Unbiased Analysis of Today's Healthcare Issues

What is evidence-based health policy?

Written By: Jason Shafrin - Jan• 16•18

Many people have heard of evidence-based medicine (EBM).  A perspective piece in the New England Journal of Medicine by Katherine Baicker and Amitabh Chandra, however, tries to define what evidence-based health policy (EBHP) is.  They list three key criteria:

  • Policies need to be well-specified.  For instance, “expand Medicaid coverage” is too general, whereas specifying that benefit package A be extended to population B at cost C, is getting to more of the specifics needed.
  • EBHP represent policies, not goals.  The same exact value-based purchasing program, for instance, could have multiple goals, (e.g., reducing cost, improving quality).  The policy itself is what is being evaluated, whereas the goals depend on the evaluator’s perspective.
  •  EBHP requires evidence of the magnitude of the effects of the policy.  Empirical evidence is needed to show a specific health policy works.  Further, just showing a statistically significant result is not enough, it also must be of sufficient magnitude to be relevant to stakeholders, especially when compared against other policy interventions.

In addition, the authors argue that the perfect should not be the enemy of the good. That is my interpretation of the authors saying that “In health policy — as in any other realm — it is often necessary to act on the basis of the best evidence on hand, even when that evidence is not strong. Doing so requires weighing the costs of acting when you shouldn’t against those of not acting when you should — again, a matter of policy priorities.”

Conducting health policy research is complex not only due to the empirical complications, but also because policy goals and priorities vary across stakeholders and empirical evidence of the effect of health policies may or may not resolve this debate.  Nevertheless, by using evidence-based approaches to identify the effect of health policies, society can make these policies decisions knowing full well the implications of their decisions.

Longitudinal Modelling of Healthcare Expenditures: Challenges and Solutions

Written By: Jason Shafrin - Jan• 15•18

Previous analyses–such as Basu and Manning 2009–have addressed the problem of mass of health care expenditures around $0. In typical economic analyses, we assume that the dependent variable is normally distributed. In the case of health care expenditures, however, a large number of people have $0 expenditures (i.e., healthy individuals). Further, among sick individuals that incur positive health care expenditures, the expenditure distribution is typically heavily right-skewed. Approaches such as a two-part model have been used to address this type of health care spending distribution for cross-sectional data.

When addressing these distributional challenges for cross-sectional data, however, there are additional challenges. As Smith et al. 2018 write

There are additional considerations with longitudinal expenditures, and less research compares strategies for modeling longitudinal expenditures. As with any longitudinal outcome,
the estimation must incorporate the correlation of repeated measurements (Basu and Manning 2009). Furthermore, the distribution of longitudinal expenditures and the proportion of zeros are dependent upon the timeframe under consideration (e.g., person-month vs. person-year).

Using 2000-2003 VA data for veterans with hypertension, they authors test the model fit of four models:

  • One-part models. “A one-part generalized linear model (GLM) fit the data using generalized estimating equations (GEEs) and treats the observed expenditures as realizations of a single process, so the model does not distinguish between zero and positive-valued expenditures.  The estimating equation is g(E(Yij)=βX.  Often the function g() is the log function.  “Similarly to GLMs fit with quasi-likelihood for cross-sectional expenditures, GLMs fit via GEEs do not require specification of a parametric distribution…Rather, one specifies only the mean and variance. Often, the variance is given as a mean–variance relationship (e.g., variance proportional to the mean, Var[E(Yij)]=ρ E(Yij), where ρ represents a proportionality constant), and a link function, as described above, provides the form of the mean model.” One can implement these regressions using standard software (e.g., SAS’s PROC GENMOD using a REPEATED statement and PROC GEE ; for Stata use xtgee and with R’s gee or geepack work
  • Uncorrelated two-part models.  These models are useful when there are a large number of 0’s or if researchers are interested in the process through which patients occur $0 of expenditures. Two-part models have a binary component–often logit–measuring the likelihood of having $0 expenditures, and continuous part modeling the expenditure distribution conditional on positive costs. One complication, however, is that the first part of the model measures the P(Yij=0) for all individuals whereas the continuous part measures spending only among individuals with positive expenditures.  One other challenges is that these two model components “…are often correlated over time, such that the probability of incurring any expense is associated with level of expenditures over time. Failure to account for this correlation leads to informative cluster sizes in the second component and biased results.
  • Correlated conditional two-part (CTP) models. “The correlated conditional two-part (CTP) random-effects model allows for estimation of correlation between the binary and continuous parts of the longitudinal expenditures by specifying a joint random-effects distribution, including correlations/covariance between the random effects. One can model these as shown below. The two-part models are the same as below, but omit the random intercepts, b1i and b2i, which are assumed to be distributed multivariate normal. Although this model can be implemented with some standard statistical packages (e.g., PROC NLMIXED in SAS) it is much more complex than the two options above.
    • logit(Pr(Yij>0)=αX+b1i
    • E(log(Yij|Yij>0)=δX+b2i
  • Correlated marginalized two-part (MTP) models.  These models are similar to the CTP, but the continuous, second part of the model is not continuous as shown in the example below. Like the CTP, this is computational difficult to implement, although it has been done with Bayesian approaches. “Similar to the correlated CTP model, parameter estimates in the binary component are subject-specific estimates. Specifically, exp(αk) represents the subject-specific odds ratio for incurring positive expenditures associated with a one-unit increase in the kth covariate. Parameter estimates in the second component represent effects on the overall mean, and those corresponding to covariates not included as random effects are both subject-specific and population average.”
    • logit(Pr(Yij>0)=αX+b1i
    • E(log(Yij)=δX+b2i

You can read the paper to see the results of the authors application to the VA data.  However, the authors do provide some useful guidance in the discussion section regarding model selection.

First, is there interest in what influences the probability of incurring expenditures? If so, a two-part model may be appropriate. Secondly, is the primary interest in overall mean expenditures of the entire population or is more interest in the level of expenditures conditional on them being incurred? If the former, the one-part GLM or the MTP model is preferable; if the latter, one should consider a CTP model. The uncorrelated two-part GLM should only be considered if the analyst feels confident of no correlation between the two components, often an untenable assumption.


The gold standard of scientific evidence

Written By: Jason Shafrin - Jan• 15•18

That is the title of my latest article in Pharmaceutical Market Europe. An excerpt is below.

Randomised controlled trials (RCTs) are regarded as the gold standard of scientific evidence,
and for good reason. By randomizsing a treatment across study arms, RCTs eliminate patient-treamtent selection bias, resulting in reliable causal inference. In contrast, in the real world patients woh are sicker may be more (or less) likely ot receive certain treatments. Because treatments are given selectively, the true causal effect of a treatment on patient health in most real-world studies cannot be determined…Nevertheless, in recent years, there has been a call to increase the use of real-world evidence (RWE)…

The rest of the article discusses some of the limitations of RCTs and places where RWE can be used to help improve health care decision-making. Do read the whole article.

Weekend Links

Written By: Jason Shafrin - Jan• 14•18

Friday Links

Written By: Jason Shafrin - Jan• 04•18
  • Science facts from 2017.
  • Peak pharma?
  • 2017 year in charts.
  • Nature’s 10.
  • Physicians ignore an important source of data…the patient.  “If you don’t sit and talk with a patient for a half hour, in terms of your job description no one is going to be mad at you. But if you don’t know what the hemoglobin is on the patient, the chief of medicine is going to be very upset with you.”

The Healthcare Economist will be on vacation next week.

The Mediterranean Diet is supposed to be very healthy. Can you write me a prescription for a vacation in Greece?

How does cost sharing rules influence drug prices in Germany?

Written By: Jason Shafrin - Jan• 02•18

Typically, we look at how changes in cost sharing affect patient demand.  However, rules regarding patient cost sharing also influence life sciences firms’ decisions about what price they should use for their products.  A paper by Herr and Suppliet (2017) looks at the effect of changing cost-sharing rules in Germany in their latest paper.

The study first reviews how drug prices are set in Germany.

…drug prices are uniform across all pharmacies and the same co-payment scheme applies to all those who are publicly insured. Pharmacists receive reimbursements directly from the health insurance companies, namely a fixed fee per package plus a fraction of the drug’s price (3 percent)…In general, drug co-payments in Germany are defined as 10 per-cent of the pharmacy’s selling price (or the reference price if the price lies above) with a minimum of €5 and a maximum of €10, plus the difference between that and the reference price, if applicable.

The authors discuss how the use of copayment exemption levels (CEL) affect demand for pharmaceuticals.  These CELs are a set price, below which patients do not have to pay a copayment for these pharmaceuticals.  The CEL depends on the reference price which is set at the 30th percentile of the prior year’s prices for drugs within a specific therapeutic basket.  Most generics (99%) and brands (77%) are set below this reference price.

The authors attempt to measure the causal effect of prices on demand using a difference-in-differences approach that using the sequential introduction of the CEL across different therapeutic groups as the source of variability. The key assumption for the difference-in-difference approach is that the timing of the introduction of CEL is exogenous, in other words, the CEL was not applied to therapeutic classes where drug prices were rising or falling more or less quickly in a systematic way.  The authors also use a 2SLS with instruments on the supply side of packaging costs (i.e., the cost of paper, plastic, and ink) and the introduction of CEL as a demand side instrument.

In their empirical case study, the authors use a nested logit to model demand for anti-epileptics drugs. The authors find that:

…the introduction of tiered co-payments leads to decreasing drug prices and to market segmentation in the German reference price market…The policy has a negative effect of 5 percent on generic prices while brand-name drugs’ prices increase by 4 percent due to the new regulation.  This pattern is similar to the price increases of brand-name drugs after generic entry,the “generic competition paradox”.


How did the Affordable Care Act affect the U.S. labor supply?

Written By: Jason Shafrin - Jan• 01•18

The Affordable Care Act (ACA) aimed to increase health insurance coverage largely through two pathways: (i) raising the income limits for individuals to qualify for Medicaid, (ii) creating new health insurance exchanges and health insurance subsidies to encourage the purchase of private health insurance among individuals that were not eligible for Medicaid.  Other provisions, such as the health insurance mandate, clearly also had an impact on coverage.  One study found that 19.2 million non-elderly individuals gained health insurance coverage as a result of ACA between 2010 to 2015.

A separate question is whether the increased insurance generosity increased or decreased individual labor supply.  On the one hand, the ACA may reduce the labor supply.  Reducing one’s income increases the likelihood of qualifying for Medicaid.  Also, within the exchanges, the subsidies are decline with income and thus additional income is in essence taxed at a higher rate (through subsidy reduction), thus discouraging work.  Additionally, the Medicaid expansion and subsidies are transfers and in essence increase individual wealth.  Since leisure is desirable, people may decide to work less if it is easier to purchase insurance after the ACA.  On the other hand, people may place more value on private health insurance relative to Medicaid.  If the private health insurance becomes more affordable with the subsidies, people may want to work more to be able to afford to purchase private insurance.

An NBER working paper by Mark Duggan, Gopi Shah Goda, Emilie Jackson (2017) aim to determine how the ACA affected labor supply. They use variation in whether states expanded Medicaid to conduct a difference-in-difference analysis using data from the American Community Survey (ACS). They find that:

…the average labor supply effects of the ACA were close to zero but that this average masks important heterogeneity in its effects. More specifically, we find that in areas with a high share uninsured and eligible for private insurance subsidies, labor force participation fell significantly. In contrast, in areas with a high share uninsured but with incomes too low to qualify for private insurance subsidies, labor force participation increased significantly. These changes suggest that middle-income individuals reduced their labor supply due to the additional tax on earnings while lower income individuals worked more in order to qualify for private insurance. In the aggregate, these countervailing effects approximately balance.

An important question since although expanding health insurance likely increases utility for those who now gain coverage, when measuring the societal benefit we must not only take into account the deadweight loss from the additional taxation needed to fund the system, but also how labor supply response to the changing health insurance market.


Friday Links

Written By: Jason Shafrin - Dec• 28•17

Top Health Economics Stories of 2017

Written By: Jason Shafrin - Dec• 26•17

What were the top stories at the intersection of health and economics stories in 2017?  Here is the Healthcare Economist’s take.

Obamacare repeal. One of the top stories clearly must be the on-going debate around the repeal of the Affordable Care Act (ACA, a.k.a. Obamacare).  Although the ACA was not fully repealed, the most recent Republican tax legislation will repeal the individual mandate. Donald Trump today predicted that the repeal of the individual mandate will eventually cripple Obamacare leading to the need to replace it.  This prediction is not crazy as by repealing the individual mandate may cause healthy individuals to drop out of the health insurance market, leading to rising premiums, and the potential for an adverse selection death spiral.

Digital medicine is here.  The FDA this year approved the first digital medicine, an antipsychotic with an embedded sensor linked to patient’s smartphone’s and the cloud to better track patient adherence.  My own research has shown that giving physicians real-time, accurate patient drug adherence information can improve treatment decisions for providers treating patients with serious mental illness.  In fact, these types of digital medicines could lead to over $2000 in cost savings if physicians used this improved adherence information.

CAR-T launches with value-based pricing.  Two life sciences firms have gotten FDA approval for new CAR-T (chimeric antigen receptor T-cell) therapy.  The treatments have shown dramatic improvements in patient outcomes, but and are administered one time–rather than in multiple doses across months or years as would be the case for many chemotherapies.  Although the treatments cost $475,000, some payers–such as Medicare–will only pay for the treatments if they demonstrate improved patient outcomes. Not only is this medical breakthrough newsworthy, but so is this innovative pricing scheme.

The opioid epidemic is here.  The CDC reported that opioid deaths are rising among America’s teens.  Donald Trump declared that a state of emergency due to this crisis.  One film I saw, Heroin(e), documents not only the opioid epidemic but the first responders, police officers, judges, and others who are helping to fight this epidemic.   Unfortunately, this health story is likely to make the list again in 2018 unless there are serious efforts to combat the epidemic.

Antibiotic resistant bacteria represent a major public health challenge.  One report identified a numerous outbreaks of antibiotic resistant bacteria, known as CREs.  Seth Seabury and Neeraj Sood argue for a new model to incentivize life sciences firms and researchers to develop new antibiotics to fight this new and deadly threat.  Antibiotic resistance is a top priority at the World Health Organization as well.

Follow the money. An interesting paper by Sood, Shih, Van Nuys and Goldman tracks the flow of funds through the pharmaceutical system.  They find that “…for every $100 spent at retail pharmacies, about $17 compensates for direct production costs, $41 accrues to the manufacturer ($15 of which is net profit), and $41 accrues to intermediaries in the distribution system: wholesalers, pharmacies, pharmacy benefit managers and insurers (with $8 of net profit split among them).”

The Innovation and Value Initiative launches its Open-Source Value Project.  With value frameworks gaining more an more traction, IVI has launched an open-source tool to help a variety of stakeholders evaluate treatment value.  The tool is completely transparent, is based on the latest scientific findings, and allows users to measure treatment value based on their own preferences or preferences over the population which they are managing.  The first Open-Source Value Project covers DMARDs treating patients with moderate to severe rheumatoid arthritis.


What happens to CHIP?

Written By: Jason Shafrin - Dec• 26•17

The Children’s Health Insurance Program (CHIP) is a federal program that provides matching funds to states in order for them to provide health insurance to children.  The program was designed to cover uninsured children in families with incomes that are modest but too high to qualify for Medicaid.

Currently, however, the program is in jeopardy.  In fact, federal funding for CHIP expired September 30.  Some states, such as Colorado, have supplied emergency funding to keep CHIP afloat in their states.  On December 21, the federal government did pass $2.85 billion of emergency funding.  As funding was backdated to October 1, the 6 months of funding will only cover expenses through the end of March, 2018.

There are arguments on both sides of the aisle about the merits of CHIP.  Provide health insurance to kids has clear value to these children and their families.  Also, CHIP makes the U.S. society more equitable.  Those opposed to CHIP may prefer other ways to help kids get insurance, such as vouchers to purchase private insurance, or may even prefer to end CHIP and spend the tax money on other federal programs (e.g., education, defense), or tax cuts or may fear that CHIP may decrease the labor supply of parents who otherwise would work to get coverage for their kids.

Regardless of where you stand on the political spectrum, this uncertainty is clearly bad for all involved.  States have a hard time planning their budgets if they are unsure whether federal funds will materialized.  The federal government’s need to dip into emergency funding leads to a lack of fiscal discipline.  These expenses should be planned for.  Most importantly, parents and their children need to be able to plan for the future.  If CHIP is available, clearly many parents will want this coverage for their children.  If the program is not available, however, then perhaps some would return to work at suboptimal jobs rather than continue a job hunt in order to get health insurance for their family.  Large purchases (e.g., house, car, durable goods) clearly depend on the amount disposable income families have and this amount would decreases significantly if they had to buy insurance rather than be covered through CHIP.

Clear planning, transparency, and predictability are the hallmarks of good government.  This debate over CHIP funding–similar to the debates over the “doc fix” over physician reimbursement–shows that these emergency funding approaches are not only bad for the budget, but they hurt the people they are supposed to help by increasing the amount of uncertainty in their lives.