His return presents an opportunity to reflect a bit on the first four years of Sensible Medicine. The site launched in the summer of 2022 as a shared site featuring the voices of leading physicians, scientists, and thinkers. The founding editors were John, Adam, Vinay, Zubin Damania, and Marty Makary.

The goal of Sensible Medicine has been, and remains, to showcase a range of ideas and opinions about all things bio-medicine. We strive to feature contrasting ideas and opinions, believing that progress occurs through dialogue. We have worked to nurture a home for this type of engagement. We are amazed by and grateful for the site's growth. We have well over 100,000 subscribers. Every article is read by many tens of thousands of people. Thank you to our readers, our subscribers, and all the people who have submitted work to us over the last four years.

Sensible Medicine 1.0 encompassed our first couple of years, from the launch until Vinay left us to join the FDA. During this time, we went from five editors to three, began accepting submissions, and found our voice. We featured a series of written debates and a lecture series on how to read the medical literature.

Sensible Medicine 2.0 saw John and Adam editing the site while Vinay was on his sabbatical. We were thrilled by the increase in high-quality submissions during this period. We also launched the This Fortnight in Medicine podcast with .

What will Sensible Medicine 3.0 look like? Some things will continue unchanged. John will post his weekly article dissections, simultaneously telling us what’s going on in medicine and teaching us how to read the literature. Every other Wednesday, we will release a This Fortnight in Medicine podcast. There will continue to be 3 hosts, but John, Vinay, Andrew, and Adam will all cycle through. Adam will do what he usually does on Friday, which seems to be writing about whatever comes to mind.

One of our goals, perhaps too lofty, is that SM creates a community of skeptical-but-not-cynical medical people large enough to hold a meeting. A congress of our own, if you will. Vinay’s foray into renting a hotel conference room at an oncology meeting to provide learners with a less biased review of the major trials is a start. All of medicine desperately needs such an alternative venue.

There will be some exciting changes. Vinay is returning with knowledge of the inner workings of the federal government and US drug and biologic regulation, and will bring this perspective to his writing. He remains committed to the central question of our era: How do we ensure medical decisions are made by good and reliable evidence, at a time of widespread excitement and innovation? He promises to only seldom mention being correct about Paxlovid.

We will also post editor debates in which we will argue important topics in medicine. We will add to our catalog of How to Not Get Fooled videos. We will continue to post submissions from our subscribers. We might even add some new Merch!

We hope that you are as excited about SM 3.0 as we are.

‘No Way To Prevent This,’ Says Only Nation Where This Regularly Happens

This article will be my version of The Onion's headline, to be trotted out every time I sit through a particularly bad lecture.

In my lifetime of listening to lectures, most have been forgettable, leaving me bemoaning an hour of life wasted. Thankfully, in recent years, at least I can be productive — or distracted — by my phone.

I have sat through a few transcendently bad lectures, true fiascos. The most memorable one, given by a plastic surgeon during medical school, was so misogynistically offensive that students walked out.

Then there are the few lectures that I remember years, or even decades, later. I have written about a talk by my Haverford College chemistry professor. I remember a talk that Booker Bush, a mentor during my residency, gave on counseling behavioral change. There was also a talk that George Daley gave — to a nearly empty room — at a AAMC meeting about the ethics of gene therapy.

There are also lecturers who stand out as memorable, people who were clearly dedicated to the craft, whose every lecture was entertaining and informative.

After a busy season of lecturing and listening to lectures, I thought I could give a few pointers on how to avoid giving a forgettable lecture.

The person who should have written this article died over 10 years ago. I never had Herbert Friedmann as a professor, but I did watch him lecture a few times. He must have been a joy to learn from. It might seem absurd to compare Friedmann — a Jewish biochemist, born in Germany, raised in India — to Muhammad Ali, but they had the same twinkle in their eyes when they spoke. You could tell that Professor Friedmann was enjoying himself when he lectured. You could tell that he knew you were enjoying listening to him. And you could tell that he was enjoying entertaining you. I get the same sense when I watch old films of Ali.

I have met only a few people who took their teaching as seriously as Friedmann. In 1990, he published a masterpiece in the Journal of Chemical Education: Fifty-Six Laws of Good Teaching. Please read it now. If you don’t come back to me, you’ll have gotten 99% of the wisdom this piece has to offer.

Standing on Friedmann’s shoulders, I have 5 recommendations inspired by some not-so-great lectures. Following these won’t guarantee you’ll give a great lecture, but it will at least help you, in the words of Joe Madden, try not to suck.

(This is one of our rare, pay-walled posts, a gift to paid subscribers.)

Read more

Adam Cifu

The 2026 American College of Cardiology/American Heart Association Guideline on the Management of Dyslipidemia was released last month. If you work in medicine, you’ve already heard some version of the same four words: just follow the guidelines.

I want to push back on that, not because the guidelines are wrong, but because “follow the guidelines” has become a thought-terminating cliché that short-circuits exactly the kind of clinical reasoning it is meant to support.

This is not an argument to ignore or dismiss guidelines. The 2026 dyslipidemia guideline was a substantial effort. Its literature review is broad, its clinical questions are relevant, and much of its guidance will help clinicians. But “follow the guidelines” treats an expert consensus document as if it were a verdict. It isn’t.

The process is more transparent than you think — and that’s the problem

The development of ACC/AHA guidelines is a remarkably transparent process, governed by a detailed methodology on the process. But transparency about process is not the same as transparency about deliberation. The methodology manual describes voting procedures and peer review, but the guideline does not show the range of opinion within the committee, how disagreements were resolved, or whether any members dissented from specific recommendations. What appears in print is a unified document; the uncertainty and disagreement behind it are largely invisible. Anyone who has practiced medicine and reviewed the literature knows that you can ask 3 doctors for their opinions on treating a patient and get 4 answers. Uncertainty is a part of medicine, but is obscured by a guideline conclusion.

Who’s in the room — and why it matters

The ACC/AHA guideline process draws heavily from the cardiovascular community. Writing committees are composed largely of cardiologists and lipid specialists, alongside pharmacists, nurses, dietitians, and a small number of generalists and patient representatives.

This is a governance choice that has predictable consequences.

A clinician managing complex lipid disorders in a subspecialty setting operates in a fundamentally different environment than a primary care physician managing multiple chronic conditions in a 15–20 minute visit. That difference directly shapes what recommendations are feasible.

The imbalance is not just about the care setting. Determining whether a trial demonstrates clinically meaningful results is a clinical and methodological judgment. It requires statistical fluency, familiarity with bias, and comfort interrogating uncertainty.

Specialists bring clinical experience that includes the consequences of undertreatment. That perspective is valuable, but it can introduce what is sometimes called specialty bias. Specialty bias encourages physicians to preferentially recommend treatments aligned with their own training and practice. For example, surgeons are more likely than nonsurgeons to recommend surgery, and these recommendations influence patient choices.

A more balanced committee would include greater representation from clinical epidemiologists, biostatisticians, and evidence-based medicine specialists — individuals whose primary expertise is in evaluating evidence rather than treating a specific disease. Their relative underrepresentation is common across guideline development.

The length problem

The 2026 dyslipidemia guideline, according to its authors, is a “comprehensive one-stop shop.” That description is accurate — and revealing. The document is accompanied by slide sets, population-specific summaries, pocket guides, and plain-language versions. This is not just accessibility design; it is a recognition that the core document is too large (134 pages) to serve as a primary clinical tool for most practicing physicians.

There is no plausible workflow in which a busy primary care physician reads and integrates a document of this scope into routine practice. The practical result is predictable: clinicians either follow simplified versions embedded in clinical decision support tools without understanding the reasoning behind them, or they disengage entirely from the full document.

Not all recommendations are equal

The ACC/AHA classification system distinguishes between the strength of recommendation and the quality of evidence. Some recommendations are supported by multiple randomized trials (Level A). Others rely on more limited data or expert opinion (Level B or C). Many of the newer guidelines are “B-NR”, meaning there has been no new randomized trial evidence to inform the change in guidance; they are based on a combination of observational data and expert opinion. A Class I-A recommendation and a recommendation based primarily on expert consensus are fundamentally different forms of knowledge. Yet “follow the guidelines” flattens these distinctions into a single instruction.

The most important recommendation is the one often ignored

There is an irony embedded in the “just follow the guidelines” framing: it overlooks one of the guideline’s central principles. Shared decision-making is not an afterthought in the ACC/AHA framework. It is explicitly emphasized in the methodology and embedded in the guideline’s overall approach. The process of risk calculation, personalization, and recommendation is inherently collaborative.

That collaboration matters most in the gray zones — where evidence is uncertain, benefits are modest, or patient preferences vary. A patient with a 10% 10-year ASCVD risk who strongly prefers to avoid daily medication has a legitimate claim on the guideline, just as one who prefers aggressive risk reduction does.

In that sense, “just follow the guidelines” is itself a failure to follow the guidelines. The document does not replace clinical judgment or patient preference; it is designed to inform both. However, the guideline itself offers little help in shared decision-making. For instance, it is sparse on Numbers Need to Treat (NNT) and does not mention that for a patient with a 10-year risk of 15%, the NNT is 25. Thus, most patients who take a statin will not benefit.

When guidelines matter most — and when they matter least

The usefulness of guidelines tracks closely with the strength of the underlying evidence.

When evidence is strong, guidelines are less important. Clinicians have access to the data through primary articles, systematic reviews, and continuously updated clinical resources.

Where guidelines matter most is where evidence is weakest — when clinical questions are real, but trials are limited or absent. In those settings, structured expert opinion is genuinely valuable.

But this is also where transparency matters most. If a recommendation rests primarily on expert consensus, readers should know how that consensus was reached. Formal methods such as the RAND/UCLA Appropriateness Method or Delphi processes make disagreement visible and quantify uncertainty. Many guideline processes do not report using these approaches, and rarely make the range of opinion explicit.

An uncomfortable conclusion

The ACC/AHA is among the more transparent guideline-producing bodies in medicine. It publishes its methods, discloses conflicts, and makes a genuine effort to standardize its process. And yet, even here, familiar concerns remain: committee composition that leans toward intervention, evidence of varying quality presented under a unified framework, documents whose scale limits real-world usability, and limited visibility into how expert consensus is formed when evidence is insufficient.

Primary care physicians are expected to navigate guidelines from many specialties — cardiology, oncology, gastroenterology, psychiatry, and others. Many of these organizations publish far less about how their recommendations are developed.

The question is not whether any single guideline is trustworthy. It is whether “follow the guidelines,” applied reflexively across this entire ecosystem, is a coherent instruction

“Follow the guidelines” is a reasonable starting point, but a poor goal, especially when the guidelines do not provide the numbers that clinicians and patients need to make decisions.

Lenny Lesser is a Family Physician in San Francisco, CA.

Share

Photo by Hal Gatewood

This is a story about how changing disease environments can change drug efficacy. And why trials need to be repeated when times change.

During the early parts of the pandemic, before vaccination and before enough people had been exposed to the SARS-CoV-2 virus, a trial called EPIC HR found that treatment of symptomatic Covid-19 with nirmatrelvir plus ritonavir (Paxlovid) resulted in a risk of progression to severe Covid-19 that was 89% lower than the risk with placebo. This highly positive trial led to strong recommendations to use this medicine in such patients—despite the large number of potential drug-drug interactions.

The pandemic then changed. People were vaccinated; others were infected and developed immune protection to the virus. The next question became: would the antiviral medication have the same effect in vaccinated or previously infected individuals.

The EPIC-SR (standard risk) trial, published in 2024, randomized 1296 patients to placebo or nirmatrelvir–ritonavir and found no difference in the primary endpoint of time to sustained alleviation of all COVID-19 symptoms. A cursory look at this seems anodyne. Patients were low-risk; the drug therefore had no significant effect.

But the timeline is bothersome. I asked CLAUDE for some help sorting this out.

The Current Paper (April 2026)

The NEJM manuscript published last week included results of the British PANORAMIC and Canadian CanTreatCOVID trials. Both enrolled higher risk older patients with a coexisting condition who tested positive for SARS-CoV-2 and had been unwell for 5 days or less.

The randomization was to usual care plus nirmatrelvir (300 mg)–ritonavir (100 mg) twice a day for 5 days or to receive usual care alone. The primary outcome was hospitalization or death from any cause within 28 days after randomization.

Enrollment was carried on between Sept 2021 and Sept 2024.

In the PANORAMIC trial, 14 of 1698 participants (0.8%) in the nirmatrelvir–ritonavir group and 11 of 1673 participants (0.7%) in the usual-care group were hospitalized or died. Clearly not a significant difference. In the CanTreatCOVID trial, 2 of 343 participants (0.6%) in the nirmatrelvir–ritonavir group and 4 of 324 participants (1.2%) in the usual-care group were hospitalized or died.

Serious adverse events with nirmatrelvir–ritonavir were reported in 9 participants in the PANORAMIC trial and in 4 participants in the CanTreatCOVID trial.

One secondary finding of note: Symptom recovery was faster with nirmatrelvir–ritonavir. Median time to recovery was 14 days vs. 21 days in PANORAMIC, and 6 vs. 9 days in CanTreatCOVID. The problem, of course, was that these were pragmatic open-label trials without a placebo, and symptom recovery is subjective and susceptible to placebo and nocebo effects.

The authors concluded:

In two open-label trials, nirmatrelvir–ritonavir did not reduce the incidence of hospitalization or death among vaccinated higher-risk participants with SARS-CoV-2 infection.

Comments

Read more

I’m embarrassed to admit that this question came to me not in clinic, where my patients bring up their MyChart alerts with me every day, but after I read the latest RSV vaccine article and discussed it with John Mandrola and Andrew Foy on This Fortnight in Medicine XXIV.

The NEJM article described a beautifully conducted RCT in which 131,276 people were randomized to receive the RSV vaccine or remain unvaccinated (no diluent or saline comparator here). The primary endpoint was hospitalization for RSV-related respiratory tract disease. The secondary endpoint was hospitalization for RSV-related lower respiratory tract disease and hospitalization for respiratory tract disease from any cause.

The randomized patients had a mean age of about 69, and about 80% had the flu vaccine the year before and the year of the study. 42% of the patients were being treated for a chronic disease.

The RSV vaccine was, as we like to say, safe and effective. Over six months of follow-up, the vaccine effectiveness for the primary endpoint was 83.3% (42.9-96.9). For the secondary endpoint, hospitalization for respiratory tract disease from any cause, the effectiveness was 15.2% (0.5-27.9). The incidence of serious adverse events was similar in the two groups.

When we look at absolute differences, rather than the relative differences described by vaccine efficacy, the numbers tell a somewhat different story. Hospitalization for RSV-related respiratory tract disease occurred in 3 of 65,642 participants in the vaccine group compared with 18 of 65,634 participants in the control group. That is a number needed to vaccinate (NNV) over six months of 4376. (For the secondary endpoint, the NNV is 1286).

The TLDR of all this is that the RSV vaccine is safe and effective, but the real question is whether RSV is a big enough problem to vaccinate every human over 70. As always, there are two arguments that can be made.

If you are pro-RSV vaccine, you’d argue that while a 1/1300-4400 chance of avoiding a hospitalization doesn’t seem impressive, it is impressive from a population standpoint. An intervention that decreases hospitalization by 1/1300 is pretty good. You’d also point out that these NNVs probably understate the benefit. If these numbers repeat themselves every 6 months, after 10 years, you’re at a NNV of 65. Preventing a hospitalization in the next 10 years in 1 of every 65 people you vaccinate is incredible. While we’re madly extrapolating data, you might also suggest there is a 3^rd party effect.

If you’re anti-RSV vaccine, you’d argue that a doctor’s role is to care for individuals, not populations, and that the data at hand should guide us, not extrapolations that may or may not be reasonable. In this case, it should be the patient’s choice whether he wants a shot that will decrease his risk from 1 in 3612 to 1 in 21,667.

Let’s get back to the MyChart alert. The RSV recommendation stems from our current approach to healthcare: if a safe treatment is beneficial, even if the benefit is small, we recommend it. This rationale is behind the new lipid management recommendations for low-risk people and explains the heaps of money the US spends on health care for outcomes worse than those of our peers. I think that taking this approach to healthcare is probably right for some people, wrong for others, and wrong for society. But that is just one opinion.

When a patient comes in asking for a test or treatment because it has been recommended by the electronic health record (a stand-in for someone making clinical recommendations), the doctor could respond in two ways.

She could say, “Yes.” This respects the national guideline. It also assumes that the patient has considered the recommendation and decided that it is right for him. It is also patient-centered. You asked me for something, it is not an unreasonable request, and I am going to satisfy it.

Or the doctor could say, “You can certainly get it today, but would you like to hear more about the magnitude of the shot’s benefits?” This response leans into patient education and shared decision-making. It is also time-consuming and undercuts the authority of national bodies. It also might be confusing to a patient who assumes the EMR alert reflects his doctor’s decision-making. It also might come off as antivax. “You really want a vaccination? Maybe you should reconsider.”

So, loyal reader, how would you want your doctor to respond if you said, “Doc, I got a MyChart alert that I am due for an RSV shot. Can I get it today?”

A. Yes.

B. “You can certainly get it today, but would you like to hear more about the shot’s benefits?”

Share

Some personal follow-up

Read more

Ultrasound-Facilitated, Catheter-Directed Fibrinolysis for Acute Pulmonary Embolism

• Endovascular management of intermediate-risk pulmonary embolism: evidence, outstanding questions, drivers of utilization, and the horizon

  Share

Adam Cifu

For years, opening a blocked coronary artery seemed like an obvious imperative, but the COURAGE Trial challenged that assumption. The ISCHEMIA Trial furthered the argument. The reversal is now settled, but these shifts don’t happen overnight. In the years immediately following COURAGE, a clinical decision support tool searching for evidence on stable angina would have surfaced more papers supporting stenting than opposing it. This is a structural problem that persists in how modern AI retrieval systems operate. That early window, where the new evidence exists but is outnumbered, is where today’s AI clinical reference tools are likely to fail.

A brief primer

Most AI systems that answer clinical questions rely on a framework called Retrieval Augmented Generation (RAG) to ground responses in actual evidence. Instead of answering from memory alone (patterns captured during training), the model first searches a database of medical literature, retrieves the most relevant articles, and then synthesizes an answer from what it finds.

The “relevant” part is where these systems can display sub-standard deduction skills. Standard retrieval ranks articles using cosine similarity, which measures how closely the words and meanings of an article match those of the question. It is a measure of topical overlap, not of evidence quality, recency, or whether the article has since been contradicted. This is fine when the evidence base is stable and coherent, but becomes a problem when it isn’t.

The medical reversal connection

Readers of Sensible Medicine are likely familiar with medical reversal: the phenomenon in which an accepted therapy is found, upon rigorous study, to be no better (or worse) than what it replaced. Prasad and Cifu published an article stating that roughly 13% of articles making claims about medical practice in a single year of the New England Journal of Medicine constituted reversals.

We need to start thinking about how AI will handle reversals. When a medical practice has been standard for years and is then overturned by a handful of newer studies, the numerical balance of the literature favors the old practice. Cosine similarity retrieves the most textually similar query, not what represents the current best evidence. Recent work by Javadi et al. demonstrated this empirically: when retrieved documents contain contradictory evidence, RAG accuracy drops by up to 20%. Instead of the contradiction being surfaced to the user, it gets fused with the answer, leading to a drop in overall accuracy.

What a fix might look like

The solution is to stop treating retrieval as a single, final step and to start engineering the pipeline to account for evidence quality. A multi-stage approach could include:

• recency weighting to prevent newer evidence from being drowned out,

• relevance pruning to remove topically adjacent but clinically irrelevant results,

• and a prioritization layer that weights by study design, sample size, and source credibility at both the journal and individual article level.

Ideally, the prioritization layer would contextualize what it retrieves, since metrics like absolute risk reduction carry different weights depending on clinical context. In other words, the same evidence hierarchy that every medical student learns in her first epidemiology course is applied programmatically to the retrieval step.

For example, if you are searching for information about percutaneous coronary intervention for stable angina soon after the publication of the Ischemia Trial, you would want the output to include more than aggregated results: it should surface the reversal, lay out the evidence on both sides, and anchor its recommendation in the strongest recent data.

But this is a research direction, not a deployed fix. So what can clinicians do today?

• First, pay attention to the citations. If the tool is citing papers from 2011 and nothing from the last two years, that’s a signal.

• Second, ask yourself whether the AI’s confidence matches your own. If the tool gives you a clean, unhedged answer on a topic you know is contested, that’s not reassurance. That’s a retrieval system that didn’t surface the debate.

• Third, use the tool the way you’d use UpToDate ten years ago: as a starting point, not a stopping point.

Arthur C. Clarke observed that ‘sufficiently advanced technology is indistinguishable from magic,’ but a good clinician might try to distinguish it anyway.

Why this matters now

A growing number of AI-powered clinical decision support tools are entering physician workflows: OpenEvidence, UpToDate Expert AI, DoxGPT, etc. Though their architectures vary, most share a common structure: a commercial language model provides the reasoning, domain-specific fine-tuning shapes how the model handles medical information, and a retrieval layer determines which documents actually get seen. If that last step is biased, the reasoning inherits that bias.

The bottom line

This problem in AI retrieval systems is important well beyond the handling of reversals. Emerging infectious diseases, newly approved therapies, updated screening guidelines: any domain where the evidence is actively evolving is one where standard RAG is likely to lag. The attending who trained during the ‘go slow on sodium’ era can instantly update her priors when she reads a new study. The retrieval system and vector database cannot.

The practical takeaway is not to avoid these tools but to resist treating them as oracles. The hierarchy of evidence exists precisely because not all studies deserve equal voice, and our retrieval systems haven’t quite mastered that. It’s a solvable engineering problem, but until then, it’s important to recognize these shortcomings and keep a critical eye on the AI in your pocket.

Alex Knapp is a joint MBA/MS Biomedical Sciences student at the University of Chicago, where she studies how digital innovation in healthcare delivery can improve patient outcomes. In a previous life, she was an EMT, a big tech employee, and a consultant. She’ll be at a surgical robotics company this summer, working at the intersection of healthcare, AI/ML, and product strategy.

Share

There are thousands of fatal diseases that can take you out. It doesn’t even have to be disease; a car wreck can be fatal. This fact has always made me see screening for disease in asymptomatic people as foolish. In other words, even if a screening test had perfect sensitivity and specificity it would still fail to prolong life.

A recent observational study of more than 90,000 US veterans’ only strengthens this take. The specific study question obscures the larger message of competing causes of death—AKA other shit that can kill you.

The Study

The UCSD-led team focused on the risk of colorectal cancer (CRC) and all-cause mortality in people age 75 years or older who had (or did not have) an adenoma on a previous colonoscopy. Current guidelines recommend stopping screening colonoscopy at age 75, but what if the patient at age 72 had an adenoma?

[Quick background: adenomas are polyps — abnormal growths projecting from the mucosal lining of the colon or rectum that have undergone dysplastic (pre-cancerous) changes. They matter because colorectal cancer almost universally follows the adenoma → carcinoma sequence. This process takes years.]

Using the VA database the authors made two groups of patients: those with and without an adenoma on colonoscopy before age 75. The main outcome measures were the incidence of CRC, CRC death and non-CRC death and all-cause death.

Of the 91,000 patients (mean age 71), 28% had an adenoma and 72% did not have one.

• At 10-year follow-up, the cumulative incidence of CRC was 1.1% in those with adenoma vs 0.7% in those without adenoma.

• At 10-year follow-up, the cumulative incidence of CRC death was 0.5% in those with adenoma vs 0.4% in those without adenoma.

• The cumulative incidence of non-CRC death ranged from 46.9% to 48.4% at 10 years.

I’ve attached a screenshot from the paper. Look at the red rectangles. In the 25,000 pts with adenomas before age 75, there were 45 patients who died of colorectal cancer over the next decade. But there were 4,408 patients who died of something else.

Many of you may ask about higher risk adenomas. There was no difference in future cumulative CRC risk.

Comments

The authors spend a lot of gentle words in their discussion explaining how it is reasonable to stop doing colonoscopy after age 75—even when there is an adenoma.

I would look at the larger picture.

From the Table above, the 10-year risk of dying from colorectal cancer at age 71 (the average age of this study) was 0.5% even if you had an adenoma. The risk of dying from something other than colon cancer was 48%.

Now do fractions: 48% divided by 0.5% = you are 96 times more likely to die of something else. Not 50% higher; not double the risk, but 96 times more likely.

This study was done in 71-year-olds and we all know that death comes to us all, and the older we get, the higher the odds of all diseases. So a screening advocate might argue that we should keep screening but do it in younger people.

My rebuttal would be what do you think the ratio of CRC death to other death looks like at age 60 or age 50 years?

Well… we can look at the NordICC trial of screening colonoscopy published in NEJM in 2022.

In the 56,000 individuals in the control arm of that study, 157 people died of colorectal cancer (0.31% over 10-years) vs 6079 people who died of any cause (11%). Younger people therefore only have a 36x greater risk of dying from non-colorectal cancer. (By the way, the ratio was the same in the screening arm: 11% vs 0.28%).

Bottom Line:

If you have blood from the rectum, or a strong family history of colon cancer, you should be checked.

But if you are a regular person, how does it make sense to look for one disease when you are 36-90 times more likely to die from something else. I am asking. Tell me.

Subscribe now

“Brain fog,” she said. “From COVID. I had it two years ago. Mild, thank God.”

She’d read the articles, followed the research, and found a name for what was happening to her. Long COVID. A thief that entered through the lungs and stayed in the brain. It made sense. It had a beginning and implied an ending, or at least a plateau. It was something she’d survived. It was her story.

I listened. I ordered the workup.

The MRI showed hippocampal atrophy. The PET scan lit up with amyloid. I sat across from her in the exam room, the images on my screen, and took her story away.

“It’s not long COVID,” I said. “It’s probably Alzheimer’s disease.”

I watched her face as the words landed. She didn’t cry. She didn’t argue. She went quiet; the silence of someone rearranging their future.

Her story had been about recovery. Mine was about decline. Her story had a cause she could point to, a week in 2021 when the virus got in. Mine had no origin, nothing to blame, just plaques on a computer screen. Her story let her stay who she’d always been: a woman who trusted her own judgment, who’d raised her daughters on skepticism and home remedies, who didn’t take pills she didn’t need. Mine made her a patient.

Her daughters came to the next appointment. We discussed medications: donepezil, memantine. I explained the modest benefits and the side effects. She listened, asked questions, then shook her head.

“I’ve never been one for pills,” she said.

The daughters nodded. They knew her. I was beginning to.

I told her we’d monitor and stay in touch. I didn’t push. She’d accepted the diagnosis. Or, at least, she hadn’t rejected it aloud. She wasn’t going to let it change how she lived. I respected that. Autonomy, we call it. Patient-centered care.

Then I mentioned lecanemab.

I didn’t recommend it. I explained the trials, the slowing of decline, the brain swelling, and the bleeds. I was careful. Balanced. I presented the option and waited.

She looked at me for a long moment.

“Do you ever make a recommendation?” she asked.

The question startled me.

I’d been trained to inform, not direct. To offer options, not opinions. Shared decision-making, not paternalism; the new covenant between doctor and patient.

But she wasn’t asking for options. She was asking for guidance. She’d spent seventy-eight years trusting her own judgment. Now she had a disease that would eventually take her judgment away, and she was sitting across from a doctor who wouldn’t tell her what to do.

I said, “If you want to explore lecanemab, I’d recommend seeing a neurologist.”

I was opening a door. She could walk through or not.

She walked through.

The neurologist had his own story. Lecanemab causes brain swelling and microbleeds. She was on Eliquis. The combination was dangerous. Before she could have the Alzheimer’s drug, she’d need a procedure, left atrial appendage closure, a device placed in the heart. Then she could stop the blood thinner. Then she could start the infusions.

I read his note and traced the path. Her story: long COVID. My story: Alzheimer’s. The neurologist’s story: treatable, if we intervene. The cardiologist’s story: procedurally manageable. Each of us adding a chapter.

My referral was the first domino. My non-recommendation was a recommendation.

“I don’t know about this,” she said. “It’s become a lot.”

I asked what she was thinking.

“I went to the neurologist because you said I could. Now there’s a heart procedure. I didn’t want any of this.”

I heard it then: clarity. She knew what she wanted. She’d known all along. She wanted to be the woman who didn’t take pills, who trusted her own body, who’d survived a virus and was living with the fog.

“What should I do?” she asked. “What would you tell me to do if I were your mother?”

Two months ago, I’d been neutral. I did not give guidance. I went with the system’s defaults. In fact, my neutrality didn’t protect her autonomy. It abandoned her to a story she hadn’t chosen. I’d been neutral. And neutral, I was learning, had a direction. It points toward the system’s defaults.

“If it were my mother,” I said, “I’d stop. I’d tell the neurologist, ‘Thank you, but no.’ I’d go home and live the way you’ve always lived.”

I saw her three months later, routine follow-up. She was the same. Slower, maybe, but herself. She told me about her garden, her grandchildren, and a book she was reading about the immune system.

“I’ve been thinking,” she said. “I think a lot of this is long COVID. The brain fog. I had it, you know. Two years ago.”

She’d taken her story back. I don’t know which story is true. I know her story lets her live. I know mine was accurate. I know the system’s story would have made her into a full-time patient, which is the one thing she never wanted to be.

She had asked me if I ever make a recommendation. I recommend the story that lets you keep being yourself, for as long as you can.

Sometimes that’s the medical one. Sometimes it isn’t.

She still calls it long COVID.

I let her.

Michael Sarai is an internist and geriatrician at PeaceHealth Center for Senior Health, Bellingham, Washington

Share

Subscribe now

I didn’t move after 3 years, or 5 years, or 10 years. As I push up against 30 years, many of the people I started caring for in the late ’90s still see me. Although I haven’t aged a single day, these patients are now in their 70s, 80s, and 90s. There are even a few who have reached their 11^th decade. Here is the Age/Gender column from my Epic schedule on the day I begin writing this post.1

The one thing guaranteed to happen to all of us eventually, despite what the longevity gurus might tell us, is that we die. It is fortunate that I’ve had decades of experience before my practice reached its current level of maturity, as I am not sure I’d have been able to tolerate the frequency of deaths early in my career, when I blamed myself for every untoward event.

I’ve written about how I feel when a patient dies.

I’ve also written about saving the face-sheets of all the patients I have lost.

This is about a way of memorializing patients that I have, only recently, stumbled into.

I almost always receive some documentation of the lives of my patients after they die. Sometimes these are emails from the University memorializing a former faculty member. Sometimes these are memorial pamphlets from a funeral. When I don’t receive anything, I usually search for obituaries.

Although it may sound morbid, I treasure reading about the lives of my former patients. These pieces are usually written by those who loved them, learned from them, respected them. After reading them, I usually spend a few minutes thinking about my relationship with the person I knew. My mini-personal-memorial service ends with snapping the face sheet and the obituary — broadly defined — into my Memory Binder.

I never leave this private mini-memorial service without feeling like I didn’t know my patient well enough. The “social histories” that I bury in epic usually say something like:

Retired English professor (Quantrell awardee)

Grew up in Brooklyn (Prospect Lefferts Gardens)

St. Ann’s, Smith, Harvard

Married to SW

3 kids, live in NY, DC, and Austin

Or

Originally from Mississippi

Moved to Chicago in 1960s for work

Worked in steel and automotive

Married over 60 years

Grandchildren: “too many to count”

These summaries are embarrassing, the encapsulation of a life in a 5-line doggerel. They are meant to jog my memory about who a person is for the first couple of visits. With time, I learn more, but seldom update the social history; there is no need. I’ll hear about hobbies and vacations. I come to understand how people relate to healthcare in general and to me in particular. I see how they deal with disease and progressive disability. I learn about the important relationships in their lives. We often discuss end-of-life plans.

What I learn from obituaries is all that I missed. Early career accomplishments, hobbies, volunteering, and interests that enriched lives but never came up in the office. Their qualities for which they were identified and loved.

My job as a doctor, even a primary care doctor, is to care for, and about, people. My job is not to be their friends. Although that sometimes happens — as we get older, more of our patients become friends and more of our friends become patients — mostly, my relationship is close enough to engage in collegial patient care, the shared decision-making that most of us, patient and doctor, aim to achieve.

Still, I regret not having known more about what amazing people my patients were when I knew them.

Today we have rising star, Dr. Mohammed Ruzieh summarizing his recent observational study on left atrial appendage closure in a typical older Medicare population. He’s honest about the study’s limitations, but the findings do two things: a) make you think, and b) reduce enthusiasm for the safety of this procedure. JMM

On the matter of knowledge: In medicine, the most rigorous way to assess the efficacy or harm of an intervention is a randomized trial. In this design, patients are randomly assigned to receive either the investigational treatment or the standard of care, ensuring equal probability of allocation. This randomization minimizes bias and balances both known and unknown confounders between groups, making it the gold standard for determining causal effects.

What are confounders? Confounders are variables that are associated with both the exposure (treatment) and the outcome, and can distort the true relationship between them if not properly controlled. For example, people who carry lighters have a higher rate of lung cancer than people who do not. So, does carrying a lighter cause lung cancer? No, smoking is the confounder. It causes both the behavior of carrying a lighter and the development of lung cancer. The association between lighters and lung cancer is real but spurious, entirely explained by the confounding variable of smoking.

Randomized trials are time-consuming, costly and often enroll healthier patients. Because of these limitations, observational studies are sometimes conducted, as they allow evaluation of large populations in real-world settings. However, their major limitation, especially when comparing treatment versus no treatment, is residual confounding. Treatment decisions are influenced by multiple patient- and physician-level factors; while some can be measured and adjusted for, others are unmeasured or unknown, leading to potential bias. Despite this, observational studies remain valuable for hypothesis generation and real-world evidence.

Today, we discuss our recently published study comparing Watchman versus oral anticoagulation.

Background

What is Watchman? A percutaneous left atrial appendage occlusion (pLAAO) device designed to occlude the left atrial appendage. (Pictures like this belie the complexity of the human body).

What is oral anticoagulation? Medications taken by mouth (e.g., warfarin or direct oral anticoagulants like apixaban or rivaroxaban) that reduce blood clot formation.

Patients with atrial fibrillation can have a fivefold increased risk of stroke, and physicians often prescribe anticoagulation to these patients. Anticoagulants also increase the risk of bleeding. But in patients with enough stroke risk, net benefit occurs because stroke reduction is larger than bleeding increase.

In some cases, the stroke originates from a structure in the heart called the left atrial appendage, a small, pouch-like extension of the left atrium where blood can pool and form clots. The Watchman device occludes this appendage, potentially lowering stroke risk and reducing the need for long-term oral anticoagulants in an effort to reduce bleeding.

The Study

Our study compared pLAAO with oral anticoagulants in Medicare patients with atrial fibrillation to evaluate which approach may better prevent stroke. We included 3,692 patients in the pLAAO group and 11,076 in the oral anticoagulants group (14,768 total – compare this to the largest trial of pLAAO, CHAMPION-AF which included 3,000 patients).

We found that patients who received pLAAO experienced more bleeding leading to hospitalization (5.30/100 person-year vs. 3.25/100 person-year, HR: 1.58, 95% CI: 1.38 – 1.79) but similar rates of stroke and death.

Discussion

We acknowledge that residual confounding may still be present and that our study has limitations, as outlined in our manuscript. Nonetheless, several key findings stand out and support causality.

First, mortality rates were comparable between the two treatment groups, reflecting results seen in randomized trials and reinforcing that we adequately addressed many of the main confounding factors.

More importantly, the mortality rate in our study was approximately 8.4% per year compared to 4.6% over 3 years in the CHAMPION-AF trial. Why is this important? Because CHAMPION-AF didn’t enroll the typical higher-risk Medicare Watchman-recipient and its findings may not apply to them.

Older high-risk patients have more reasons (competing causes) to get stroke other than the left atrial appendage (such as atherosclerosis) and a localized intervention is less likely to be helpful. Furthermore, these patients may be more likely to experience complications from device implantation.

Second, the higher bleeding rate observed with pLAAO in our study is noteworthy. This can be explained by the fact that patients typically receive intensified “blood-thinning” therapy in the first few weeks after device implantation to prevent clot formation on the device, which increases bleeding risk

Additionally, patients typically continue aspirin indefinitely after pLAAO, which may not be safer than apixaban. The device implantation itself can cause bleeding too. While early separation of curves favoring the intervention in observational studies could reflect residual confounding, a point we previously discussed, this is unlikely in this case.

Why? In the case of pLAAO, the early signal of increased bleeding is explained by the procedural risks of device implantation and the subsequent intense anticoagulation regimen. See the figure below from CLOSURE-AF trial, which demonstrates early separation of the curves too. Procedural interventions inherently carry risks, and it is not uncommon to observe initial harm following such procedures.

Finally, the results of our study show a similar direction of signal as the randomized CLOSURE-AF trial, which also included a high-risk population. Since our study was conducted prior to CLOSURE-AF, this consistency further strengthens our confidence in the findings.

In summary, while observational studies have inherent limitations, they can still provide valuable insights. In this case, the real world data support the notion that pLAAC does not reduce bleeding largely due to the early increase in bleeds. It also supports the RCT observations that older more ill patients likely garner less net benefit.

Subscribe now

Dr Ruzieh is too nice to say this but I will. This observational study was rejected by numerous journals before landing in the Journal of Interventional Cardiac Electrophysiology. You might argue that it was rejected because it was an observational nonrandomzied comparison study, which are often marred by bias. That’s possible, but in the many months it was being rejected at various journals, these same journals published a slew of other similar observational studies. Publication bias is not obvious but is important to consider when reading the medical literature. JMM

Adam Cifu

This is my second attempt to write this. My first attempt came out full of anger. In the early days of COVID, we didn’t act soon enough. When we finally did act, many of the restrictions and measures that might have helped us months earlier were too late and unreasonable for the time. We were let down by institutions that we trusted. We were lied to by our government. As an emergency medicine doctor in a major New York City hospital, I have been hailed as a hero. But I have also been denigrated as a liar who would shill for the government to promote a “plandemic” as an excuse to deprive us of our freedoms, and who would falsify death certificates to profit from taxpayer money. I am neither a hero nor a liar.

As I reflect on COVID from six years ago, my two overwhelming emotions are gratitude and sadness.

I’m thankful for the relationships I have with my colleagues throughout the hospital. We supported one another through something we can barely describe to an outsider. I work less often now, and when I step back into the hospital, I greet patient care techs, janitors, security guards, nurses, staff who stock our supplies, doctors in various departments, and others with a real sense of camaraderie, with fist bumps and sometimes hugs. We all went through a trial together, survived, and conquered. We showed up despite our fears and did our jobs as best we could. We helped each other.

I am sad for my colleagues who died. Yes, more than one of my coworkers was killed by COVID in the spring of 2020. And several lived, but only just. Imagine standing at the foot of the bed of a man whose COVID-infected lungs were drowning him when he had stood with you two days before at the foot of a bed of another man whose COVID-infected lungs were drowning him. Except this time, my colleague-turned-patient knew how little we knew and that he would most likely die just as we had helplessly watched many of our patients die. Months later, when he miraculously returned to work, we had a new crop of residents who had missed the worst of COVID. This critical care doctor, survivor, came down from his intensive care unit (ICU) covered from head to toe in protective gear (PPE) when most of us had switched to eye protection and an N95 mask. My new resident, who didn’t know him, didn’t know how we had all watched him nearly die, said, “Maybe he’s overdoing it with the PPE?”

I’m thankful for the ways in which we came together as a country, as a city, as a hospital. New York City was hit hard. I dare you to tell me it wasn’t. And as a city, we came together in the early days. Restrictions were tough, even punishing, but over April and May of 2020, cases dropped to manageable levels. Across the city and hospital, everyone pitched in. People stayed indoors. People wore masks. Schools figured out online instruction. Restaurants brought food to the hospital. Emergency medicine docs worked ICU shifts. Orthopedists and otolaryngologist and podiatrists worked emergency department shifts. People paid their domestic workers to stay home. Restaurants began delivering family meals, and families ordered them to help keep neighborhood restaurants afloat. Instead of dinners and brunches, we met friends on Zoom. Eventually, we met for walks in Central Park.

I am sad for our country that is so politically divided. I think COVID worsened that division. If we were to face another pandemic in our lifetime, I’m not sure we would come together to fight it as well as we fought COVID. In many ways, we didn’t do enough, and in many others, we did too much for too long. Maybe only the vaccine push, Trump’s Operation Warp Speed, was done well. And then that was politicized, and the messaging was bungled.

I am sad that I lost confidence in our scientific and public health institutions. I learned to be a doctor at Emory School of Medicine, on the doorstep of the Centers for Disease Control and Prevention (CDC). Many of their researchers taught my medical school courses. I revered the CDC as an institution that guided us during public health emergencies, and that reverence and trust are gone. Other agencies didn’t present much better. Why was there such reluctance to admit what we didn’t know? Or that we needed help? How will we ever recover this lost trust?

I’m thankful for the experience of treating COVID patients; it made me a better doctor. I have taken care of so many patients with acute respiratory distress syndrome (ARDS) — the basic problem with COVID patients’ lungs. And because the hospital was so overwhelmed, I managed these complicated patients over days in our makeshift emergency department ICU. Since COVID, I have had a few patients who developed ARDS from other illnesses. Recently, I cared for a pregnant young woman, horribly septic with bacterial emboli to her extremities. I saw her X-ray and knew. But I was able to manage her ARDS almost without thinking, which allowed me to spend more mental energy on dealing with the rare, complex infection threatening her and her pregnancy. She lived.

I am sad for the compromises we were forced to make in caring for our patients during COVID. For their discomfort — stuck sitting in folding chairs in the emergency department for days. Because we had converted hospital beds to ICU beds, if all you needed to live was oxygen, there was no bed for you. Hospitals that weren’t overwhelmed were forced to accept transfers of these patients. There were times when I delayed intubating patients because I had one ventilator available and two people on the verge of needing it. I will never know whether that delay led to a death. I will have to live with that. I’ve made a lot of difficult decisions, but was that what I signed up for when I went to medical school, when I became a doctor?

I’m thankful that COVID forced me to re-evaluate my priorities. I work less, I focus on my own physical and mental health, and I spend more time with my children. In a few years, they will have moved on, and medicine will still be there. Emergency departments will still need doctors. Someone once said to me that it’s okay for your life and work to have seasons, and COVID pushed me to take that advice. This season of my life still includes being a doctor. I’ll always be a doctor - and a plague doctor at that. But I’m focusing much more on being a mom. I’ll always be a mom too.

I am sad for the way that we couldn’t protect our children. Schools were closed for so much longer than necessary. The playgrounds closed. My own children were afraid of me, trained not to go near me after I came home from work. This lasted for years. My teenager has only just begun to let me hug her again. She still sometimes asks, “Have you been to work today? Did you shower?”

I’m thankful that COVID has become less virulent and that the vaccine prevents deaths (it really did save us at the time). I’m thankful that I have evolved as a doctor and as a person. I’m thankful for what we learned, overcame, and survived. But I am also filled with sadness for everything we lost.

Caption: Modern Plague Doctor. By Colleen Smith. Colored Pencil

Dr. Colleen Smith is an Emergency Medicine physician and educator who practices in New York City. She is a visiting fellow for the Center for Modern Health think tank, a Roots of Progress Institute Fellow, and writes a Substack about medicine, health, and related topics.

On March 27^th, I saw CW, a 60-year-old man with an LDL cholesterol of 224 and CAC at the 52^nd percentile.

I’m a total EBM devotee. EBM gives us the best chance of practicing effective, patient-centered, parsimonious medicine. EBM is the process of integrating clinical experience and expertise with the best available evidence from systematic research. EBM tends to de-emphasize intuition, unsystematic clinical experience, and pathophysiologic rationale as sufficient grounds for clinical decision making. EBM stresses the examination of evidence from clinical research.1

EBM has its shortcomings, many of which were predicted by Alvan Feinstein. Feinstein was a product of the University of Chicago, receiving a bachelor’s in 1947, a master’s in mathematics in 1948, and his MD in 1952. Although we will, of course, claim him, he made a name for himself at Yale. He was a giant in the field of clinical epidemiology and a founding editor of the Journal of Clinical Epidemiology

Feinstein was a skeptic of EBM from the start. He noted that a minority of clinical decisions can be addressed by RCTs – the ultimate source of truth in EBM. He argued that EBM mostly focuses on hard outcomes, while most of medicine focuses on improving soft, difficult-to-measure outcomes. For me, he was most articulate at highlighting that the evidence generated in studies is for an average patient, one who does not exist. The patients we care for, like all of us, are unique. Two patients might have the same disease, or risk factor, but they differ in the severity of their symptoms, their co-morbidities, their genetics, and everything we pay attention to in the past medical, social, family histories, including previous treatments and treatment trials, ability and interest in adhering to therapy, functional status, and on and on.

Clinical practice guidelines, which have only become important since the passing of Feinstein in 2001, grew out of EBM. Recommendations come with a strength and a quality of evidence. The strength of a recommendation considers the balance of benefits and harms, while the quality is based on the type of study supporting the recommendation.

The pros and cons of guidelines have been discussed endlessly. The longer I spend with them, the less of a fan I am. I understand that they help standardize care, and they certainly make it easier to find basic recommendations on things you know little about. But the more I read, the more I feel that guidelines are overly influenced by people with either financial or intellectual conflicts of interest. The power on the guideline writing committees is seldom held by experts in research analysis, but rather by content experts. Too many recommendations are “strong but based on weak evidence.” This is how we parent, but it should not be how we practice medicine. I also see trainees relying on guidelines and not thinking enough about the patient in front of them

I have finally finished reading the new cholesterol guidelines. I don’t think Feintstein would be surprised by how divorced they are from any consideration of value. As noted, a strong recommendation is given when the benefits of a treatment far outweigh the risks. In the cholesterol guidelines, we are mostly talking about statins, exceedingly safe drugs. Thus, most recommendations are strong. This figure is telling.

The authors define treatment threshold based on one side effect that could be quantified, the balance of cardiovascular benefit against the likelihood of incident DM. These are not equivalent, or even comparable, outcomes. They neglected to consider so many things. There is the cost of the medication, the monitoring, the visits, and the turning an enormous swath of the population into patients. Treating a patient with a statin who has a 3% 10-year risk might reduce the risk to 2% — helping someone with a 97% chance of doing well reach 98%. Is this how we want to spend our time, our effort, and our money?

Of course, the guideline suggests shared decision-making.

With a different value set, a different view of medicine, you could imagine the guidelines being a fraction of their 134 pages. If we concentrated on treatment and only the highest yield prevention, the guidelines would focus on secondary prevention and primary prevention in those with familial dyslipidemia. I recognize this might sound like the ramblings of a crazed minimalist — heck, I am not even sure I agree with myself — but it seems less crazy when you see algorithms that include this:

If we are serious about spending less money on healthcare AND getting better outcomes, we might start by thinking seriously about the actual clinical effect of the money we spend.

Let’s get back to CW and Dr. Feinstein. CW adamantly refuses to begin a statin. The irony of seeing CW on the day I read the guideline was glaring. Anyone familiar with preventative cardiology data would agree that he should be on a statin. However, given his “Feinsteinian clinical nuances,” he is utterly unaffected by evidence or guidelines. There are certainly people who err (my value judgment) in the other direction, people more aligned with the guideline writers, those with a 0-3% 10 year risk who choose to take medications (as long as they are not directly charged for them) or change their lifestyle for the possibility of a 1% risk reduction.

When I think about how EBM actually affects my practice, I recall this figure from a 2024 Sensible Medicine post by Mariana B. Caiado Ferreira.

I could re-label this diagram like this:

My practice depends on my practicing EBM but, most of the time, my patients’ clinical state, circumstances, preferences, and actions crowd out EBM. I often end up doing something only vaguely related to the studies or the guidelines.

• World Mosquito Program

Antibiotic Therapy for Uncomplicated Acute Appendicitis

Here, Dr. Develtere takes us through a recent article in which an improved PFS masks more important findings.

Adam Cifu

Modern cancer care increasingly relies on combination therapies—stacking multiple drugs together in the hope that more treatment upfront leads to better outcomes down the line. The phase III COSMIC-313 trial is a good example of this trend. It tested whether adding a third drug to an effective regimen could further improve outcomes for patients with advanced kidney cancer.

The study results looked encouraging, with the triple therapy showing improvements in progression-free survival (PFS). A closer look reveals a more complicated story—one that raises broader questions about how we interpret clinical trials.

The COSMIC-313 Trial

The COSMIC-313 study enrolled patients with previously untreated advanced clear-cell renal cell carcinoma, the most common type of kidney cancer. Patients were randomly assigned to receive one of two treatments:

Standard treatment (doublet)

• nivolumab + ipilimumab (two immune checkpoint inhibitors)

Experimental treatment (triplet)

• cabozantinib (a targeted therapy)

• nivolumab

• ipilimumab

The hypothesis was that adding cabozantinib, a drug that targets tumor growth pathways, might enhance the immune therapy and improve outcomes.
The primary endpoint was progression-free survival (PFS)—the time until the cancer worsens. After a median follow-up of 45 months, the results showed an improvement in PFS 11.2 months in the doublet group compared to 16.6 months in the triplet group, Notably, there was no difference in overall survival between the groups. Patients receiving three drugs experienced a delay in tumor progression, but they did not live longer.

The Hidden Cost: Much More Toxicity

The triplet treatment came with a substantial downside: far more toxicity. Severe treatment-related side effects occurred in 75% of patients receiving the triplet and “only” 43% of patients receiving the doublet. Nearly half of the patients on triplet therapy had to stop at least one drug because of side effects. In the end, patients receiving the triplet therapy experience more severe side effects, spend more time on a more toxic treatment, but do not live longer.

Why PFS Without OS Can Be Misleading

Progression-free survival is commonly used in cancer trials because it can be measured earlier than overall survival. This can translate into faster approval and earlier patient access to the therapy. The progression part of PFS can be a surrogate endpoint, an endpoint that is invisible to patients, maybe only visible to radiologists.

PFS does not necessarily translate into meaningful patient benefit. One way to understand this is to think about the entire treatment journey. If patients stay longer on a first-line treatment that is more toxic, but eventually receive similar second- and third-line treatments, then the overall sequence of therapies remains the same. This is fine if patients live longer (and maybe better) with the new treatment. But if they don’t, at least the extended time on the first-line treatment should be better tolerated compared to the alternative. If the treatments are equally tolerated, the new therapy regimen should be cheaper or more convenient. But if there is no apparent benefit, you might want to stick with the well-established treatment regimen.

In COSMIC-313 trial, treatments that were given after there had been disease progression were similar between the two groups. Without an OS benefit, this means patients receiving the triplet therapy spent more time on the toxic first-line treatment and consequently spent less time receiving later therapies.

From a patient perspective, the result is straightforward: More toxicity, without longer life. Although the positivity of cancer trials often rests on improvements in PFS, meaningful outcomes for patients remain: Do I live longer? Do I live better?

In COSMIC-313, the answer appears to be that patients don’t live longer, but possibly live worse, due to more side effects. There is also a societal dimension. Triple therapy means higher drug costs, more toxicity management, and longer exposure to expensive medications. All without clear evidence that patients ultimately benefit.

A Reminder for the Future

COSMIC-313 is not a failed trial—it provides valuable information. But it also reminds us that more treatment is not always better treatment. When new combinations increase PFS but fail to improve survival, we need to look for another benefit before accepting a new treatment regimen. We want patients to live longer and/or better, and if that’s not the case, keep “First, do no Harm” in mind.

Share

Dries Develtere, MD, is a Urologist at the General Hospital of Ypres, Belgium. He specializes in robotic surgery, with a focus on prostate and bladder procedures, as well as urological oncology. He is the founder of Surgical Vision, a video platform that provides high-quality surgical training videos designed to advance surgical education.

CHAMPION AF was the second trial; Hi PEITHO went first. I was only partly paying attention to the trial of an invasive ultrasound-facilitated catheter-directed fibrinolysis treatment of intermediate-risk patients with pulmonary embolism. What caught my ear was the jubilant presenter and discussants. This comment stuck in my brain:

A new era in PE care

I thought to myself: John, you’d better look under the hood of this trial.

Some brief background: a pulmonary embolism occurs when a clot gets lodged in the large pulmonary arteries. If the clot is large enough, oxygen levels drop and the right heart can fail. The mainstay of therapy is anticoagulation, which facilitates the body’s clearance of the clot. When clots are large, and patients sick enough, invasive measures have been proposed for more direct dealings with the clot.

The invasive measure tested in HI-PEITHO is Boston Scientific’s EkoSonic catheter which delivers ultrasound and fibrinolytic therapy right into the clot-ridden pulmonary arteries. The ultrasound is the novel part: it’s supposed to create tiny fluid currents within the clot, thereby disrupting the clot and facilitating the clot-busting fibrinolytic. (I know; it’s a nice story; Medicine overflows with these.)

Stories are nice but they require testing in randomized trials.

The Trial

HI-PEITHO was a multinational open label trial of ultrasound facilitated catheter directed fibrinolysis for intermediate-risk PE patients. Inclusion required an RV/LV ratio >1 and an elevated troponin and two indicators of cardiorespiratory (CR) distress (SBP <110, HR ≥ 100 or RR > 20.

One group gets US-facilitated alteplase plus anticoagulation (AC) and the other gets AC alone. The primary outcome is a composite of PE-related death, cardiorespiratory (CR) decompensation or symptomatic PE recurrence.

We need to talk about the primary endpoint components. PE-related death and PE recurrence are straightforward; CR decompensation was more complex. It included five items: cardiac arrest, signs of shock, placement on ECMO, intubation or initiation of noninvasive mechanical ventilation, or a National Early Warning Score (NEWS) > 9 by two consecutive measures 15 minutes apart.

Four of the five components of the CR decompensation endpoint are obvious. But I had never heard of the NEWS score. It’s an ordinal score with each item ranging from 0-3 for bedside physiologic signs including respiratory rate, oxygen saturation, supplemental oxygen use, temperature, blood pressure, heart rate, and level of consciousness. Here is a picture:

The key aspect of this score is that a clinician (possibly a research nurse) makes these judgements, such as respiratory rate of 19 vs 21. I looked up the NEWS score and learned that it was designed as a surveillance and escalation trigger rather than an endpoint for trials. One more fact: HI-PEITHO was open-label so clinicians knew the patients treatment assignment.

The authors screened 4300 patients and enrolled only 544. The maker of the EkoSonic endovascular system, Boston Scientific, sponsored the trial and had representation on the executive committee. Co-sponsors include the University of Mainz and the PERT consortium—which is a nonprofit receiving money from industry. The statistical analyses were performed by a contract research organization, with biostatistics planned by a Boston Scientific employee.

Results

The two arms had about 270 patients.

A primary-outcome event occurred in 11 patients (4.0%) in the intervention group and 28 (10.3%) in the control group (relative risk, 0.39; 95% CI, 0.20 to 0.77; P=0.005).

The drivers of the primary.

• PE related death higher in the active arm (3 vs 1)

• Recurrent PE were each 1

• All the difference was lower rates of CR decompensation in the active arm 10 vs 28

So let’s look into the supplement and see about this 10 vs 28 difference.

• Cardiac arrest and CPR 6 vs 4

• Signs of shock better 5 vs 8

• ECMO 1 vs 3

• Intubation 5 vs 6

• NEWS score > 9 2 vs 19

The primary endpoint was completely driven by bedside score of items judged by a human clinician, like level of consciousness, taken twice 15 minutes apart. Hard outcomes, such as PE-related death was higher, and there was no difference in PE recurrence.

14 of the 19 patients in the control arm who had [NEWS > 9] had only that as the endpoint.

On safety, major bleeding at 72 hours was higher in the Ekosonic arm [3.7% vs 1.5% (RR 2.5, 95% CI 0.8–7.9, p=0.17)]. The p-value is non-significant but there were 2.5x more bleeds in the intervention arm. It used thrombolytics, so it should not be a surprise nor cast off as noise.

At 30 days there were 6 deaths in the invasive arm vs 3 in the control arm. There was also no difference in the 6-minute walk test was (405 vs 393 meters).

The authors declared victory and the discussants called it a new era in PE care. The editorialists wrote that this trial “represents a major advance in the foundation of evidence for thrombolytic approaches.”

Comments

I am not sure why there was so much celebration.

• The “positive” result is driven by a soft physiologic score (NEWS >9) in an open label trial. All hard endpoints were clearly not different and trended against the intervention.

• Bleeding was 2.5x higher in the invasive arm.

• The trial enrolled a highly selected population where 87% screened were not enrolled.

• There was no overall mortality benefit and if anything, numerically went the wrong way for the EkoSonic catheter.

• Functional outcomes were also nearly identical.

To be fair. PE trials are very hard to do, because presentation is often variable, and, by nature the selection has to be rigorous or you stand a chance to have too much noise—that is, patients who are not sick enough to benefit or too sick to benefit.

That said, however, the lack of signal of improvement in hard outcomes stands out. The driver of the positive endpoint was a prediction score that requires clinician judgement. Clinicians, in this case who are both proponents of invasive measures and aware of treatment assignment.

I would not call this a win for ultrasound directed fibrinolysis. Not at all. If I were a regulator I would reject this therapy.

I mention this trial here on Sensible Medicine because it is another great example of looking deeper than topline results and studying the supplement.

There is also a written review and podcast of this trial over at Cardiology Trials Substack

Subscribe now

I thought about this story while reading Pradyumna Sapre’s essay. I have written about how I worry that our application process leads us to miss some promising students. I am glad Mr. Sapre has found a home, and I look forward to following his trajectory.

Adam Cifu

In the first draft of my medical school personal statement, I referred to Kazimir Malevich —my favourite Russian modernist painter’s—formalist compositions, Tolstoyan wisdom about tending to the dying in Anna Karenina, a story about humming Dvorak’s Ninth Symphony at a patient’s bedside, and how the Bergsonian notion of consciousness and time might be relevant in neurology.

One of my best friends from college was my first reader. He was unsurprised by the essay. He told me to cut “some of the more arcane humanities references” in the gentlest possible way, aware that he was killing my darlings. I begrudgingly culled Bergson and Malevich, though I couldn’t bear to sacrifice Tolstoy and Dvorak. To compensate for the loss, I crafted my personal statement around how the humanities call me to medicine—how the humanities have enabled me to connect most with patients in moments of great vulnerability.

When I showed it to a less biased reviewer, he said, “What is this? You should write about basic science instead.” I ignored him, added a sentence about what the aesthetic self-consciousness of Realist art teaches us about intellectual humility in medicine, and sent in the essay.

***

I entered college in 2020, aspiring to be a physician but with an enduring love of the humanities. I knew early on that I didn’t want to fit the typical pre-med archetype: perennially stressed and always complaining about an upcoming orgo exam.

Early on, I invested in the humanities and in a diverse range of experiences. I spent more time writing opinion columns than on EMS shifts. I joined a sketch comedy group. When the sight of micropipettes began to make me feel nauseous, I “quiet quit” my wet lab. I loved volunteering every week at Yale New-Haven Hospital, chronicling the stories of patients I met there and uploading them to their charts, but I took on an informal leadership role instead of running for President of the club.

Buoyed by the adulation of well-wishers and a loose philosophical conviction whose origin I can’t place, that “well-rounded people make good doctors,” I placated my own concerns about whether I was doing enough to prepare for application season. I vowed to tick the necessary premed boxes but to focus more on developing an identity outside of medicine.

I viewed pre-med culture as antithetical to the liberal arts model, where college years offer a broad intellectual, social, and cultural education that expands the mind and heart and lasts far beyond college’s cloisters. Focusing disproportionately on basic science research I didn’t enjoy, accruing titular positions in the name of padding my CV, chasing additional service experiences in pursuit of a 15^th AMCAS activity felt like spending four years at an intellectual carnival on a single, rickety ride, without even chit-chatting with my co-riders.

And yet, in the months after I graduated, as I began to stare down the barrel of impending applications, I was gripped by anxiety around whether I had done enough. By working in a lab for two years instead of four, or by refusing to dedicate my gap year to more typical clinical experiences, was I avoiding unnecessary “weed out” tasks or just dressing hedonism—a prioritization of things I enjoyed doing—in the extravagant, inflated costumes of a “liberal arts education”?

I spent my gap year more smoothly balancing these dual identities with greater ease: the dedicated future doctor who volunteered at a hospice facility every weekend, with the sensualist who spent Tuesday evenings basking on a picnic blanket and reading Elena Ferrante in Central Park.

When I eventually applied to medical school, I did so with top grades, scores, and pedigree, and yet I worried my extracurricular profile was more “interesting” than “impressive”—from a purely “premed” lens. My conviction to be a physician had never wavered, and I hoped that my clear love of the humanities and my investment in community would only help contextualize my clinical experiences. Wouldn’t you want a doctor who was also a good friend to the people in his life?

My future success in the medical school application process began to feel like a referendum on my decision-making. Faced with skeptical mock interviewers who told me that most doctors believe the humanities are a mere luxury, stories of medical students who published scores of articles on diabetes management before their first year of medical school, and the general sense that there was nothing special about me from a cynic’s standpoint, I began to feel like an imposter to my own convictions.

A lack of success would be incontrovertible evidence that the notion “interesting people make interesting applicants,” which I had blithely parroted to myself, was fraudulent.

If I could go back in time, I would do many things differently. I’d have started publishing health policy research—as I did in my senior year—earlier in my college career. I’d probably have sought out service experiences that better fit my personal identity. I would have played the game more.

I don’t believe that the ability to enjoy wet lab research is a prerequisite for future success as a physician. Nor do I doubt the thrust behind the idea that people who’ve spent more time honing soft skills or gaining interdisciplinary knowledge might be better able to relate to patients. But I wish I had saved myself the heartache of knowing I did less than I could have to optimize my application for success.

On the other side of the process, as I look forward to matriculating to medical school, I’m filled with nothing but excitement. I can’t wait to spend my days thinking about physiology and pathology; I can’t wait to feel like everything I am studying could tangibly change a life. I already read Sensible Medicine in my free time; I’m excited to understand more of it.

I’m grateful to have interviewed at many excellent schools and committed to a program where both my interviewers told me to dream big, that their school is the perfect place for future physicians with interests beyond the canonical triad of clinical work, research, and education: for physician-innovators, physician-writers, and physician-humanists.

Sometimes, I still think about the programs I didn’t interview at. The admissions officers who tossed my application on the no pile because I had more research hours in Russian art history than in RNA biology. The interviewers who thought that my humanistic understanding of medicine was “precious” or that my love of the humanities was “pretentious.” In rare moments, it is still a hard pill to swallow. It remains easy to caricature myself as a lazy sybarite, not having made the most of every pre-med opportunity to optimize for the selection criteria of my future career, opting instead for things that fulfilled me.

And then I remind myself, even if I had made every traditional choice or done one more interview, I’m not sure I would have thrived at a traditional program that doesn’t encourage a diversity of interests—within and beyond medicine. I’m excited to nerd out over heart murmurs, septic shock prevention, and the art of diagnosis with new friends in medical school. But I’m also excited to share my reflections from attending the Ben Shahn exhibit at the Jewish Museum with them, as well as with the art historian whom I care for during my clinical clerkship.

I could have been a much better pre-med. But I don’t think it would have made me a better doctor.

Share

Pradyumna (Pradz) Sapre is an incoming medical student at Weill Cornell Medical College in New York City. He is a recent graduate of Yale University with a B.S. in molecular biochemistry and a BA in the humanities, focusing on Russian literature and art history. He wrote a regular opinion column of personal essays and anthropological observations while in college, a practice he hopes to continue throughout his medical career.

The idea behind the course was this: By the time a medical student reaches her senior year, her medical knowledge has come from textbooks, review articles, and lectures. From that point forward, however, continuing medical education depends on the ability to learn from the cases and to assimilate new data as it appears in the medical literature. CALML was meant to refine students’ skills in this type of learning.

I also hoped the course would provide a foundation in the medical literature by reviewing key studies. The syllabus for the course included an index of “landmark articles.” These were articles people frequently referenced, should be referencing, or were great for teaching critical appraisal. I edited the list every year, adding articles that students had presented and ones I had come across.1

When I read, Sodium Correction Rates and Associated Outcomes Among Patients With Severe Hyponatremia: A Retrospective Cohort Study, I thought, “Now this is an article I would have added to the syllabus.” We discussed it on This Fortnite in Medicine XXI, but I wanted one more say about it.

Background

Not a week goes by on a general medicine service that someone isn’t admitted with hyponatremia. The differential usually includes heart failure, cirrhosis, and SIADH, with some kidney disease and potomania thrown in for good measure. Any internist worth her salt will excitedly retreat to the whiteboard, calculate sodium and fluid needs, and design the fluid and electrolyte resuscitation.2 Repleting the sodium slowly means the patient spends more time in the hospital (and if it is really low, more time at risk of seizure). Repleting the sodium rapidly increases the risk of osmotic demyelination syndrome (ODS).3 What is interesting about these two competing outcomes is that the former is kind of the fault of the disease, while the latter is kind of the fault of the doctor. Hence, the chagrin factor for ODS is higher than for a prolonged, complicated hospitalization.

Design

The Annals article was impressive. It is a retrospective cohort study done at 21 hospitals over 15 years. Patients hospitalized with a serum sodium under 120 mEq/L were included. The exposure was the maximum 24-hour rate of sodium correction: slow (< 8 mEq/L), medium (8-12 mEq/L), or fast (> 12 mEq/L). The endpoint was a composite of 90-day death or delayed neurologic events (from 3-90 days). This is a perfect endpoint as it includes the negative outcomes of going too fast or too slow.

Results

Almost 14,000 patients were analyzed in this study. About 6,000 patients were in the slow group, 3,600 in the medium group, and 2,500 in the fast group. 1,800 patients had missing data. The fast repletion group had lower initial mean sodium levels and more people with sodium levels < 115 mEq/L. The primary outcome occurred in 21% of the patients – this is a sick group. The primary results are shown in the table below.

In short, the patients whose sodium returned to normal the fastest did best, though there was no statistical difference between the fast and medium groups. Demyelinating disease was exceedingly uncommon, accounting for < 1% of outcomes in each group. There were no subgroups in which the faster groups did worse.

Analysis

There are two ways of interpreting this data. First, it might be a true positive, indicating that patients whose sodium is corrected more quickly do better. If this were true, it would be a great example of a medical reversal; a practice that we have embraced for years would be wrong. What would have driven our faulty therapy would have been our fear of doing harm. This might be similar to years of withholding seizure medications, which we later learned were safe, in pregnant women.

This study could also be a false positive, probably due to confounding. Because this was an observational study, the patients in the three groups differed by far more than how quickly their sodium was repleted. Doctors who repleted sodium faster were different from those who went slow. As shown in Table 1, patients who had their sodium repleted faster received different medications. Thus, it might not have been the rapid sodium correction that helped the patients but the rest of their care. The patients themselves were also different. We might not be comparing a strategy of rapid vs. slow correction, but patients who rapidly correct themselves vs. those who slowly correct.4 As they always say, the smartest nephrologist knows less than the dumbest kidney.

The authors made extensive adjustments, but residual confounding remains possible. We know that only about 80% of observational studies predict the results of RCTs of the same question. Which brings us to what we do with this study. These results certainly open the door for an RCT comparing strategies of repleting sodium in patients with hyponatremia. A study published in the NEJM of patients with mild hyponatremia suggests that a future RCT of treatments for severe hyponatremia will give similar results as this observational study.

What do with patients now? It is a tough question. On the one hand, we should not change care based on observational studies. On the other hand, our standard of care is probably based more on tradition — and the worry about causing harm — than on treating patients the right way.

Read more

I have always been put off by other terms for my job other than doctor. I’ll accept ‘physician’, though it sounds a little self-important. I hate the term ‘provider’ and ‘practitioner’ almost as much as I despise it when people refer to my patients as ‘clients’. However, when I read the ACP paper that Dr. Huddle discusses here, I was put off. I could not really articulate why I felt that way until Dr. Huddle did it for me.

Adam Cifu

The American College of Physicians (ACP) views itself as a guardian of professional identity—important for professionals and their work. In its latest attempt to encourage physicians to be their best selves, the ACP is concerned about how physicians are named in public discourse and how those names may alter professional self-conception. “Physicians are not providers,” we are told. “Provider” fits those offering goods or services to buyers in commercial transactions. It is an inaccurate and reductive term for professionals who care for patients. Physicians must be careful to avoid “motivations economic and not ethical.” The forces of commerce threaten the professionalism of professionals; they must resist their blandishments, including any labels suggesting an economic aspect of clinical work.

The ACP’s position paper reflects a longstanding professional conceit—the assertion that professionals care about their clients, unlike the unfortunate creatures in the world of business and commerce who care only about profit. The ACP grudgingly admits that commerce exists in medicine but insists that it must be kept firmly within bounds—in this case, by avoiding any language connoting or alluding to its existence in professional work. Any whiff of the commercial endangers the ethical in professional life—or so we are led to believe.

This erroneous stance—valorizing professionals as champions of the ethical, not to be confused with those whose motives are purely acquisitive—issues from a long and not wholly mistaken tradition of thought about professional work. It is nevertheless an error and, paradoxically, subversive rather than protective of a proper professional self-conception. The grain of truth in it is that physicians are not merely providers; that is, the care they provide is an economic good—valuable and scarce—but they offer it in a different key than do providers in the world of commerce. The essential difference is in the degree to which ends are shared between buyer and seller and in the seller’s regard for the good of the buyer. In ordinary commercial transactions, sellers must deal fairly and consider the customer’s interest, but need not make the customer’s good their own in the way a physician must.

The physician’s provision is more demanding. She must not only provide competent care; she must discern the patient’s individual good and tailor her care to that good. Patients are vulnerable, and physicians must respond to that vulnerability with judgment, nurturing, and self-restraint at the margin of professional actions, possibly beneficial to the physician but not the best way forward for the patient. In the world of commerce, the customer determines her good not unaided but mostly uncontested; in the world of medicine, the physician partners with the patient to clarify as well as to further that good, keeping her own pecuniary interest in check. [AC1]

That said, it is deceptive to suggest that physicians are not economic as well as ethical actors or that physician-patient interactions are not transactional as well as relational. And it is a caricature of business ethics to assert that “business transactions need not focus on values or consider the interests of consumers before those of stockholders and owners.” The ACP is not wrong to contend that commerce and its norms may subvert professionalism. The remedy is not, however, to assert that commercial considerations should play no role in professional work.

In place of a supposed contrast between “ethics” and profit-seeking, the ACP would do better to articulate the more nuanced differences that actually distinguish professional from business norms. In the course of business, businessmen and women seek to gain, but in doing so also help customers to advance their own interests. They must bargain fairly and deliver the promised goods for the agreed-upon price. Professionals share self-interest with businesspeople, but unlike them, subordinate that interest to more demanding fiduciary obligations. That both are paid for services demonstrates their kinship without abolishing their differences. Ideally, both are ethical; both are also (economically) self-interested.

Physicians should welcome analyses of professional work that focus on its economic aspects, including the physician’s role as “provider”. The task for professional ethics is to discern practice arrangements that can be economically competitive and efficient while also encouraging physicians to offer patients the wise and compassionate guidance their profession demands in the course of providing excellent care. It is also to encourage an accurate view of professional identity. Physicians are and ought to be self-interested; but that self-interest should be capacious enough to include professional ideals alongside economic concerns, with the latter properly disciplined and subordinated. That might be achievable if these aspects of professional self-interest are allocated their due and proper limits in the professional’s self-conception. It is less likely to be if the commercial aspects of professional work are obscured by the suggestion that commercialism is evil or, at best, deeply suspect.

Physicians need not be threatened by being named “providers” unless the suggestion is that the patient visit is merely a commercial transaction—which, to my knowledge, is almost never the case when physicians are named “providers.” In the right context, the term serves a useful purpose in highlighting a particular economic aspect of professional work. The danger is not in acknowledging that aspect, as the ACP contends; it is in supposing that professional ethics are somehow strengthened or protected by denying its existence. Physicians should be secure enough to admit that they care about income while also aspiring to Francis Peabody’s ideal of really caring for the patient in their work. Their task is to live in the tension between these aspects of who they are, not to deny the existence of one of them.

Thomas Huddle, MD is Professor Emeritus of Medicine at the UAB Heersink School of Medicine. He served on the ACP Ethics, Professionalism, and Human Rights Committee from 2018 to 2022.

Share

It’s both one of the most biased and most consequential studies I have reviewed in years.

The question: for the millions of people with atrial fibrillation (AF) who take oral anticoagulation (OAC), does the Watchman Flx left atrial appendage closure (LAAC) device provide an alternative to OAC for stroke prevention?

The stakes for society, patients, the medical profession and of course the company are massive, since millions of people take OAC for stroke prevention.

It’s also a reasonable question because if mechanical occlusion of the left atrial appendage worked then a one-time procedure might be able to provide lifelong protection from stroke without taking tablets for anticoagulation.

You would think such a consequential question would be studied in a rigorous flawless trial. Alas, that was definitely not the case.

CHAMPION Trial

The authors randomized 3000 patients into either a Watchman arm or OAC arm. Modern direct acting oral anticoagulants (DOAC) were used for the medical arm.

Patients were fairly low stroke risk at age 71 and CHADSVASC score of 3.5. They also had a low HAS-BLED score of 1.3, indicating a very low bleeding risk.

You find most of the trouble in this trial design in the choice of endpoints and their analyses.

Two Problem with the PRIMARY EFFICACY ENPOINT

The primary efficacy endpoint is stroke, systemic embolism and cardiovascular death tested with non-inferiority. Recall that it in a noninferiority design, no difference in outcomes is considered a positive.

The issue here is adding CV death to the endpoint; everyone accepts the fact that neither Watchman nor OAC will affect CV death. Therefore… adding an outcome to a composite endpoint that will not be affected by either treatment simply adds outcomes (noise) in both arms making noninferiority easier to reach. (Thing is that all LAAC trials do this, so CHAMPION AF authors aren’t deviating from bad practice.)

The far more problematic issue was the choice of noninferiority margin. In a noninferiority trial, the authors, often in discussion with regulators, choose a margin that the new therapy should be no worse than. But to do this, you first have to estimate an event rate in the control arm. Here, they expected a 12% rate of primary outcome events in 3 years. They then chose 4.8% as the NI margin. IOW: if the upper bound of the 95% confidence interval for the absolute risk difference was less than 4.8%, the Watchman would be declared noninferior for efficacy.

Now we tip-toe into fractions. Sorry. 4.8% higher than 12% is equal to 40% or 1.4. Translation: the relative risk margin is 1.4. In most if not all drug noninferiority trials, the authors set out NI margins in both absolute and relative terms. CHAMPION AF authors did not do this. Their noninferiority margin was just 4.8% in absolute terms.

Can you guess the problem?

In CHAMPION AF, the event rates came in much lower than 12%. They were 5.7% in the Watchman arm and 4.8% in the OAC arm. That risk difference was 0.9% higher in the Watchman arm; the 95% confidence intervals were -0.8-2.6. Since 2.6< 4.8, the authors declared noninferiority. And there it sits in the NEJM, on the scoreboard if you will, as a win.

The problem though is that when event rates come in that much lower, the 4.8% margin is much too lenient. In normal trials, the authors would have also tested NI with a rate ratio or relative risk.

As I said above, the margin in relative terms is 1.4. The rate ratio (or relative risk difference) comes out with a Hazard Ratio of 1.20 (20% higher) but 95% confidence intervals of 0.87-1.66). Since 1.66 > 1.4, the device does not make NI on relative terms. Why the editors or regulators let the trialists get away with not using both rate ratio and risk difference is hard to explain.

Two Problems with the PRIMARY SAFETY ENDPOINT

The authors chose nonprocedural clinically relevant nonmajor bleeding as the primary safety endpoint. The results favored Watchman (10.9% vs 19.0%) and easily met superiority. Again, another win on the scoreboard.

The first problem is that exactly zero patients can exclude bleeding from the invasive procedure, which could be bleeding from venous access or pericardial tamponade. Both are bad. The authors just exclude that.

The second problem is that CHAMPION AF is an unblinded trial wherein patients know the treatment assignment. Patients on oral anticoagulants are far more likely to complain of nonmajor bleeding like bruising, nose bleeds or gum bleeds. This was well demonstrated in the OPTION trial, wherein the authors listed all nonmajor bleeds in the supplement. CHAMPION AF did not list these bleeds.

The Secondary Safety Endpoint is Actually the Proper Safety Endpoint but its Statistical Test was Incorrect

The secondary safety endpoint counted all major bleeds including the procedure. It was 83 in the Watchman arm and 87 in device arm. The HR was 0.92; 95% CI 0.68-1..24). This easily met noninferiority but it was not tested for superiority. So, there it sits, another win on the scoreboard.

The problem is that the standard for noninferiority trials is to test the safety endpoint with superiority. The reason for this is that if you are giving up some efficacy, the new treatment should offer something important, such as safety. In this case, when bleeding is counted properly, the device is clearly not superior.

Looking at the Actual Numbers is Even More Sobering

The authors provide a “net benefit” analysis where strokes are placed against bleeds. Their calculation is positive but that’s because they use nonprocedural bleeding.

Let me show you what I feel is a fairer way to assess net benefit: Stroke rates were higher in the watchman arm: 50 vs 33. Ischemic strokes were 45 vs 27. Hemorrhagic stroke was 5 in each arm. Major bleeds were lower in the Watchman arm but there was a difference of only 4 (83 vs 87).

A sober patient looking at these numbers would see that there are 17 more strokes and only 4 less bleeds. Since strokes are literally one of the worst outcomes a person can have (because of disability), this is not a good trade.

Placing CHAMPION AF in context with previous Watchman vs OAC trials.

The original trials of Watchman vs Warfarin were essentially negative for Watchman. PROTECT did not pass FDA muster; PREVAIL found Watchman not noninferior to warfarin (due to increased strokes) in the co-primary endpoint of stroke, systemic embolism and CV death. In November, I reported on the German CLOSURE AF trial, which compared left atrial appendage closure with Watchman (and other devices) vs best medical therapy. CLOSURE AF found that LAAC was not only not noninferior but inferior to best medical therapy.

In Bayesian terms, the prior evidence for LAAC is extremely pessimistic. If CHAMPION AF was to change practice, it would have had to been an utter grand slam. I hope to have shown you that it was not. In fact, when you look at the actual data, and the way it should have been analyzed, it also was a negative trial.

Theme at ACC

Proponents of Watchman said in the main presentation and on social media, that CHAMPION AF results can be discussed with patients who may choose to have left atrial appendage closure. They use the term shared decision making.

I would argue that if we use shared decision making with patients for this decision then shared decision making is dead.

Why? Because the the prior data clearly show that left atrial appendage is worse than anticoagulation. CHAMPION AF does nothing to change that, and I don’t think we should ever offer inferior procedures. Despite what’s on the scoreboard, Watchman is clearly worse than direct acting oral anticoagulation.

Possible Explanation of Higher Stroke Rates and Similar Bleeding Rates

In Table s12 of the supplement, the authors show that at 4 months post procedure, 21% of patients in the Watchman arm had a peri-device leak. Any leak is bad because it increases the risk of clot forming and increases stroke rate.

Worse, though, is that device-related thrombus was seen in 4.8% of patients. That, too, is bad because it increases the risk of stroke, and warrants taking oral anticoagulation—which then negates the purpose of the device.

CHAMPION AF authors were critical of the German teams who did CLOSURE AF because that trial reported a much higher procedural complication rate. Yet, I don’t find these numbers on peri-device leaks reassuring at all.

Spin is Coming

Boston Scientific reps brought lunch to our office last week. They had flyers made about CHAMPION AF. They did not tell the results but it was obvious the trial was positive from the enthusiasm. One of the comments on the flyer was “Is your practice ready?”

There will be an extreme push to use these results to say Watchman can be an alternative to oral anticoagulation. Proponents will point to the scorecard they have in the NEJM. But as I have shown you, CHAMPION AF is clearly not a win for Watchman and does not come close to changing practice.

Other Links to Read:

1. I wrote a more detailed piece on Medscape for a doctor audience.

2. We also have coverage at Cardiology Trials Substack

3. My Twitter thread with 83K views.

4. When a win is not a win by Sanjay Kaul, MD

5. Perhaps the most enthusiastic Watchman proponent, Christopher Ellis, MD with this sobering comment.

Subscribe now