Tuesday, March 27, 2018

Myth and Measurement — The Case of Medical Bankruptcies

Myth and Measurement — The Case of Medical Bankruptcies. Carlos Dobkin et al. N Engl J Med 2018; 378:1076-1078, 10.1056/NEJMp1716604. March 22, 2018

Though there is compelling evidence that medical expenses do cause bankruptcies in the United States, they may cause far fewer than has been claimed. Overemphasizing such events may distract from an understanding of the true nature of health-related economic hardship.

During the push to pass the Affordable Care Act, President Barack Obama often described the “crushing cost of health care” that was causing millions of Americans to “live every day just one accident or illness away from bankruptcy” and repeatedly stated that the high cost of health care “causes a bankruptcy in America every 30 seconds.” Stories of illnesses and injuries with financial consequences so severe that they caused households to file for bankruptcy were used as a major argument in support of the 2010 Affordable Care Act. And in 2014, Senators Elizabeth Warren (D-MA) and Sheldon Whitehouse (D-RI) cited medical bills as “the leading cause of personal bankruptcy” when introducing the Medical Bankruptcy Fairness Act, which would have made the bankruptcy process more forgiving for “medically distressed debtors.” But it turns out that the existing evidence for “medical bankruptcies” suffers from a basic statistical fallacy; when we eliminated this problem, we found compelling evidence of the existence of medical bankruptcies but discovered that medical expenses cause many fewer bankruptcies than has been claimed.

Policymakers’ beliefs about the frequency of medical bankruptcies are based primarily on two highprofile articles that claim that medical events cause approximately 60% of all bankruptcies in the United States.1,2 In these studies, people who had gone bankrupt were asked whether they’d experienced health-related financial stress such as substantial medical bills or income loss due to illness. People were also asked whether they went bankrupt because of medical bills. People who reported any of these events were described as having experienced a medical bankruptcy. This approach assumes that whenever a person who reports having substantial medical bills experiences a bankruptcy, the bankruptcy was caused by the medical debt. The fact that, according to a 2014 report from the Consumer Financial Protection Bureau, about 20% of Americans have substantial medical debt, yet in a given year less than 1% of Americans file for personal bankruptcy, suggests that this assumption is problematic.  Clearly, many people face medical debt but do not go bankrupt.  Even after correction for overly broad definitions of “medical” expenses, 3 the existing, widely cited evidence on medical bankruptcy is built on the fallacy that when two things occur together there is necessarily a causal relationship between them.

To understand the problem, consider an analogous line of inquiry: suppose we want to know which factors increase a person’s chances of becoming a technology billionaire. Investigation of recent technology giants might suggest that dropping out of college is a high-return strategy (think: Bill Gates, Steve Jobs, and Mark Zuckerberg [dropping out of Harvard seems to have a particularly high payoff]). By examining only college dropouts who have already became technology billionaires rather than all college dropouts, this analysis misses the fact that most college dropouts do not go on to lucrative careers in the tech business. A similar problem pervades the current literature on medical bankruptcy. The studies mentioned above examine the experiences only of people who went bankrupt, but it is impossible to infer the role of medical expenses in causing bankruptcy without information on the proportion of the population with large medical expenses that did not go bankrupt.

To estimate the share of bankruptcies actually caused by medical factors, we therefore selected a sample of people who were admitted to the hospital in California and tracked information on their annual credit reports, including whether and when they filed for bankruptcy. Because we examined the relationship between when people go to the hospital and the timing of any bankruptcy, we were able to estimate the increase in bankruptcy filings caused by illness or injury, rather than the fraction of people filing for bankruptcy who happen to have substantial medical expenses.  Our study was based on a random stratified sample of adults 25 to 64 years of age who, between 2003 and 2007, were admitted to the hospital (for a non–pregnancyrelated stay) for the first time in at least 3 years. We linked more than half a million such people to their detailed credit-report records for each year from the period 2002–2011. The graph shows the results of our analysis.  The results show a clear effect of hospital admission on bankruptcy: the rate of bankruptcies rises sharply in the years after hospital admission, and this change is statistically significant (at conventional levels) both 1 and 4 years after the admission, after which bankruptcies appear to level off. This finding indicates that the expenses that result from the illness or injury that caused the hospital admission — for example, out-of-pocket medical costs and lost labor income — cause some people to file for bankruptcy.  However, the magnitude of the bankruptcy effect is much smaller than previously thought: we estimate that hospitalizations cause only 4% of personal bankruptcies among nonelderly U.S. adults, which is an order of magnitude smaller than the previous estimates described above.

We calculated this estimate as follows: the graph shows that, on average, a hospitalization increases the annual probability of going bankrupt in the following 4 years by 0.004. Multiplying this figure by the annual hospitalization rate of 7.8% for our population (which we calculated using the 1999–2010 Medical Expenditure Panel Survey) reveals that 0.031% (0.004 × 0.078) of the population goes bankrupt each year as the result of a hospitalization.  Given that the annual household bankruptcy rate is 0.8% among the nonelderly,4 hospitalizations cause about 4% (0.031 ÷ 0.8) of bankruptcies among nonelderly adults. A similar calculation for a subsample of uninsured adults reveals that even in that population, hospitalizations are responsible for only 6% of bankruptcies4; for this population, the effect of a hospitalization on the likelihood of bankruptcy is (not surprisingly) larger, but the hospitalization rate is lower than in the overall nonelderly population.

Of course, these results do not cover all potential medical bankruptcies.  They do not consider hospitalizations for children or for the elderly — although in other work we found that hospitalizations have no effect on bankruptcy rates among the elderly.4 Our results are also specific to our population — people in California hospitalized for non–childbirth- related conditions who have not had a hospital admission in the previous 3 years (although they may, and often do, have additional admissions over the subsequent years). However, as we have described elsewhere, recent related research using different sample populations (but also using large administrative data sets and a similar research design) also revealed a limited effect of health shocks on bankruptcy rates.4 Perhaps most obviously, our analysis excludes illness and injuries that do not result in a hospital admission. However, our sample of hospitalized people is likely to include most people with large medical expenses: in the Medical Expenditure Panel Survey, we estimated that about 63% of people in the top 5% of annual medical spending (at least $8,433) had had a hospitalization in that year. This finding suggests that focusing on hospitalized people probably does not lead to vast underestimation of the effect of all illness and injury on bankruptcy rates.

The Effect of Hospitalization on the Likelihood of Filing for Bankruptcy.

The x axis shows time relative to the index hospital admission. Each data point represents the proportion of people who filed for personal bankruptcy between the year before the start of our credit-report data and the indicated date, after adjustment for any patterns in bankruptcy rates by calendar year. The dashed line shows the estimates from fitting a flexible, nonlinear function quantifying the relationship between the timing of hospital admission and the bankruptcy rate, again with adjustment for calendar-year trends. (More detail on the sample and estimators can be found in Dobkin et al.4)

Our results also do not speak to the financial costs of hospital admissions outside the bankruptcy- filing decision. We have found that hospitalizations cause increased out-of-pocket spending on medical care, increased medical debt, and decreased employment and income.4 These costs may have considerable adverse consequences, and evidence from the Oregon Health Insurance Experiment indicates that they can be partially ameliorated by health insurance.5 But our findings suggest that medical factors play a much smaller role in causing U.S.  bankruptcies than has previously been claimed. Overemphasizing “medical bankruptcies” may distract from an understanding of the true nature of economic hardship arising from high-cost health problems.

1. Himmelstein DU, Warren E, Thorne D, Woolhandler S. Illness and injury as contributors to bankruptcy. Health Aff (Millwood) 2005; 24: Suppl Web Exclusives: W5- 63–W5-73.
2. Himmelstein DU, Thorne D, Warren E, Woolhandler S. Medical bankruptcy in the United States, 2007: results of a national study. Am J Med 2009; 122: 741-6.
3. Dranove D, Millenson ML. Medical bankruptcy: myth versus fact. Health Aff (Millwood) 2006; 25(2): w74-w83.
4. Dobkin C, Finkelstein A, Kluender R, Notowidigdo MJ. The economic consequences of hospital admissions. Am Econ Rev 2018; 108: 308-52.
5. Finkelstein A, Taubman S, Wright B, et al. The Oregon Health Insurance Experiment: evidence from the first year. Q J Econ 2012; 127: 1057-106.  DOI: 10.1056/NEJMp1716604

With a few hundred twins, authors examined genetic & environmental influence on all pairwise functional connections between 264 brain regions (~35,000 functional connections) & found high non-shared environmental influence across the entire connectome, moderate heritability in roughly half of all connections, & weak-to-moderate shared environmental influences

Genetic and environmental influence on the human functional connectome. Andrew E. Reineberg, Alexander S. Hatoum, John K. Hewitt, Marie T. Banich, Naomi P. Friedman. bioRxiv, https://doi.org/10.1101/277996

Abstract: Detailed mapping of genetic and environmental influences on the functional connectome is a crucial step toward developing intermediate phenotypes between genes and clinical diagnoses or cognitive abilities. Historical attempts to estimate the genetic etiology of the connectome have focused on large-scale brain networks - obscuring possible heterogeneity among or novel communities of small network subcomponents. In the current study, we analyze resting-state data from two, adult twin samples - 198 twins from the Colorado Longitudinal Twin Sample and 422 twins from the Human Connectome Project - to examine genetic and environmental influence on all pairwise functional connections between 264 brain regions (~35,000 functional connections). We find high non-shared environmental influence across the entire connectome, moderate heritability in roughly half of all connections, and weak-to-moderate shared environmental influences. The pattern of genetic influence across the connectome is related to a priori notions of functional brain networks but also highly heterogeneous as confirmed by a hierarchical clustering analysis of the genetic profile of all 264 regions. Additionally, we confirm genetic influences on connections are independent of genetic influences shared with a global summary measure of brain connectivity - an important validation analysis for future, high-dimensionality genetic neuroimaging studies. Together, our analyses reveal a novel genetic taxonomy of brain regions and have implications for studies employing multivariate signals for prediction purposes. Variation across the population in those neurobiological signals is influenced by genes and the environment in different spatial locations and to different degrees suggesting genetic risk factors may be limited to a subset of the connectome.

Behavioral evidence suggestive of frontal lobe pathology in the amnesic H.M.

Behavioral evidence suggestive of frontal lobe pathology in the amnesic H.M. William Winter. Brain and Cognition, Volume 123, June 2018, Pages 136-141, https://doi.org/10.1016/j.bandc.2018.03.005

•    H.M. displayed behaviors that cannot be explained by temporal lobe amnesia.
•    Such behaviors have previously been attributed to amygdalar damage.
•    Frontal lobe lesions were detected by MRI and in post-mortem examination.
•    These behaviors are most parsimoniously explained as due to frontal lobe pathology.

Abstract: From the earliest published reports, Henry Gustav Molaison—who until his death in 2008 was known simply by his initials H.M.—was characterized as having a profound anterograde amnesia subsequent to mid temporal lobe resection, and that this amnestic condition was uncomplicated by other cognitive or behavioral impairments. Post-mortem neuropathological examination has detected—in addition to the expected temporal lobe lesions—previously unreported frontal lobe and white matter pathology, inviting questions concerning the behavioral and cognitive consequences that might result from such lesions. The purpose of this article is to recount published descriptions of a range of anomalous behaviors by H.M. that can not be explained by the memory impairments typically associated with anterograde amnesia, to counter previous claims that these behaviors are attributable to amygdalar damage, and to advance the interpretation that these behaviors are instead consistent with well-documented effects of frontal lobe pathology. Transcripts of interviews with H.M. which feature disjointed, often contradictory, and arguably confabulatory responses are presented in support of this argument.