. Using simulation to compare intervention classes for reducing cardiovascular disease risks: results for the United States and a less-advantaged county. Manuscript pending clearance at Centers for Disease Control and Prevention (available from Dr. Homer); August 2012. Homer J. ReThink Health: a simulation model of health system transformation. Reference guide for model version 2c, Anytown, USA calibration. For Fannie E. Rippel Foundation, Morristown, NJ; May 2013. Homer J. Introduction to the ReThink Health Dynamics model: simulating local health reform in Anytown USA. For Fannie E. Rippel Foundation, Morristown, NJ; August 2013.
Work&in&Progress DoesPlaceMatterforPolicy?&&Page9 Homer J, Milstein B, Wile K, Trogdon J, Huang P, Labarthe D, Orenstein D. Simulating and evaluating local interventions to improve cardiovascular health. Preventing Chronic Disease 2010; 7(1). Available at: http://www.cdc.gov/pcd/issues/2010/jan/08_0231.htm. Homer J, Trogdon J, Wile K, Cooper L. PRISM: the Prevention Impacts Simulation Model of cardiovascular risk. Manuscript pending clearance at Centers for Disease Control and Prevention (available from Dr. Homer); April 2013. Marmot M, Friel S, Bell R, Houweling TA, Taylor S. Closing the gap in a generation: health equity through action on the social determinants of health. The Lancet 2008; 372:1661-1669. Mechanic D, Tanner J. Vulnerable people, groups, and populations: societal view. Health Affairs 2007; 26(5):1220-1230. Mohney G. Wrong zip code can mean shorter life expectancy. ABC News, Good Morning America, July 19, 2013. Available at: http://gma.yahoo.com/wrong-zip-code-mean-shorterlife-expectancy-213124529--abc-news-topstories.html. Rose G. Sick individuals and sick populations. Intl J Epidem 1985; 14:32-38. Smedley BD, Stith AY, Nelson AR, editors. Unequal treatment: confronting racial and ethnic disparities in health care. Institute of Medicine, The National Academies Press: Washington DC; 2002. Wenger M. Place matters: ensuring opportunities for good health for all; a summary of Place Matters community health equity reports. Joint Center for Political and Economic Studies. September 2012. Available at: http://www.jointcenter.org/research/place-matters-ensuringopportunities-for-good-health-for-all. Wennberg JE. In health care, geography is destiny. Chapter 1 of Tracking medicine: a researcher s quest to understand health care. Oxford University Press: New York; 2010.
Work&in&Progress DoesPlaceMatterforPolicy?&&Page10 Table 1. Local health system metrics for 2010: Anytown, Atlanta, and Morris POPULATION STATUS & HEALTH CARE, 2010 Metric Anytown Atlanta Morris Disadvantaged % 34% 34% 14% Uninsured % 15% 20% 9% Youth (0-17) % 24% 24% 24% Senior (65+) % 13% 9% 13% Risky behavior % 62% 53% 52% High crime neighborhood % 25% 13% 9% Hazardous environment % 24% 14% 8% Severe chronic physical illness % 12% 11% 8% Uncontrolled mental illness % 8% 9% 8% Average income per capita $21,372 $22,575 $25,801 Absenteeism/presenteeism productivity loss %* 5.2% 5.2% 5.2% Death rate (% per year) 0.81% 0.66% 0.67% Age-standardized death rate (% per year) 0.81% 0.85% 0.68% Disadvantaged % of deaths 46% 55% 22% PCPs per 10,000 population 7.5 7.5 10.5 Specialists per 10,000 population 12.5 12.4 13.4 Adequacy of PCP capacity for Disadv. Uninsured* 76% 88% 71% Hospital ER visits per capita 0.42 0.41 0.23 Non-urgent fraction of ER visits* 17% 15% 14% Hospital inpatient stays per capita 0.12 0.11 0.10 Readmission rate (within 30 days of initial stay) 15.5% 15.4% 17.6% Nursing facility census per 10,000 population 52 42 49 Home health census per 10,000 population 48 41 59 Health care costs per capita* $7,188 $6,761 $6,854 Age-standardized health care costs per capita* $7,177 $7,054 $6,869 * These estimates calculated for all locations using national coefficients applied to local data These Atlanta estimates calculated (entirely or in part) using national coefficients applied to local population
Work&in&Progress DoesPlaceMatterforPolicy?& Table 2. Ranking of 21 interventions along 5 cumulative outcome metrics through 2040 (1=best, 21=worst) RANKING OF INTERVENTIONS (1=best, 21=worst) ALONG 5 OUTCOME METRICS, BASED ON CUMULATIVE IMPACT 2012-2040 Outcome Metric Healthcare Costs Death Rate Health Inequity Economic Productivity Program Spending Intervention Anytown Atlanta Morris Anytown Atlanta Morris Anytown Atlanta Morris Anytown Atlanta Morris Anytown Atlanta Morris Coordinate Care 3 3 2 13 12 12 9 9 7 8 7 8 11 11 13 Coordinate Care (with Updating) 1 1 1 11 11 11 8 7 6 6 5 7 12 12 14 Crime Reduction 6 11 11 4 8 8 7 8 8 9 10 12 16 14 12 Environmental Hazards 5 10 12 3 3 7 6 6 10 7 9 11 15 16 11 Family Pathways 8 4 14 8 4 9 1 1 1 1 1 1 21 21 21 Generic Drugs 4 5 4 16 15 17 10 10 13 11 11 15 3 3 4 Healthier Behaviors 2 2 3 1 1 1 3 3 5 4 4 4 17 17 18 Hospice 17 17 17 21 21 21 19 18 19 20 20 21 1 1 1 Hospital Infections 14 15 9 5 5 4 20 20 20 10 8 6 8 8 6 Malpractice Reform 11 12 5 18 18 16 14 14 12 16 16 13 9 7 9 Medical Home 20 20 10 9 10 6 18 19 9 21 21 10 6 6 8 Mental Illness Care 18 18 19 7 6 5 5 4 4 3 3 2 19 20 20 PCP Efficiency 15 13 15 15 16 15 16 15 17 18 18 19 5 5 7 Post-Discharge Care 9 9 8 17 17 18 13 13 14 13 12 16 7 9 5 Preventive/Chronic Care 19 19 20 6 7 3 21 21 21 12 15 9 10 10 10 Pre-Visit Consultation 7 6 6 14 14 14 12 12 11 14 14 14 14 15 15 Recruit PCP (FQHC) 10 7 13 12 13 13 11 11 16 15 13 18 4 4 3 Recruit PCP (General) 16 16 18 20 20 20 17 17 18 19 19 20 13 13 16 Self-Care 21 21 21 2 2 2 4 5 3 5 6 5 18 18 19 Shared Decision Making 13 14 7 19 19 19 15 16 15 17 17 17 2 2 2 Student Pathways 12 8 16 10 9 10 2 2 2 2 2 3 20 19 17 Green shading indicates 5 best interventions for each of the four health metrics. Pink shading indicates 5 with the highest program spending.