What does insurance buy you?
A serendipitous experiment that teaches both about health and good science.
You can buy insurance, but what does it buy you?
Insurance plays a central role in accessing quality medical care in the US, so getting more people access to health insurance is a recurrent policy dream (and/or fight). For example, consider the past 10 years of Medicaid expansions painfully implemented state-by-state or the repeated legislation calling for “Medicare for All.”[1] But, ahem, all of this is expensive, and good decisions require answers to questions like, “If people gain coverage, will they use more care? Or do they change the kind of medical care they use?” For example, if the ER is currently the only non-insurance service available, gaining insurance might divert people from this expensive and last-ditch choice into primary care instead. “Would gaining insurance change the kind of providers people see? If specialist care is covered, would people ditch their more cost-effective primary care?” Or finally, dare I suggest, we should know the answer to “Does insurance even improve health outcomes!?”
Turns out these questions aren’t easy to answer. Why not? Can’t we just look around and observe people enrolled in insurance versus those left out? Write down insured people’s spending and health outcomes and report back? The crux of the problem is that people choose an insurance plan. They choose based on their health status, needs, and expectations. For example, people with chronic conditions or expectations of high health care use (say, a rambunctious little one who’s already broken an arm once…) logically search out plans with good coverage for these conditions. Conversely, healthy patients may tolerate plans with less coverage, but lower out-of-pocket expenses, since they don’t expect to use any care, anyway.
Besides health status, people with and without insurance coverage look different on measures of income, employment, race, and gender. Part of an individual’s insurance choice is tied to their socioeconomic environment. This is a big part of the drive to increase coverage in the first place! If low-income people are more likely to be uninsured (which they are[2]), but low-income individuals also face more income uncertainty or unsafe working conditions, now it becomes hard to separate their worse health outcomes from other things happening in their environment.
Suppose we want to know if access to primary care and pharmaceutical coverage helps improve diabetic outcomes. If we find that diabetes is poorly controlled in uninsured, low-income populations, can we say this is from a lack of primary care guidance? Or is it caused by food insecurity and no time to exercise while working three jobs? Are they not getting access to insulin medication? Or has their electricity been cut off, rendering their insulin ineffective because of a lack of refrigeration?[3] If the non-medical health risks are different for those in versus out of insurance, it becomes a mess to understand the role of insurance in health when we’re simultaneously trying to sort out heaps of other environmental differences.
A Serendipitous Solution
But couldn’t we do a randomized control trial? Like clinical trials in medicine? This is the gold standard for discovering truth in science, where random samples are separated into those that get treatment and those that don’t. Since getting the treatment is not related to any important characteristics of outcomes, it becomes clear (and computationally simple) to determine the impact of the treatment.
Think about pitching this for health insurance. We would need the treatment (being given insurance coverage) to be unrelated to other factors like income, education, job, or, most importantly, health status. Here’s how the pitch would go: “Let’s see…we’ll be arbitrary on who gets covered… in particular, no, being sick isn’t a valid reason for you to be covered…..and then, well, we’ll see if those people get even sicker by the end!” Most people aren’t jazzed up by this proposal.[4] Until…a few clever economists who understood these obstacles, the incomparable Amy Finkelstein[5] and Katherine Baicker, noticed an interesting opening in the world of insurance: Enter the biggest and best insurance project of the past 50 years—the Oregon Health Insurance Experiment! (Oh man, get excited!)
Oregon had been wanting to expand its Medicaid program to more low-income adults. Unfortunately, their budget wasn’t large enough for their big dreams—they could only afford to cover approximately 10,000 new enrollees out of the nearly 90,000 that might be eligible and interested. The state decided the fairest way to deal with this problem was to have interested and eligible people send in their names, and Oregon’s health officials would essentially draw names from a hat—random assignment to the program from the interested people.
But wait, did you see what they did? Oregon created a random experiment, that elusive dream for health insurance research! Finkelstein and Baicker realized this would be a perfect opportunity to partner with Oregon’s health officials to study those who “lucked out”- received the random coverage- versus those who didn’t get picked. These researchers followed both those that received insurance and those that missed out, tracking their use of healthcare services, surveying their financial and mental health, and even taking physical health measures for two years.
(Well-measured) Health Insurance Outcomes
What did we learn about health insurance? First, people indeed use more healthcare when they gain insurance coverage.[6] Outpatient visits increased by approximately 50 percent compared to those without insurance and preventive care also jumped- particularly mammograms and cholesterol monitoring. Hospitalization increased by 30 percent in the covered versus non-covered group. What about that idea of the uninsured being forced to use the ER because they lack primary care access? Yes, the insured used more primary care, but, unfortunately, they also used more ER visits! ER visits rose as much as the other medical services, by approximately 40 percent over a year and a half.
Use (and cost) of healthcare rose across the board; What about improvements in health status? Diabetes diagnoses increased in the covered group, which could lead to improved health outcomes in the future if patients learn how to manage their condition. The covered group self-reported a 25 percent increase in rating themselves in “good to excellent” health and rates of depression dropped 30 percent compared to the uninsured group.
In contrast, there were no statistically significant differences in physical measures of health. No changes in blood pressure, cholesterol, hemoglobin measures, or a 10-year aggregated measure cardiovascular risk. What!? People self-reported they felt healthier, but were they kidding themselves? Granted, two years of study may not be enough time to observe meaningful changes in physical health. However, one interesting link between these two seemingly opposite findings on health status is the study’s findings on financial hardship. Insured individuals reported that Medicaid coverage eliminated catastrophic medical expenditures and reduced the probability they needed to borrow money or skip other bills to pay for medical care by more than 50 percent. If you’ve read my work before, I stress that the most basic function of insurance should be to smooth out risk—that we hate uncertain, expensive events and would pay to simply smooth them out. Reducing financial hardship and anxiety may improve individuals’ general mental health which feeds into their self-reported health status.
Looking to the Future
The Oregon Health Insurance Experiment was a big-scale answer to “How do I get real evaluation of policy in a messy world?” You might share Finkelstein and Baicker’s desire to improve policy in your organization, but not their budget. How could an organization harness this on a small scale to understand what policies work best? The first insight is that funding shortages may be lemons, but you can make lemonade by thinking carefully about how to implement your policy. Besides advancing our understanding of insurance generally, Oregon now can forecast where cost burdens would be heaviest in any future expansion. The second insight is that the “fairness” of randomization, even for a short time, can help improve policy by allowing the true effects of a policy to be isolated from other causes. If this is socially problematic, even addressing these questions on a small scale- i.e. focus groups or a short time period- could yield big insights. People often bristle at the idea of advantages being given unrelated to our characteristics, but small-scale implementation of these kinds of experiments can help us choose the best policies to implement on a large scale. Let me know if you stumble across any in your healthy settings- here’s one researcher ready to join you!
Best,
TMD
You can read more at Oregon Health Experiment homepage—charts, tables, summaries and all kinds of fun!
If you’re interested in North Carolina’s expansion, check out this roundup from Kaiser Family Foundation.
[1] Check out my other post discussing who Medicaid covers and the role of these expansions in the ACA.
[2] Most uninsured people are in low-income families and have at least one worker in the family. Nonelderly adults are more likely to be uninsured than children because Medicare covers the elderly and the most basic Medicaid coverage always covers children. Despite gains across groups over time, racial and ethnic disparities in coverage persist. (Tolbert et al., “Key Facts About the Uninsured Population,” Kaiser Family Foundation. Dec 2023)
[3] I have spoken with medical professionals in Georgia that cited exactly this as a problem plaguing effective care in Georgia’s summer heat.
[4] Unless you’re a health economist- heh heh!
[5] Researcher to know: For her and her colleagues’ work on the Oregon Health Insurance Experiment, plus several other papers on insurance and public interventions, Amy Finkelstein was awarded the “Pre-Nobel of Economics,” the Clark Medal.
[6] Learn more at the Oregon Health Insurance Experiment reporting site.
I always struggle with equating Medicaid with insurance. At least in cancer, Medicaid does provide the safety net for catastrophic costs, but it does not cover important services like lymphedema therapy (more than 3 visits a year) and breast reconstruction after a mastectomy. What would really be interesting is if people were randomized to Medicaid versus a tricked out commercial plan.