Research Translation Podcast

The Ghost On the Balance Sheet of Cancer Screening

Cancer screening has always hidden its dark side. Behind the simple messaging and pink ribbons is a long history of complicated truths, failed trials, and titanic spending. Now, with a novel blood test flooding [https://www.zacks.com/stock/news/2919135/hims-hers-expands-digital-healthcare-platform-via-new-care-offerings] the market and media misrepresenting [https://www.theguardian.com/society/2026/may/15/prostate-cancer-screening-save-lives-benefit-small-study] new data, it’s a good time to clarify the simplest truth of all about cancer screening: the balance sheet. To read it correctly, however, you must first learn a jargon term: all-cause mortality. Let us start with a scenario described in the book Hippocrates’ Shadow [https://www.simonandschuster.com/books/Hippocrates-Shadow/David-H-Newman/9781416551546] (ahem, my book): Imagine a treatment so effective that no heart attack patient ever dies—we call it the HeartSaver 3000. When the HeartSaver touches a patient’s chest their heart attack is immediately halted. Historically, up to 10% of such people die but with the HS-3000, heart attack mortality drops to zero. Unfortunately, the Heartsaver also induces fatal strokes—in about 10% of people. The good news is ‘heart attack mortality’ drops by 10%. The bad news is ‘stroke mortality’ rises by 10%. And all-cause mortality stays the same. In trials, cancer screening is like a Heartsaver. Mammograms have never reduced [https://www.bmj.com/content/352/bmj.h6080] all-cause mortality in any trial or combination of trials. The same is true for PSA and colonoscopy screening. Advocates for screening know this and instead tout reductions in breast cancer mortality or prostate cancer mortality (see, for instance [https://www.statnews.com/2026/05/14/psa-tests-reduce-risk-prostate-cancer-deaths-new-study-cochrane-review/], this week [https://www.pulsetoday.co.uk/news/clinical-areas/cancer/cochrane-reviewers-conclude-psa-testing-reduces-deaths-from-prostate-cancer/]’s ebullient headlines [https://www.thetimes.com/uk/healthcare/article/study-benefits-psa-prostate-cancer-screening-9rdnqx0zv] on PSA). But all-cause mortality is the endpoint that matters because it measures deaths, not death-certificate labels. Breast cancer mortality asks whether fewer women died with breast cancer listed as a cause. All-cause mortality asks whether fewer women died. If screening improves the first but not the second, the benefit may be a reshuffling of causes rather than a reduction in deaths. So why, in trials of screening, does breast and prostate cancer mortality go down but NOT all-cause mortality? What other deaths balance out the ledger? Frustratingly, we don’t know. Researchers did not painstakingly account for each exact cause of death in millions of participants, so all we know is that overall deaths did not drop, but breast and prostate cancer mortality did. Which means we’re left to guess about why. And, predictably, those guesses run in two opposing directions. Here’s the guess most popular among screening advocates: There aren’t enough people in screening trials. Since only 2% of people die of breast or prostate cancer, the argument goes, it is very difficult for that 2% to sway the overall numbers. Advocates often argue, therefore, that all-cause mortality is an unfair metric, because even 600,000 women in mammogram trials and 800,000 men in PSA trials isn’t enough. The benefit, they argue, is so small that it would take millions of people in studies before enough breast or prostate cancer deaths accrued to move the needle on all-cause mortality. For skeptics, a different guess is more popular: Either screening doesn’t save lives, or it kills enough people to balance out any lives saved. This argument is grounded in the fact that biopsies, chemotherapy, and major surgeries like mastectomy are substantially increased by screening. Fatal complications of these treatments, even if rare, thus counterbalance any lives saved. These are guesses, not facts. But they highlight a point that is irrefutable, and must be dealt with: Either trials haven’t been large enough to show screening saves lives, or screening doesn’t save lives. Those are the only options. With this fact established, the pro versus con ledger for cancer screening takes on a different hue. On the ‘pro’ side there is a single claim, and it is hypothetical. On the ‘con’ side there are proven harms, and they are legion. For example, more than half of women experience a false cancer scare in ten years of mammograms. MORE THAN HALF. Meanwhile, even if you believe the claimed benefit is real, it amounts to roughly 1 person per thousand. Below is a table of benefits and harms with screening mammography per 1,000 women over ten years, according to the United States Preventive Services Task Force: This balance sheet is, I think, worth putting price tags on. False positives leading to biopsy cost roughly $3,000 each. Unnecessary cancer diagnosis checks in at about $78,000 each. Also of note: More than half of women will suffer the personal cost of weeks to months of anxiety when they’re told they may have cancer (though 95% of the time [https://www.bcsc-research.org/index.php/statistics/screening-performance-benchmarks/performance-measures] or more testing will show this was wrong). For prostate cancer screening in men, this number is roughly a quarter, or 1 in 4. Here is a similar table for prostate cancer screening: Each false positive is, again, about $3,000. Each needle biopsy episode is roughly $10,000. One unnecessary cancer diagnosis costs about $100,000. In addition, treating impotence costs in the range of $5,000 each, while urinary incontinence generates an average bill of $15,000 each. And while that’s an astonishing amount of money, here is the expense virtually no one says out loud: Screening does not merely find disease, it manufactures patients. A woman with a false-positive mammogram becomes a patient until the extra images, ultrasound, MRI, biopsy, pathology report, and follow-up visit say otherwise. A man with an elevated PSA becomes a patient through repeat blood tests, MRI, prostate biopsy, pathology, urology visits, and sometimes a cancer diagnosis—that would never have harmed him. Many then become surgical patients, radiation patients, incontinence patients, impotence patients, surveillance patients. This is not free, and it is not even close. The screening test is just the first bill of many. The real business model in cancer screening is the cascade. A positive screen creates appointments. Appointments create images. Images create biopsies. Biopsies create diagnoses. Diagnoses create procedures. Procedures create complications. Complications create more appointments. The cancer screening machine does not consume one dollar at a time. It consumes in cascades of care that cost thousands to hundreds of thousands each. One estimate is that cancer screening directly costs the U.S. about $43 billion a year. [https://www.eurekalert.org/news-releases/1053067] Add the downstream fallout—false-positives, workups, biopsies, complications, overdiagnosis, overtreatment, surgery, radiation, chemotherapy, surveillance, and treatment of harms—and the total rises to an estimated $70 billion in medical spending. Cancer screening is often sold as a cheap front door to prevention. It is not. The real bill is in the same neighborhood as stroke care—less like a preventive service and more like a chronic care industry. And of course, these numbers do not include anxiety, or lost work, or travel, or the hours spent on hold with billing departments (the most American cancer of all?). The usual response to all of this from screening advocates is that screening is worth it because it saves lives. But the best evidence has found no such thing. What it shows is a small shift in cancer-specific death certificates, purchased with a gargantuan increase in false alarms, procedures, diagnoses, and treatments. This is the difficult truth behind the easy math: When a medical intervention does not clearly reduce all-cause mortality, but clearly creates millions of downstream medical events, the affordability problem is not mysterious. We are not paying to prevent deaths—that is a wishful ghost that cannot be found on the balance sheet. We are paying to diagnose, chase, biopsy, irradiate, cut out, monitor, and medically manage vast numbers of manufactured ‘patients’. That is not prevention. It is a multi-billion-dollar ghost story that sells reassurance, manufactures disease, and then bills for the cleanup. Get full access to Research Translation at researchtranslation.substack.com/subscribe [https://researchtranslation.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

Surgery To Treat Knee Degeneration Increased Knee Degeneration

Dr. Katz, the attending ER doctor, walked ahead of me as we left the bedside of a woman with abdominal pain and vomiting. “What’s the most common reason for emergency surgical admission to the hospital?” Wanting to impress him, I stumbled. “Um, appendicitis?” “In the top three, but no.” Dr. Katz sat down and began typing. I tried again. “Gall bladder?” He didn’t look up. “Also top three—one left.” Defeated, I mumbled. “Small bowel obstruction?” Attending: “Correct. And the most common reason for small bowel obstruction?” This one I knew. “Adhesions from prior surgery.” Dr. Katz kept typing but freed a hand briefly to point at me. “Nailed it. So, in summary, what’s the most common reason for emergency surgery?” Finally understanding, I shook my head in amazement. “Prior surgery.” “Yessssss” he said, still typing. ------ Research Translation is 100% reader supported—to help me continue, become a paid subscriber. This is not just a clever teaching pearl. It goes to the core of modern medicine’s deepest problems. Because medicine has a habit of creating self-sustaining ecosystems. And nowhere is this more visible than orthopedic surgery. Each year in the U.S. hundreds of thousands of people undergo arthroscopic surgery for meniscal tears and degenerated knees. Yet for decades sham-controlled trials have shown the surgeries to be roughly as effective as sham surgery, during which surgeons only pretend to operate. But the surgeries continue. On April 29th the 10-year follow-up of the FIDELITY trial was published [https://www.nejm.org/doi/10.1056/NEJMc2516079]. The trial assigned people to surgery versus fake surgery for meniscal tears of the knee. First reported in 2013, there was no benefit after one year [https://www.nejm.org/doi/full/10.1056/NEJMoa1305189]. Then, same results at two years [https://ard.bmj.com/content/77/2/188] and five years [https://blogs.bmj.com/bjsm/2020/09/09/arthroscopic-partial-meniscectomy-for-degenerative-knee-disease-just-sham-or-does-it-potentially-harm/]. Through it all, the only obvious harms seemed to be the cost, pain, inconvenience, and surgical risk of the procedure itself. By ten years, however, things changed. People in the real surgery group had more arthritis, more knee pain, and needed more surgeries. Major corrective surgery including knee replacement was roughly three times more frequent. Disaster. This is the orthopedic equivalent of the cobra effect. In colonial India British officials, alarmed by venomous cobras, offered bounties for dead snakes. Enterprising citizens promptly began killing cobras. Then they began breeding more of them, in order to kill them. Eventually the government canceled the program. Whereupon the now-worthless cobras were released into the wild. The result: More cobras than ever. Modern orthopedics is eerily similar. Knee pain leads to MRI, which reveals ‘abnormalities’: torn meniscus, ratty cartilage, degeneration. Surgery follows. Then complications, accelerated arthritis, persistent pain, more imaging, more surgery. The system feeds itself. Cardiology has long struggled with what’s called the oculostenotic reflex—the irresistible urge to open any narrowed artery once it’s seen. Orthopedics suffers from its own version: the orthoquixotic reflex, the irresistible urge to heroically repair structural abnormalities. See a tear, repair it. See degeneration, shave it. See asymmetry, align it. Like Don Quixote charging windmills, modern orthopedics often mistakes visible imperfection for an enemy that must be defeated. In a Finnish study I covered recently [https://researchtranslation.substack.com/p/orthopedic-surgerys-big-problem], 96% of MRIs in healthy adults with perfectly functioning shoulders had surgically ‘fixable’ findings. That’s a lot of windmills. And yet trials show we are aggressively tilting at them—roughly a million or more elective surgeries each year in the U.S. that are done to fix ‘abnormalities’ seen on imaging, despite randomized trials repeatedly failing to show meaningful benefit. This includes surgeries for meniscus [https://www.nejm.org/doi/10.1056/NEJMoa1305189] degeneration, acute meniscal [https://evidence.nejm.org/doi/10.1056/EVIDoa2100038] tear, osteoarthritis [https://www.nejm.org/doi/full/10.1056/NEJMoa013259], rotator cuffs [https://pubmed.ncbi.nlm.nih.gov/27385156/], ACL [https://www.nejm.org/doi/full/10.1056/NEJMoa0907797] repair, shoulder decompression [https://www.bmj.com/content/391/bmj-2025-086201], and more. One of the most extraordinary recent examples came not in elderly knees, but in children. In the CRAFFT trial [https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(26)00409-5/fulltext] children with dramatically displaced wrist fractures were randomly assigned to surgical fixation or casting with no manipulation. The result: No important differences between groups including, incredibly, for short-term function. Above is one example of a nine year old’s awful-looking wrist fracture. Pictures A and B are front and side views at the time of injury, showing both bones are displaced and ‘off-ended’. Two years later, panels C and D, there’s no trace of injury—after no manipulation, operation, hardware, or anesthesia of any kind. Kids, man. One would think adults, and degenerating joints, might be different. But adult versions of the CRAFFT trial keep giving us the same answer. To be clear, orthopedic surgery is a crucially important specialty. When bones are shattered and joints disrupted, surgery can be miraculous. But the FIDELITY trial now suggests the orthoquixotic reflex is not merely generating unnecessary surgeries. It may be creating a vast new population of patients harmed by the surgeries themselves. Get full access to Research Translation at researchtranslation.substack.com/subscribe [https://researchtranslation.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

Paxlovid: A Requiem

Paxlovid is dead. Again. Last month Pfizer’s pill for Covid failed in two large trials [https://www.nejm.org/doi/full/10.1056/NEJMoa2502457] published together in the New England Journal of Medicine. The bigger one, with roughly 1,700 participants per group, found 14 people assigned to Paxlovid were hospitalized versus 11 in the control group. No benefit—and leaning in the wrong direction. This was a long time coming. But it’s definitely been coming. I’ve written [https://researchtranslation.substack.com/p/the-secret-studies-of-paxlovid] a handful of pieces [https://researchtranslation.substack.com/p/secret-study-revealed] on Paxlovid, including the hidden studies, nasty side effects [https://researchtranslation.https://researchtranslation.substack.com/p/fact-checking-the-times-on-paxlovid], FDA path [https://researchtranslation.substack.com/p/secret-study-revealed], and more. Which is why a post-mortem might tell us something about our information ecosystem, and how careful translation of research has the potential to change the world. To illustrate, below is a timeline of Paxlovid’s major trials. Green dots indicate a trial that found benefit, red dots mean a trial in which the drug failed. RT is 100% reader-supported, and I’d love to keep doing it. To help that happen, please become a paid subscriber. The first trial, a green dot from 2022, represents the trial that led to the FDA’s Emergency Use Authorization. It was preliminary, rushed, and never intended as a final answer. In it, researchers tested the drug exclusively in people who were unvaccinated and had never been exposed to Covid—a group that no longer exists. In every trial since then, done in populations relevant to today, the drug has failed. The red dots, representing six additional trials, all found no benefit. To my mind the key moment occurred in July 2022 (the dashed vertical line). This is the date of a press release notifying Pfizer’s investors that Paxlovid’s second trial, which enrolled people vaccinated or previously exposed to Covid, was stopped early for futility. The drug failed to reduce hospitalization or death, and failed to reduce symptoms of Covid. No benefit in any outcome. This result should have forced everyone to wonder: Can Paxlovid help people in the current world? The answer was no. Paxlovid failed not just for high-risk people with Covid, it also failed as Covid prevention, for severe Covid, and in Long Covid. By mid-2023 the data were consistent. Worse yet, the evidence for rebound, a relapse illness commonly caused by Paxlovid, was piling up [https://www.acpjournals.org/doi/10.7326/M23-1756]. With that in mind, now look at the blue dots. These are NY Times headlines, which were telling a starkly different story. Paxlovid, according to the Times, was beneficial and underused. Which raises a critical point about medical evidence. For serious research translators, randomized trials are not one input among many. They are at the top of a hierarchy. They are the method we use to strip away bias—to neutralize the hidden distortions that make ineffective treatments appear useful. This is why when randomized trial evidence is available, it replaces weaker forms of evidence. It doesn’t sit beside them. It supersedes them. And yet with Paxlovid the NY Times repeatedly used weaker studies to rhetorically discredit randomized trials. Their 2025 headline, “Paxlovid Improved Long Covid Symptoms In Some Patients” is shocking, partly because it was based on a 13-person case series [https://www.nature.com/articles/s43856-024-00668-8]. Case series are—literally—the lowest form of scientific evidence available. Two years earlier, Paxlovid had failed in a double-blinded randomized trial [https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2819901] for Long Covid. In fact, just months after the headline, it failed again in a second randomized trial [https://www.thelancet.com/journals/laninf/article/PIIS1473-3099%2825%2900073-8/fulltext], which the Times neither covered nor even acknowledged. In this way, quietly, the NY Times inverted the evidence hierarchy, trumpeting weaker studies in order to discredit and eclipse the results of much stronger ones. But the Times is not alone. They simply co-opted and amplified the opinions of ‘experts’, the CDC, the FDA, the AMA, and more. How did these people and institutions get Paxlovid so wrong? Institutions optimize for their own incentives: Media organizations optimize for engagement. Pharmaceutical companies optimize for sales. Professional societies optimize for relevance and authority. Public health agencies optimize for actionable guidance. Conventional wisdom is therefore held hostage by the institutions with the power and influence to sculpt it—even when they prefer the opposite of truth. But this is the path to irony, because science will always move toward truth, and falsehoods will be revealed. And when they are the institutions that used their influence for self interest will find that influence slipping away. Which is how we got here. The real lesson, therefore, is not that medicine gets things wrong. Of course it does. Urgent first studies of potentially profitable pandemic cures will often be wrong. Think remdesivir, molnupiravir, and others. But rigorous evidence translation sees those errors in real time, and can hold institutions accountable before they dig in—saving them, and the rest of us, from themselves. Which raises an interesting question for today: The mad scramble to restore institutional influence is now underway—who will blink first? Which major institution will formally correct before the others? It has been four years since Pfizer announced [https://www.fiercepharma.com/pharma/pfizer-stops-paxlovid-work-less-vulnerable-covid-19-patients-after-no-benefit-symptom-relief] their drug failed in the only relevant population. Two years since those data were formally published [https://www.nejm.org/doi/full/10.1056/NEJMoa2309003]. And two weeks since large new trials beat a dead horse by burying Paxlovid yet again. As of this writing, perhaps unsurprisingly, the NY Times has offered zero news coverage of the new trials. Nor is the Times unusual. Yale Medicine still strongly promotes [https://www.yalemedicine.org/news/13-things-to-know-paxlovid-covid-19] Paxlovid on its public website. Google’s AI overview [https://www.google.com/search?q=paxlovid&client=firefox-b-d&hs=QBqp&sca_esv=97234541236c36f9&biw=1102&bih=665&sxsrf=ANbL-n7SteRs_8ncFyVC8xI_dEOcEkzBwg%3A1778084612575&ei=BGv7aeDcItO5mtkP4KjRkQ8&ved=0ahUKEwjg7tmtiaWUAxXTnCYFHWBUNPIQ4dUDCBE&uact=5&oq=paxlovid&gs_lp=Egxnd3Mtd2l6LXNlcnAiCHBheGxvdmlkSMcoUABYAHABeAGQAQCYAQCgAQCqAQC4AQPIAQCYAgCgAgCYAwCIBgGSBwCgBwCyBwC4BwDCBwDIBwCACAE&sclient=gws-wiz-serp] still says the drug reduces hospitalization and death. The first page of a google search is dominated by FDA, CDC, Pfizer, and Wikipedia entries, all presenting Paxlovid as a lifesaver. Which helps to clarify the lesson of the Paxlovid timeline. Bad science is not the greatest danger. Science usually self-corrects. The greater danger is institutional pride and inertia: the years-long gap between when science settles a question and when institutions are willing to absorb the answer. That gap is costing billions, distorting public understanding, and exposing millions to a harmful drug that simply did not—and does not—work. But the evidence was there all along. Research Translation is totally reader-supported. If you dig it, please become a paid subscriber so I can keep it up! Get full access to Research Translation at researchtranslation.substack.com/subscribe [https://researchtranslation.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

Cholesterol 5: The Power at Your End of the Stethoscope

During the Cholesterol Mistake series, we’ve walked through the unraveling of a medical dogma. First, we saw [https://researchtranslation.substack.com/p/the-first-cholesterol-mistake] how the Lipid Hypothesis violates the seminal rule of risk factors. Next, we watched [https://researchtranslation.substack.com/p/cardiologys-nonfatal-fatal-mistake] the goalposts shift from extending life to chasing non-fatal laboratory endpoints. Then, in trial appendices, we found [https://researchtranslation.substack.com/p/cholesterol-the-big-mistake] the Big Mistake: tallying checkbox diagnoses while ignoring what matters to people. Along the way, with new eyes we saw the subterfuge in a breaking cholesterol trial [https://researchtranslation.substack.com/p/a-cholesterol-study-straight-out] that failed to help humans, yet claimed victory. Then we unwound the stilted logic and murky math [https://researchtranslation.substack.com/p/checking-the-math-in-new-statin-guidelines] that allows a guideline to quietly serve itself. Finally, we discussed [https://researchtranslation.substack.com/p/cholesterol-what-went-wrong] ideologic gumption, the driver of a religious faith—cholesterolism [https://researchtranslation.substack.com/p/the-cholesterol-myth-unraveled-for?utm_source=publication-search]—that eschews reason, logic, and scientific method. How do we stop the momentum? How do we right-size cholesterol culture and help 25 million healthy people avoid a lifetime of pills that, if informed, most would not choose? The answer lies in a simple truth: There is more power at your end of the stethoscope than you have ever been led to believe. For decades, the clinic dynamic has been asymmetrical. The doctor holds the clipboard, interprets the guidelines, and hands down a prescription. But reform doesn’t start with new guidelines from the AHA. It starts with information symmetry (the goal of this Substack). The solution to the Cholesterol Mistake—and nearly every other systemic problem in modern medicine—is to stop asking for the prescription and start asking for the information. Here, then, is a playbook for your next appointment. When your doctor suggests taking—or staying on—a statin, do not accept or refuse. Instead, ask for the math. Right there in the exam room ask your doctor to pull up the guideline’s new calculator [https://professional.heart.org/en/guidelines-and-statements/prevent-calculator]. Together, calculate your 10-year risk for ASCVD. With the result in hand insist on absolute truth, not relative. If the drug offers a ‘30% relative reduction’ and your baseline risk is 5%, together you can calculate your absolute reduction as 1.5% over ten years. With that 1.5%—a 1 in 67 chance of benefit—hanging in the air, ask the critical follow-up: “What is my chance of getting diabetes or muscle damage from the drug during that same decade?” To make the numbers transparent, use the Research Translation calculator [https://statin-decision-calculator.vercel.app]—crafted just for statins, and just for this conversation: Doctors overwhelmingly have good intentions, but they’re trapped in an experiential delusion, conditioned to treat the guideline rather than the person. By forcing them to articulate the absolute benefits alongside harms, you can break the spell. Any physician unable to explain the data behind the pill will suddenly have to figure out how. Evidence-based medicine isn’t just for doctors. The data belongs to you. By insisting on seeing the science, you restore the balance of power. The era of blindly swallowing the Lipid Hypothesis is over. It is time to do the math—and show it. Get full access to Research Translation at researchtranslation.substack.com/subscribe [https://researchtranslation.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]

Checking The Math in New Statin Guidelines

Today’s piece is an interlude in the Cholesterol Mistake series (parts 1 [https://researchtranslation.substack.com/p/the-first-cholesterol-mistake], 2 [https://researchtranslation.substack.com/p/cardiologys-nonfatal-fatal-mistake], 3 [https://researchtranslation.substack.com/p/cholesterol-the-big-mistake], and 4 [https://researchtranslation.substack.com/p/cholesterol-what-went-wrong]) which was inspired by the new AHA lipid guidelines. I got sidetracked by the figure below from those guidelines, and ended up in the weeds—deep enough to go much longer than normal. Apologies for length, hope the payoff is worthy. Next week part 5, back to brevity! “He says the chest pain started suddenly an hour ago, it feels ‘tearing’, and radiates to his back, shoulders, and arms.” Jeff, my senior resident, raised an eyebrow as he presented the case of a 48 year-old man. “He’s in a lot of pain, too. I ordered dilaudid—he says that’s worked for him in the past—and I put him first in line for CT.” Jeff ended with a question, “So, maybe I’ll call surgery?” I nodded. “Let me talk to him first” I said, walking toward room 3. “Ok,” Jeff hesitated. “I feel like we should hustle, you know? I think he’s dissecting.” A tear in the wall of the body’s main artery—aortic dissection—can be rapidly fatal. “I get it. Give me two minutes.” Turning the corner, I glimpsed the man in room 3. He was well dressed and appeared comfortable. In the room I examined him carefully and listened intently to his story. It was a good story for aortic dissection. Too good. RT is 100% reader-supported and I’d love to keep doing it. If you’re enjoying, please consider becoming a paid subscriber. Parts of the new AHA Lipid Guideline remind me of the man’s act. One was highlighted [https://www.sensible-med.com/p/guidelines-evidence-based-medicine] recently by the wonderful Adam Cifu of Sensible Medicine [https://www.sensible-med.com/] (which I read religiously for honest, often fearless commentary from the front lines). Last week Cifu wrote [https://www.sensible-med.com/p/guidelines-evidence-based-medicine] about Figure 4 in the new AHA guideline. He lamented how poorly such graphs translate in the exam room because humans have preferences, and quirks, and rarely if ever fit the average numbers envisioned by guidelines. What he did not discuss—and we will—is the veracity of the math in those graphs. This is not a small matter. Figure 4 is titled, “The Logic For Defining…” 3% or more as the 10-year risk threshold for starting statins. Indeed it is the only place in the 123-page document where they show us how they arrived at this number. Again, this is not academic. The 3% threshold recommends statins for an additional 25 million U.S. adults. They deserve to know why. So let’s find out. Both graphs plot cardiac risk versus benefit. The x-axis (bottom line of the graphs) shows a person’s 10-year heart risk, which anyone can get from the AHA’s online calculator [https://professional.heart.org/en/guidelines-and-statements/prevent-calculator]. The y-axis (vertical) shows possible benefits of statins expressed as Numbers Needed to Treat, or NNT. For those new to the concept, NNTs get smaller as benefits get bigger because fewer people ‘need to be treated’ for one of them to experience a benefit. This is nicely illustrated by the graph’s red curve which shows NNT dropping (i.e. benefits getting bigger) as the risk of heart problems rises. That’s because the bigger the chance of a benefit the lower the NNT; and the higher a patient’s risk the more likely they’ll benefit. The dotted vertical line in Panel A represents someone with a 3% risk of a heart problem—the new threshold. Wording above the graph tells us that statins reduce risk by 35%, or about a third. When a 3% risk drops by about a third, it’s dropping by 1%. That means exactly 100 people ‘need to be treated’ for 1 to benefit, which is why the dotted line and the NNT curve converge at 100. Crucially, they tell us an ‘NND’, which is the same as NNT but refers to statin-induced diabetes. Their “NND=100” means they assume exactly 1% of people will get diabetes from 10 years of statins—the same as the chance of benefit at 3% coronary risk. What’s it all mean? They are saying (obtusely and indirectly) that a 1% diabetes rate and a 1% coronary benefit is a break-even point and it happens at a 3% risk of heart problems. Anyone with less risk will see less benefit, and thus statins will more likely harm than help. Anyone at higher risk sees more benefit and is more likely helped than harmed. Get it? And again, to be explicit: This is the only place in their guideline where they explain the new 3% threshold. Which suggests to me that the committee likely saw this graph and agreed it should drive the new recommendations. All of which makes the assumption of 1% for the diabetes number a little weird. The guideline’s most cited source on diabetes is an 2024 analysis [https://www.thelancet.com/journals/landia/article/PIIS2213-8587(24)00040-8/fulltext] that found statins cause diabetes at 0.12% per year. For 10 years that’s 1.2% (not 1%), which means harms are greater than benefits at the new threshold. But you know what? Whatev. Let us not quibble. Perhaps they estimated downward to 0.1% per year, and although it shifts the graph (their threshold should be 4%) who cares?! It’s only 8 million more people mistakenly recommended for statins. It’s cool, it’s cool. We won’t panic. Until Panel B. Panel B is the same calculation, only for high-dose statins. Here the AHA says 3% get diabetes from the drugs after ten years. The break-even point is a 7% risk, because they say high-dose statins reduce risk by 45%, turning 7 into 4—a 3% drop, matching the increase in diabetes. Get it? But this time their diabetes estimate departs from the data by a lot. The guideline’s preferred source [https://www.thelancet.com/journals/landia/article/PIIS2213-8587(24)00040-8/fulltext] found that high-dose statins cause diabetes in 1.27% per year, or 12.7% after ten years. Where did they get 3%, pray tell? It may come from their last guideline [https://www.ahajournals.org/doi/10.1161/01.cir.0000437738.63853.7a] 13 years ago, which used the 3% number but cited a single 2010 study as their evidence source. That would be strange because the updated 2024 analysis includes data from 23 trials and is plainly the new guideline’s go-to source, cited and quoted throughout the document. So why didn’t they use it? Shrug. Inexplicable. In any case, using their logic—that diabetes is the one harm worth calculating, and the benefit of high-dose statins is a 45% risk reduction—the correct break-even point would not be at 7% coronary risk, but 28%. Which is interesting because very few without heart disease have a risk that high. Levels above 20% are seen overwhelmingly in secondary prevention. Based on this corrected calculation, therefore, the AHA should be cautioning against high-dose statins in primary prevention. Again, that is using their numbers and logic, but correcting the outdated input for statin-induced diabetes. But of course, it is not just that they forgot to update their diabetes numbers. There are lots of other flaws in the AHA logic. For instance, that diabetes is the only downside. As Adam Cifu points out, “There is the cost of the medication, the monitoring, the visits, and the turning an enormous swath of the population into patients.” Not to mention muscle pain and damage in at least 5% [https://www.ahajournals.org/doi/full/10.1161/CIRCULATIONAHA.112.136101], or 1 in 20 (which is certain to be an under-count). And then there’s the prickly issue of ‘benefits’. Last week in part 4 [https://researchtranslation.substack.com/p/cholesterol-the-big-mistake] we walked through the raw numbers to better define statin benefits. Briefly, in the AHA graphs above the benefit consists of three parts: roughly 45% nonfatal MI, 25% nonfatal stroke, and 30% death. But as we discussed there is no proven reduction in death at risk levels <20%. The 2012 paper—the AHA’s source—shows clearly that in this risk group mortality was unaffected (see the figures and tables in part 4 [https://researchtranslation.substack.com/p/cholesterol-the-big-mistake]). And yet the guideline is quietly counting deaths as a benefit. That is, to put it mildly, inconsistent with the evidence. If we therefore remove this portion of the benefit and recalibrate, the AHA’s treatment threshold shifts from 3% to 5%. Then we make the big move. Now we remove what patients don’t take statins for: to prevent ‘nonfatal MI’. As discussed in part 2 [https://researchtranslation.substack.com/p/cardiologys-nonfatal-fatal-mistake], this endpoint is a historical mistake, and in modern trials often represents little more than a lab abnormality. This explains why it is proven [https://jamanetwork.com/journals/jamainternalmedicine/fullarticle/2785560] not to be associated with death or disability, the two outcomes people actually take statins to prevent. When we remove nonfatal MI, leaving nonfatal stroke as the one proven, patient-centered benefit, the new break-even point occurs at a 14% baseline coronary risk. Below is a table showing mathematically correct break-even points using the AHA’s logic (diabetes is the one harm; statins reduce risk 35% and 45%), both with and without their outdated inputs. The table suggests that, by their own framework, the AHA should recommend only moderate-dose statins, and only for those at >14% risk. High-dose statins, meanwhile, are more likely to harm than help virtually everyone in primary prevention. But it gets worse. Presumably, the AHA agrees that patients should also be informed about common, established harms like muscle pain and damage. If those are incorporated into the same break-even framework, the picture changes again: Essentially, no one in primary prevention would be eligible—and few in secondary prevention. At this point, however, the model itself should be questioned. The AHA’s framework assumes that all harms and benefits carry equal weight—that a 1% reduction in coronary risk perfectly offsets a 1% increase in diabetes, or a 1% increase in muscle injury. That is not how people feel outcomes. Rather than impose those values, we can simply show people the numbers and let them decide. For someone at a 14% baseline risk considering moderate-dose statins, the likely effects look like this: * Mortality: None * Avoiding a nonfatal stroke: 1 in 83 * Getting diabetes from the drug: 1 in 83 * Muscle pain and damage: 1 in 20 This—finally—is the conversation we should be having. People decide what matters most using their values. Not the AHA’s, or their doctor’s, or mine. But theirs. When I examined my possible dissection patient I found no signs of illness—the third red flag. The first was a request for opiates, the second was an unlikely story. Skeptical, I dug one layer deeper, offering him non-opiate pain control and a full work-up for possible aortic dissection. But alas, the lack of opiates was a deal breaker for the man and he walked out. While flipping me the bird. The AHA guideline is, in my opinion, littered with red flags. One is a blind faith in the Lipid Hypothesis. Another is an unlikely story—that low risk healthy people should take cholesterol drugs for the rest of their lives. And one more is a graph with data inputs that don’t match their own citations. All of which should make us dig deeper. And Figure 4 is where ‘X’ marks the spot. What is most striking about the graphs, to me, is that they’re offered as the scientific justification for a major expansion in treatment—and yet the underlying assumptions are opaque and untraceable. Why wouldn’t they show us their work? The answer, it seems, is that their model uses numbers that can’t be reconciled with the underlying data, obscures key calculations, and relies on outcomes that people taking statins don’t care about. This is not how science fails. It is how it drifts—quietly, incrementally, until the appearance of rigor replaces the thing itself. The guideline sounds good, looks good, and has fancy, science-y graphs that blind and confuse. Figure 4 is, in this sense, the centerpiece of the illusion—a masterstroke of form over function, style over substance, and spectacle over science. Which is why their conclusions look convincing, even when they are not. And why people considering statins might want to dig a layer deeper. Get full access to Research Translation at researchtranslation.substack.com/subscribe [https://researchtranslation.substack.com/subscribe?utm_medium=podcast&utm_campaign=CTA_4]