Brent Gibbons’s journal round-up for 9th April 2018

Every Monday our authors provide a round-up of some of the most recently published peer reviewed articles from the field. We don’t cover everything, or even what’s most important – just a few papers that have interested the author. Visit our Resources page for links to more journals or follow the HealthEconBot. If you’d like to write one of our weekly journal round-ups, get in touch.

The effect of Medicaid on management of depression: evidence from the Oregon Health Insurance Experiment. The Milbank Quarterly [PubMed] Published 5th March 2018

For the first journal article of this week’s AHE round-up, I selected a follow-up study on the Oregon health insurance experiment. The Oregon Health Insurance Experiment (OHIE) used a lottery system to expand Medicaid to low-income uninsured adults (and their associated households) who were previously ineligible for coverage. Those interested in being part of the study had to sign up. Individuals were then randomly selected through the lottery, after which individuals needed to take further action to complete enrollment in Medicaid, which included showing that enrollment criteria were satisfied (e.g. income below 100% of poverty line). These details are important because many who were selected for the lottery did not complete enrollment in Medicaid, though being selected through the lottery was associated with a 25 percentage point increase in the probability of having insurance (which the authors confirm was overwhelmingly due to Medicaid and not other insurance). More details on the study and data are publicly available. The OHIE is a seminal study in that it allows researchers to study the effects of having insurance in an experimental design – albeit in the U.S. health care system’s context. The other study that comes to mind is of course the famous RAND health insurance experiment that allowed researchers to study the effects of different levels of health insurance coverage. For the OHIE, the authors importantly point out that it is not necessarily obvious what the impact of having insurance is. While we would expect increases in health care utilization, it is possible that increases in primary care utilization could result in offsetting reductions in other settings (e.g. hospital or emergency department use). Also, while we would expect increases in health as a result of increases in health care use, it is possible that by reducing adverse financial consequences (e.g. of unhealthy behavior), health insurance could discourage investments in health. Medicaid has also been criticized by some as not very good insurance – though there are strong arguments to the contrary. First-year outcomes were detailed in another paper. These included increased health care utilization (across all settings), decreased out-of-pocket medical expenditures, decreased medical debt, improvements in self-reported physical and mental health, and decreased probability of screening positive for depression. In the follow-up paper on management of depression, the authors further explore the causal effect and causal pathway of having Medicaid on depression diagnosis, treatment, and symptoms. Outcomes of interest are the effect of having Medicaid on the prevalence of undiagnosed and untreated depression, the use of depression treatments including medication, and on self-reported depressive symptoms. Where possible, outcomes are examined for those with a prior depression diagnosis and those without. In order to examine the effect of Medicaid insurance (vs. being uninsured), the authors needed to control for the selection bias introduced from uncompleted enrollment into Medicaid. Instrumental variable 2SLS was used with lottery selection as the sole instrument. Local average treatment effects were reported with clustered standard errors on the household. The effect of Medicaid on the management of depression was overwhelmingly positive. For those with no prior depression diagnosis, it increased the chance of receiving a diagnosis and decreased the prevalence of undiagnosed depression (those who scored high on study survey depression instrument but with no official diagnosis). As far as treatment, Medicaid reduced the share of the population with untreated depression, virtually eliminating untreated depression among those with pre-lottery depression. There was a large reduction in unmet need for mental health treatment and an increased share who received specific mental health treatments (i.e. prescription drugs and talk therapy). For self-reported symptoms, Medicaid reduced the overall rate screened for depression symptoms in the post-lottery period. All effects were relatively strong in magnitude, giving an overall convincing picture that Medicaid increased access to treatment, which improved depression symptoms. The biggest limitation of this study is its generalizability. Much of the results were focused on the city of Portland, which may not represent more rural parts of the state. More importantly, this was limited to the state of Oregon for low-income adults who not only expressed interest in signing up, but who were able to follow through to complete enrollment. Other limitations were that the study only looked at the first two years of outcomes and that there was limited information on the types of treatments received.

Tobacco regulation and cost-benefit analysis: how should we value foregone consumer surplus? American Journal of Health Economics [PubMed] [RePEcPublished 23rd January 2018

This second article addresses a very interesting theoretical question in cost-benefit analysis, that has emerged in the context of tobacco regulation. The general question is how should foregone consumer surplus, in the form of reduced smoking, be valued? The history of this particular question in the context of recent FDA efforts to regulate smoking is quite fascinating. I highly recommend reading the article just for this background. In brief, the FDA issued proposed regulations to implement graphic warning labels on cigarettes in 2010 and more recently proposed that cigars and e-cigarettes should also be subject to FDA regulation. In both cases, an economic impact analysis was required and debates ensued on if, and how, foregone consumer surplus should be valued. Economists on both sides weighed-in, some arguing that the FDA should not consider foregone consumer surplus because smoking behavior is irrational, others arguing consumers are perfectly rational and informed and the full consumer surplus should be valued, and still others arguing that some consumer surplus should be counted but there is likely bounded rationality and that it is methodologically unclear how to perform a valuation in such a case. The authors helpfully break down the debate into the following questions: 1) if we assume consumers are fully informed and rational, what is the right approach? 2) are consumers fully informed and rational? and 3) if consumers are not fully informed and rational, what is the right approach? The reason the first question is important is that the FDA was conducting the economic impact analysis by examining health gains and foregone consumer surplus separately. However, if consumers are perfectly rational and informed, their preferences already account for health impacts, meaning that only changes in consumer surplus should be counted. On the second question, the authors explore the literature on smoking behavior to understand “whether consumers are rational in the sense of reflecting stable preferences that fully take into account the available information on current and expected future consequences of current choices.” In general, the literature shows that consumers are pretty well aware of the risks, though they may underestimate the difficulty of quitting. On whether consumers are rational is a much harder question. The authors explore different rational addiction models, including quasi-rational addiction models that take into account more recent developments in behavioral economics, but declare that the literature at this point provides no clear answer and that no empirical test exists to distinguish between rational and quasi-rational models. Without answering whether consumers are fully informed and rational, the authors suggest that welfare analysis – even in the face of bounded rationality – can still use a similar valuation approach to consumer surplus as was recommended for when consumers are fully informed and rational. A series of simple supply and demand curves are presented where there is a biased demand curve (demand under bounded rationality) and an unbiased demand curve (demand where fully informed and rational) and different regulations are illustrated. The implication is that rather than trying to estimate health gains as a result of regulations, what is needed is to understand the amount of demand bias as result of bounded rationality. Foregone consumer surplus can then be appropriately measured. Of course, more research is needed to estimate if, and how much, ‘demand bias’ or bounded rationality exists. The framework of the paper is extremely useful and it pushes health economists to consider advances that have been made in environmental economics to account for bounded rationality in cost-benefit analysis.

2SLS versus 2SRI: appropriate methods for rare outcomes and/or rare exposures. Health Economics [PubMed] Published 26th March 2018

This third paper I will touch on only briefly, but I wanted to include it as it addresses an important methodological topic. The paper explores several alternative instrumental variable estimation techniques for situations when the treatment (exposure) variable is binary, compared to the common 2SLS (two-stage least squares) estimation technique which was developed for a linear setting with continuous endogenous treatments and outcome measures. A more flexible approach, referred to as 2SRI (two-stage residual inclusion) allows for non-linear estimation methods in the first stage (and second stage), including logit or probit estimation methods. As the title suggests, these alternative estimation methods may be particularly useful when treatment (exposure) and/or outcomes are rare (e.g below 5%). Monte Carlo simulations are performed on what the authors term ‘the simplest case’ where the outcome, treatment, and instrument are binary variables and a range of results are considered as the treatment and/or outcome become rarer. Model bias and consistency are assessed in the ability to produce average treatment effects (ATEs) and local average treatment effects (LATEs), comparing the 2SLS, several forms of probit-probit 2SRI models, and a bivariate probit model. Results are that the 2SLS produced biased estimates of the ATE, especially as treatment and outcomes become rarer. The 2SRI models had substantially higher bias than the bivariate probit in producing ATEs (though the bivariate probit requires the assumption of bivariate normality). For LATE, 2SLS always produces consistent estimates, even if the linear probability model produces out of range predictions. Estimates for 2SRI models and the bivariate probit model were biased in producing LATEs. An empirical example was also tested with data on the impact of long-term care insurance on long-term care use. Conclusions are that 2SRI models do not dependably produce unbiased estimates of ATEs. Among the 2SRI models though, there were varying levels of bias and the 2SRI model with generalized residuals appeared to produce the least ATE bias. For more rare treatments and outcomes, the 2SRI model with Anscombe residuals generated the least ATE bias. Results were similar to another simulation study by Chapman and Brooks. The study enhances our understanding of how different instrumental variable estimation methods may function under conditions where treatment and outcome variables have nonlinear distributions and where those same treatments and outcomes are rare. In general, the authors give a cautionary note to say that there is not one perfect estimation method in these types of conditions and that researchers should be aware of the potential pitfalls of different estimation methods.

Credits

 

Brent Gibbons’s journal round-up for 22nd January 2018

Every Monday our authors provide a round-up of some of the most recently published peer reviewed articles from the field. We don’t cover everything, or even what’s most important – just a few papers that have interested the author. Visit our Resources page for links to more journals or follow the HealthEconBot. If you’d like to write one of our weekly journal round-ups, get in touch.

Is retirement good for men’s health? Evidence using a change in the retirement age in Israel. Journal of Health Economics [PubMed] Published January 2018

This article is a tour de force from one chapter of a recently completed dissertation from the Hebrew University of Jerusalem. The article focuses on answering the question of what are the health implications of extending working years for older adults. As many countries are faced with critical decisions on how to adjust labor policies to solve rising pension costs (or in the case of the U.S., Social Security insolvency) in the face of aging populations, one obvious potential solution is to change the retirement age. Most OECD countries appear to have retirement ages in the mid-60’s with a number of countries on track to increase that threshold. Israel is one of these countries, having changed their retirement age for men from age 65 to age 67 in 2004. The author capitalizes on this exogenous change in retirement incentives, as workers will be incentivized to keep working to receive full pension benefits, to measure the causal effect of working in these later years, compared to retiring. As the relationship between employment and health is complicated by the endogenous nature of the decision to work, there is a growing literature that has attempted to deal with this endogeneity in different ways. Shai details the conflicting findings in this literature and describes various shortcomings of methods used. He helpfully categorizes studies into those that compare health between retirees and non-retirees (does not deal with selection problem), those that use variation in retirement age across countries (retirement ages could be correlated with individual health across countries), those that exploit variation in specific sector retirement ages (problem of generalizing to population), and those that use age-specific retirement eligibility (health may deteriorate at specific age regardless of eligibility for retirement). As this empirical question has amounted conflicting evidence, the author suggests that his methodology is an improvement on prior papers. He uses a difference-in-difference model that estimates the impact on various health outcomes, before and after the law change, comparing those aged 65-66 years after 2004 with both older and younger cohorts unaffected by the law. The assumption is that any differences in measured health between the age 65-66 group and the comparison group are a result of the extended work in later years. There are several different datasets used in the study and quite a number of analyses that attempt to assuage threats to a causal interpretation of results. Overall, results are that delaying the retirement age has a negative effect on individual health. The size of the effect found is in the ballpark of 1 standard deviation; outcome measures included a severe morbidity index, a poor health index, and the number of physician visits. In addition, these impacts were stronger for individuals with lower levels of education, which the author relates to more physically demanding jobs. Counterfactuals, for example number of dentist visits, which are not expected to be related to employment, are not found to be statistically different. Furthermore, there are non-trivial estimated effects on health care expenditures that are positive for the delayed retirement group. The author suggests that all of these findings are important pieces of evidence in retirement age policy decisions. The implication is that health, at least for men, and especially for those with lower education, may be negatively impacted by delaying retirement and that, furthermore, savings as a result of such policies may be tempered by increased health care expenditures.

Evaluating community-based health improvement programs. Health Affairs [PubMed] Published January 2018

For article 2, I see that the lead author is a doctoral student in health policy at Harvard, working with colleagues at Vanderbilt. Without intention, this round-up is highlighting two very impressive studies from extremely promising young investigators. This study takes on the challenge of evaluating community-based health improvement programs, which I will call CBHIPs. CBHIPs take a population-based approach to public health for their communities and often focus on issues of prevention and health promotion. Investment in CBHIPs has increased in recent years, emphasizing collaboration between the community and public and private sectors. At the heart of CBHIPs are the ideas of empowering communities to self-assess and make needed changes from within (in collaboration with outside partners) and that CBHIPs allow for more flexibility in creating programs that target a community’s unique needs. Evaluations of CBHIPs, however, suffer from limited resources and investment, and often use “easily-collectable data and pre-post designs without comparison or control communities.” Current overall evidence on the effectiveness of CBHIPs remains limited as a result. In this study, the authors attempt to evaluate a large set of CBHIPs across the United States using inverse propensity score weighting and a difference-in-difference analysis. Health outcomes on poor or fair health, smoking status, and obesity status were used at the county level from the BRFSS (Behavioral Risk Factor Surveillance System) SMART (Selected Metropolitan/Micropolitan Area Risk Trends) data. Information on counties implementing CBHIPs was compiled through a series of systematic web searches and through interviews with leaders in population health efforts in the public and private sector. With information on the exact years of implementation of CBHIPs in each county, a pre-post design was used that identified county treatment and control groups. With additional census data, untreated counties were weighted to achieve better balance on pre-implementation covariates. Importantly, treated counties were limited to those with CBHIPs that implemented programs related to smoking and obesity. Results showed little to no evidence that CBHIPs improved population health outcomes. For example, CBHIPs focusing on tobacco prevention were associated with a 0.2 percentage point reduction in the rate of smoking, which was not statistically significant. Several important limitations of the study were noted by the authors, such as limited information on the intensity of programs and resources available. It is recognized that it is difficult to improve population-level health outcomes and that perhaps the study period of 5-years post-implementation may not have been long enough. The researchers encourage future CBHIPs to utilize more rigorous evaluation methods, while acknowledging the uphill battle CBHIPs face to do this.

Through the looking glass: estimating effects of medical homes for people with severe mental illness. Health Services Research [PubMed] Published October 2017

The third article in this round-up comes from a publication from October of last year, however, it is from the latest issue of Health Services Research so I deem it fair play. The article uses the topic of medical homes for individuals with severe mental illness to critically examine the topic of heterogeneous treatment effects. While specifically looking to answer whether there are heterogeneous treatment effects of medical homes on different portions of the population with a severe mental illness, the authors make a strong case for the need to examine heterogeneous treatment effects as a more general practice in observational studies research, as well as to be more precise in interpretations of results and statements of generalizability when presenting estimated effects. Adults with a severe mental illness were identified as good candidates for medical homes because of complex health care needs (including high physical health care needs) and because barriers to care have been found to exist for these individuals. Medicaid medical homes establish primary care physicians and their teams as the managers of the individual’s overall health care treatment. The authors are particularly concerned with the reasons individuals choose to participate in medical homes, whether because of expected improvements in quality of care, regional availability of medical homes, or symptomatology. Very clever differences in estimation methods allow the authors to estimate treatment effects associated with these different enrollment reasons. As an example, an instrumental variables analysis, using measures of regional availability as instruments, estimated local average treatment effects that were much smaller than the fixed effects estimates or the generalized estimating equation model’s effects. This implies that differences in county-level medical home availability are a smaller portion of the overall measured effects from other models. Overall results were that medical homes were positively associated with access to primary care, access to specialty mental health care, medication adherence, and measures of routine health care (e.g. screenings); there was also a slightly negative association with emergency room use. Since unmeasured stable attributes (e.g. patient preferences) do not seem to affect outcomes, results should be generalizable to the larger patient population. Finally, medical homes do not appear to be a good strategy for cost-savings but do promise to increase access to appropriate levels of health care treatment.

Credits

Brent Gibbons’s journal round-up for 30th January 2017

Every Monday our authors provide a round-up of some of the most recently published peer reviewed articles from the field. We don’t cover everything, or even what’s most important – just a few papers that have interested the author. Visit our Resources page for links to more journals or follow the HealthEconBot. If you’d like to write one of our weekly journal round-ups, get in touch.

For this week’s round-up, I selected three papers from December’s issue of Health Services Research. I didn’t intend to to limit my selections to one issue of one journal but as I narrowed down my selections from several journals, these three papers stood out.

Treatment effect estimation using nonlinear two-stage instrumental variable estimators: another cautionary note. Health Services Research [PubMed] Published December 2016

This paper by Chapman and Brooks evaluates the properties of a non-linear instrumental variables (IV) estimator called two-stage residual inclusion or 2SRI. 2SRI has been more recently suggested as a consistent estimator of treatment effects under conditions of selection bias and where the dependent variable of the 2nd-stage equation is either binary or otherwise non-linear in its distribution. Terza, Bradford, and Dismuke (2007) and Terza, Basu, and Rathouz (2008) furthermore claimed that 2SRI estimates can produce unbiased estimates not just of local average treatment effects (LATE) but of average treatment effects (ATE). However, Chapman and Brooks question why 2SRI, which is analogous to two-stage least squares (2SLS) when both the first and second stage equations are linear, should not require similar assumptions as in 2SLS when generalizing beyond LATE to ATE. Backing up a step, when estimating treatment effects using observational data, one worry when trying to establish a causal effect is bias due to treatment choice. Where patient characteristics related to treatment choice are unobservable and one or more instruments is available, linear IV estimation (i.e. 2SLS) produces unbiased and consistent estimates of treatment effects for “marginal patients” or compliers. These are the patients whose treatment effects were influenced by the instrument and their treatment effects are termed LATE. But if there is heterogeneity in treatment effects, a case needs to be made that treatment effect heterogeneity is not related to treatment choice in order to generalize to ATE.  Moving to non-linear IV estimation, Chapman and Brooks are skeptical that this case for generalizing LATE to ATE no longer needs to be made with 2SRI. 2SRI, for those not familiar, uses the residual from stage 1 of a two-stage estimator as a variable in the 2nd-stage equation that uses a non-linear estimator for a binary outcome (e.g. probit) or another non-linear estimator (e.g. poisson). The authors produce a simulation that tests the 2SRI properties over varying conditions of uniqueness of the marginal patient population and the strength of the instrument. The uniqueness of the marginal population is defined as the extent of the difference in treatment effects for the marginal population as compared to the general population. For each scenario tested, the bias between the estimated LATE and the true LATE and ATE is calculated. The findings support the authors’ suspicions that 2SRI is subject to biased results when uniqueness is high. In fact, the 2SRI results were only practically unbiased when uniqueness was low, but were biased for both ATE and LATE when uniqueness was high. Having very strong instruments did help reduce bias. In contrast, 2SLS was always practically unbiased for LATE for different scenarios and the authors use these results to caution researchers on using “new” estimation methods without thoroughly understanding their properties. In this case, old 2SLS still outperformed 2SRI even when dependent variables were non-linear in nature.

Testing the replicability of a successful care management program: results from a randomized trial and likely explanations for why impacts did not replicate. Health Services Research [PubMed] Published December 2016

As is widely known, how to rein in U.S. healthcare costs has been a source of much hand-wringing. One promising strategy has been to promote better management of care in particular for persons with chronic illnesses. This includes coordinating care between multiple providers, encouraging patient adherence to care recommendations, and promoting preventative care. The hope was that by managing care for patients with more complex needs, higher cost services such as emergency visits and hospitalizations could be avoided. CMS, the Centers for Medicare and Medicaid Services, funded a demonstration of a number of care management programs to study what models might be successful in improving quality and reducing costs. One program implemented by Health Quality Partners (HQP) for Medicare Fee-For-Service patients was successful in reducing hospitalizations (by 34 percent) and expenditures (by 22 percent) for a select group of patients who were identified as high-risk. The demonstration occurred from 2002 – 2010 and this paper reports results for a second phase of the demonstration where HQP was given additional funding to continue treating only high-risk patients in the years 2010 – 2014. High-risk patients were identified as having a diagnosis of congestive heart failure (CHF), chronic obstructive pulmonary disease (COPD), coronary artery disease (CAD), or diabetes and had a hospitalization in the year prior to enrollment. In essence, phase II of the demonstration for HQP served as a replication of the original demonstration. The HQP care management program was delivered by nurse coordinators who regularly talked with patients and provided coordinated care between primary care physicians and specialists, as well as other services such as medication guidance. All positive results from phase I vanished in phase II and the authors test several hypotheses for why results did not replicate. They find that treatment group patients had similar hospitalization rates between phase I and II, but that control group patients had substantially lower phase II hospitalization rates. Outcome differences between phase I and phase II were risk-adjusted as phase II had an older population with higher severity of illness. The authors also used propensity score re-weighting to further control for differences in phase I and phase II populations. The affordable care act did promote similar care management services through patient-centered medical homes and accountable care organizations that likely contributed to the usual care of control group patients improving. The authors also note that the effectiveness of care management may be sensitive to the complexity of the target population needs. For example, the phase II population was more homebound and was therefore unable to participate in group classes. The big lesson in this paper though is that demonstration results may not replicate for different populations or even different time periods.

A machine learning framework for plan payment risk adjustment. Health Services Research [PubMed] Published December 2016

Since my company has been subsumed under IBM Watson Health, I have been trying to wrap my head around this big data revolution and the potential of technological advances such as artificial intelligence or machine learning. While machine learning has infiltrated other disciplines, it is really just starting to influence health economics, so watch out! This paper by Sherri Rose is a nice introduction into a range of machine learning techniques that she applies to the formulation of plan payment risk adjustments. In insurance systems where patients can choose from a range of insurance plans, there is the problem of adverse selection where some plans may attract an abundance of high risk patients. To control for this, plans (e.g. in the affordable care act marketplaces) with high percentages of high risk consumers get compensated based on a formula that predicts spending based on population characteristics, including diagnoses. Rose says that these formulas are still based on a 1970s framework of linear regression and may benefit from machine learning algorithms. Given that plan payment risk adjustments are essentially predictions, this does seem like a good application. In addition to testing goodness of fit of machine learning algorithms, Rose is interested in whether such techniques can reduce the number of variable inputs. Without going into any detail, insurers have found ways to “game” the system and fewer variable inputs would restrict this activity. Rose introduces a number of concepts in the paper (at least they were new to me) such as ensemble machine learningdiscrete learning frameworks and super learning frameworks. She uses a large private insurance claims dataset and breaks the dataset into what she calls 10 “folds” which allows her to run 5 prediction models, each with its own cross-validation dataset. Aside from one parametric regression model, she uses several penalized regression models, neural net, single-tree, and random forest models. She describes machine learning as aiming to smooth over data in a similar manner to parametric regression but with fewer assumptions and allowing for more flexibility. To reduce the number of variables in models, she applies techniques that limit variables to, for example, just the 10 most influential. She concludes that applying machine learning to plan payment risk adjustment models can increase efficiencies and her results suggest that it is possible to get similar results even with a limited number of variables. It is curious that the parametric model performed as well as or better than many of the different machine learning algorithms. I’ll take that to mean we can continue using our trusted regression methods for at least a few more years.

Credits