Tomorrow, as mandated by the Patient Protection and the Affordable Care Act (PPACA, often called just the Affordable Care Act, or ACA, or “Obamacare”), the government-maintained health insurance exchanges will open for business (that is, assuming the likely government shutdown doesn’t stop them temporarily). This post will be basically a followup to a post I did almost a year ago that used a statement by Mitt Romney during the height of the Presidential campaign as a jumping off point to look at the relationship between health insurance status and mortality. Unfortunately, it is not entirely possible to disentangle science from politics, and we have to dive in. Also, politics is the art of the possible; so, policy-wise, what is best as determined by science might well not be what is possible politically.
The reason I wanted to revisit this topic is because of a political battle that went on for quite some time over the last several months to expand Medicaid in Michigan according to the dictates of the ACA. The reason that this battle is occurring in many states is because when the Supreme Court ruled last year that the individual mandate requiring that citizens have health insurance was Constitutional, one provision that it ruled unconstitutional was the mandatory expansion of Medicaid in states participating in the Medicaid program to cover all people under 65 up to 133% of the federal poverty level. States thus had to decide whether or not they would accept the Medicaid expansion. In our state, Governor Rick Snyder supported the expansion. Even though he is Republican, he is also a businessman and realized that it was a good deal, with the federal government covering 100% of the cost for the first three years and then phase down to 90% of the cost in 2020. The bill to expand Medicaid managed to pass the House of Representatives, but then it stalled in the Senate. Unfortunately—and this is what got me involved—my state senator led the opposition to the Medicaid expansion in the Senate, much to my chagrin, disappointment, disapproval, and even anger. His argument, which is being repeated elsewhere in the blogosphere, is that Medicaid is worthless and doesn’t improve health outcomes. Instead, he endorsed an alternative that (or so he claimed) place Medicaid-eligible patients into in essence low cost, high deductible concierge practices, with health savings accounts like this one. Ultimately, my state senator lost, and Medicaid was expanded in Michigan in a plan that was characterized by John Z. Ayanian in this week’s New England Journal of Medicine as “a pragmatic pathway to link Republican and Democratic priorities for health care.”
The drama in my own state notwithstanding, the whole kerfuffle got me to thinking. In my post a year ago, I basically asked what the evidence was that access to health insurance improves health outcomes, but I didn’t really stratify the question into kinds of health insurance. Rather, I just looked at being uninsured versus having health insurance. After my little Facebook encounter with one of my elected representatives, I wondered what, exactly, was the state of evidence. So I decided to do this post. In the U.S., currently we have in essence three kinds of health insurance, broadly speaking: Private insurance, Medicare, and Medicaid. Medicare, for those of our readers from other countries, is a plan that covers the medical care of people 65 and over and those receiving Social Security disability benefits. It is funded through payroll taxes and directly paid for by the federal government. Medicaid, in contrast, is a plan designed for low income people who fall below certain income levels. Also in contrast, it is jointly funded by the states and the federal government with each participating state administering the plan and having wide leeway to decide eligibility requirements within the limits of federal regulations that determine minimal standards necessary for states to receive matching funds. Indeed, the loss of this leeway to determine the income level at which a person is eligible for Medicaid is one of the reasons the provision for Medicaid expansion was part of the Supreme Court challenge to the ACA. These days, most Medicaid plans hire private health maintenance organizations (HMOs) to provide insurance. Finally, what needs to be understood is that, compared to private insurance, Medicare reimbursement rates tend to be lower and Medicaid reimbursement rates are lower still, which is part of the reason why a lot of doctors don’t accept Medicaid. Increases in reimbursement under the ACA might well help this situation.
With that admittedly lengthy introduction behind us, let’s look at the science of the question of whether Medicaid is as useless as my state senator and others claim that it is. It’s a huge set of studies, many conflicting and many also unable to control completely for confounders that interact with insurance status, such as socioeconomic status, risk factors like tobacco and alcohol abuse, and the like. Not surprisingly, unlike claims of Medicaid detractors, it’s complicated, and, like most complicated issues in medicine, there are studies that can be cherry picked by anyone to make whatever point he wants (which is exactly what my state Senator tended to do). A good place to start (for me, at least) is my back to my original post from a year ago, because there was one study to which I alluded that showed differences in outcomes in cancer patients that correlate with their insurance status, specifically the University of Virginia study from 2010, which found that Medicaid and uninsured status were independently associated with increased risk-adjusted mortality. I mentioned that, but didn’t dwell on it, because the question I was examining was not Medicaid versus private insurance but rather having health insurance versus no health insurance. In this study, however, the authors noted the multifactorial causes of poorer outcomes in Medicaid patients. There are other studies that find similar outcomes disparities. For instance, using the National Inpatient Sample (NIS) database, which is a stratified random sample of all hospital discharges in the United States maintained by the Agency for Healthcare Research and Quality as part of the Healthcare Cost and Utilization Project, a 2003 study from the University of Michigan found that for abdominal aortic aneurysms (AAA) uninsured and Medicaid status were associated with higher mortality and rupture rates:
Operative mortality rates after elective AAA repair were greater (P =.04) for patients with no insurance (2.6%) or Medicaid (2.7%), compared with patients with private insurance (1.2%). Similarly, operative mortality rates for AAA repair after rupture were greater (P =.001) in patients without insurance (45.3%) or Medicaid (31.3%), compared with patients with private insurance (26.2%).
Of course, in this study, Medicaid patients did better than the uninsured for some outcomes. For instance, if you look at the adjusted odds ratio for AAA rupture, Medicaid status had no effect, with an odds ratio of 0.84 (0.55-1.3, p=0.41), while no insurance produced an odds ratio of 2.3 (1.5-3.5, p=0.001). From the point of view of this study, it is better to have Medicaid than to be uninsured but not as good as having private insurance. This is not entirely surprising because of the low reimbursement rates of Medicaid, which limit the choices of physicians and institutions for Medicaid patients and make it prohibitive for many private primary care doctors to care for Medicaid patients. Similar results were found in a 2011 study of cardiac valve replacement from the University of Virginia, which showed the best outcomes in terms of mortality and in-hospital complications in patients with private insurance, followed by Medicare patients, Medicaid patients, and the uninsured, who had the worst outcomes of all. They also reported that Medicaid patients accrued the longest hospital stay and highest costs. Consistent with this, a 2013 study from the University of Virginia of pediatric surgery patients undergoing mostly urgent surgery for appendectomy, intussusception, decortication, pyloromyotomy, congenital diaphragmatic hernia repair, and colonic resection for Hirschsprung’s disease showed that uninsured patients were at increased risk for mortality, while Medicaid patients were at increased risk of morbidity.
Of course, the question really boils down to whether it is equivalent to be uninsured versus having Medicaid, which some studies seem to indicate, or not. Again, this is a very complicated question, because studies looking at specific outcomes can be confounded by uninsured patients getting Medicaid. For instance, uninsured patients who are diagnosed with cancer in our state frequently qualify immediately for Medicaid and are no longer uninsured. This leads to a rather frustrating situation for some of my patients who are uninsured and suspected of having cancer but can’t afford the biopsy necessary to prove it and make them eligible for Medicaid. We have other resources, including limited state funds and charitable funds administered through our cancer center that can fund such biopsies, but it’s very frustrating to cancer doctors that such resources are even necessary, given how great the need is.
All of this leads to a potential explanation, and quite a reasonable explanation at that, as to why Medicaid patients do more poorly than patients with private insurance, in some studies (the ones touted by my state senator as indicating that Medicaid is worthless and doesn’t improve health outcomes), and that’s delay in treatment. This was alluded to as a possible cause in the AAA study from 2003, but it’s suggested more explicitly as a cause in a 2012 study from the Brigham and Women’s Hospital examining outcomes after surgery for spinal metastases. This study found higher mortality rates for the uninsured and Medicaid patients, as well as higher complication rates. However, these were crude estimates. When the investigators adjusted for acuity of presentation, there was no significant differences in the risk of death or complications between privately insured patients and Medicaid patients or the uninsured, leading the authors to conclude that, “This nationwide study suggests that disparities based on insurance status for patients undergoing surgery for spinal metastases may be attributable to a higher acuity of presentation.” A recent systematic review of outcomes in lung cancer patient mortality came to similar conclusions:
The mechanisms underlying care disparities for patients without insurance and for those who receive Medicaid are unclear but probably multifactorial (6 6. Woods LM, Rachet B, Coleman MP. Origins of socio-economic inequalities in cancer survival: a review. Ann Oncol 2006;17:5–19. View Full Reference List , 41 ). There are likely patient-related factors, such as individual differences in health behaviors such as smoking, income, education, and comorbidities. Others may stem from a differential ability to interact with the healthcare system, differences in the care provided by institutions that serve Medicaid and uninsured patients, and less access to better-quality care. Our review indicates that some of these latter mechanisms may be important based on studies that show differential rates of receiving guideline-concordant care (31, 32), receiving care at a university cancer center (29), receipt of surgery or radiation therapy (20), and receipt of care at high-volume centers (27, 28). Although uninsured patients and those with Medicaid may be more likely to be treated at certain centers, no studies directly adjusted for center-level effects, so we cannot determine their role on the assessed outcomes (42).
Taken together, from my perspective, the evidence is consistent with a conclusion that having Medicaid results in better health outcomes than not having Medicaid but that those outcomes are not as good as those associated having private insurance (although one study did find a paradoxical result). Most likely this is due to a combination of socioeconomic status and lack of primary care resources, all leading to their presenting at a later stage in their disease, as these studies, which are just a more recent sampling of existing studies, clearly indicate. For some conditions, Medicaid patients do as poorly as the uninsured, and those are the studies cited by legislators like my state senator to argue against Medicaid expansion. Sometimes they are intermediate in their outcomes, not as good as patients with private insurance but not as bad as the uninsured. These tend to be the studies ignored by legislators like my state senator and the pundits that he cites. However you analyze the evidence, however, it is clear that Medicaid patients do have worse outcomes, sometimes a lot worse, than patients with private insurance, and that the cause is almost certainly multifactorial in such a way that simply getting access to bare-bones health insurance like Medicaid can’t remedy. Many of these studies have unmeasured confounders that resort in worse outcomes in Medicaid patients. As Frakt et al have argued that selection bias in these studies explains a lot of the results:
It’s far more likely that such results are driven by selection bias. Medicaid enrollees (including dual-eligible recipients of both Medicaid and Medicare) tend to be sicker than uninsured patients and to have lower socioeconomic status, poorer nutrition, and fewer community and family resources. Medical and social service providers may also help the sickest or neediest patients to enroll in Medicaid — a more direct cause of selection bias. Few of these potential confounders can be completely addressed using commonly available clinical or population data.
Health economists use an alternative approach in analyzing Medicaid’s outcomes that seeks to eliminate selection bias related to unobservable factors affecting enrollment and health outcomes. By exploiting the variation in Medicaid eligibility rules or other program characteristics influencing states’ enrollment rates, scholars have consistently found that Medicaid coverage leads to health improvements.4,5 The assumption behind these “instrumental variables” approaches is that Medicaid enrollment depends on state-level eligibility rules but patients’ health status does not.
Personally, I suspect that there is a lot of selection bias in these studies. Young healthy adults without insurance who are eligible might not enroll in Medicaid because they don’t think they need it and also tend not to need the procedures examined in the studies I discussed. In contrast, if you’re sick and eligible for Medicaid, you’ll be more likely to enroll because you need the treatment. Also, as I’ve pointed out before, in studies of cancer, there is a lot of crossover between the uninsured and Medicaid recipients, because in some states like mine a cancer diagnosis makes an adult who might not have been eligible for Medicaid (remember, some states restrict Medicaid eligibility using more than just income) will suddenly become eligible. I’ve lost count of the number of patients I’ve taken care of for whom this was true. The same thing happens in emergency rooms all over the place, where hospitals, confronted with an uninsured patient, help him apply for Medicaid during the course of an acute illness.
Would expanding Medicaid result in better health outcomes?
Of course, the name of this blog is Science-Based Medicine, and the question that results from the confusing and conflicting mass of studies sampled above is whether expanding Medicaid would result in better health outcomes for the people who receive the expanded coverage. This is a much more difficult question to answer, but there have been studies designed to address this question. The two most prominent are frequent “targets” of discussion. The previous studies that I’ve cited are all retrospective studies, with all the attendant shortcomings of retrospective studies, and none of them address this question. They simply found correlations, and some of them tried to explain these correlations.
One study that is often touted as strong evidence that expanding Medicaid eligibility will likely result in better health outcomes was a study published in the NEJM a year ago. What the investigators did was to identify states that had expanded Medicaid to cover childless adults (in many states childless adults have not been eligible for Medicaid coverage) between 2000 and 2005 to allow comparisons for a five year period before the expansion and after the expansion:
Three states met our criteria: Arizona, which expanded eligibility to childless adults with incomes below 100% of the federal poverty level in November 2001 and to parents with incomes up to 200% of the federal poverty level in October 2002; Maine, which expanded eligibility to childless adults with incomes up to 100% of the federal poverty level in October 2002; and New York, which expanded eligibility to childless adults with incomes up to 100% of the federal poverty level and parents with incomes up to 150% of the federal poverty level in September 2001.
The controls selected were neighboring states without Medicaid expansions that were closest in population and demographic characteristics to the three states with Medicaid expansions. The primary outcome examined was annual county-level all-cause mortality per 100,000 adults between the ages of 20 and 64 years (stratified according to age, race, and sex), obtained from the Compressed Mortality File of the Centers for Disease Control and Prevention (CDC) from 1997 through 2007, totaling 68,012 observations specific to an age group, race, sex, year, and county. Secondary outcomes included percentages of persons with Medicaid, without any health insurance, and in “excellent” or “very good” health (from the Current Population Survey, a total of 169,124 persons) and the percentage unable to obtain needed care in the past year because of cost (from the Behavioral Risk Factor Surveillance System, a total of 192,148 persons). Multivariate analyses were carried out, and prespecified subgroup analyses, and as an additional test the same analyses were carried out for people over 65, who were eligible for Medicare and whose Medicaid eligibility was therefore not affected by the Medicaid expansion. Overall, the investigators found a 6.1% relative reduction in the risk of death among adults between the ages of 20 and 65, leading them to estimate this:
A relative reduction of 6% in population mortality would be achieved if insurance reduced the individual risk of death by 30% and if the 1-year risk of death for new Medicaid enrollees was 1.9% (Table S4 in the Supplementary Appendix). This degree of risk reduction is consistent with the Institute of Medicine’s estimate that health insurance may reduce adult mortality by 25%, though other researchers have estimated greater35 or much smaller36 effects of coverage. A baseline risk of death of 1.9% approximates the risk for a 50-year-old black man with diabetes or for all men between the ages of 35 and 49 years who are in self-reported poor health. The lower end of our confidence interval implies a relative reduction in the individual risk of death of 18%.
This study did, of course, have a fair number of confounders and shortcomings. For one thing, it was not a randomized design, and it was an ecological study, which tends to overestimate effects. Also, as the authors point out, states tend to decide to expand Medicaid when the economy is doing well and they can afford to do it. Also, there is a correlation between states willing to expand Medicaid and investment in other measures designed to improve public health. On the other hand, the authors reported found that new Medicaid enrollees tended to be older, disproportionately minorities, and twice as likely to be in fair or poor health as the general population, all of which to them suggested a higher risk of mortality. In other words, this study was promising, but by no means slam-dunk evidence that Medicaid expansion will result in better health outcomes.
The Oregon study
No discussion of this issue is complete without a consideration of a study in Oregon designed to look at the effect of Medicaid expansion. Its most recent results were reported in the NEJM five months ago and were seized upon by advocates on all sides, but in particular the “Medicaid expansion doesn’t work” and “Medicaid is harmful” side. It’s a curious study in that, had it been proposed to me before I already knew that it had been begun, I would have seriously questioned whether the study was ethical and would pass an institutional review board. Obviously it did, and the reason is that it wasn’t the investigators who did the randomization. Rather, Katherine Baicker and her colleagues took advantage of an existing randomization. What happened is this. In 2008, because its legislature found the money to fund additional Medicaid coverage, Oregon used a lottery system to determine who of a waiting list of 90,000 would have a chance at getting Medicaid. Selected adults won the right to apply for Medicaid and got it if they met the eligibility requirements. A sample of adults who won the Medicaid lottery were compared to adults who participated in the lottery but didn’t win. Outcome measures examined included blood-pressure, cholesterol, and glycated hemoglobin levels; screening for depression; medication inventories; and self-reported diagnoses, health status, health care utilization, and out-of-pocket spending for such services. This sample was limited to the Portland metropolitan area because of logistical constraints and consisted of 20,745 people: 10,405 selected in the lottery (the lottery winners) and 10,340 not selected (the control group), of which a total of 12,229 persons in the study sample responded to the survey. Interviews were conducted between September 2009 and December 2010 and took place an average of 25 months after the lottery began.
This was a rare opportunity to take advantage of an existing natural experiment in whether providing Medicaid coverage to the uninsured actually does what it is intended to do. Unfortunately, given its timing, the results of the study have become a political punching bag with the Oregon study in essence being used as a weapon against the whole of Obamacare and misrepresented as being slam dunk evidence that Medicare is at least useless. In essence, as Ashish Jha put it, the Oregon Study became a Rorschach test of sorts, confirming people’s biases about whether Medicaid is “good” or “bad.”
So what did it show?
The results were mixed and rather disappointing in some respects but not entirely unexpected given the short followup time of only two years:
We found no significant effect of Medicaid coverage on the prevalence or diagnosis of hypertension or high cholesterol levels or on the use of medication for these conditions. Medicaid coverage significantly increased the probability of a diagnosis of diabetes and the use of diabetes medication, but we observed no significant effect on average glycated hemoglobin levels or on the percentage of participants with levels of 6.5% or higher. Medicaid coverage decreased the probability of a positive screening for depression (−9.15 percentage points; 95% confidence interval, −16.70 to −1.60; P = 0.02), increased the use of many preventive services, and nearly eliminated catastrophic out-of-pocket medical expenditures.
As has been pointed out at The Incidental Economist, this led those opposing Obamacare to declare the experiment a failure. Some go so far as to declare that the ACA has completely failed, that Medicaid is a horrible thing, and that anyone who tries to argue against this is a “Medicaid denier.” (I kid you not; unfortunately, this is a person known for making seriously bad arguments.) Unfortunately, my state Senator buys into these arguments. I also agree with Jha when he challenges physicians to go through all 62 pages of the supplementary appendix describing the methodology in detail because this study is as good as any study on the matter likely to be done in a generation.
So is it really as bad as that. Of course not. The study is, as are most studies of this type, messy and early results are disappointing, but by no means is it time to declare failure yet. For one thing, improved mental health outcomes are improved outcomes. The way critics of Medicaid ignore this result or pooh-pooh it to me is consistent with how we short shrift mental health in this country, viewing mental illness as somehow not being “real” illness. Also, the marked decrease in financial distress reported in this study is no small thing. For another thing, lack of statistical significance doesn’t necessarily mean that there is no treatment effect. It can mean that, but it can also mean that the numbers are too small and the study is underpowered, which it could well be. Indeed, the authors themselves concede as much:
Hypertension, high cholesterol levels, diabetes, and depression are only a subgroup of the set of health outcomes potentially affected by Medicaid coverage. We chose these conditions because they are important contributors to morbidity and mortality, feasible to measure, prevalent in the low-income population in our study, and plausibly modifiable by effective treatment within a 2-year time frame. Nonetheless, our power to detect changes in health was limited by the relatively small numbers of patients with these conditions; indeed, the only condition in which we detected improvements was depression, which was by far the most prevalent of the four conditions examined. The 95% confidence intervals for many of the estimates of effects on individual physical health measures were wide enough to include changes that would be considered clinically significant — such as a 7.16-percentage-point reduction in the prevalence of hypertension. Moreover, although we did not find a significant change in glycated hemoglobin levels, the point estimate of the decrease we observed is consistent with that which would be expected on the basis of our estimated increase in the use of medication for diabetes. The clinical-trial literature indicates that the use of oral medication for diabetes reduces the glycated hemoglobin level by an average of 1 percentage point within as short a time as 6 months.15 This estimate from the clinical literature suggests that the 5.4-percentage-point increase in the use of medication for diabetes in our cohort would decrease the average glycated hemoglobin level in the study population by 0.05 percentage points, which is well within our 95% confidence interval. Beyond issues of power, the effects of Medicaid coverage may be limited by the multiple sources of slippage in the connection between insurance coverage and observable improvements in our health metrics; these potential sources of slippage include access to care, diagnosis of underlying conditions, prescription of appropriate medications, compliance with recommendations, and effectiveness of treatment in improving health.
The Incidental Economist thinks that most likely the reason that the results, although trending in the right direction, didn’t achieve statistical significance was because the study was underpowered. Jha thinks it’s because Medicaid only addresses access to care and not quality of care. Both are likely contributors, but I tend to think that The Incidental Economist is likely to be closer to being correct. Contrary to analyses claiming that this study was not underpowered, for a controlled study of this type, the fraction of subjects with each condition very much matters. If only 5% of the population have a condition, that’s only around 300 subjects in each group. When looking at all these subgroups, absent dramatic improvements in certain parameters or high prevalence of the conditions being examined, it is hard to detect statistically significant differences, particularly in a short time frame. Finally, these measures are all surrogate measures. What will really be interesting to determine will take considerably more time than two years to determine, namely whether Medicaid coverage decreases morbidity and overall mortality. Indeed, The Incidental Economist asks a very pertinent question: “What is reasonable to expect? How much does private insurance affect these values? Do we know? No. There is no RCT of private insurance vs. no insurance. No one claims we have to have one. We just “know” private insurance works.”
The bottom line
As I stated earlier, as much as we would like political policy to be science-based, particularly in health care, it can’t be said enough that politics is the art of the possible, and the ACA is what was possible at the time it was being negotiated. As a product of a messy political process, it is far from perfect. The question is whether the claims made for the ACA and its provisions are supported by evidence. Because a large part of mechanism by which the ACA will decrease the number of uninsured is through the Medicaid expansion, studies like the Oregon study are very important for determining whether it has a reasonable chance of succeeding in improving the health outcomes of these people. As is all too frequently the case, the the error bars are large surrounding the data relating Medicaid and health outcomes, and unfortunately the most recently reported results of the Oregon study come after too brief a time to make any definitive pronouncements, attacks on it by anti-Obamacare pundits notwithstanding. Moreover, when it comes to public policy, science is certainly a major consideration, but so are economics and justice. Reasonable people might disagree on where the balance should be struck, but nothing is served by distorting the science for political ends. Unfortunately, there is a lot of that going on right now, and, I suspect, it will only get worse before it gets better.