Better Living through Chemicals? Addressing Substitution Effects in Medicine and Public Health

An interesting new study published in next month’s Journal of Health Economics examines the moral hazard involved in prevention through pharmaceutical therapy, and it has implications for the ACA’s goals of population health improvement. A convincing body of research shows that the dramatic rise in statin use over the past two decades has improved cardiovascular health by reducing the incidence of high cholesterol, serious cardiac events like AMI, and mortality from cardiovascular disease. But do statin users slack off on their nutrition and physical activity, consistent with a substitution effect, possibly leading to other health problems? Or do statin users redouble their healthy living because the drugs increase life expectancy and therefore boost the value of investing in future health status, consistent with a complementary effect?

Robert Kaestner and colleagues use longitudinal data from the famous Framingham Heart Study to investigate these questions. Using data collected on people with moderate to high cholesterol both before and after the introduction of statins in the U.S. market (1991-2001), the researchers observe changes in health behaviors associated with statin use.

Naturally, this study faces considerable risks of bias due to unobserved confounding and endogeneity because people may initiate statins due to unobserved health shocks that also influence their subsequent health behaviors. The authors use one of my favorite techniques to mitigate this bias: fixed effects estimation with a novel instrumental-variables technique. The IVs used here reflect the gradual time trend in the availability and use of statins among people who eventually become users of the drug, capturing the presumably exogenous forces that influence the diffusion pattern of a new drug. And these IVs stand up to the usual tests of strength and excludability (well, mostly…).

Results indicate that statin use led to a small BMI increase and a larger increase in the probability of being obese, consistent with the substitution hypothesis. The effect on physical activity, however, was mixed, with women reducing and men increasing their activity levels in response to statin use. Smoking was not consistently responsive to statin use, but moderate alcohol use increased among males. The authors conclude that statins serve as a strong substitute for healthy dietary intake but not for exercise or other general health behaviors that have a weaker association with cholesterol levels.

Taken as a whole, these results provide another powerful reason for combining clinical and non-clinical approaches when attempting to promote health and prevent disease on a population-wide basis. One approach used in isolation may generate behavioral offsets and unintended consequences that slow progress toward larger population health goals. As a result, medicine and public health strategies applied together may have larger effects than either strategy acting alone – the true definition of synergy. The ACA includes provisions to facilitate and incentivize these types of combined strategies, but more research is needed to determine whether these provisions work as intended. This blog will track the progress of this research, so stay tuned.


The Comparative Effectiveness of Preparedness

The CDC’s anthrax event earlier this week provides a reason to reflect on the uncertainties and difficult policy choices involved in reducing the health risks posed by large-scale disasters. Modern societies face a complex and evolving mix of potential disasters, ranging from intentional acts of violence and terrorism, to uncontrolled viral pandemics, to the weather-produced emergencies fueled by climate change. A new paper from economists at the London School of Economics and MIT offers some useful insight into a critical but too-often obscured question in the preparedness field: how should governments decide which potential disasters to prevent and prepare for, and how much to invest in such strategies?

Ian Martin and Robert Pindyck’s paper shows that cost-benefit analysis of a single strategy to avert a catastrophe, considered in isolation, is unlikely to give us the correct answer about the health and economic value of pursuing that strategy. The policy interdependence of potential catastrophic events means that it is often not socially or economically optimal to avert all such events, even when an apparently cost-effective strategy is available for each hazard. To use preparedness speak, the optimal “all hazards” approach to risk mitigation may not, in fact, seek to avert all hazards.

A key reason for this result is that the existence of one hazard often increases the benefit of averting another hazard, due to the law of diminishing returns. For example, a policy to contain the threat of pandemic influenza will produce more social benefit when other hazards also threaten the population, like catastrophic flooding or industrial accidents. As these background or competing hazards diminish, so too does the value of averting the primary hazard. Martin and Pindyck’s theoretical analysis shows that, when seeking to avert multiple hazards, “the benefits are additive but the costs are multiplicative,” so policymakers should seek to identify the subset of hazard mitigation strategies that maximize net benefit to society.

In practice, identifying this optimal subset of hazard prevention and preparedness strategies can be difficult in the presence of many possible hazards and many ways of reducing risk, which vary across communities and population groups. To address these decision challenges, the authors point to a need for: (1) better measures of risk, vulnerability and resilience; and (2) more research on the effectiveness and cost of prevention and preparedness strategies.

Fortunately, initiatives such as the National Health Security Preparedness Index program are pushing on both fronts – for improved metrics and for expanded research – promising eventually to enable vast improvements in evidence-informed preparedness policy. And even though the CDC’s federal Preparedness and Emergency Response Research Centers Program is sunsetting, this initiative has spawned numerous lines of inquiry that are producing valuable evidence about the effectiveness of preparedness strategies (see for example work that Mary Davis and I have done with colleagues at UNC’s preparedness center to explicate the roles of local public health capabilities in producing preparedness). Some of the most promising recent progress in preparedness research is exploring what makes communities resilient to various types of hazards, and how to promote and reinforce such resilience. See for example Malcolm William’s latest research on this topic presented at last week’s AcademyHealth meeting.

Continued progress in these lines of inquiry will bring us closer to understanding the comparative effectiveness of preparedness: which combination of strategies produce the greatest net benefits, for which populations, and in which community contexts. This type of CER evidence is needed not just for treatment choices facing individual patients, but also for health policy choices facing populations.

Costs, Collective Actions and Crowd-Out in Public Health: New Evidence from ARM

AcademyHealth’s Annual Research Meeting (ARM) convened in San Diego this week to unleash the latest findings from a host of new health services research studies, many with a strong orientation toward economics and public health. I cannot begin to summarize all the great work from this meeting in a single post, so for now I will take a random thought-walk through a couple of the studies I found most striking. (Note: I will update this post in a few days with links to meeting slides once they are posted online).

Front of mind for me is the fascinating pragmatic randomized trial conducted in local health department settings by Allison Saville and her colleagues at the University of Colorado, Colorado Children’s Hospital and Colorado’s state health department. This trial tested the cost-effectiveness of a novel strategy for boosting child vaccination rates using a reminder and recall (R&R) intervention delivered centrally by local health departments in collaboration with community-based primary care practices. The study found that the collaborative, health department-delivered R&R model outperformed a standard physician practice-based R&R model both in terms of vaccination rates and in terms of costs, clearly showing the value of collective action involving public health agencies and primary care practices.

Another elegant study of vaccination practices was presented by LA County Health Department’s Ricardo Basurto-Davila and colleagues, focusing on the health and economic effects of school-based vaccination programs used during the 2009 H1N1 influenza outbreak. The researchers find that school-based strategies produced significantly higher child vaccination coverage rates than did traditional community-based and clinic-based vaccination programs. These higher rates of child coverage, however, came at the expense of coverage among adults 65 years of age and older. In fact, the cost reductions achieved by school-based vaccination programs were attributable mostly to the lower vaccination rates they produced among seniors, rather than to lower disease transmission rates among children. The findings suggest that public health agencies must carefully consider the characteristics of the outbreak and the population groups most at risk when choosing to use mass vaccination strategies for influenza.

Other economically interesting studies focused on food and diet-related public health risks. The University of Washington’s Betty Bekemeier linked financial data on local expenditures for food safety and sanitation with county-level surveillance data on the incidence of food-borne and water-borne diseases in two states over 8 years, finding that some of these diseases are less common in higher-spending communities. A very different study presented by Washington’s Davine Wright used a Markov model to estimate the cost-effectiveness of a school-based screening program designed to detect eating disorders in children, finding an attractive cost per QALY gained of about $57,000.

Penn’s famed behavioral economist and physician Kevin Volpp presented findings from two new trials that explored whether financial incentives tied to weight loss tend to crowd-out the intrinsic motivations people have to achieve a healthier weight. He finds that voluntary participants in randomized weight loss trials tend to have high levels of intrinsic motivation for weight loss at baseline, and that randomly assigned financial incentives do not erode these motivations significantly during the course of a trial. Money does not appear to crowd-out other forms of motivation in these studies.

The new study I presented at ARM this year examined financial crowd-out at a higher scale of operation − one that is particularly relevant now that 26 states are in the midst of implementing the ACA’s Medicaid expansions. Specifically, we asked: does growth in state Medicaid spending crowd out the ability of state governments to invest in other types of beneficial public health programs? And if so, what are the health consequences of such crowd-out? Using almost 20 years of data from 1993-2011, we find evidence of significant crowd-out, consistent with the federal matching formula that rewards states for Medicaid spending but does not provide commensurate incentives for other types of public health outlays. We project that, over time, this crowd-out phenomenon is large enough to result in elevated rates of preventable mortality, offsetting the mortality gains produced by expanded Medicaid coverage. (We were flattered for this work to receive designation as one of the “Best of ARM” studies and as such to receive critiques from noted health economists Steve Parente and Jack Needleman during the meeting).

And speaking of the best, UC-Berkeley’s Tim Brown received the AcademyHealth Article of the Year Award in Public Health Systems Research for his research examining the health effects attributable to public health spending by California’s local agencies. In a pair of recent analyses using careful panel data methods and instrumental-variables techniques, Tim finds sizable reductions in all-cause mortality and notable gains in self-reported health status among California counties that experience growth in per-capita local health department spending during 2001-2008. These studies add to a growing evidence base regarding the health and economic value of public health activities.

Equally importantly, these studies and the larger ARM meeting (including the Public Health Systems Research Interest Group meeting) showcase the value of investing in rigorous research designs and analytic methodologies to answer important questions about public health policy and practice. These studies show that creativity, persistence, and a deep commitment to inquiry and analysis can produce valuable public health knowledge, even when time, money and data are constrained.

Murky Data Undercut the ‘Underfunding’ Argument in Public Health

The Trust for America’s Health latest annual report on U.S. public health expenditures was released last week, reminding us that good data on this topic are so frustratingly elusive. Consistent with past years, the report concludes that the U.S. public health system is “chronically underfunded.” But the sad fact is that available data are woefully inadequate to support this conclusion with any reasonable level of confidence and empirical precision.

As noted in previous posts, a high degree of uncertainty surrounds our best estimates of governmental public health spending. Three basic questions of public finance go mostly unaddressed in public health:

1. How much does the nation currently spend on public health activities?

2. Who pays for these activities?

3. How are funds allocated and used across programs and population groups?

Without solid answers to these questions, it is impossible to make a strong case that public health is “underfunded” in the sense that current spending falls below a given threshold defined by objective measures of need, risk, or efficiency. Moreover, answering the basic public finance questions are necessary prerequisites for determining the health and economic value of investments in public health strategies. TFAH deserves credit for soldiering through this data desert, scraping information from public budget documents, websites, and interviews with public health officials. But these sources cannot yield the completeness, comparability and granularity of data needed for sound policy analysis and deliberation, much less for credible economic research and evaluation.

What makes this fiscal opacity even more frustrating is that mechanisms currently exist that could be employed to generate high-quality data on governmental public health spending: the U.S. Census Bureau’s annual Survey of Government Finances and the quinquennial Census of Governments. These mechanisms could be used to produce government-wide data on how much is spent on public health activities, who pays, and how funds are used. These mechanisms could be used to capture and analyze data on the incidence and level of spending in high-priority areas like tobacco prevention, obesity prevention, infectious disease control and food safety protection – information that could be combined with existing disease and risk surveillance data to help society make better decisions about where and how to deploy resources, and to help us learn faster what’s working and what’s not.

Of central importance, such data would fill in the many unknowns inherent in existing financial data captured from periodic surveys of state and local public health agencies, which are subject to the vagaries of how public health responsibilities are divvied up within state and local government bureaucracies, and how these bureaucracies keep their books. (Although their financial measures are imperfect, these periodic surveys are extremely valuable for conducting research on the implementation and impact of public health strategies, as my own studies and many others in the field of PHSSR illustrate).

Why aren’t public health expenditure data already regularly produced? Using the federal government’s existing governmental data collection mechanisms to produce better data on public health spending would probably require channeling some additional funding to the Census Bureau – a nontrivial but not insurmountable requirement. The larger challenge involves reaching consensus on what set of governmental activities should be defined and measured as public health responsibilities on a nationwide basis. This is a long-standing problem for the public health profession, which often relies on flexible and fuzzy conceptualizations of public health activity to build broad advocacy coalitions and to navigate ideological differences of opinion about the appropriate roles of government in the health policy arena. To counter this problem, a 2012 report by the Institute of Medicine recommended a national consensus process to identify a “minimum package” of public health services that should be available in all states and communities, and some work is now underway.

Many other social sectors benefit from regular flows of reliable financial data, including education, housing, criminal justice, and medical care. These data enhance research and evaluation, policy analysis and decision-making, managerial benchmarking and quality improvement, and governmental accountability and transparency. Maybe it is time to generate and use such data for public health.

How Did Massachusetts’ Health Reform Reduce Mortality?

An interesting new study of the 2006 Massachusetts health reform law adds to a growing body of evidence demonstrating that coverage expansions and related reforms help to improve overall health status. Specifically, this is the first study of the Massachusetts reform to suggest that its health effects are sufficiently large so as to be detectable at the population level, as reflected in the overall mortality rates for counties. These very encouraging results beg the question of how the results were achieved — which components of comprehensive reform are responsible for the drop in mortality?

The estimated 2.9% reduction in all-cause mortality represents a large and significant drop in deaths at the population level, particularly given the relatively short 5-year study period following health reform implementation. Chronic diseases are the main drivers of morbidity and mortality in the U.S., and most of these diseases chip away at our health over long periods of time, eventually causing death. Interventions that can make a dent in overall death rates in the space of just a few years are few and far between, and certainly worthy of attention (as I noted in a related commentary on this study).

The study authors focus on the expansion of health insurance coverage as the active ingredient in the Massachusetts law that is likely responsible for the drop in deaths. The results imply that for every 830 adults who gained insurance coverage, 1 death was prevented. However, I am not entirely convinced of this implication, and believe there could be more to the story of how Massachusetts lowered their mortality rates.

This study examines the population-level health effects of a health reform strategy that included not only insurance coverage expansions but also improvements in preventive services delivery and public health protections. Other elements of Massachusetts’ health reform strategy, such as reducing out-of-pocket costs for clinical preventive services and enhancing public health programs and infrastructure, may have contributed to the drop in mortality. For example, other studies have found large reductions in smoking prevalence and tobacco-associated health events among Massachusetts Medicaid recipients after the state added a comprehensive tobacco cessation benefit to its Medicaid program in 2006 as part of its health reform strategy. These preventive strategies and the resulting drop in tobacco exposure could plausibly explain some of the observed drop in mortality.

Importantly, the research design employed in this study cannot definitively attribute the mortality reductions to the gains in insurance coverage. The study uses county-level mortality data, so we cannot be sure that the people who gained coverage are the ones who experienced a reduction in their mortality risk (the classic ecological fallacy problem). The study hinges on comparing Massachusetts’ 14 counties to a statistically matched comparison group of counties outside the state, so with such a small number of intervention counties there is always the risk of systematic nonequivalence between the intervention and comparison counties—particularly when some of the covariates used for matching (like the baseline uninsured rate) are themselves subject to considerable sampling error.

Of course, other studies that have focused more narrowly on coverage expansions – such as the recent Oregon Medicaid experiment studies – have failed to find evidence of significant mortality reductions. Most of these studies use person-level rather than county-level analyses, so they are not able to detect population-level effects and they are not vulnerable to ecological fallacy. And the Oregon studies so far have been limited to shorter follow-up periods than the five years used in this latest Massachusetts study. Still, the mixed results from existing studies raise the possibility that insurance coverage by itself may not be sufficient to achieve large population-level health improvements. The mortality reductions found in Massachusetts may be attributable to the study’s population-level perspective, which captures the effects not only of insurance coverage expansion but also of other elements of reform. If so, combining coverage expansion with enhanced prevention and public health programming might just be the secret recipe for successful health reform.

These results provide encouraging news about the population-level health gains that ACA may produce. Before their 2006 reform, Massachusetts was already better off than most other U.S. states in terms of health insurance coverage and overall health status. Consequently, we might expect to see even larger health gains as similar reforms take hold in less-advantaged states. Of course, some of these less-advantaged states have chosen not to implement some of the ACA’s reforms to date – specifically the Medicaid coverage expansions. And the ACA’s prevention and public health programming has been scaled back considerably as a result of sequestration and decisions to use Prevention and Public Health Fund resources for other priorities. So, ACA’s population-level health effects remain uncertain and a very important target for ongoing study.

Publishing and Funding Research on Public Health Economics

For fellow travelers in public health economics, this post provides a quick update on resources for publishing and funding research on these topics. First, the publishing frontier: I am guest-editing an upcoming supplemental issue of the American Journal of Public Health devoted to the topic of advances in public health services and systems research (PHSSR). This issue is being developed with support provided by the Robert Wood Johnson Foundation and the U.S. Centers for Disease Control and Prevention. We are especially interested in studies examining health and economic consequences of organizational strategies, funding mechanisms, workforce models, policy/legal approaches, and health IT applications deployed in public health settings. The extended deadline for submitting manuscripts is May 15, 2014, and more information is available here.

Next, the funding frontier: I have recently added a page to this blog that highlights research funding opportunities relevant to public health economics. This page is far from a comprehensive inventory at this stage, but I will be building it over time as I have time.

Speaking of research funding, for those who may have missed it, an interesting item was included in the President’s FY2015 budget request submitted to Congress earlier this month. A $5 million request for public health systems research at the U.S. Centers for Disease Control and Prevention was included in the HHS budget request, as authorize by Section 4301 of the ACA. As the request explains: “CDC will undertake research that seeks to identify the economic and budgetary impacts of public health interventions; expand data on health care utilization and effectiveness; and inform how public health should evolve and public health and health care should collaborate as the health care delivery system transforms.” This blog will keep track of any research findings that flow from this federal initiative, should it be implemented.

Public Health Stimulus Spending and Healthcare Associated Infections

Can investments in public health infrastructure help solve big-ticket problems within the medical care system? A new study suggests an affirmative answer when it comes to healthcare associated infections. The controversial federal stimulus package known as the American Recovery and Reinvestment Act of 2009 included a small program aimed at the big problem of HAIs, the potentially lethal and increasingly drug-resistant pathogens that people acquire as a byproduct of receiving care in hospitals and other health care settings. Stimulus allocated approximately $36 million to a problem that generates an estimated $30 billion in hospital costs alone each year. Researchers from CDC and Emory analyzed the effects of this stimulus spending in a new paper appearing in the current issue of AJPH.

Notably, this federal HAI prevention program did not funnel resources directly to hospitals and other health care settings. The pool of funds was far too small for that. Rather, the outlays flowed to state public health agencies to allow them to build state-wide surveillance infrastructure for detecting and reporting HAIs, conduct trainings for health care personnel on infection control practices, form collaboratives among hospitals and other health care stakeholders to collectively plan and implement evidence-based prevention strategies, and monitor and evaluate progress.

The study finds that stimulus-funding significantly improved public health agency capacity to support HAI prevention activities, resulting in reductions in catheter-associated urinary tract infections (CAUTIs). In particular, states that received higher levels of program funding to implement CAUTI prevention collaboratives achieved larger reductions in CAUTI infection rates compared to states without these collaboratives. Infection rates for two other types of HAIs also declined during the study period, but these changes appeared to occur independently of stimulus funding levels and program activities.

That state public health agencies appear to have achieved significant reductions in HAI rates by mobilizing multi-organizational collaboratives is particularly notable in view of the fact that prior research has failed to find an effect for other high powered policy mechanisms such as mandatory HAI public reporting laws and hospital payment penalties tied to HAI events. This pattern of findings suggests that when it comes to HAI prevention, collective action by multiple community stakeholders may trump individual effort fueled by competition, public pressure or financial incentives. Moreover, the findings suggest that public health agencies can be effective in stimulating and supporting collective action through broad multi-organizational collaboratives (see my last post and IOM paper on this topic).

As is common in many PHSSR studies, the authors had to MacGyver a research strategy together from little more than kite string and chewing gum, and for this they deserve high praise. Pre-intervention measures of HAI prevention activity had to be scraped from the grant applications that state agencies submitted to compete for the funding. All states received some level of program funding, so lacking a control group the study had to exploit variation in the amount of funding received and the types of activities implemented in order to estimate possible program effects. State-specific data on HAI infection rates were not uniformly available prior to the initiation of the stimulus funding, making a strong pre-post design or difference-in-difference estimation infeasible. And then there is the problem of surveillance bias: program funding might have strengthened states abilities to detect and report HAIs, the very outcomes that the program seeks to reduce. These and other caveats indicate that the results must be interpreted with caution and should be confirmed with further study.

This study also fuels further thoughts about how best to design and finance public health programs, particularly those with uncertain effectiveness and constrained resources. This HAI program, like many federally-funded public health programs, awarded funding using competitive grant mechanisms that gave preference to settings that already had relevant resources, skills, and program capabilities in place. By attempting to maximize the chances of success, these funding mechanisms may blunt the impact of programs by failing to channel resources where they are needed most. Moreover, competitive funding mechanisms may exacerbate existing disparities in health services and health outcomes between low-resource and higher-resource settings. In this study, one can only wonder what program effects would have been observed if resources were targeted to states based on their observed HAI rates and other measures of need.

Using explicit and objective funding criteria and formulae for such programs, rather than competitive award processes, would also allow for more rigorous research on program effects. If program funding levels are determined based on objective measures, then strong quasi-experimental research designs can be used such as the regression discontinuity design to mitigate selection bias and confounding of the program exposure variable of interest. These types of designs are also advantageous when program implementation resources are constrained, allowing the targeting of resources to high-need settings based on observable and measurable criteria.

Despite its caveats, this study provides a revealing example of how modest investments in public health infrastructure and activities can have meaningful effects on a $2.8 trillion dollar health care system. More PHSSR research is underway to help elucidate and improve the interplay between public health and medical care, so stay tuned to this blog for further updates.