What this covers

Adam Cifu of the University of Chicago discusses with Russ Roberts why widely adopted medical treatments often turn out to be ineffective or harmful once subjected to rigorous testing. Cifu defines medical reversal as the discovery that therapies adopted by millions of patients—based on plausible biological reasoning and weak evidence—fail when tested against better data. He argues the pattern is distressingly common: review of a decade of articles in The New England Journal of Medicine identified roughly 100 reversed therapies, suggesting 35–40% of current medical practice may ultimately prove wrong. The core problem is that medicine repeatedly moves from biological plausibility to widespread adoption before randomized control trials confirm actual benefit, and inertia—psychological, financial, and institutional—keeps reversed procedures in use long after evidence turns against them.

The conversation covers several distinct sources of error in medical decision-making. Observational study results mislead far more often than recognized because researchers cannot fully control for confounding, and surrogate endpoints like blood pressure or cholesterol correlate poorly with outcomes patients actually experience. Cifu illustrates with coronary stents for stable angina—adopted for 80% of Medicare cases before the 2009 COURAGE trial showed they prevented no more heart attacks or deaths than medical therapy alone. He examines prostate cancer screening, where roughly 30–35 men must be treated to save one life, with one in three suffering erectile dysfunction and one in three incontinence—patients far more likely to experience harm than benefit. The discussion also addresses why doctors resist abandoning practices they trust, the design superiority of sham-controlled trials, and how to navigate genuine uncertainty: Cifu urges patients to ask whether treatments actually work and insists that where evidence is genuinely unclear, the burden of that decision should not fall on patients alone.

Sharpest takeaway

Cifu argues that 'medical reversal'—the discovery that widely adopted treatments are ineffective or harmful—is distressingly common because medicine repeatedly adopts therapies based on plausible biological mechanisms and observational data before rigorous randomized trials confirm they help, and the remedy is greater humility, evidence-based practice, and informed patient decision-making.

  • An estimated 35-40% of current medical practices may be wrong, based on review of NEJM articles identifying ~100 reversed therapies over 10 years.
  • Observational studies and surrogate endpoints frequently mislead because of confounding and incomplete understanding of mechanism, whereas randomized controlled trials remove confounding.
  • Psychological, financial, and educational incentives make doctors resist abandoning practices they believe in, so reversed procedures often persist.

The argument · threads8 threads · 36 claims
0.72

Widely adopted therapies often fail because improving surrogate endpoints does not guarantee improved clinical outcomes patients actually experience.

4 pointscentrality 5/5
  • Atenolol (marketed as Tenormin) lowers blood pressure as well as most other antihypertensives, but pooled placebo-controlled studies show it does not improve mortality or heart attack risk and only slightly reduces stroke risk, illustrating that a drug can succeed on the surrogate endpoint yet fail on clinical endpoints.

    “Atenolol does do a really wonderful job of lowering blood pressure... But when you brought together all the studies in which it was compared to a placebo, it turns out that it doesn't improve mortality. Doesn't improve the risk of heart attacks.”

  • Surrogate endpoints (blood pressure, cholesterol, blood sugar) are cheap measurable stand-ins for clinical endpoints (how you feel, living longer, avoiding stroke), and therapies are often accepted because they improve the surrogate; but improving a surrogate does not guarantee improving the outcomes patients actually care about.

    “Surrogate endpoints are stand-ins for those... So it's a lot easier to pick a surrogate marker as stand-in, like blood pressure... And we'll accept this therapy, because we know it affects the surrogate marker in a good way.”

  • Estrogen replacement therapy was widely prescribed to postmenopausal women based on observational data (the Nurses' Health Study) and a plausible biochemical rationale, but a good randomized controlled trial later showed it does not reduce cardiovascular events and may actually increase event rates in the first couple of years.

    “when a really good sort of experimental randomized control trial came out, we figured out that, 'Huh. You know, estrogen replacement therapy really doesn't help reduce the risk of cardiovascular events.' And for the first couple of years you use it, it may actually increase the rate of events.”

  • Rather than blocking access, the FDA can reasonably approve a needed drug on surrogate-endpoint evidence, but it should require that studies of real clinical endpoints be ongoing at the time of approval; the problem is these confirmatory studies often never get done, leaving us using therapies that might not work.

    “it really seems necessary that there should be studies ongoing at the time that the FDA approves that drug that will show us those real endpoints... What often happens is that these drugs are approved based on surrogate endpoints. And then we never get that final data”

0.72

Medical reversal occurs frequently: rigorous evidence later overturns therapies adopted on weak evidence, harming future patients and eroding trust.

4 pointscentrality 5/5
  • Medical reversal occurs when a new therapy replaces an old one based on weak evidence, is adopted by thousands or millions of patients, and is then found—once more robust data emerges—to be no better than, or worse than, the old therapy or doing nothing.

    “Reversal is when a new therapy comes along that replaces the old therapy. But usually that new therapy is not based on really foolproof evidence. But we only find that out after the new therapy has been adopted--used by hundreds, thousands, maybe millions of patients.”

  • Reviewing all articles in The New England Journal of Medicine over a 10-year span identified 146 articles covering about 100 widely adopted therapies that turned out to be wrong; combined with other research this suggests as much as 35-40% of what medicine does could be wrong.

    “over those 10 years, we identified 146 articles, which concerned about 100 different therapies... Our estimate, sort of taking other people's research with our research, is that maybe as much as 35%, 40% of what we do could be wrong.”

  • Coronary stents are clearly life-saving for heart attacks and unstable angina, but their adoption for stable angina—where they consumed 80% of Medicare coronary stent dollars by 2009—was overturned by the 2009 COURAGE trial, which showed stents were no better than medical therapy at preventing heart attacks or deaths in those patients.

    “this very famous trial, the COURAGE Trial... came out in 2009. Which, if you are looking at things like preventing heart attacks, preventing deaths, stents were no better than just the medical therapy we were using at the time.”

  • A rigorous trial randomized spinal-compression-fracture patients to real vertebroplasty (injecting medical cement) or a sham procedure (full prep, opening cement so patients could smell it, but injecting only saline), and found no difference in pain, quality of life, or activity scores over the first month—showing the procedure's apparent benefit was a placebo effect.

    “it turned out that over the first month after the procedure, all the endpoints were exactly the same between the sham group and the intervention group. No difference in pain. No difference in quality of life. No difference in activity scores. Nothing.”

0.68

Observational studies confound confounders; randomized trials eliminate bias by ensuring groups differ only in the intervention.

3 pointscentrality 5/5
  • Observational studies are natural experiments where groups self-select or are selected into treatments, so the groups differ in many ways besides the intervention; in the estrogen case, women who took estrogen were younger, thinner, had better cholesterol and drank more, so their better outcomes likely reflected those traits rather than the estrogen itself.

    “Women who took estrogen were younger, thinner, had actually better cholesterol levels... So it's a very different population. And so it was probably not the estrogen replacement therapy which was benefitting them, but everything else about them”

  • Controlling for confounders with statistics has a place, but you never know if you've fully adjusted for them; even sophisticated researchers controlling for a dozen-plus variables in the Nurses' Health Study couldn't capture how different the groups were, so observational results agree with RCTs only about 80% of the time and you can't tell when they're misleading.

    “those studies agreed most of the time--think about 80% is what I recall. Which is good. But it's not perfect. And you don't know when those observational studies are steering you in the right direction and when they are steering you in the wrong direction.”

  • In a randomized controlled trial, patients are randomly assigned to treatment or placebo so the groups are on average identical in all risk factors—both those you know and those you don't know yet—meaning any difference in outcome at the end is due to the intervention itself.

    “those groups are exactly the same on average in all the risk factors that you know of. But also all the risk factors that you don't know of but that we might know of in 10 years. And so, in a really well-done randomized control trial... is due to your intervention”

0.57

Screening and prevention require higher evidence thresholds than treatment because tests are imperfect and harms often outweigh benefits in healthy people.

7 pointscentrality 3/5
  • When a procedure can harm patients and turns out to be no better than sham, being randomized to the placebo group is actually beneficial—an insurance policy—and conducting the sham trial spares future patients from undergoing a useless or harmful procedure.

    “the placebo group--in a way is an insurance policy... in fact, in the vertebroplasty case, you know, those people saved--I don't know--thousands, maybe millions of people in the future from getting a procedure which is not helpful.”

  • Even when reversed treatments don't directly harm patients, they impose costs, create opportunity costs (patients miss effective alternatives), and erode patients' faith in medicine, making them slower to accept future recommendations.

    “once you've spent a year on a medication which your doctor then tells you, you should come off, because it's not doing anything: You are a little bit more slow on the uptake for future recommendations”

  • Recent data from prostate cancer screening indicates that to save one life from prostate cancer about 30-35 people must be treated, with roughly one in three suffering erectile dysfunction and one in three incontinence, so patients are far more likely to experience side effects than benefit.

    “to save a life from prostate cancer, we actually need to treat about 30, 35 people for prostate cancer. That's a lot of people being treated just to save one life.”

  • Screening is intuitively appealing ('catch it early'), but because tests are imperfect and even 'common' diseases are statistically rare, screening generates many false positives that cause anxiety and lead to unnecessary procedures and treatments.

    “even with a pretty good test, when you are looking for a rare disease you are going to come up with a lot of false positives. And those false positives cause anxiety among the patients... They probably also lead to procedures that don't need to be done.”

  • Where evidence is genuinely uncertain—such as prostate screening or mammograms between 40 and 50—clinical guidelines say to 'discuss with the patient,' which effectively puts the impossible decision on the patient precisely in the cases where doctors themselves do not know what to recommend.

    “whenever there is something like you talked about--prostate screening or mammograms between 40 and 50, the guidelines are: Well, we should definitely discuss this with your patient... those are the places where we really don't know what to recommend. So we put it on the patient to make the impossible decision.”

  • The required strength of evidence should depend on who is being treated: for a healthy person receiving screening or prevention, you must be nearly certain it helps because you risk turning a healthy person into a patient; for a sick person in pain, the bar is lower and the decision should rest on an open discussion of likelihood of benefit and quality of evidence.

    “if you are talking about a healthy person and you are talking about a screening intervention or preventative therapy... you are basically turning them into a patient and potentially making them sick with your intervention. With someone who is sick, who is in pain--well, then I think the bar is actually a little bit lower.”

  • For decisions where no patient has a predetermined reason to prefer one equally-valid therapy over another (e.g., two antibiotics for sinusitis), patients could be automatically randomized unless they opt out, applying the Nudge principle to generate large amounts of trial data cheaply and quickly.

    “We are big fans of the Nudge Principle... why not just have that person randomized? Unless they opt out. And we could get lots of data quickly in that way.”

0.54

Doctors' genuine belief in treatments combined with financial incentives makes them resistant to evidence that practices do not work.

3 pointscentrality 4/5
  • Because the NIH cannot fund everything and much drug and device research is industry-funded, super-specialists who become experts often have industry affiliations—as trial consultants, trial leaders, or paid/unpaid spokespeople—and a financial stake makes it harder for their recommendations to remain unbiased.

    “very often these super-specialists have some sort of affiliation with industry... it probably gets harder and harder to get completely unbiased in your recommendations and your thoughts about these, if you begin to have a financial stake in a medication.”

  • Doctors mostly resist reversals because they genuinely believe a practice works and have seen patients improve, but financial incentives compound this: if a procedure has put your kids through college, you are more apt to argue against evidence that it doesn't work.

    “if you've made a lot of money over the years doing a procedure on people's knees, which you believe it works but it's also helping you, and your kids through college, when you find out that that doesn't work, you're probably a little bit more apt to argue with it.”

  • Care from generalists tends to rest on solid evidence because common diseases are well studied, but as patients see more specialized physicians for rarer problems, the therapies are less well studied, patients are more willing to accept unproven treatments out of need, and specialists—often involved with the companies developing drugs and devices—are more willing to adopt therapies because they feel expert; this makes super-specialists' therapies among those most prone to reversal.

    “those are people who, probably--the people you want to see with some of the problems, but maybe ones whose therapies might be most prone to reversal.”

0.54

Biochemical rationale and mechanistic understanding are unreliable guides; empirical data frequently contradicts what theory predicts will work.

3 pointscentrality 4/5
  • Even when we are sure a treatment will work because we think we understand the biology and mechanism, real empirical data frequently shows it doesn't; the human body is so complex that we should rely on empiricism over biochemical rationale when making treatment decisions.

    “when everything lines up to say, this intervention should work... and then real empirical data shows that it doesn't work--you know, it's shocking... we need to go with empiricism more than the biochemical rationale that underlies some of these decisions.”

  • Because reversal partly stems from learning biochemistry and pathophysiology first—which convinces students everything works mechanistically as theorized—medical education should be flipped to teach evidence-based medicine first, presenting students with patients and cases and asking them to research the best-supported treatment, then teaching the underlying science later as a lens viewed through clinical experience.

    “Could doctors be trained in evidence-based medicine at the beginning? And be presented with patients and cases, and then be asked to research: What is the best treatment here?... maybe that should come later. And so you'd be seeing that sort of science through the window of what you've already learned about taking care of patients.”

  • Personalized medicine using genetics holds enormous promise but increases the temptation to assume a drug works because of how it acts on a gene; because therapies will generalize to smaller and smaller groups, each must be tested on those small groups, making randomized trials and evidence-based medicine even more important rather than less.

    “personalized medicine may make randomized control trials and evidence-based medicine even more important because we need to test each of these personalized medicine interventions on a smaller and small group of people, since our therapies will generalize to smaller and smaller groups.”

0.45

Trial design flaws like early stopping and the FDA's tolerance for unconfirmed surrogates allow ineffective therapies to remain in use.

4 pointscentrality 3/5
  • Cifu and Prasad labeled something a reversal only when the study overturning the practice was clearly more robust than the studies that had originally supported it—not merely a single contradicting study, which statistics guarantees will sometimes occur by chance.

    “we only labeled something a reversal if the study that overturned the practice was clearly a better study than what had supported the practice in the past.”

  • Trials stopped early because an effect looks dramatic tend to overestimate the benefit of a treatment, because across many studies outcomes vary by chance and stopping at a favorable moment captures the high-variance positive result rather than the true effect.

    “studies that are stopped early tend to overestimate the benefit of a treatment... maybe there are other trials which will come out which will be negative; and maybe this doesn't in fact work.”

  • Physicians rely on the FDA as a kind of insurance company because when a drug fails or causes harm, the FDA—not the physician—takes the heat, so the FDA faces pressure and potential influence from multiple directions.

    “Physicians really rely on the FDA, because the FDA is in a way our insurance company. You know--the FDA takes the heat when a drug doesn't work or causes harm. Not the physician.”

  • Rapid response teams are widely adopted and intuitively appealing, but the data supporting them are generally single-center studies, and to this point there is no clear evidence that they save lives.

    “The data that says they work are generally single-center studies... to this point, we haven't seen that rapid response teams save lives.”

0.40

Patients should ask doctors hard questions, demand evidence, and find generalists with sound judgment rather than specialists with financial stakes.

3 pointscentrality 3/5
  • Because specialists view care through the lens of their own discipline and a determined patient can always find a doctor who will agree to do what they want, patients should find a trusted generalist who has their best interest at heart and will steer them to the right specialists when needed.

    “as a patient, you can always find someone who will look at your health care in the way that you do. So, if you want something done and the first few doctors you see say it's not the right thing to do, you can always find someone who will. So you need to find a doctor you trust”

  • Patients should feel comfortable asking their doctors the real questions—not 'what are the side effects' or 'will insurance pay' but 'Does this work? How do you know it works? How likely is it to benefit me? What are the other options?'—and even ask the doctor to argue the opposite side, overcoming the power differential that makes otherwise confident people clam up in the exam room.

    “those real questions are, to begin with are not: 'What are the side-effects of this medication? Will my insurance pay for this medication?' It's: 'Does this medication work? How do you know it works? How likely is it to benefit me if you give it to me? What are the other options”

  • Patients often prefer a confident, arrogant doctor who promises certainty, but the reality of medical reversal means humility is valuable; the right approach is for patients to find a doctor whose decision-making style (directive vs collaborative) matches their own preferences.

    “you are encouraging a change in... A lot of people--we don't want a humble doctor... So, there is an interesting culture there”