By AVIK ROY

Austin Frakt has penned a reply to a recent piece I wrote on Medicaid for my health-policy blog on Forbes, The Apothecary. Austin is a guy who takes the time to address opposing points of view, to his credit, and I’ve enjoyed my back-and-forth with him over time. But while I’m grateful for Austin’s attention to an issue of high import—the degree to which Medicaid harms the poor—he didn’t respond to the core concerns I raised in my post.

For those who haven’t been following the debate on Medicaid outcomes from the beginning, let me offer a brief summary.

How Medicaid Harms the Poor: The Debate (So Far)

Last summer, on my old blog, I put up a series of posts highlighting the findings of a study published in Annals of Surgery by a group of surgeons at the University of Virginia, entitled “Primary Payer Status Affects Mortality for Major Surgical Operations.” The study evaluated 893,658 major surgical operations occurring between 2003 and 2007, stratified by primary payer status, on three outcomes endpoints: in-hospital mortality, length of stay, and total costs incurred.

Despite the fact that the authors controlled for age, gender, income, geographic region, operation, and 30 comorbid conditions, Medicaid fared poorly compared to those with private insurance, Medicare, and even the uninsured. Relative to those with private insurance, Medicare, uninsured, and Medicaid patients were 45%, 74%, and 97% more likely to die in the hospital post-operatively. The average length of stay for private, Medicare, uninsured, and Medicaid patients was 7.38, 8.77, 7.01, and 10.49 days, respectively. Total costs per patient were $63,057, $69,408, $65.667, and $79,140 respectively.

Despite Austin’s initial criticism that this was merely one study, and therefore not representative, the poor performance of Medicaid beneficiaries is well-established in a very large body of medical literature. What was striking about the UVa study was its large sample size; that it controlled for a highly validated set of background health and social factors; and its finding that Medicaid beneficiaries not only underperformed those with private insurance (and dramatically so), but also those who lacked insurance.

Given that a core feature of PPACA is its large expansion of Medicaid to those with higher incomes than current beneficiaries, I argued that it was far from clear that this expansion would improve health outcomes, and in fact was likely to harm them by crowding out the more-efficacious private sector. Furthermore, I argued for the clinical benefits of migrating Medicaid over to a premium-support or cash-assistance model, which would allow Medicaid recipients to benefit from the superior quality of care delivered by private insurance. As I’ve said all along, “There is, doubtless, a level of poverty at which Medcaid is better than nothing at all. But most people can afford to take on more responsibility for their own care, and indeed would be far better off doing so.”

In response to my point that the poor performance of Medicaid is well-established in the medical literature, Austin came up with six studies from the economic literature to support the intuitive argument that Medicaid was, in fact, better than having no insurance at all. Furthermore, Austin argued, implausibly, that the UVa study was “not credible” because it did not use an econometric technique called instrumental variables analysis to isolate factors that might conflate correlation and causation.

I reviewed those six studies, and raised a series of criticisms. The most important of these were: (1) of the six studies, only four were actually outcomes studies, and therefore germane to the topic at hand; (2) of the four remaining studies, three were authored by the same duo (Currie and Gruber), and therefore these four studies represented the work of only two groups of authors, in contrast to how widely this problem has been studied in the medical literature; (3) the first of the four studies, which measured outcomes in HIV-positive patients by payer status, used a woefully inadequate, and effectively useless, database called the HIV Cost and Services Utilization Study (HSCUS), which relied on third-party interviews and neither recorded background health status nor treatment history; (4) the three Currie and Gruber studies, using 20-30 year-old data that may be outdated, plausibly concluded that Medicaid may benefit the very poorest beneficiaries in the obstetric and pediatric populations, but also—importantly—that Medicaid does not appear to benefit those with relatively higher incomes: a point that is quite relevant to the policy merits of expanding Medicaid under PPACA.

As to the issue of correlation vs. causation in the UVa study, I pointed out that many of the factors that UVa critics like Austin seek to treat as independent variables—things like poor access to high-quality surgeons and poor social relationships—were in fact quite directly linked to Medicaid and welfare dependency, and that an attempt to separate those factors from the Medicaid population would introduce, rather than reduce, bias in the UVa findings. (For a fuller treatment of this topic, read this.)

Notably, Austin’s post addressed none of these criticisms.

His view, and that of some other pro-PPACA health economists, is that bad data sets, like that of the HIV Cost and Services Utilization Study, can be salvaged by econometric techniques such as instrumental variables analysis. While I agree that such techniques can be useful, when appropriately applied, the importance of a large, high-quality primary data set cannot be understated.

To put it perhaps too simply: if you have two data sets of equal quality, one with a sample size of ten patients, and another with a sample size of one million, there is no methodology that can make the first data set as informative as the second. If that first data set is somewhat larger, but contains no information about background health status or treatment history, the problem gets even worse. The point is: while sound methodology is important to every study, methodology can’t do much for an uninformative data set like HSCUS.

Discussion of the UVa Surgical Outcomes Study

Austin selectively quotes from the discussion section of the UVa study to make it seem like the authors were unsure about the value of their findings. Here is a more complete review of their discussion. As a reminder, the UVa study reviewed nearly one million case reports; as the authors write, it “represents the largest and most comprehensive” analysis of surgical outcomes by insurance status ever conducted:

To our knowledge, this study represents the largest and most comprehensive review of contemporary outcomes for major operations as a function of primary payer status. In this study, we have demonstrated disparate differences in short-term surgical outcomes among payer groups. The inclusion of a broad surgical population, comprising several different surgical subspecialties, allows us to more confidently comment upon trends that have been previously reported among smaller, more specific, surgical patient groups. Our results indicate that Medicaid and Uninsured payer status confers worse unadjusted and adjusted outcomes compared with that of Private Insurance. We have shown that Medicaid and Uninsured status also independently increases the risk of adjusted in-hospital mortality, and that Medicaid status further increases the risk of adjusted in-hospital complications compared with those with Private Insurance. Moreover, our results demonstrate significant differences in resource utilization among payer groups as Medicaid patients accrued the longest adjusted hospital length of stay and highest adjusted total costs. These findings bolster those of other smaller series that have been performed in select surgical populations, and it extends the examination of payer status to include a large, nationwide, diverse surgical population.

As the authors note, their findings are consistent with the broad range of literature on the subject:

The effect of insurance status on treatment allocation and surgical outcomes has been a recent focus of many investigators. In a study by Giacovelli et al (2008), insurance status was demonstrated to predict disease severity among a vascular surgery population of over 225,000 patients.4 Alternatively, Kelz et al (2004) reported that Medicaid and uninsured patients encountered worse outcomes following colorectal cancer resections.19 In their review of 13,415 patient records, Medicaid patients were found to incur a 22% increased risk of complications during hospital admission and a 57% increased risk of in-hospital death compared with those with private insurance. These findings are consistent with the results of our study. After adjusting for the potential confounding influence of several patient and hospital related factors, we found that Medicaid payer status conferred 97% increase in the odds of postoperative death compared with Private Insurance patients while Uninsured status independently increased the risk of in-hospital mortality by 74%. Interestingly, the adjusted odds of in-hospital death for both Medicaid and Uninsured patients were higher than that for Medicare patients after controlling for comorbid disease. We further demonstrated similar trends among the estimated odds of postoperative complications for Medicaid patients. Importantly, even after adjusting for socioeconomic status through mean income, primary payer status served as a significant independent predictor of risk-adjusted surgical outcomes.

The Medicaid and uninsured populations tend to have their surgeries through the emergency room rather than electively. The UVa authors adjusted their results for this bias, but may not have done so adequately. Either way, the fact that Medicaid patients tend to have their surgeries in the emergency room is a function of the way Medicaid constrains their access to doctors:

The demonstrated effect of primary payer status on outcomes in this study is likely multifactorial in origin. First, among all payer groups, elective operations were more commonly performed in patients with Medicare or Private Insurance while Medicaid and Uninsured patients more commonly underwent nonelective (urgent and/or emergent) operations. The higher incidence of emergent operations among Medicaid and Uninsured populations and the presumed negative effect on outcomes is in agreement with previously published surgical literature.4,20,21 However, in our analyses operative status was accounted for in the estimates of adjusted outcomes and the differences in payer groups were still significant. It is also likely that the confounding influence of inadequate preoperative resuscitation and planning that occurs in the emergent operative situation may have contributed to compromised outcomes for these populations.

Another way in which Medicaid’s poor access to doctors harms beneficiaries is that those with private insurance have better access to surgical specialists. The overwhelming evidence in the medical literature is that surgeons who specialize in doing one type of procedure, over and over again, have better clinical outcomes than surgeons who do a variety of procedures, each in low volume:

Second, it is plausible that the influence of healthcare provider and system bias may impact surgical outcomes for Medicaid and Uninsured payer groups. For many surgical patients, private insurance status often allows for referral to expert surgeons for their disease. Alternatively, Medicaid and Uninsured patients may have been referred to less skilled and less specialized surgeons. In this study, the most frequent operations performed were CABG, colectomy, and hip replacement. For these operations, the impact of surgeon volume on outcomes has been well established, and expert surgeons have been shown to significantly impact outcomes.22

Medicaid patients have higher incidence of drug and alcohol abuse, along with other diseases. It’s possible that these factors had an influence on the results, but the authors did adjust the results for all of them. Alcohol abuse was actually more frequent in the uninsured than the Medicaid population (5.8% vs. 5.0%), whereas drug abuse was slightly more frequent in the Medicaid population (3.2% vs. 3.4%). Obesity was most common in the privately insured population (10.2%, vs. 9.1% for Medicaid, 8.3% for the uninsured, and 6.2% for Medicare). And on and on. Note also that the study controlled for teaching hospital status and hospital geographic region. The point is, there were differences, but the differences were not large enough to account for the substantial difference in outcome, especially given that the study adjusted for them.

Third, differences in comorbid disease may serve as a proxy for larger social and lifestyle influences between payer groups. Both Medicaid and Uninsured payer groups had the highest incidence of drug and alcohol abuse. In addition, Medicaid patients had the highest incidence of acquired immunodeficiency syndrome, depression, liver disease, neurologic disorders, and psychoses. Furthermore, Medicaid patients had the highest incidence of metastatic cancer, which likely reflects the combined influence of deficits in access to care, poor health maintenance, and delayed diagnosis resulting in the presentation of advanced disease stage within this population. Another possible explanation for the differences we observed among payer groups is the possibility of incomplete risk adjustment due to the presence of comorbidities that are either partially or unaccounted for in our analyses. Nevertheless, multivariable logistic regression identified Medicaid and Uninsured payer status as the highest significant independent predictors of in-hospital mortality after controlling for all patients, hospital- and operation-related variables.

Here’s one of the parts of the discussion that Austin highlights:

Several explanations for inherent differences in payer populations have been suggested. Factors including decreased access to health care, language barriers, level of education, poor nutrition, and compromised health maintenance have all been suggested.2,23

Here is the rest of that paragraph, which Austin chose to omit. It begins, “However, there is no question that payer status has significant implications on multiple processes of health care delivery.” It goes on to discuss the serious harm that Medicaid does to beneficiaries’ health by limiting their access to physicians, and forcing them to go to the emergency room for care:

However, there is no question that payer status has significant implications on multiple processes of health care delivery. Differences exist in not only access but also in the type of primary care that Medicaid and Uninsured populations receive compared with Private Insurance patients. For example, studies have shown that Medicaid and Uninsured populations often receive the majority of primary care within Emergency Departments.24,25 In a recent study by White et al (2007), Uninsured patients visiting the emergency department were shown to have significantly lower number of radiographic studies and were less likely to be admitted to the hospital following  consultation as compared with private insurance patients.26 In addition, the Medicaid and Uninsured populations often present with more advanced stages of disease, a reflection of cost prohibitive health maintenance, delayed diagnosis, and the higher incidence of comorbid disease. In fact, type of insurance has been shown to impact access to cancer screening, treatment, and outcomes.27,28 Other social and lifestyle factors, including drug and alcohol abuse, psychiatric illness, obesity, and high-risk behavior, may further contribute to differences in payer group populations. The impact of the economic burden of poverty may also influence patients’ ability to seek medical care and to be discharged from the hospital in a timely manner due to lack of support and resources to be cared for properly at home.

The authors then go on to discuss the limitations of the database they used to come up with their findings. These limitations are important to acknowledge, but it’s even more important to acknowledge that these kinds of limitations are true of every retrospective study, especially those studies that use the smaller and thinner databases of Austin’s preferred authors. Despite the quite typical limitations of the UVa study, its authors point out, “upon sensitivity analyses our statistical models proved resilient to the presence of a potentially unmeasured confounder”:

There are several noteworthy limitations to this study. First, inherent selection bias is associated with any retrospective study; however, the strict methodology and randomization of the NIS database reduces the likelihood of this bias. Second, NIS is a large, administrative database, and the potential for unrecognized miscoding among diagnostic and procedure codes as well as variations in the nature of coded complications must be considered. Further, we are only able to comment on short-term outcomes as data collected for NIS reflects a patient’s inpatient admission. Consequently, the results reported herein may underestimate true perioperative mortality and morbidity rates that may have occurred following the patient’s discharge. Assumptions regarding payer groups and status may also impact data analyses. Among payer groups the potential for cross over exists, and the possibility for miscoded payer status must be considered. For example, the proportion of Medicaid patients may be artificially inflated due to the fact that normally Uninsured patients may garner Medicaid coverage during a given hospital admission. In addition, it is possible that a small proportion of Privately Insured patients may actually have inadequate insurance coverage and may functionally represent an Uninsured patient with respect to the effects of poor health maintenance and presentation with advanced disease. However, as the NIS dataset is validated both internally and externally for each year, we believe it is reasonable to assume that payer status is accurately represented in our data analyses. With respect to comorbid disease, we are unable to comment on disease stages or severity. Finally, in our data analyses and statistical adjustments there exists a potential for an unmeasured confounder. Due to the constraints of NIS data points, we are unable to include adjustments for other well-established surgical risk factors such as low preoperative albumin levels or poor nutrition status. However, upon sensitivity analyses our statistical models proved resilient to the presence of a potentially unmeasured confounder.

Reform, Instead of Equivocation

To repeat, Austin argues that the UVa study is “not credible.” There is simply no way to justify that statement.

To believe that the UVa study is “not credible,” you have to believe that unobserved factors are more important than the very real, and observed factors, that play a dispositive role in health care delivery: poor access to primary care physicians; poor access to specialist surgeons; treatment in the emergency room rather than elective surgery; poor access to screening for cancer and other diseases. And you have to remember that the study controlled for observed background comorbidities including AIDS, alcohol abuse, depression, diabetes, drug abuse, high blood pressure, liver disease, cancer, obesity, and psychosis: the things that are commonly described as differing the Medicaid population from others with comparable income.

Indeed, in certain ways, the UVa study was biased in Medicaid’s favor. The study, generously, controlled for teaching hospital status and elective operative status. These controls make Medicaid look good, because they are otherwise significantly driven by Medicaid’s poor access to high-quality providers.

In an important sense, the debate about Medicaid vs. the uninsured is a distraction. Despite Austin’s implication that I seek to “revoke Medicaid [and] replace [it] with nothing,” he knows quite well that what I and others propose is allowing the poor to obtain private insurance and control their own health spending. Such a system works impressively well in Switzerland. It is incontrovertible that Medicaid beneficiaries underperform those with private insurance. In the UVa study, Medicaid patients undergoing surgery were twice as likely to die before leaving the hospital than those with private insurance.

Anyone who professes to be empirical, rather than ideological, about the delivery of health care has a duty to take such findings seriously, and—at the very least—give states the freedom to try alternative approaches to caring for the poor. Those who favor empiricism and experimentation should join me in calling for conversion of Medicaid into a block-grant program, whereby the federal government gives the states free rein to compare market-oriented and socialized approaches to Medicaid.

PPACA advocates are curiously silent on this issue, many arguing instead that the solution to Medicaid’s woes is to preserve the program’s existing architecture, and spend more money: an intellectually indolent suggestion, and a wildly unrealistic one in the present fiscal climate.

Finally, it’s important that we cut through the statistical arcana and remember what’s really at stake here: the lives and longevity of people like Carol Vliet and Brian Curtis.

After Carol Vliet’s cancer recurred, this time with metastases, her primary physician of two years told her that he couldn’t see her any more, because he could no longer afford Medicaid’s penurious reimbursement rates. Carol was one of the lucky ones—at least she was able to find a primary care physician to begin with. Brian Curtis, a two-year-old boy, had a tougher time. “I called four or five doctors and asked if they accepted our Medicaid plan,” his mother, Rebecca, told the New York Times. “It would always be, ‘No, I’m sorry.’ It kind of makes us feel like second-class citizens.”

The cold, hard, truth is that, thanks to the indifference of intellectuals and the stubbornness of partisans, Brian and his parents are second-class citizens. The Brian Curtises of this world, who will grow up without access to primary care, won’t even find out they have cancer until it’s too late.

We have it in our power to change this, to give children like Brian a future—but only if we are willing to see what is in front of our noses, and liberate the states to give high-quality care to the truly needy. I hope we do.

Avik Roy is a health care analyst at Monness, Crespi, Hardt & Co., and writes on health care policy for Forbes at his blog, The Apothecary. You can follow him on Twitter at @aviksaroy.