TL;DR
“One broken leg and you lose everything” is the most repeated claim about American healthcare, and it rests on a number that has been seriously contested. Himmelstein and colleagues surveyed 2,314 bankruptcy filers in 2007 and concluded that, on a conservative definition, 62.1% of all US bankruptcies were medical.1 In 2018 Dobkin, Finkelstein, Kluender and Notowidigdo used an event-study design on hospital admissions linked to credit reports and concluded that hospitalisations cause about 4% of personal bankruptcies among non-elderly adults — “an order of magnitude smaller than the previous estimates.”2 Their critique is precise: the earlier work “is built on the fallacy that when two things occur together there is necessarily a causal relationship between them.”2 But the same paper contains the finding that matters more than the argument. After a hospital admission the earnings decline is large relative to the out-of-pocket spending increase, and only about 10% of that earnings loss is insured in the US — against almost 50% in Denmark, where the earnings hit itself is comparable.3 The catastrophe is real. It is mostly not the bill.
Paper on a table. The number that gets quoted and the number that got measured are not the same number. Photo: 2H Media on Unsplash.4
The Number Everybody Quotes
The claim entered public argument through two studies by David Himmelstein, Elizabeth Warren, Deborah Thorne and Steffie Woolhandler.
The first, in Health Affairs in 2005, surveyed 1,771 personal bankruptcy filers across five federal courts and completed in-depth interviews with 931 of them. About half cited medical causes. Among those whose illnesses led to bankruptcy, out-of-pocket costs averaged $11,854 since the start of illness — and 75.7% had insurance at the onset of illness.5 That last figure is what made the study land: this was not a story about the uninsured.
The second, in the American Journal of Medicine in 2009, scaled up. A random national sample of 2,314 filers in 2007, court records abstracted, 1,032 interviewed. Bankruptcies were designated “medical” based on debtors’ stated reasons for filing, income loss due to illness, and the size of their medical debts. The headline: 62.1%, up from at least 46.2% in the earlier five-state work. Of those medical debtors, 92% had medical debts over $5,000 or exceeding 10% of pre-tax family income.1
That 62% has been quoted in political speeches, documentaries and countless articles ever since.
The Number That Got Measured
In 2018 four economists published a paper in the New England Journal of Medicine with the title “Myth and Measurement.” Their design is the important part.
Instead of asking bankrupt people why they went bankrupt, they took hospital admissions and followed what happened afterwards — an event study on two datasets, the Health and Retirement Study and hospitalisation records linked to credit reports.3 That lets you compare the same people before and after a health shock, rather than asking people already in bankruptcy to narrate their own causation.
The result: a hospitalisation raises the annual probability of bankruptcy over the following four years by 0.004. Multiplied by the 7.8% annual hospitalisation rate for that population, hospital admissions account for roughly 4% of personal bankruptcies among non-elderly adults.2 In the underlying American Economic Review paper the split is about 4% for the insured non-elderly and about 6% for the uninsured, with no economically or statistically significant increase for those over 65.3
Why Both Numbers Are Honest
This is not a case of one team being wrong. They are measuring different things, and the difference has a name.
Himmelstein’s question is: among people who went bankrupt, how many had illness or medical debt involved? That is a question about the composition of a group.
Dobkin’s question is: if you take a person and give them a hospital admission, how much does their probability of bankruptcy rise? That is a question about causation.
The economists’ critique is that the first answer has been read as if it were the second. Their analogy is the clearest statement of the problem I have seen: if you want to know which factors increase your chances of becoming a technology billionaire, studying the recent giants would suggest that dropping out of college is a high-return strategy.2 Gates, Jobs and Zuckerberg all did it. Almost everyone who drops out does not become a billionaire.
Applied here: many people carry medical debt and do not go bankrupt. Being in the bankrupt group with medical debt does not establish that the medical debt put you there — especially when bankruptcy tends to follow a pile-up of several simultaneous shocks, and illness is one of the more memorable ones to name when a researcher asks.
The Finding That Should Have Been the Headline
Here is what makes the 2018 paper more interesting than its own debunking.
The authors report that after a hospital admission, the earnings decline is substantial compared with the out-of-pocket spending increase, and that it is “minimally insured” before age-eligibility for Social Security Retirement Income.3 For older workers in their sample, only about 10% of the earnings decline is insured through social insurance.3
Then the comparison that reframes everything. In Denmark, non-fatal health shocks to households under 60 produce comparable earnings declines — 15 to 20% — but almost 50% of that decline is insured, largely through sick pay and disability insurance.3
Read those two sentences together. Getting seriously ill costs you roughly the same amount of foregone income in Copenhagen as in Cleveland. What differs is whether anyone replaces it.
So the popular story — American healthcare is uniquely ruinous because the bills are enormous — is aiming at the wrong variable. The bills are enormous. But the mechanism doing most of the financial damage is that you stop earning, and almost nobody covers that.
There is a second, subtler finding in the same paper. A hospital admission raises unpaid medical bills by about $6,000 for the uninsured against about $300 for the insured, and the decline in credit limits after an admission is over half the decline that follows an unemployment spell.3 The insured are not protected from the event; they are protected from the invoice.
What This Changes
For anyone arguing about American healthcare: stop quoting 62% as a causal figure. It is not one, and the economists who said so published it in the New England Journal of Medicine. Using it invites the strongest possible rebuttal to the strongest possible case.
For anyone arguing the opposite: 4% is not reassurance. It is 4% of all personal bankruptcies caused by hospital admissions alone — not all illness, not chronic conditions, not the care people forgo before it gets to an admission. And the same paper documents large, largely uninsured earnings losses. The catastrophe survived the correction; it just moved.
For policy: if the damage is mostly foregone wages, then expanding coverage of bills addresses the smaller share of the problem. Sick pay, disability insurance and job protection are the instruments the Danish comparison points at, and they are almost absent from the American debate, which is conducted almost entirely about insurance premiums and deductibles.
This is the same shape as an argument this site has made about structural versus behavioural health policy: the intervention everyone argues about is not the one with the most leverage. And it is the same measurement error as confusing what a group looks like with what caused it — composition read as causation.
Where I’d Hold This Loosely
Five limits, and the third is the one I’d most want kept.
The 4% figure covers hospital admissions only. Chronic illness managed without admission, mental health crises, long-term disability, and care deferred until it becomes catastrophic are all outside the estimate. It is a lower bound on medical contribution to bankruptcy, not a total, and the authors are measuring what their data can see.
Second, both literatures are pre-pandemic and pre-ACA-maturity in different ways. Himmelstein’s data is 2001 and 2007; Dobkin’s runs through the early 2010s. Coverage expansion, high-deductible plan growth and the 2005 bankruptcy law changes all move these numbers, in directions that are not all the same.
Third, and against the tidy story I have told: bankruptcy is a bad proxy for financial ruin. Filing requires money, legal access and assets worth protecting. Many people destroyed by medical costs never file — they take on informal debt, skip care, sell things, lean on family, or simply live with unpayable bills. A measure that counts court filings will systematically miss the poorest victims, which means both estimates understate harm in the same direction.
Fourth, the Danish comparison comes to me through Dobkin et al.’s summary of Fadlon and Nielsen rather than from that paper directly, and the populations, definitions of health shock and time windows are not identical. I have used it as an order-of-magnitude contrast, which is what it can bear, and not as a precise 5:1 ratio.
Fifth, an event study on hospital admissions has its own assumption: that the timing of admission is not itself driven by the financial trajectory. Someone sliding toward bankruptcy may delay a hospital visit, which would push the estimated effect around. The design is much stronger than a survey of filers, and it is not an experiment.
