The Incidental Economist

I’ve mentioned this before, but it’s worth a whole post. There is actually a full Cochrane review of pay-for-performance for primary care:

Background: The use of blended payment schemes in primary care, including the use of financial incentives to directly reward ‘performance’ and ‘quality’ is increasing in a number of countries. There are many examples in the US, and the Quality and Outcomes Framework (QoF) for general practitioners (GPs) in the UK is an example of a major system-wide reform. Despite the popularity of these schemes, there is currently little rigorous evidence of their success in improving the quality of primary health care, or of whether such an approach is cost-effective relative to other ways to improve the quality of care.

Objectives: The aim of this review is to examine the effect of changes in the method and level of payment on the quality of care provided by primary care physicians (PCPs) and to identify:

1. the different types of financial incentives that have improved quality;
2. the characteristics of patient populations for whom quality of care has been improved by financial incentives; and
3. the characteristics of PCPs who have responded to financial incentives.

Search Methods: We searched the Cochrane Effective Practice and Organisation of Care (EPOC) Trials Register, Cochrane Central Register of Controlled Trials (CENTRAL) and Cochrane Database of Systematic Reviews (CDSR) (The Cochrane Library), MEDLINE, HealthSTAR, EMBASE, CINAHL, PsychLIT, and ECONLIT. Searches of Internet-based economics and health economics working paper collections were also conducted. Finally, studies were identified through the reference lists of retrieved articles, websites of key organisations, and from direct contact with key authors in the field. Articles were included if they were published from 2000 to August 2009.

Selection Criteria:Randomised controlled trials (RCT), controlled before and after studies (CBA), and interrupted time series analyses (ITS) evaluating the impact of different financial interventions on the quality of care delivered by primary healthcare physicians (PCPs). Quality of care was defined as patient reported outcome measures, clinical behaviours, and intermediate clinical and physiological measures.

Data Collection and Analysis: Two review authors independently extracted data and assessed study quality, in consultation with two other review authors where there was disagreement. For each included study, we reported the estimated effect sizes and confidence intervals.

I will freely admit I’ve been skeptical of these types of reforms. It’s not that I don’t think quality is important, or that – in theory – paying for quality is better than paying for quantity. My problem is that I think we haven’t yet figured out how to measure quality well in many cases, and the metrics we pick are often too flawed. People can game the system, rather than really improve outcomes.

Regardless of my views, this review looked for all reasonably well-designed studies looking at the impact of financial incentives for quality of care in primary care practice. There were seven of them. They included a variety of financial mechanisms. Outcomes of interest included smoking cessation, cervical screening, mammography screening, HbA1c, childhood immunizations, chlamydia screening, appropriate asthma medication, and even diabetes outcomes and patients’ assessments of quality. Six of the seven studies found some modest improvements in some outcomes, but certainly not all. One study found no difference at all. Unfortunately, there was plenty of bias in  some of the studies as well. For instance, none of them considered that the physicians who agree to participate in these types of plans are going to be much different than those that do not.

The table with the results is simply too large to post. But if it’s not behind a paywall, it’s here and it’s Table 1. Go look. You won’t be that impressed.

The conclusions were these:

The use of financial incentives to reward PCPs for improving the quality of primary healthcare services is growing. However, there is insufficient evidence to support or not support the use of financial incentives to improve the quality of primary health care. Implementation should proceed with caution and incentive schemes should be more carefully designed before implementation. In addition to basing incentive design more on theory, there is a large literature discussing experiences with these schemes that can be used to draw out a number of lessons that can be learned and that could be used to influence or modify the design of incentive schemes. More rigorous study designs need to be used to account for the selection of physicians into incentive schemes. The use of instrumental variable techniques should be considered to assist with the identification of treatment effects in the presence of selection bias and other sources of unobserved heterogeneity. In randomised trials, care must be taken in using the correct unit of analysis and more attention should be paid to blinding. Studies should also examine the potential unintended consequences of incentive schemes by having a stronger theoretical basis, including a broader range of outcomes, and conducting more extensive subgroup analysis. Studies should more consistently describe i) the type of payment scheme at baseline or in the control group, ii) how payments to medical groups were used and distributed within the groups, and iii) the size of the new payments as a percentage of total revenue. Further research comparing the relative costs and effects of financial incentives with other behaviour change interventions is also required.

I don’t think there’s much more for me to add. Spot on.

@aaronecarroll

An enormous literature documents that better off people are healthier than worse off people (entry points here or here). But that’s not the same thing as showing that transferring resources from the rich to the poor would improve the health of the poor.

In an ideal world — or at least in my ideal world — we would run controlled experiments that transferred resources to the poor to get objective evidence about the effects of redistribution. However, there are few if any such experiments. You can get close though, as shown by Hilary Hoynes, Douglas Miller, and David Simon in an NBER working paper.

Hoynes and her colleagues looked at the effect of a change in the Earned Income Tax Credit (EITC) on infant health. The EITC is a tax credit available to poor and middle income Americans. Moreover, if the credit is greater than your tax liability, you get a refund. The amount you get depends on the size of your family.

Of course, infants don’t earn income or receive tax credits. But their parents do and it is plausible that being born in a family that has received a tax credit would be to the infant’s advantage.

So how can we measure how much a poor infant would benefit from the family’s EITC? (WARNING: If you don’t care about the methods here, skip way down to the paragraph beginning “Hoynes and her colleagues looked…”)

The EITC was increased as part of the Omnibus Budget Reconciliation Act of 1993 (OBRA 1993). We could test whether there was an improvement in infant well-being in poor families after 1993. The problem, however, is that if we found such an improvement, it could have resulted from some other change that made the years after 1993 different from the years before.

But the OBRA 1993 didn’t just increase the EITC, it also increased the size of the credit for families with two or more children relative to the credit for one-child families. So imagine four families.

• Family_1990_2nd had its second child in 1990. Because this family had one child in the previous tax year, it would have received a $1250 EITC during the year of the pregnancy.
• Family_1990_3rd had its third child in 1990. Because this family had two children in the previous tax year, and because the EITC treated families with one and two children the same, this family would have also received a $1250 EITC.
• Family_1996_2nd had its second child in 1996. The EITC has now increased and this family would have received a $2250 EITC during the year of the pregnancy.
• Family 1996-3rd had its third child in 1996. Because the tax code now distinguishes between families with one child and families with two or more children, this family received a $3750 EITC.

Notice that EITC(Family_1990_3rd) = EITC(Family_1990_2nd). So if the EITC benefits infants, it wouldn’t benefit Family_1990_3rd’s 1990 newborn relative to Family_1990-2nd’s newborn. There might be a parity effect (e.g., second children might be healthier than third children). But there wouldn’t be an EITC effect.

However, EITC(Family_1996_3rd) > EITC(Family_1996_2nd). So if you compared birth outcomes for the 1996 families, if the EITC matters there would be a parity effect and an EITC effect. Hoynes et al. cleverly saw that this meant that if the EITC matters, then

Outcomes(Family_1996_3rd) - Outcomes(Family_1996_2nd) > 

Outcomes(Family_1990_3rd) - Outcomes(Family_1990_2nd),

and they could use this relationship to estimate the benefit to infants of the EITC. The key insight is that by comparing within-year differences in parity groups before and after 1993, they have controlled for other factors that changed across that span of history.

Hoynes and her colleagues looked at these patterns using US Vital Statistics data for birth outcomes. They found that

increased EITC income reduces the incidence of low birth weight and increases mean birth weight. For single low education (<= 12 years) mothers, a policy-induced treatment on the treated increase of $1000 in EITC income is associated with a 6.7 to 10.8 percent reduction in the low birth weight rate.

Low birthweight is likely a proxy for prematurity, which is much harder to measure. Premature infants are at risk for pulmonary, vision, and neurological problems. These problems can persist through childhood and adulthood. Premature infants can require expensive hospitalizations in neonatal intensive care units.

How would the EITC income improve infant health?

Our results suggest that part of the mechanism for this improvement in birth outcomes is the result of more prenatal care, and less negative health behaviors (smoking).

The bottom line is that redistributing income to poor families improves the health of their infants. It is, in effect, a form of prenatal care.

Thanks to Aaron Schwartz (@aschwattie) for the link to the Hoynes article.

UPDATE: Here’s an earlier and excellent post on Hoynes et al. by Brendan Saloner (@brendansaloner). How did I miss this, given that he posted it on a blog I write for? Umm…

@Bill_Gardner

Since the May 2nd publication of the NEJM paper on the Oregon Health Insurance Experiment (OHIE), there has been an uptick in commentary about the effects of poor mental health on physical health. In a new post on the AcademyHealth blog, I round up some of that commentary and toss in some additional review of the relevant literature. Go read it!

@afrakt

In NEJM, Nick Seddon and Thomas Lee make a claim worth considering and discussing.

Only when societies commit to covering all their citizens with their limited resources do they take on the difficult work of improving the value of care. [...]

Universal coverage creates challenges — notably, the rationing that results from competition for scarce funds. But without commitment to universal coverage, it’s too easy to “solve” financial problems by not insuring or underinsuring people. Universal coverage forces discipline. It also shapes social solidarity, community responsibility, and even audacious aspirations. In U.S. institutions, for example, “stroke teams” think about giving great care to people who’ve had strokes. In the English National Health Service (NHS, which is administered separately in England, Northern Ireland, Scotland, and Wales) stroke teams do the same but also think about how to reduce strokes in a given population.

It is remarkable to me how much squabbling we’ve endured over the coverage question (a century and no signs of letting up) relative to the value question. Though it’s true that the nature of the coverage regime imposes some constraints on and distortions to value, I don’t think they’re as large as often suggested. Meanwhile, it is hard for me to reject the hypothesis that the haphazardly accreted U.S. status quo delivers less value per dollar, on average, than just about any of the designed systems of countries to which it is often compared (from Singapore to Switzerland to the U.K. to Canada).

The authors go on to describe aspects of U.S. and U.K. health systems that should make each country the envy of the other. I was surprised to read that the U.S. offers great outcomes and service, two areas I find lacking, especially given the price. We have not committed to universal coverage. Hence, by the logic above, we haven’t fully taken on “the difficult work of improving the value of care.” Amid all the bickering and obstruction (which goes “both” ways), it’s hard to imagine we soon will.

The paper is ungated.

@afrakt

More than one of you sent me the following study at the end of last week. “Caloric effect of a 16-ounce (473-mL) portion-size cap on sugar-sweetened beverages served in restaurants“:

Background: New York City recently proposed a restriction to cap the portion size of all sugar-sweetened beverages (SSBs) sold in food-service establishments at 16 oz (473 mL). One critical question is whether such a policy may disproportionally affect low-income or overweight individuals.

Objective: The objective was to determine the demographic characteristics of US individuals potentially affected by a 16-oz portion-size cap on SSBs and the potential effect on caloric intake.

Design: We analyzed dietary records from the NHANES 2007–2010. We estimated the proportion of individuals who consumed at least one SSB >16 fluid oz (473 mL) in restaurants by age, household income, and weight status.

Results: Of all SSBs >16 oz (473 mL) purchased from food-service establishments, 64.7% were purchased from fast food restaurants, 28.2% from other restaurants, and 4.6% from sports, recreation, and entertainment facilities. On a given day, the policy would affect 7.2% of children and 7.6% of adults. Overweight individuals are more likely to consume these beverages, whereas there was no significant difference between income groups. If 80% of affected consumers choose a 16-oz (473-mL) beverage, the policy would result in a change of −57.6 kcal in each affected consumer aged 2–19 y (95% CI: −65.0, −50.1) and −62.6 kcal in those aged ≥20 y (95% CI: −67.9, −57.4).

Conclusion: A policy to cap portion size is likely to result in a modest reduction in excess calories from SSBs, especially among young adults and children who are overweight.

This was a retrospective study comparing those who drank at least one 16 oz or greater sugar sweetened beverage per day to those who did not. The results aren’t surprising. Most of the beverages were purchased from fast food restaurants. And, more than 7% of adults and children partook of at least one such beverage each day. Moreover, the study found that if a policy banning such beverages were enacted, and people decided to abandon these large beverages, they’d consume about 60 less calories per day.

I don’t doubt any of this. But it all misses the point. I’m concerned that the policy won’t result in what the researchers hope will happen. In fact, simulations show things could get worse. There’s lots of evidence that it’s more than just the sugar sweetened beverages that’s the problem. Plus, this:

How does it make sense to ban a 225-calorie soda when a “Gotta Have It”–sized PB&C (peanut butter and chocolate ice cream) shake at Cold Stone Creamery clocks in at 2010 calories alone? That drink got the top prize in the 2011 Xtreme Eating Awards sponsored by the Center for Science in the Public Interest. Or perhaps you could wander over to the Cheesecake Factory and get yourself a Farmhouse Cheeseburger (for 1530 calories) and a nice piece of Ultimate Red Velvet Cake Cheesecake (for an additional 1540 calories). Without even touching a French fry, you’ve consumed more than 3000 calories.

Fancier food isn’t immune either. A Morton’s steakhouse Porterhouse with mashed potatoes and half a side of creamed spinach rates 2570 calories, 85 g of saturated fat, and 2980 mg of sodium. As the CSPI noted, “That’s the calories of eight pieces of Original Recipe chicken plus mashed potatoes and gravy, coleslaw, and four biscuits at KFC [Kentucky Fried Chicken], with an extra 1½ days’ [saturated] fat on the side.”

I’m sorry, but banning the soda and leaving the rest untouched is silly to me. It’s a distraction. I’m all for trying to get people to reduce their caloric intake. I don’t think another study showing us that some of those calories come from sugar sweetened beverages convinces me that a ban on this one thing is worthwhile.

@aaronecarroll

In two weeks Google Reader will shut down. If that’s how you read this blog, you’ll have to find a new way. My suggestion: sign up for the daily email digest. That way, you’ll keep getting TIE posts (bundled in one email, once per day, around 7AM) even as you shop around for a new reader.

As for your new reader, I have recommended and still recommend NewsBlur. There are other options,* but I have not tested them as extensively as NewsBlur, with which I am totally satisfied. I’ve been using it on the web and on Android and iOS devices exclusively for a couple of months. It’s worth the fee, though you can get a trial version for free, despite the appearance that you can’t on the site (or so I’m told).

You’ll want to transfer your feeds from Google Reader. Do so with Google Takeout.

* To find other options, do a web search on “Google Reader replacements” or similar.

@afrakt

I cracked open the latest MedPAC report to read its recommendations on hospital readmissions. There wasn’t much new in it beyond what I could infer about the Commissions’ thinking in March. Still, there’s a lot more detail. You’ll find it in Chapter 4 (PDF) and the accompanying appendix (also a PDF). Some of it was reported on by Jordan Rau. All of this is ungated, which is one reason I’m giving it short shrift here.

The other reason is that I was more surprised by what I found in Chapter 1 (PDF).

Consistent with the goal of encouraging beneficiaries to make cost-conscious choices, this chapter presents an overview of a model based on government contributions toward purchasing Medicare coverage—an approach we call competitively determined plan contributions (CPCs). The Commission uses the term CPC to broadly describe a federal contribution toward coverage of the Medicare benefit based on the cost of competing options for the coverage, including those offered by private plans and the traditional FFS program. Specifically, CPC has two defining principles: First, beneficiaries receive a competitively determined federal contribution to buy Medicare coverage; second, beneficiaries’ individual premiums vary depending on the option they choose.

As far as I know, this is the first time the Commission has considered Medicare plan competitive bidding (aka, premium support) — their CPC — in this way before. To be clear, it is not recommending CPCs. It’s merely exploring the idea. The chapter hits many of the issues related to competitive bidding that have been discussed on this blog (look here and here).

Competing private plans, however, do not necessarily lower the cost to the Medicare program if the rules defining how they get paid do not encourage them to compete based on cost or premiums. For example, the current Medicare Advantage (MA) program produces a higher cost to Medicare than the traditional FFS program. Therefore, whether a CPC approach can lower overall Medicare spending depends on the specific design of the model and how different components of the model interact. [...]

Medicare Part D provides a working example of a CPC approach and illustrates the range of the detail and specificity of the rules that a CPC approach requires. [...]

The Federal Employees Health Benefits (FEHB) Program also illustrates different applications of the CPC principles.

Again, you can read the chapter for details. There you’ll also find an exploration of these questions:

• Should the benefit package be standardized?
• Should a CPC model be based on competitive bidding?
• Should a CPC model include FFS Medicare?
• How should the federal contribution be determined?
  

These are just the “first-order” questions. A presumably high-order question, “How does the federal contribution grow over time?” was raised in the report but not addressed.

What’s interesting to me is not so much what MedPAC addressed or how they did so, since, again, I’ve covered it all here in some form. What’s interesting is that it has taken a small but significant step toward competitive bidding/premium support, not by endorsement, but just by consideration. It’s now clearly on the table for discussion by the Commission, though I don’t think this necessarily moves the political needle at all. Meanwhile, as far as I know, the Commission’s prior MA payment recommendation still stands: pay plans 100% of average fee-for-service cost.

UPDATE: MedPAC considered competitive bidding in a 2009 report. See Chapter 7 here.

@afrakt

Based on my N=1 survey and despite what it says below, I suspect this will actually be fairly easy even for some readers with advanced degrees. Still, I like this problem. To make it more interesting, see how many examples you need to deduce the algorithm.

I grabbed this off Google+. For some reason, when I went back a second time (just moments later) to try to find who posted it, I couldn’t re-find it. So, apologies for not citing the source.

@afrakt

No, not “here,” but at AcademyHealth.

AcademyHealth seeks thought leaders with innovative experiences in the translation, dissemination and implementation of health services research (HSR) for its Innovators-in-Residence Program, a new effort to learn from those who demonstrate unique approaches to moving technical information into policy or practice. Supported by a grant from Kaiser Permanente, the inaugural year of the program will address the uptake of new research findings in the delivery of safety net care, with the goal of improving care for vulnerable populations. During their one- to three-month tenure in the program, Innovators will serve as internal consultants for AcademyHealth while completing specific projects related to the science or process of translation and dissemination.

More here and here (PDF).

Disclosure: I serve on AcademyHealth’s Translation and Dissemination Institute’s Advisory Committee.

@afrakt

Helen Mayberg and her colleagues have a paper in JAMA Psychiatry (see also here) reporting progress on the development of a treatment-specific biomarker for major depressive disorder (MDD). A treatment-specific biomarker is a biological assessment that indicates what treatment is likely to succeed for a patient. The paper is important because MDD is a devastating illness for which the first treatment often fails. If we could identify the right treatment for the right patient, care could be made both more efficient and more effective. This is also the first proposed psychiatric biomarker based on brain imaging — most proposed biomarkers are based on genomic technologies. Finally, the paper is important because there are currently no treatment-specific biomarkers in psychiatry. Treatment-specific biomarkers are a key idea in personalized medicine, so this is a good opportunity to think about the possible benefits and risks of such assessments.

Here’s how the study worked. There were 82 patients with a diagnosis of MDD and a sufficiently high score on a depression severity measure to ensure that they were currently depressed. The researchers performed positron emission tomography (PET) brain scans of all patients. They looked at the images and identified locations in each patient’s brain that were more or less active relative to the rest of that patient’s brain.

The patients were then randomly selected into two groups, one to receive Lexapro (a commonly-used selective serotonin reuptake inhibitor antidepressant) and the other to receive cognitive behaviour therapy (CBT, the standard psychotherapy for depression). After 12 weeks of treatment, the researchers identified which patients got better (‘remitters’) and which did not (‘non-responders’). The key question was whether any pattern of activity or inactivity in the PET images predicted whether you were more likely to get better with one treatment versus the other.

The encouraging finding was that there did seem to be such a pattern. The insula is a portion of the cerebral cortex tied to many cognitive and metabolic functions. The upper left graph in the Figure plots brain activity in the insula (vertical axis) as a function of which treatment the patients received (the two lines) and whether they got better (the two points on the left) or did not (the two points on the right). The key result was that if you had low insular activity you were likely to do better on Lexapro than on CBT, whereas if you had high activity the reverse was true. This suggests that clinicians might begin depression care with a PET scan and then use the imaging data to decide whether the patient should get psychotherapy or the drug.

The authors carefully noted that the data are far too preliminary for clinical use. Above all, the signal from the data is unusually weak. The key data pattern does not appear to be statistically significant. Moreover, the investigators explored a large space of possible brain scan patterns to find the one that best fit the treatment outcome data. I don’t think this study would have been published in a JAMA journal were it not for the originality of the question and the potential significance of the result. To be credible, the results have to be replicated prospectively in a much larger sample.

For the sake of discussion, however, let’s suppose they and others rerun this study many times and replicate these findings. Where are we then?

We would still need to think carefully about when we should use this technology. PET has a hard science patina that might be attractive to patients. Many physicians and all PET manufacturers would be pleased if an expensive scanning technology became a part of routine care for a relatively common mental illness.

This should not happen, however, without clear evidence that the benefits that patients receive from getting the better treatment are greater than the cost of the treatment and the small radiation exposure from the PET scan. This benefit might be much less than we imagine looking at the dramatic cross-over pattern in the data in the Figure. The graph represents averaged data from many patients, but the signal from an individual patient’s PET scan will be much noisier. In addition, many patients seen in routine care are less depressed than those examined in this study. The PET scan is likely to be less informative for such patients. Therefore, despite impressive results in a controlled study like this one, it could be that treatment choices based on biomarkers would be better than standard psychiatric decision making for only a few patients in a thousand.

The concern, then, is that unless we are careful in how we validate and implement the technology, even ‘successful’ biomarker technologies could add considerable expense to routine care with little or no benefit for most of the patients who receive it.

It doesn’t have to happen that way. We could introduce personalized technology with a careful evaluation of the costs and benefits both when it is implemented and as practice evolves. We could limit insurance payments for these assessments to cases where the supporting evidence is clear. We should be capturing data from routine practice for ongoing evaluations of the cost-effectiveness of care. I’ll have to save it for another post, but I don’t think we are doing enough of any of these things.

@Bill_Gardner