Tag Archives: Medicine

The Oregon Health Insurance Experiment

Two weeks ago the first paper was released reporting results from a new important medical study, the Oregon Health Insurance Experiment. Many are comparing it favorably to the RAND Health Insurance Experiment, which I call our best medical data ever. These new results are officially here, ungated here, and commentary is here, here, and here. I’ve now had time to read the new paper, talk to one of its authors, and ponder.

This new data is possible because for a short time Oregon assigned a limited number of available Medicaid slots by lottery. 89,824 people signed up for the lottery, about 70,000 of whom were plausible eligible. 35,169 of these folks won the lottery, and of those 8,704 (~30%) were enrolled in Medicaid medical insurance.

As Oregon ended the lottery within two years, we will at most see two years worth of data. This is probably too little data to see anything but implausibly large mortality effects, but they have been collecting many direct health measures like blood pressure. In this first paper, however, all we have are the results of surveys, which include self-reported health.

The big news is that lottery winners had substantially and significantly better self-reported health. The overall health difference is significant at a 10-4 level. Lottery winners reported, for example, being healthy (= unimpaired) an average of a half day more per month. If one assumes that being a lottery winner influences health mainly via giving health insurance, then health insurance gives people 1.6 more healthy days per month.

Sounds like solid proof that medicine is healthy, right? Not so fast. First, over two thirds of the health gains that appeared on the one-year-later survey also appeared on the very first survey, done before lottery winners got additional medical treatment. So clearly at least two thirds of the health gains here are due to the comfort of knowing one has insurance. (And since they’ll only directly measure health once per person, we may never get the timing data to see if any gains in direct measures also appeared right from the start.)

Second, the folks in this study aren’t remotely comparable to the folks in the RAND experiment. The RAND experiment was mainly on random people, though it over-sampled from people with the lowest 20% of income. The Oregon experiment, in contrast, is on very sick and poor folks. For example, “healthy days per month” above refers to to how people answered a survey question on the number of days in the last 30 that poor physical or mental health impaired their usual activity. On average people in this study were impaired for ten days per month! 28% of them have asthma, 30% high blood pressure, and 56% depression.

They are also very poor, with an average yearly income of $11,790. 67% have a high school education or less, and 55% are unemployed. While only 13-17% of Americans spend less that the federal poverty income level, 70% of these folks spend below that level, and 40% of them spend less than half that level. Just how poor, sick, and just plain dysfunctional these folks are is shown by the fact that only 30% of lottery winners actually managed to get insurance. For example, “only about 60 percent of those selected sent back applications.”

But if medicine is good for the poor and sick, can’t we presume it is good for everyone? No, because the most significant overall health result found in the RAND experiment was that medicine hurt the poor and healthy! The main pre-determined health measure in the RAND experiment was a “general health index” and they reported on how getting free medical insurance influenced four different groups: people were split by income, low (= lowest 20%) vs high, and by initial health, low (= lowest 20%) vs. high. At about a 6% significance level the RAND experiment found that poor but initially-well people got sicker when given free insurance (see Table 6.12, Page 210 of Free for All?).

My interpretation: such folks went wild getting stuff checked out that they’d been ignoring, and all that extra treatment of symptoms they could have ignored for longer led to lots of false positives on tests and over-treatment, making them worse on average.

Bottom line: So far, the new Oregon Health Insurance Experiment shows that for very poor and sick folks who go out of their way to request medical insurance, giving them such insurance makes them report feeling healthier. Two-thirds of this effect appears immediately on granting their request, and before they actually got more medical treatment. It remains to be seen if these healthy feelings will be reflected in more direct health measures, though that seems plausible, and we’ll probably never see mortality effects. The main results of the RAND experiment, which looked at all sorts of people, suggests doubts about presuming that if medicine helps the very poor and sick, it on average helps everyone.

GD Star Rating
loading...
Tagged as:

New Is Not Better

“As a non-American, I don’t completely understand it, but there is a phenomenon in the U.S., the latest and the greatest. … There was a patient demand to get these implants on the misconception that the latest was the best.” …“The vast majority of the ‘innovations’ on which we have spent money with respect to orthopedics over the past two decades have not resulted in improved patient outcomes.” (more; HT Tyler)

Assuming no side-effects, if users gain from innovations then innovators must gain less than the social value of their innovations, which risks their having insufficient incentives to innovate. This effect can be countered, however, by giving extra social status to the creators and users of innovative products, services, and behaviors.

United States culture gives such extra status to creators and users of innovations, and so probably deserves some credit for encouraging innovation. But alas much of this is wasted via merely rewarding things things that are new, rather than innovative. And if your reaction to reading that was “what is the difference?,” that just shows the depth of the problem.

Innovative things must be new, but new things need not be innovative. To be usefully innovative is to be better some how. Innovators try many new things, most of which are not better, but a few of which are. On average new things are w0rse, but those that are eventually retained are hopefully on average better. And with the right incentives, the retained better things are so much better that they pay for all the other new worse things.

If our culture waited until it was clear which new things were actually better, and gave more status to the creators and early adopters of those things, culture would promote innovation. But alas culture instead mainly showers status on those who merely create and use new things, regardless of whether they are better. While in small amounts even this status effect can promote innovation, in larger amounts it can hurt. After all, when there is too little added reward for creating or using something that is both new and better, relative to something that is just new, people will mainly focus on the new part.

The problem comes from an excess focus on current behavior, relative to past track records. In enforcing social status norms, it is relatively easy to just see that someone is today affiliated with with something that is new today, and give that person credit for their newness. It is much harder to remember that a person was once affiliated with something that was then new, and which later turned out to actually be better. A mechanism that made it easier to collect and view such track records could be of great social value, at least if combined with new matching social norms on who deserves social credit for being “innovative.”

GD Star Rating
loading...
Tagged as: , , ,

First Cryonics Hour

Me two years ago:

I hereby offer to talk for one hour on any subject to anyone who can show me they’ve newly signed up for cryonics. You can record the conversation, publish it, and can sell your time to someone else.

Stuart Armstrong has signed up for cryonics, and then redeemed my offer. Congrats Stuart! We talked for an hour, and he recorded the conversation. If he does something with that recording, I’ll post a link here.

Any other takers?

GD Star Rating
loading...
Tagged as: ,

Promises Signal

Promises can both 1) help others predict and rely on our future behavior, and 2) signal our current feelings to others. The signaling function seems to dominate:

People who had the most positive relationship feelings and who were most motivated to be responsive to the partner’s needs made bigger promises than did other people but were not any better at keeping them. Instead, promisers’ self-regulation skills, such as trait conscientiousness, predicted the extent to which promises were kept or broken. … Participants who were [caused to] focused on their feelings for their partner promised more, whereas participants who generated a plan of self-regulation followed through more on their promises. …

When people make promises to address a point of contention with their partner, they seem to get swept up in what they want to do for their partner. A promise situation might appear as an opportunity to be responsive to a partner’s needs and demonstrate the loving feelings they experience for the other person, and it is these feelings they are thinking of when they make promises. … If promised behaviors can be completed immediately after promising rather than be sustained over a longer period of time, then the link between positive relationship feelings and extent of followthrough reemerges. (more)

If relationship promises mainly function to signal our current feelings, that makes it more plausible that paying and pushing for medicine for our associates serves a similar function of showing that we care. We humans apparently do relatively little checking later of whether promises were kept, or if medicine helped.

GD Star Rating
loading...
Tagged as: , ,

Let Brits Do US Med Eval

Here is my entire 300 word NYT oped:

Medicare should stop paying for treatments that the British Medical Journal says probably don’t work.

Today, United States agencies that try to not pay for ineffective treatments face the wrath of Congress, egged on by the surgeons and drug companies whose revenue is threatened. So far, U.S. agencies have pretty much always backed down, and just paid for everything.

The United Kingdom, where, on average, people live longer than in the U.S., spends only about 9 percent of gross domestic product on medicine, compared with our 18 percent. The British control costs in part by having the will to empower a hard-nosed agency, the National Institute for Health and Clinical Experience (N.I.C.E.), to study treatments and declare some ineffective. Some hope the United States will create a similar agency, but I fear it would be hopelessly politicized and declawed.

My solution: admit we are cost-control wimps, and outsource our treatment evaluation to the U.K. Pass a simple law saying Medicare (and Medicaid) won’t cover treatments considered but not positively appraised by the Britain’s national health institute.

Even better, use clinical evidence evaluations of the British Medical Journal. They’ve classified more than 3,000 treatments as either unknown effectiveness (51 percent), beneficial (11 percent), likely to be beneficial (23 percent), trade-off between benefits and harms (7 percent), unlikely to be beneficial (5 percent) and likely to be ineffective or harmful (3 percent). Let’s at least stop paying for these last two categories of treatments! And to put pressure on doctors to collect evidence, let’s stop paying for “unknown effectiveness” treatments after 10 years of use.

Yes, eventually, this evaluation source would become corrupted, as were the asset risk rating agencies that contributed to the recent financial meltdown. But we’d at least have a few more years to come up with a better solution.

Interestingly, two of the nine other opeds on “What Medicare Services to Cut, Now” wanted more hospice care.

Added 4June: Will Wilkinson:

This reminds me of another proposal, similar in spirit, to de-nationalise the drug-approval process … [by] Daniel Klein …:

One idea is to recognize the drug approvals of, say, 15 other governments. That is, we reform the U.S. system so that if the drug-approval agency of even one of those 15 countries approves a drug for that country, then the drug is automatically approved in the United States.

GD Star Rating
loading...
Tagged as: ,

Blocked State Innovation

I’ve complained that regulation usually slows innovation. For example, huge driverless cars gains seem needlessly delayed by excess regulation (Tyler agrees). The problem, however, is not government per se, but the citizens to whom government defers. Politics is not about policy; voters are far more interested in showing off symbolic stances than in giving citizens more of what they want.

But to be fair, citizens hinder not only private innovation, but also government innovation. Long ago when people were imagining a future of cheap computing and communication, they imagined dramatic gains from government databases and citizen monitoring. But then some warned of how such data and monitoring could support tyranny, and ever since most voters have been so eager to signal their disapproval of such Big Brother domination that they are unwilling to consider the most promising government innovations in data and monitoring.

For example, me a year ago:

Overall my students oppose change, moderately favoring whatever is the status quo. So I was quite surprised to see … 85% of my students said yes to: Should all medical practice data be published, aside from data identifying patients?

I assigned this paper topic again this year and, combining the two years, 76% of 76 students favored the change, which correlated 0.29 with student ability to identify relevant pro/con arguments. Again, I don’t grade students on their position, I don’t say what I support, and students usually oppose change. (For example, they overwhelmingly opposed stricter public-place policies on hand washing after sneezing or using restrooms.)

Last year, commenters’ main complaint was that it is impossible guarantee privacy. And this is true. In principle, any piece of info you publish about someone could be the last little clue someone else needs to uncover a great secret about them. It all depends on what other info people reveal, and to whom. The only safe policy is to never publish anything about anyone. And since info supposedly only visible to government employees are often leaked via bribes, the only really safe policy is to never collect any info.

But note that this same argument applies to every piece of info the government reveals about anyone, including date of birth, addresses, who/when they marry or divorce, professional licenses, lawsuits, bankruptcies, tax liens, criminal records, etc. The reason few complain about privacy leaks due to such revelations is that most folks have adapted their other info behavior to expecting this info to be made public.

Similarly, if we gave sufficient advanced warning on a new regime of revealing all med info (minus directly identifying info), most people could adapt their other info behavior to preserve the privacy they want. Don’t let friends drive you to the doc if you don’t want them to know who is your doc. Of course some would mistakenly reveal themselves, and illegal bribery would reveal more. But that can be a price worth paying if there is much to be gained.

Alas, while even my undergrads can see that revealing all med info could easily meet a cost benefit test, voter distaste for anything smacking of Big Brother will probably long block this innovation. This even though recent legal changes go a long way to actually enabling future dictators:

Our Presidents can now, on their own: order assassinations, including American citizens; operate secret military tribunals; engage in torture; enforce indefinite imprisonment without due process; order searches and seizures without proper warrants. (more)

Citizens don’t make a careful tradeoff between social value and preventing future dictators. Instead, thoughtless voters enable Big Brother while symbolically opposing him, and block useful government innovation in the process.

GD Star Rating
loading...
Tagged as: , ,

School Death Puzzle

When Ken Lee looked at 367,101 folks followed over 11 years, he found (in Table 12) that richer folks consistently died less. But for education, the trend wasn’t as consistent:

KenLeeSchoolPuzzle

Folks who graduate from high school die less than those who only start high school, those who graduate from college die even less, and those who have some grad school live even more. But, folks with no school at all do as well as college graduates! And compared with those who stop at three years of high school, those who get less school than three years of high school seem to suffer no health penalty – if anything they die less!  This fits with seeing higher mortality in states with more high school graduates, after controlling for college grads, but is odd. What gives?

Could this be a status effect, where those who didn’t buy into school as an ideal don’t mind that they didn’t get so much school?

GD Star Rating
loading...
Tagged as: , ,

Death Cause Correlates

Over the years I’ve seen many studies correlating overall death rates with other features, and also seen studies on correlates of particular causes of death, but until Ken Lee’s thesis I’d never seen how death correlates change with broad categories of death causes. Yesterday I pointed to one disturbing correlate: more med spending correlates with more cancer deaths, but not with more deaths from other causes.

That data also found injury deaths increasing more with alcohol use, which makes sense. While no population density estimates were significant, density’s most positive correlation with death was for “other” deaths, which contains most known contagious conditions. This also makes sense, as density increases contagion.

That was all from Lee’s chapter 2, where he looks at 50 states over 28 years. In chapter 3 Lee turns to a much larger data set, 367,101 adults from the National Longitudinal Mortality Study, followed over 11 years during which 9.1% of them died. Here are a few selections from Lee’s Table 14, where he breaks down deaths into cancer, heart attack, injury, and other:

KenLeeCauseCorrelates

If docs are especially bad at treating cancer, then we should expect those who use docs more to do worse at cancer. And in fact women, the rich, and the well educated do worse at cancer. Since there are many more dangerous objects in rural and poor lives, it also makes sense that such folks suffer injury deaths more.

If the main reason rural folks die less is that lower density reduces contagion, we’d expect the rural effect to be largest for “other” deaths, and that is what we find. Interestingly, that is also the kind of death which marriage best prevents – does married life prevent contagion compared with single life?

GD Star Rating
loading...
Tagged as: , , ,

Skip Cancer Screens

Ken Lee’s result that high med spending states tend to have more cancer deaths inspired me to look up the med lit on cancer screening.  I turned to Cochrane Reviews, high quality med lit reviews.  Here are the reviews I found on cancer screening:

Breast cancer:

Eight eligible trials were identified. We excluded a biased trial and included 600,000 women in the analyses. Three trials with adequate randomisation [with 260,000 women] did not show a significant reduction in breast cancer mortality at 13 years; four trials with suboptimal randomisation showed a significant reduction in breast cancer mortality with an RR [risk ratio] of 0.75 (95% CI 0.67 to 0.83). … Significantly more breast operations (mastectomies plus lumpectomies) were performed in the study groups than in the control groups: RR 1.31 (95% CI 1.22 to 1.42) for the two adequately randomised trials. … Breast cancer mortality was an unreliable outcome that was biased in favour of screening, mainly because of differential misclassification of cause of death. The trials with adequate randomisation did not find an effect of screening on cancer mortality, including breast cancer, after 10 years (RR 1.02, 95% CI 0.95 to 1.10) or on all-cause mortality after 13 years.

Colorectal cancer:

Four RCTs [randomized controlled trials] … involved 327,043 participants in Denmark, Sweden, the United States, and the United Kingdom. … Combining the four RCTs show that screening results in a statistically significant relative reduction in CRC mortality of 16% (fixed and random effects models: RR 0.84, 95% confidence interval [CI] 0.78–0.90) … Combining the four RCTs did not show any significant difference in all-cause mortality between the screening and control groups.

Prostate cancer:

Five RCTs with a total of 341,351 participants were included in this review. … The methodological quality of three of the studies was assessed as posing a high risk of bias. Our analysis of the five studies showed no statistically significant reduction in prostate cancer-specific or all-cause mortality among the whole population of men randomised to screening versus controls.

Lung cancer:

We included seven trials (six randomised controlled studies and one non-randomised controlled trial) with a total of 245,610 subjects. There were no studies with an unscreened control group. Frequent screening with chest x-rays was associated with an 11% relative increase in mortality from lung cancer compared with less frequent screening (RR 1.11, 95% CI 1.00 to 1.23).

Wow.  While cancer screening does consistently lead to more cancer detection and more cancer treatment, it consistently doesn’t lead to lower mortality.

GD Star Rating
loading...
Tagged as:

Beware Cancer Med

Chapter 2 of Ken Lee’s thesis compares med spending and age-adjusted deaths across the 50 US states from 1980 to 2007. Lee’s baseline model finds that deaths increases with smoking use, alcohol use, population density, and med spending: a 10% increase in med spending increases deaths by 0.85%. Breaking down this med spending death effect by drug vs. non-drug spending, and by four causes of death (cancer, heart attack, injury, and other), Lee finds (in Tables 5,6) that med spending hurts mainly because increasing non-drug med spending by 10% increases cancer deaths by 2.1%:

Cause of Death, Drug vs Non-Drug Med Spending

The apparent lesson: avoid cancer docs, and especially their non-drug cancer treatments. It seems some places tend to spend more on med overall, and when they spend more on cancer patients, those patients die no less, and maybe more. That fits with cancer patients living longer when they go to hospice and get no cancer treatment and with randomized trials of cancer screening consistently showing no effect on total mortality. Other explanations, however, are that high med spending places tend to classify more deaths as due to cancer, or that med treatment of all sorts tends to cause cancer.

For you stat whizzes, Lee uses state and year fixed effects, and uses per capita physicians, beds, and dental spending as med spending instruments to disentangle the direction of causation.  He picked that instrument set because it had the smallest bootstrap variance, and passed many tests. Here is Lee’s baseline model (from Table 3):

Continue reading "Beware Cancer Med" »

GD Star Rating
loading...
Tagged as: