Yet on Sensible Medicine, we said plainly. Hantavirus was nothing to waste time thinking about. It would not be a pandemic. End of story.

Now the verdict is in: Sensible medicine - 1; media - 0

The media has quietly stopped talking about the topic. This is the main stream media playbook. Present a misleading story— and then never follow up your error.

Imagine saying this:

Aside: Why is that name familiar?

We did follow up on this topic:

The list goes on and on. School closure was horrific. The inaccurate NYTimes coverage aided it. Masking kids didn’t work. Vaccine mandates damaged trust. Healthy people don't need boosters. Yet the media took the opposite position at the time, and has never honestly followed up.

Yet looking backwards only gets us so far. Each day, more false narratives appear in the media.

What are some current science narratives the media is getting wrong:

1. Every single dollar spent on science is spent optimally, and any spending cuts will result in a loss of cures. This narrative is almost certainly incorrect. Vast amounts of taxpayer money are wasted, and we have no idea the impact of cuts, particularly if those cuts preferentially fall on disparities research (which largely documents disparities over and over, rather than offering any solution). Cutting low value science may have negligible impact.

2. In America we have great healthcare; the problem is access. Again, false. In America, the majority of people have access to a healthcare system that offers many interventions with clear and indisputable benefit, and then a vast array of interventions that may not work, probably don't work, definitely don't work, and are on balance harmful. We tax people to pay for the entire portfolio. Overall, we spend like a drunken sailor. Access is a problem for some. Access to every intervention that may or may not work is not financially viable for the nation and is not a reasonable goal.

3. Experts are superior to non-experts. This is only true in fields where experts predictions are compared against indisputable external benchmarks. If you build a rocket ship, you can be judged by whether it takes off. In biomedicine, experts can distort analyses to hide their own shortcomings. As seen with Hantavirus, monkeypox, covid, cancer medicines, etc. In medicine, experts can create mountains of propaganda— I mean observational studies— to support claims that lack reliable evidence.

4. Universities are unbiased arbiters of truth. Unfortunately universities have lost their way and are no longer politically neutral. As such. It should not be surprising that the general public seeks further reforms.

Just like Hantavirus, be careful with mainstream media science coverage.

Subscribe now

ARISE FLUIDS

The Surviving Sepsis Campaign began 25 years ago after a 2001 small trial found that early goal-directed care reduced mortality in patients with sepsis. Goal-directed care was aggressive: making invasive measurements (central lines) and targeting these numbers with IV fluids, vasopressors and blood transfusion when needed. The mortality benefit was huge (16% in absolute terms). The invasive components of that trial were later reversed but early resuscitation with fluids remained.

Then came the tragic case of 12-year-old Rory Staunton from Queens who grazed his arm playing basketball and was sent home after an initial evaluation and later died of sepsis. This tragedy ushered in the mandating of early aggressive IV fluid therapy for sepsis via performance measures that had financial penalties.

Thing is: this is a lot of fluid, and not all sepsis patients are the same. Another way to manage the low blood pressure and under-perfusion of sepsis is with vasopressors—drugs that raise blood pressure.

The ARISE FLUIDs trial investigators randomized 1000 patients with early sepsis to the either high-dose IV fluids or early vasopressors with less IV fluid. The primary endpoint of days alive and out of the hospital was 76 in both groups. Adverse events were largely similar, except pulmonary edema which was 9x more likely in the early fluids arm (0.6% vs 5.0%).

SODa-BIC

Perhaps even more entrenched than the aggressive fluid strategy for sepsis is the notion of “fixing” metabolic acidosis with sodium bicarbonate. The sky is blue and hospital doctors give bicarb for acidosis. The HCO3 ion is a buffer to the excess acid.

A brief background: metabolic acidosis occurs because of low perfusion of tissues, usually from shock of one sort or another or kidney failure. Acidic blood is bad because it can cause kidney injury. You try to avoid metabolic acidosis, and when it occurs, treatment is directed at the underlying cause.

But. But. Acidosis is a number; and bicarb corrects that number. Yet it does nothing to fix the underlying cause. No matter, doctors give it; because they do. Previous trials have been inconclusive.

NEJM recently published the SODa-BIC trial where patients with metabolic acidosis (pH < 7.30) who were receiving vasopressors in the ICU were randomized to either a bicarb infusion or placebo.

The primary outcome of a major adverse kidney event (death, dialysis, or persistent kidney dysfunction) did not significantly differ (40.2% of the bicarb arm vs 39.4% in the placebo arm). Four patients (1.6%) in the sodium bicarbonate group had an adverse effect, as compared with none in the placebo group (P=0.06).

BIHCA

Another use of the acid buffer sodium bicarbonate is the treatment of patients who have in-hospital cardiac arrest. The thinking is that cardiac arrest causes massive under-perfusion of the organs and acidosis is a) likely and b) detrimental.

The background here differs from bicarb use in metabolic acidosis in the ICU. In ICU acidosis, there is at least some data to suggest benefit—albeit weak because of flawed trials. In the case of giving bicarb to cardiac arrest survivors, there is no data, and guidelines recommend against using it. No matter, habits die hard in Medicine; doctors still give the acid buffer.

JAMA published the BIHCA trial, which randomized 913 patients who had in-hospital cardiac arrest. One group got bicarb; the other a placebo. The primary outcome of return of spontaneous circulation did not differ (39% in the bicarb arm vs 37% with placebo). The risk ratio was 1.05 [95% CI, 0.88-1.24]; P = .62).

Comments

The specific lessons are easy: in sepsis, doctors can use judgement. In some cases, early vasopressors with less aggressive fluid may be appropriate. Routine use of super aggressive fluids in all patients should not be a performance measure. In the bicarb trials, it’s even easier: treat the underlying cause and resist the urge to give interventions to make lab values look better.

The larger evidence-based lessons though are far more important. Fluid resuscitation in sepsis and acid-base management join a long list of what I call therapeutic fashions that have been overturned by proper trials.

Examples include temperature management, oxygen targets, prevention of contrast-induced kidney injury, levosimendan in cardiac surgery and of course the older story of Swan-Ganz catheters.

Todays’ studies and this list should teach us that therapies that a) make sense and b) target physiology or c) make lab values look better should always be questioned.

The longer I practice, the more I realize that the surer we are of something, the more we should test it in trials. We are not humble enough about equipoise.

Empathy, compassion, and sitting down when taking a history; these things require no trial. Almost everything else, randomize or later look foolish.

Subscribe now

Recently, a video went viral showing several diabetes researchers being thrown out of the American Diabetes Association meeting for distributing their editorial, which laments this administration’s science funding cuts. The episode was covered widely.

On Sensible Medicine today, three 3 takes on this issue.

Read more

Adam Cifu

Prediction markets such as Kalshi and Polymarket have become commercially successful by allowing users to trade on the outcomes of real-world events, including elections, court decisions, economic data releases, sports results, and policy questions. The premise is straightforward: When people put money behind their expectations, the market price can be interpreted as a real-time estimate of the probability of an event. This has given prediction markets a prominent role in political forecasting.

More recently, there have been efforts to build prediction markets for clinical trials, some of which have attracted attention on social media. Proponents argue that such markets could democratize biotech forecasting, make clinical-development risks more legible, and aggregate dispersed information more efficiently than isolated expert commentary. They also note that clinical trial “betting” already occurs indirectly through biotech equities and event-driven trading, and that a dedicated prediction market might simply make these judgments more explicit.

This framing is too optimistic.

Clinical development can be slow, expensive, and inefficient. There is certainly room for improvement. The FDA has, for example, explored “real-time trials” with the goal of accelerating decision making and reducing reliance on rigid, discrete trial phases. But real-time regulatory review is not the same as public real-time speculation. The former can be governed by protocols, confidentiality rules, prespecified monitoring plans, and regulatory accountability. The latter may alter the informational and ethical conditions under which the trial is conducted.

One reason prediction markets sometimes perform well is that they may reward private information. In politics, for example, someone may have information about local campaign activity, turnout, polling quality, or institutional dynamics that is not yet fully reflected in public commentary. Whether this is desirable is debatable. In clinical trials, it is more clearly problematic.

Consider a multicenter trial that has just started recruitment. One site has enrolled five patients. By chance, all five have had unfavorable outcomes: Early discontinuation, toxicity, or apparent lack of response. Even if the trial is blinded, staff at that center may form a negative impression. If someone with access to that local experience places negative predictions on the trial, the displayed odds may fall.

To an outside observer, those odds may appear to represent a global probability of trial success. In reality, they may reflect the early, incomplete, and potentially misleading experience of one site.

This matters because trial participation is sensitive to perceived benefit, perceived burden, and risk information. A patient considering enrollment may reasonably look for information about the trial online. If they find a prediction market showing unfavorable odds, they may decide not to participate. Physicians may also become less enthusiastic about referring patients. Recruitment may slow. Retention may be affected. In this way, the market could influence the outcome it claims only to forecast.

Clinical trial systems already recognize the risk that interim information can change conduct. FDA guidance on data monitoring committees notes that investigators who know early trends might alter recruitment or monitoring behavior, and it recommends that data monitoring committee members for a given trial not include investigators in that trial. The same guidance states that knowledge of unblinded interim comparisons can bias trial conduct or analysis, and that such data should generally be accessible only to the data monitoring committee or statisticians performing the interim analyses.

Prediction markets would create a public, tradable, continuously updated signal about an ongoing trial. This is difficult to reconcile with the principles that motivate restricted access to interim information.

The argument that insider trading already occurs through traditional financial instruments is only partly convincing. Traditional financial platforms usually have stricter know-your-customer requirements and more established surveillance mechanisms. In addition, the stock price of a larger biotech or pharmaceutical company is rarely determined by one trial alone. Other products, earnings, regulatory developments, and market-wide factors can all affect the share price. A trial-specific prediction contract is much more directly tied to one outcome.

Even if insider trading were not a concern, manipulation would remain a problem. An entity with a financial or strategic interest in a trial’s failure could place large negative bets, especially in a thin market, in order to move the displayed odds. Those odds could then deter enrollment or reduce confidence in the study. In this respect, clinical trials differ from many other events commonly traded on prediction markets. Elections, court decisions, and economic data releases may be forecast by markets, but the displayed odds are less likely to directly determine whether the event can be completed. Ongoing clinical trials are different. Publicly visible negative odds could become self-reinforcing.

Clinical development needs better forecasting, faster regulatory learning, and more transparent reporting. But public prediction markets for ongoing clinical trials are a poor solution to those problems. Trials are not passive events waiting to be predicted. They are systems whose validity depends on controlled information flows, participant trust, adequate recruitment, and sustained equipoise. Once trial-specific odds become public and tradable during recruitment, the market may no longer merely forecast the outcome. It may help produce it.

That risk, however, argues for regulation rather than a blanket rejection of the underlying technology. Sensible rules could preserve useful forms of forecasting while limiting the cases most likely to distort trial conduct. At a minimum, trial-specific betting should not be allowed while recruitment is still ongoing. Markets should also require know-your-customer procedures to reduce the risk that trial staff or others with access to non-public information trade on early, incomplete, or unblinded data.

Under such constraints, prediction markets may actually have a positive impact after recruitment is complete or after initial results are public. In oncology, for example, a market on whether an overall survival benefit will emerge after immature progression-free survival data have been reported could aggregate expert expectations about durability, clinical relevance, and downstream practice change without directly influencing patient enrollment. The goal should therefore not be to reject forecasting altogether, but to introduce sensible regulation so that it supports evidence interpretation rather than distorting evidence generation.

Paul Windisch, MD, is a researcher in radiation oncology at the Cantonal Hospital Winterthur in Switzerland, working on the implementation of artificial intelligence to structure information and to automate clinical workflows.

Share

I'm 40 years old and healthy. I exercise 5 days a week, and I don't have hypertension or diabetes. My doctor says I am good. I've got a BMI of 29 (5'10", 200 pounds). I want to start a GLP-1 as I'd love to weigh 175, like I did when I started college. Is this a good idea?

Read more

Before I clicked the link (they baited me), I already knew one important and obvious fact about the world. People do not have kids at random.

People who are in relationships, married, have kids, have jobs to support those families, etc, are different than those who don’t pursue these endeavors. While there are some perfectly happy and well-adjusted people without kids, there is undoubtedly a link between being agreeable, health seeking behavior, mental stability and having a family.

I am willing to bet people who have kids are less likely to smoke, more likely to do some physical activity (apart from chasing after kids), more likely to be able to get along with others (their co-parent/ spouse), and probably better off financially.

My question is: assuming that a “good brain”-parent relationship is true, did the authors do anything (anything at all) to separate the chicken from the egg? Did they try to answer the question: Do the kids come first, or is the type of person who has kids more likely to also… have good brain health (whatever that means)?

I click on the link to find out….

The op-ed writer begins, “as I discovered when researching my new book”

In my experience, people who discover something while writing a new book, and not before, seldom persuade me. Often, they twist the facts to fit their thesis (already sold to the publisher), rather than fashion the thesis upon the facts.

Very soon in the essay the author makes her claim causal. Having kids will improve your brain. Kids—> brain health, she claims, and not vice versa.

At last, I get to the key data (and a link) to support this claim.

I pull the paper. It analyzes “brain metrics” by number of kids.

Question 1: are people who have kids otherwise similar to those who don’t?

The authors have hidden this information. They provide demographic data for men and women, but do not separate the data by the number of offspring. They deprive the reader of the most important information. Both in the paper and supplemental tables. It is astonishing the editors permit this choice— it violates basic scientific standards— which requires displaying demographics among the groups you will compare. It’s astonishing peer reviewers did not ask for this (a commentary on asleep-at-the-wheel peer review).

Then they show their key finding.

Yes, there is a relationship. No, the paper can’t separate the chicken and egg.

I read the full paper wondering if the authors employ any clever natural experiments. Just like economists used car seat laws to see if people were more likely to have 2 vs 3 kids (yes). They do not.

Any effort besides adjusting for some poorly measured covariates?

Nope.

The authors even concede this.

Note: I didn’t even explore whether the authors’ metric is actually a reliable, global indicator of brain health.

Who benefits and loses from science like this?

1. Winners are the authors of the paper- they increase their CV. They may even justify receiving federal funds. The journal editors— they fill their pages, and typically earn a small amount. The journal owners. They benefit from open-access fees and free content. They make more profit (in % terms) than the owners of oil and natural gas companies. The reviewers probably don’t benefit: they got paid nothing to not do their job. Like hiring a volunteer lifeguard that wanders away. The author of the op-ed wins big— gets to promote a book in the pages of the Times.

2. Who loses. The taxpayers who fund this work and the researchers’ salaries. The readers of the New York Times, who are not educated about good vs bad science.

What can be done?

The only solution is for newspapers to hire technical experts. The New York Times should have a John Mandrola. The Washington Post should have an Adam Cifu. Well, not Adam or John because Sensible Medicine owns them forever, but you get the idea; It is inconceivable to me that major news outlets operate without internal expertise in clinical appraisal if they are truly committed to the truth. Instead, when scientific questions arise, they look on Twitter for doctors they follow (who share their worldview) to provide commentary. A dangerous echo-chamber.

Bottom line: It is possible that all else being equal, deciding to have kids improves a man’s brain health. That might be the worst reason to have kids, but it is possible it is true. Yet, every bit of research presented in the op-ed fails to address the question. The op-ed author relies on the fact that the uninformed readers won’t question a narrative that fits their worldview.

The op-ed also offers many other evidence-free claims.

Being an engaged father improves kids self-regulation? Or… kids whose parents are disengaged do poorly for reasons besides the parenting? Maybe the parent is disengaged for the same reason the kid is doing poorly— they are in a tough life circumstance.

Hands on dads reduce postpartum depression…. let’s click the link. Wow, the paper measures the father’s performance based on the mother’s diary!

Is it possible that a mother who is more depressed may also feel like the partner is spending less time, and inaccurately document father’s presence?

Also, how does time spent by father in a small sample in the NICU generalize into “hands-on” parenting?

Maybe families who are poorer, have harder life circumstances, might have to have a man return to work sooner, so he can’t be at the NICU.

Like pulling on a thread on a cheap suit, the essay starts to fall apart entirely upon even superficial scrutiny. Strangely, it appears in the paper of record without any of this explaination.

JAMA has published the TRACK trial of low-dose (2.5 mg) rivaroxaban every 12 hours vs placebo in patients with advanced CKD and either established vascular disease or high-risk for vascular disease.

The thinking

The most common cause of death in patients with CKD is cardiovascular. Yet numerous trials of interventions which have shown benefit in standard cardiac populations have failed in CKD—re statins.

Low-dose rivaroxaban has been shown (in two trials) to modestly lower major adverse cardiac events (MACE) in patients with coronary artery disease (COMPASS) and patients with symptomatic peripheral artery disease (VOYAGER PAD). Why not try it in high-risk patients with CKD?

The trial

The TRACK trial screened nearly 6000 patients to include nearly 1500 who were randomized to receive low-dose rivaroxaban or placebo. Pause there and note the highly selective nature of the trial. (Good on the investigators for showing this important trial feature.)

Inclusion required the patients to have had dialysis-dependent kidney failure or CKD stage 4 or 5 not receiving kidney replacement therapy (estimated glomerular filtration rate ≤29 mL/min/1.73 m²). Patients also had to have actual coronary, peripheral or cerebrovascular disease or be at high-risk for it.

Exclusion criteria were even more extensive. Bleeding, indication for anticoagulation (AF), drugs that could interfere with rivaroxaban metabolism, liver disease, heart failure, anemia, kidney transplant and many more conditions were causes for exclusion.

The primary outcome was a composite of CV death, MI, stroke or PAD event. The primary safety event was major bleeding.

The trial was powered to find a 25% reduction in the primary endpoint assuming an event rate of 10 per 100 patient-years in the placebo arm (which was quite reasonable).

Patients were 63 years old; 30% female and half were on dialysis.

Results

After a prespecified interim review of the data after only half the expected events had occurred, the data and safety monitoring board recommended early termination of the trial due to concerns over net harm and an extremely low probability of efficacy.

A primary outcome occurred in 164 patients (22.6%) in the low-dose rivaroxaban group and 151 (20.7%) in the placebo group (13.0 vs 11.8 events per 100 person-years; hazard ratio, 1.09 [95% CI, 0.87-1.36]; P = .46).

Rivaroxaban also showed no signal of benefit in any of the secondary outcomes. Nearly every outcome was numerically higher in the active arm.

Major bleeding was higher in the rivaroxaban arm (5.1 vs 3.4 events per 100 person-years; hazard ratio, 1.51 [95% CI, 1.02-2.22]; P = .04).

Comments

Whenever you have a nonsignificant trial, it’s always worth asking whether the signal is not there or did we not detect it. In other words, what are the chances of a false negative? That is one of the risks of early trial termination.

Here, with 95% confidence intervals having a lower bound of 0.87, I think it’s unlikely to have found a 25% reduction. What’s more, even if benefit was underestimated it would be countered by a 50% greater risk of major bleeding.

Clearly low-dose rivaroxaban should not be used in this patient population.

The teachable critical appraisal point is that high-risk does not guarantee greater benefit. The control arm event rates were highest in TRACK (12% per year) vs VOYAGER PAD (8% per year) and COMPASS (5% per year).

But low-dose rivaroxaban worked in COMPASS and VOYAGER PAD. In COMPASS, low-dose rivaroxaban plus aspirin vs aspirin alone, MACE reduction was 24%. In VOYAGER PAD (PAD patients), rivaroxaban + aspirin vs aspirin alone, MACE reduction was 15%.

CKD therapeutics teaches this high-risk-does-not-guarantee-benefit lesson well.

The authors and editorialists posit that non-thrombotic cardiovascular pathways may dominate in CKD patients; they also note the higher bleeding risk in CKD patients. I would argue the cause is less important than the observation that you cannot assume a therapy’s relative risk reduction applies to higher risk patients.

There are many other examples of interventions that work in the non-CKD populations but do not work in patients with CKD.

Statins are the classic example. More than 25 RCTs have shown statin benefit in primary and secondary prevention, but the 4D and AURORA trial both found no benefit for statins in patients with advanced CKD. The internal cardioverter defibrillator has shown benefit in patients with heart failure, but the ICD2 trial of the ICD in patients with CKD was stopped for futility.

Conclusion

Trials in patients with CKD are always worth your attention.

The first reason is that of course we want to find therapies to help this group of high-risk patients. And the only way to prove benefit is with the RCT.

The EBM reason to pay attention to CKD trials is the concept of treatment effect heterogeneity. Most therapies have a sweet spot of benefit.

In the case of low risk, it’s hard for a therapy to make low risk lower. Harm can outweigh benefit. See our upcoming post on GLP1a.

In the case of high risk patients, competing risks, disease heterogeneity, and higher adverse events (lower drug clearance or drug interactions) can lead to lack of benefit.

The extension of this concept is that we should be very cautious extending trial results to patients unlike those enrolled in trials.

E.g 1 - Paxlovid

Vinay Prasad's Observations and Thoughts

Paxlovid was a 10.6 billion dollar taxpayer funded mistake

Paxlovid (Nirmatrelvir–Ritonavir) is an oral anti-covid medication made by Pfizer. The original randomized study (EPIC HR) evaluated the product in adults who had a high risk of hospitalization from COVID-19, who were unvaccinated, and who largely had not had covid before (i.e. largely, they were un-immune). EPIC HR found the drug lowered the rate of hospitalization or death (from 6.3% to 0.8%). By 2021, however, these data were obsolete, and there was a debate raging in bio-medicine…

Read more

18 days ago · 259 likes · 11 comments · Vinay Prasad

Example 2 -

Here I would like to present the findings of an empirical analysis Vinay Prasad and I conducted on the idea of evidence expiration.

In our paper – titled Quantifying the Age of Evidence: Lessons From Cardiovascular Drugs – we sought to develop a method to measure the age of evidence for medical interventions. In order to do it, we used 2 widely used drugs as examplars – statins and sodium-glucose co-transporter protein 2 inhibitors (iSGLT-2).

Some background

Patient characteristics and clinical contexts change over time. Thus, even strong evidence favouring a certain medical intervention can become outdated. Therapies that were once supported by robust evidence may later prove less effective or not effective at all.

For example, aspirin was once recommended for cardiovascular disease (CVD) primary prevention. However, when new studies were conducted, the recommendation was revoked because aspiring was more likely harmful than beneficial. The decline in efficacy might be explained partly by healthier lifestyles and improved risk-factor control among patients enrolled in more recent trials compared with those in older trials.

So, when one aims at practising evidence-based medicine, it is important to consider the age of evidence. In principle, the older the evidence the less likely it is to reflect current patient populations, and therefore less likely to remain applicable to present-day patients. While time itself should not be judged alone – for instance, RCT evidence supporting the use of steroids in severe COVID is only 5 years old, yet it is arguably outdated –, the accuracy of the “age of evidence” as a proxy for outdatedness seems to improve as time increases.

Our study

We performed a cross‐sectional analysis of all Cochrane reviews evaluating the effects of statins and SGLT-2 inhibitors on CVD management up to August 2024. We summarised the evidence and calculated the lag in time between patient enrolment in trials and August 2024.

Reviews were included if they focused on statin/ iSGLT-2 therapy in adults, or were broader in scope but included meta-analyses for statins/iSGLT-2 alone. The primary outcomes had to include CV mortality, CV events or death from a major CV event.

First, we calculated the age of evidence for each trial as the time between the midpoint of the enrolment period and August 1, 2024. Then, the age of evidence was summarised for statins and iSGLT-2 as drug classes using the median age across all trials.

Results

Nine reviews of statins and two reviews of iSGLT-2 were included. Table 1 shows a list of all included reviews, their topics and the number of trials included in each review.

The 9 reviews of statins include a total of 153 different trials with enrolment periods spanning 29 years (between 1988 and 2017). The figure below illustrates the enrolment periods for individual trials, along with the overall enrolment period for each review. For context, the declining annual CVD death rates in the US during the same time period are also displayed. Study periods are unspecified or unavailable for 56% of trials.

In turn, the 2 reviews of iSGLT-2 include a total of 67 different trials spanning 13 years (patients enrolled between 2008 and 2021). Study periods are unspecified or unavailable for 19% of trials.

In our paper, we also provide a table summarising both drugs’ effect estimates on all-cause mortality and CV outcomes. Among the statins review topics, evidence was most robust for statins in the context of primary CVD revention and chronic kidney disease (CKD) not requiring dialysis. iSGLT-2 drugs have also demonstrated significant benefits in CVD and patients with CKD and diabetes.

Finally, the overall age of evidence for statins is 24.1 years and 8.6 years for iSGLT-2.

Implications of our findings

First, we developed a method to quantify the age of medical evidence and a graphical way to convey it. This way, one can more objectively and effectively evaluate the potential outdatedness of medical evidence.

Second, the fact that the age of evidence supporting statin use is 24 years old suggests that this evidence may be outdated. Over the last few decades (after most statins trials were conducted), the patient population has changed considerably. Risk factor management has improved significantly and newer drugs have become part of the CV therapeutic armamentarium (e.g. SGLT-2 inhibitors, GLP-1 agonists). High BMI and obesity have become the leading behavioural risk factors for disability in the western world, whereas a few decades ago other risk factors, like smoking, were preponderant. In contrast, the evidence supporting SGLT-2 inhibitors is far more likely to apply to present-day patients.

Third, our findings are relevant because statins are among the most widely prescribed drugs. Between 2019 and 2021, atorvastatin was the most prescribed drug in the UK and the second most in the US between 2017 and 2019.

Given the simplicity of our method, we propose that all future Cochrane reviews report the age of evidence not only numerically but also visually. We believe this can be readily implemented, as it does not require any additional data collection.

Ultimately, we believe these tools can be highly valuable not only for clinicians, but also for researchers and policymakers committed to evidence-based medicine.

He employed pharmaceuticals in his practice throughout his career -- antipsychotics, tranquilizers, early antidepressants – but psychiatry in general, and his practice in particular, was heavy on therapy and light on drugs.

This changed after the release of fluoxetine in 1988. The expectation of patients and specialty societies was that care would be more about pharmaceuticals. My father did not think that the change was for the worse, but in his last decade of practice, he saw his field becoming unrecognizable.

I sometimes feel like this is happening to me. I chose my field and trained to treat the sick. I learned to (lightly) embrace my role in disease prevention. But now, more and more, people expect interventions that are more about lifestyle. I am not even sure that lifestyle is the right term; performance, wellness, or even cosmetics might be more appropriate. I don’t like offering this type of treatment; I wonder if I should; I wonder if not offering it is acceptable.

There are four types of concerns that people present to me, a general internist. For thee of these concerns, I know where I stand. For the fourth, I am conflicted.

I care for people who are sick. These are people who come to me feeling poorly, or with a diagnosis, or with abnormal test results. This is why I do what I do. I love the intellectual challenge of making a diagnosis and crafting a treatment plan. I love collaborating with colleagues toward improving the health of my patients. I love working with and caring for the patient during the evaluation, the treatment, and beyond. When I eventually stop working, I will read the Case Records of the Massachusetts General Hospital every week just for sport.

I didn’t become a primary care general internist to do screening and health promotion, but I have embraced the role. I retain healthy skepticism about the benefit of treating the healthy. I remain a medical conservative and recognize the risks of screening.1 But I also know that I can help people live longer and better through education and assisting them in changing their behaviors. Simple, evidence-based counseling on diet, exercise, smoking cessation, and harm reduction pays dividends, as does some screening and risk factor modification.

As clear as it is to me that I have a role in treating the sick and counseling the healthy, it is equally clear to me that I want no role in people’s wellness hobbies. Some of these hobbyists feel poorly, and some feel well, but they are all interested in remedies or enhancers being sold to them. The products come with stories but no actual evidence. The elixirs often come with three-word names: Jamaican noni juice, wild Canadian chaga, red yeast rice, apple cider vinegar.2 Some are sold anonymously (or at least algorithmically) online, while others are sold personally through an array of wellness practitioners. When people try to involve me in their hobby, I express my doubts and counsel them about the absence of evidence, the lack of regulation, and the presence of cost. I will get involved when ill effects occur; otherwise, I demur.

The fourth category is the one that challenges me. These are patients who are not sick but recognize that they are not perfect. Some people are experiencing signs of aging: they have less energy, they take longer to recover from an injury, they don’t sleep as well, their libido has waned. Some people are young and feel like they should be better. They are seeking prescription medications, usually for an off-label use. Common requests are GLP-1s, testosterone, stimulants, and 5-alpha reductase inhibitors.

These requests usually lead me to think:

I’m not writing for that. I am not here to prescribe performance enhancers. This is not what I trained for. I am supposed to be helping the sick, helping to keep people well, not prescribing medications to those who have researched what might make them feel a bit better.

But when I reflect more thoughtfully, the complexity presents itself. What separates these requests from some of the treatments we don’t question? I don’t hesitate to treat menopausal symptoms. I would never withhold an SSRI for someone with mild dysthymia. I’ve prescribed odd cocktails of medications to help people with their jet lag.

While my father saw his field change from one based on talk therapy to one based on medications, I am seeing mine change from a focus on the treatment and prevention of disease to one that includes helping people improve. The origin of this evolution is complex. It includes a change in what people expect from medicine and what they think is possible. It includes the strange way we pay for our healthcare. It comes from savvy marketing by drug makers and companies like Hers, Hims, or Ro.

I admit to feeling a bit lost. I can still fall back on evidence-based medicine. Have the enhancers you are requesting actually been proven to be safe and effective? Quite often, they haven’t, as there is little incentive to test an already approved drug for a new indication for which it is already being used. Mostly, though, I need to consider how far I want to stretch my role. Personally, I am fortunate to be at a place in my career where I can refuse to evolve. That is not true for my younger colleagues.

Share

Adam Cifu

The American Society of Hematology (ASH) recommends raising the lower limit of normal for ferritin to 30 μg/L for men and women alike. ASH is also proposing a new diagnosis: Iron deficiency without anemia. This diagnosis would be based solely on ferritin values. This follows a trend we see across medicine of turning healthy people into patients.

Increasing the lower limit of normal from 25 to 30 in men won’t change much, but going from 15 to 30 in women and labeling low ferritin as a pathological state would be an important change. According to population prevalence in North America, up to 38% of menstruating women could be defined as having the new condition, iron deficiency without anemia.

ASH has not released screening recommendations, but other hematology associations and experts are already advocating for screening menstruating women, given the high prevalence of this condition. Some labs are already flagging ferritin levels below 30 as abnormal.

We are at a turning point in the overdiagnosis of iron deficiency because ferritin is not a perfect test for it, and iron deficiency without anemia is probably unimportant.

First, the flaws of using ferritin to diagnose iron deficiency. A Cochrane review from 2021 found insufficient evidence to recommend ferritin as a screening tool for iron deficiency in healthy populations. Ferritin is an indirect marker of iron deficiency. To assess true iron deficiency, bone marrow sampling is required. Original studies on ferritin show us how poorly ferritin performs as a test for iron deficiency. In young healthy women specifically, iron depletion with bone marrow staining correlated with a median ferritin level of 9 µg/L – well below the new recommended threshold of 30 µg/L. Furthermore, up to 25% of women were iron replete at a ferritin level of 27 µg/L. Many women would thus be misdiagnosed as iron-deficient using ferritin in this context, especially given the new threshold.

Ferritin levels also fluctuate in young women with menses and with iron-rich food intake. Labeling normal ebbs as a disease state contributes to overmedicalizing normality.

Then there is the issue that treating iron deficiency without anemia has not been shown to be beneficial. Low ferritin levels are poorly correlated with reported symptoms in the absence of anemia – hair loss and restless leg syndrome are the only symptoms that reach statistical significance in a well-designed study. Most notably, reported fatigue does not seem to make the cut. Do we know what we are treating then? How can we measure treatment efficacy in the absence of reliable symptoms?

I’m not minimizing the symptoms or the burden of iron deficiency anemia. Are iron deficiency and anemia prevalent in menstruating women? Yes. Should we treat symptomatic women presenting for care? Absolutely. Should we screen for and treat iron deficiency in the absence of anemia? I’m not so sure.

Ferritins should be ordered and interpreted when there is suspicion of symptomatic iron deficiency anemia (or iron overload). Iron deficiency without anemia should not be a new diagnosis for which healthy women are tested.

Defining ferritin below 30 as being ‘abnormal’ while it is found in 38% of people seems aberrant.

So then, what should normal ferritin levels be? Normal values should be normal by definition, i.e. ‘’occurring naturally’’, ‘’approximating the statistical average or norm’’, or by scientific definition, normals should be the lower and higher limits of a bell curve of what is found in a sample of the general, otherwise healthy, population.

Let’s not make ferritin what it’s not.

Share

Julie Laurence is a registered adult care nurse practitioner in Montreal who specializes in internal medicine at a community hospital outpatient clinic. She is affiliated with the Choosing Wisely campaign and trains future nurse practitioners at both the Université de Montréal and McGill University.

Photo Credit: Filip Mroz

GLP-1 Receptor Agonists or SGLT2 Inhibitors and Nonarteritic Anterior Ischemic Optic Neuropathy

• Previously discussed on Fornight XXII: Glucagon-like peptide-1 receptor agonists and risk of substance use disorders among US veterans with type 2 diabetes: cohort study

Everyone agrees that reducing hospitalizations due to heart failure is a good goal. One way is to have patients be mindful of things like weight and presence of swelling; they can then be taught to give themselves extra diuretics to alleviate the congestion. This however requires a quite skilled and motivated patient. I’d be lousy at it.

Here is an even better idea: place a little insertable monitor under the skin in the left chest. It’s tiny and takes only seconds to implant. That device has a fluid sensor capable of real-time transmission to clinicians who can act on the numbers to head off congestion episodes.

Here are some images of an implantable loop recorder:

The trial was called ALLEVIATE-HF (Algorithm Using LINQ Sensors for Evaluation and Treatment of Heart Failure) and it sought to assess the safety and efficacy of the Reveal LINQ (Medtronic) ICM with investigational HF risk-status software to guide individually protocolized interventions in patients with HF.

By testing a structured care pathway, this trial represented a pivotal evaluation of the surveillance-driven management paradigm.

A total of 711 patients received the device and were then randomized to active management vs standard care. The intervention was an individualized 4-day as needed increase in diuretic prescription. A prespecified algorithm was set in motion when the device alerted the care team that fluid levels were elevated. Specially trained nurses first ruled out any confounding medical condition or patient safety issues before imitating the higher dose diuretics.

In sum, this was a highly technical substitute for a patient seeing he had gained 5 lbs. and his belt got tight, so it was time to take some extra tablets.

The primary outcome was a 5-component hierarchical composite including cardiovascular death or HF hospitalization or outpatient HF event within 60 days of high-risk onset, Kansas City Cardiomyopathy Questionnaire Clinical Summary Score, and 6-minute walk distance, analyzed using win ratio.

JACC published the study results to little fanfare last week. Why? You guessed it: the trial was negative.

The primary composite endpoint did not significantly differ between groups (win ratio: 0.79; 95% CI: 0.62-1.01; P = 0.06). Win ratios should be positive. This was not. Actually, rates of CV death and HF events were 43% higher (133 vs 93) in the intervention arm.

The authors spent 1900 words in the discussion explaining how and why this result was clearly negative. I can explain it with four words:

It does not matter.

The reasons why or how this good idea failed is not relevant.

I was in Denmark last week to speak at the Danish Cardiac Society meeting in Nyborg. I had many great conversations with Danish doctors. One topic that came up was the implantable loop recorder. We use bunches of these devices in the US. The Danes use almost none.

Why? Because in Denmark, healthcare is paid by taxpayers, and government authorities do careful analyses of cost and efficacy. Their conclusion on implantable loop recorders is that they are a) costly and b) have little impact on health.

Here in the US, I call the devices cash machines. They have an outsized reimbursement for implant—which belies common sense because they can be injected under the skin with local anesthesia in 3 minutes. No sedation required. It’s the easiest procedure you can imagine. But that’s not the half of the cost issue: now, the patient is billed every 30 days for a download of the data. The doctor and hospital receive money every month—like a cash machine.

The current indications for using these devices have been detection of AF after a patient has a stroke, and detection of causes of infrequent fainting. Neither indication has any RCTs showing benefit.

The use in stroke is especially dubious with the new knowledge that short-duration AF episodes are common and associated with lower stroke risk than clinical AF detected with symptoms.

The companies who make these little cash machines wanted to expand their indication. And since they can also be equipped with a fluid sensor, the idea came to use them to help adjust medications in patients with heart failure. That was the origin of the ALLEVIATE HF trial. But it failed.

The answer is simply don’t spend dollars on these devices for this indication. Good, we learned something from ALLEVIATE HF.

The trial makes the Study of the Week for multiple reasons: one is that all good ideas require testing in RCTs. And not all marketing studies turn out positive.

Yet will need to be on alert, because negative marketing studies don’t often die peacefully.

On the same day that the authors published negative results, comes a post-hoc analysis from the trial showing that the recorder placed for managing heart failure also detected arrhythmias, and some of them were bad, and required interventions.

The authors spent another 1800 words explaining this finding with the implication being…maybe we should still insert these cash machines because even if they don’t help manage heart failure, they pick up arrhythmias.

Of course, this would be a foolish conclusion, because to test that conclusion we would want a separate trial with its own prespecified primary endpoint. You can’t change endpoints after seeing the data.

Subscribe now

Footnote: Thanks for all the new subscriptions. We appreciate your support and have big plans to expand Sensible Medicine, including recording more educational videos this summer. JMM

Keeping up is hard because there are so many things competing for your time. Ignore the journals for a week, or a month, or a year, and you (and your patients) are likely to be fine. But continue to ignore incoming information, and you will fall behind. The promise of a “review course” that will get you “up to date” is as false a promise as the latest from a longevity guru.

Everybody finds their own way to keep up. Here are some suggestions to consider. I’ve also included what I do as an example (understanding that these apply to a generalist, my patient panel, and my idiosyncratic interests; in other words, they apply to me).

Set Goals

Every job requires a different type of expertise. You need to decide the breadth of knowledge that you need based on who you are and what you do. A department chair once told me that she read the introduction to every article in the NEJM, JAMA, Lancet, and BMJ because this gave her an understanding of a broad swath of medicine. I generally skip introductions. A generalist will read differently from a sub-sub-specialist. Start by defining where you aim for excellence, where you aim for adequacy, and what you can ignore.

For me, I go for detailed knowledge of articles that will immediately affect my practice. I also look for important examples of medical reversals and, since I am supposedly an academic, I try to maintain a cursory understanding of major trends in the clinical studies from subspecialty medical and surgical fields.

Choose your journals

Start with major journals in your field and then whittle them down as you get a sense for what they publish. You always need to balance breadth and depth. A generalist is likely to skim multiple journals, while a subspecialist may read one or two journals cover to cover. Have the tables of contents of your chosen journal emailed to you. Do not delete the email until you have reviewed the issue.1

I read the NEJM, JAMA, and JAMA IM pretty closely. I look at the titles of BMJ, Lancet, and Annals of Internal Medicine. Even doing this, I miss a lot of interesting articles. This is where Twitter and reading John’s Study of the Week come in. They help me to catch articles I would have otherwise missed.

Skim effectively

Nobody can read every article with the attention I give to the two articles we cover every other week on Fortnight. Thus, you need to learn to effectively skim. Effective skimming requires you to use abstracts to do an initial appraisal of an article to determine if it is worth reading. My suggestions (biases included):

• For RCTs, consider whether the results, positive or negative, might change your practice. If so, is the study large and the endpoints clinically robust?

• I almost never read cohort studies unless the results are truly surprising; they will not change my practice, and there is only so much time.

• I think case-control studies are worth reading if the topic is important and the odds ratios are large.

• There are not enough diagnostic test (or clinical decision rule) studies. When one is published, I use the abstract to determine whether the test is usable and whether the study’s design limits spectrum and verification bias.

When I skim, I read all the titles and the abstracts of articles that seem clinically useful. If an article is worth reading, I skim the intro, read the methods, figures, tables, and the results. I do not read discussions. I read editorials if I am confused. I only closely read articles that will change my practice or that I plan to write about, talk about, or teach from.

Summary tools

There are many good summary tools out there: Journal Watch, ACP Journal Club... I have colleagues who use these effectively. I have never found them very useful, as they seem to feed me more articles rather than more important articles.

Colleagues

I find colleagues invaluable sources of information. I ask them questions. I eavesdrop when they talk about topics I don’t know enough about. Throughout my career, I have found it useful to review journals with a group. When we were most organized about this, we met once a month. Each member of the group was responsible for one journal. Important articles were summarized on a single, highly structured page. The task of summarizing and discussing an article helped me to retain the information.

On-the-go searches

It’s rare that I don’t need to look something up. UpToDate and OpenEvidence are my go-to, in-clinic resources. They are great for answering a question, but learning about a topic takes more time. I make sure to return to the page later in the day. I usually print a few key papers to read.2 If there is one absolute truism in medical education, it is that you learn best when the topic is associated with an actual patient. I remember everything I learned about cirrhosis, caring for JR in medical school. I remember little of what I read in JAMA last week. In the last six months, I have completely retooled my knowledge of the following topics based on the patients I have cared for.

• Diagnosis of light chain disease

• Prophylactic surgery in people with BRCA mutations

• Novel diagnosis tests and treatments for patients with Alzheimer’s disease

• Differential diagnosis of renal tubular acidosis.

Conclusion

Keeping up is hard. Lots of people do it well, and everyone does it differently. The only real truism for me is that you learn best when you are reading about a topic that is directly related to the care of an actual patient. Please chime in with your own tricks of the trade.

Share

Earlier this year, I appeared on a cable news broadcast to comment on yet another AI study. At the end of the segment, the host asked a question that caught me off guard: When will AI start saving lives? I smiled for the camera and tried my best, but I couldn’t shake the question.

Since ChatGPT’s launch, PubMed has indexed over 178,000 articles mentioning AI in their title or abstract. Many are interesting; some are promising. But when will this AI wave truly save lives?

That question reframed my thinking: AI’s biggest impact isn’t in documentation, prediction, or optimization, but in high-stakes moments when seconds count and existing human systems can fail.

The answer became clear: out-of-hospital cardiac arrest.

I’m not a clinician; I’ve never performed or trained in CPR. I’m also part of a supermajority of Americans where only 2% have been certified to perform CPR within the past year. When someone collapses in front of us, we call 911 and wait.

Dispatchers take about 75 seconds just to recognize a cardiac arrest, with chest compressions only starting nearly 3 minutes into the call. Across more than 350,000 out-of-hospital cardiac arrests each year in the United States, that delay is measured in lives where survival sits at roughly 9%.

Closing That Deadly Gap with AI

What if, the moment I needed help rescuing a loved one in arrest, I immediately had an expert CPR coach with me, telling me exactly what to do?

Our Study

Working alongside UC San Diego's Altman Clinical and Translational Research Institute — the very institution that pioneered CPR — my co-leads and I took on this challenge. Our team brought together Nimit Desai, a medical student and longtime collaborator, and Christopher Horvatt, a critical care physician who also happens to be a former high school classmate. Our findings are published in JAMA Internal Medicine.

We benchmarked popular AI models, including ChatGPT, Claude, Grok, Gemini, Llama, and Mixtral, on CPR coaching in simulated emergency scenarios against a checklist of criteria for delivering guideline-concordant CPR in out-of-hospital cardiac arrest. “Minimally viable” criteria captured the non-negotiable basics, e.g., coaching on where to compress and at what rate. “Maximally effective” criteria represented everything needed to optimize survival, e.g., instructing rescuers to allow full chest recoil between compressions. The benchmark results were encouraging but also sobering.

Across scenarios varying by cause (e.g., drowning vs. collapse during a jog) and patient (e.g., toddlers vs. seniors), AI models averaged 90% on minimally viable criteria, ranging from 79% (Gemini) to 97% (Grok, Claude), and 70% on maximally effective criteria, ranging from 61% (Llama) to 75% (ChatGPT). But in cardiac arrest, good is not good enough. Missing 10-30% of steps can be the difference between life and death.

Building ChatCPR for Real Emergencies

Those gaps informed how we developed ChatCPR, an open-source AI agent for CPR coaching. Grounded in 911 dispatcher training materials and decades of evidence on CPR best practices, we iteratively engineered the system to fix the specific failures we had observed in the benchmark. In these same simulations, ChatCPR scored 100% on both checklists.

But does ChatCPR work in reality? We obtained a set of de-identified 911 calls where dispatchers had provided CPR instruction, blindly comparing the dispatchers to ChatCPR’s responses. This is precisely the setting where ChatCPR could be deployed: a real emergency, a panicked bystander, help-seeking in progress.

ChatCPR won every head-to-head comparison with human dispatchers: +15 percentage points on minimally effective criteria (85% of guideline criteria met by dispatchers vs. 100% for ChatCPR) and +36 on maximally effective criteria (63% vs. 99%).

Performance differences were most pronounced for minimally viable criteria related to assessing if the patient was awake or responding, providing initial chest compression instructions, and instructing compression quality (depth and rate) as minimally viable CPR criteria. For maximally effective criteria, the largest gaps were in directing the caller to retrieve and use an AED if available, instructing full chest recoil between compressions, and ensuring proper continual CPR positioning.

This wasn’t about style, like our widely cited study on, AI for patient messaging, it was about strict adherence to CPR guidelines where precision matters most. Specifically, the only unmet checklist criteria occurred in one call and involved a maximally effective item requiring assessment of patient responsiveness and breathing. Although ChatCPR addressed both elements, the questions were not asked in the guideline-recommended order. ChatCPR excelled in patient assessment, depth/rate instructions, and recoil, areas where stressed, multitasking dispatchers faltered.

A Free AI that Can Save Lives—With Safeguards

ChatCPR proves AI can deliver precise, guideline-based CPR coaching on demand. The remaining challenge is implementation.

We have open-sourced ChatCPR and made it freely available, publishing the complete system, evaluation framework, and all supporting materials so that any developer or organization can freely use, adapt, and deploy the system.

This open-source approach is what makes practical deployment realistic: because the full system is publicly available, technology companies can integrate it directly into the devices people already carry, such as smartphones, search engines, and voice assistants, without building CPR coaching from scratch, while researchers can test, refine, and improve it, especially as multimodal AI tools rapidly advance. The goal is simple: make expert CPR coaching available instantly, anywhere, to anyone who needs it.

Importantly, systems like ChatCPR could add value across the cardiac arrest response continuum. They could help bystanders start CPR sooner (especially in difficult to reach or remote locations), support dispatchers with standardized guidance (a kind of coach’s coach) and assist clinicians and first responders with complex or scenario-specific instructions during training thereby making familiarity with the tool part of initial training. The real promise is closing the deadly gap between a person collapsing and lifesaving care beginning.

Yet, rigorous real-world trials are essential for safety, usability in chaos, and supporting how people interact with AI guidance. Safeguards such as integration with existing 911 systems and human oversight will be a necessary first step. Moreover, clear regulatory frameworks will be essential as our tool moves from research to real-world use.

For example, good samaritans receive legal protections for good-faith resuscitation efforts; how protections extend to developers and deployers of AI-enabled CPR coaches remains an open question.

Ultimately our work grounds AI hype in life-or-death reality. If AI is going to earn its place in medicine, it should start by helping people save the person right in front of them.

Subscribe now

John W. Ayers is a computational epidemiologist focussed on getting the public back in public health. He is a professor (medicine), vice chief of innovation (infectious disease and global public health), and head of AI (Altman Clinic and Translational Research Institute) at UC San Diego.

Read more

A company called LivaNova has a novel VNS system. It looks a bit like a pacemaker with a lead to the carotid area for vagal nerve stimulation.

A pilot study was encouraging. FDA gave the company a breakthrough device designation and allowed an adaptive statistical design to minimize the number of patients in the trial. The idea behind this special designation is to speed the availability of innovative products.

JACC published the ANTHEM trial results as well as the trial’s story.

Trialists designed ANTHEM like many other HFrEF trials: a device arm vs a non-device arm (Note: no sham control). The primary endpoint was standard—CV death and hospitalization for HF.

The planned enrollment was 1000 patients. The trial had a data safety monitoring board that looked at interim results.

After enrolling 532 patients the sponsor—not the DSMB—stopped enrollment.

The primary efficacy endpoint occurred in 34.6% in the treatment group and 40.2% in the control group. The HR was 0.84; 95% CI: 0.62-1.12; P = 0.115), which did not meet statistical significance. The primary safety endpoint achieved 96.7% freedom from procedure- or device-related serious adverse events.

The device seemed to improve vagal effects: 24-hour ECGs showed increased heart rate variability.

The authors modified the statistical analysis plan and added a hierarchical analysis using a win-ratio which was positive for VNS but inconclusive. It was inconclusive because a) it was not prespecified and b) included a quality of life assessment and there was no blinding.

The authors noted that the trial’s DSMB evaluated results after 500 patients (as planned) and did not find reason to terminate the trial for futility. Two editorialists added this comment (italics mine)

At this point, the trial’s DSMB recommended proceeding with further enrollment because prespecified conditions for early FDA filing or futility had not been met.

A few months later, the company, however, did their own analysis and stopped the trial. At the trial’s presentation, lead author of ANTHEM, Marvin Konstam, MD described the decision this way:

“not associated with either study futility, safety concerns, or device malfunction,” it was financial.”

Comments

I may be wrong but my read of this was that FDA gave the company a good shot at approval. The device did pretty well, but not good enough for early filing. The trialists did not find futility and called for continuing the trial. The company weighed the costs of the trial against the probability of profit and made a financial decision to stop.

It does not seem like any rule was broken, but it’s bad for all involved; bad for patients with HFrEF who may benefit from VNS, super bad for the patients enrolled in the trial, and ugly for the research enterprise.

The story highlights the tension from having so much of our progress in cardiology dependent on profit from industry. This is not a statement against profit; clearly, profit motive has brought us many advances in modern medicine. I would not want to go back to the 1970s. But profit motive is not exactly congruent with scientific inquiry.

Drs. Dhruva and Kadakia writing in an editorial note that the company’s actions undermined the justification behind the FDA’s upfront investment of regulatory resources and risks. They suggested that FDA attach explicit conditions to breakthrough designations in the pivotal trial setting. Namely, that there be prospective agreement on permissible grounds for sponsor-initiated termination.

That sounds like a good tweak. But perhaps a better tweak would be that there be less phase 3 (regulatory) trials conducted by parties who stand to gain from positive results of the trial. How to accomplish that is not an easy task.

Yet until we rework the evidence-generation system, I suspect we will continue to see such things.

Subscribe now

I am hesitant to share essays that feel like a list of memories. I know that great memoirs can make readers nostalgic for experiences they never had; a feeling sometimes referred to as anemoia. I am not a good enough writer to achieve this, but the memories I’m going to write about are of medical training, so maybe there will be something in them for those who’ve experienced it, expect to experience it, or are vaguely interested in it.

The first rotation of my internship was at the Brockton VA Hospital. The intensity of my memories of The Brockton can partly be attributed to it being my first experience as a doctor. But the memories are so lasting that I think there was something special about the place.

Clinical Experience, Responsibility, and Learning

I ranked a few excellent residency programs lower than The BI because I got the impression that, when you showed up at those programs, the program director handed you the hospital keys and told you to lock up when you were done. I felt like I would need a little more support and nurturing. To paraphrase Captain Willard from Apocalypse Now, “I tried to dodge a mission, and for my sins, they gave me one.” My mission was to begin my internship at The Brockton.

My memory is that we were lightly supervised. This bred both fear and responsibility. The first night of my internship, I signed out a basic metabolic panel1 to my co-intern. The patient had been admitted with cellulitis, and his creatinine had unexpectedly bumped. I came back the next morning and was informed that not only had the creatinine continued to climb, but that the urine sediment suggested post-strep glomerulonephritis. My fellow intern asked why I had not examined the urine before I’d left.2

That same week, my resident told me to increase a patient’s fluid from 75 to 100 cc/hr before I left. I forgot. These were pre-cellphone times, so there was a barrier to calling the on-call intern. I only fell asleep when I realized that 25 cc/hour for the 8 hours I’d be away amounted to less than a can of Coke. I made it up with a 250-cc bolus during pre-rounds the next day.

This independence led to mistakes, and probably some less-than-ideal outcomes, but I learned things I’ll never forget. One night, when I was a resident, we intubated a patient with a COPD exacerbation and ventilatory failure. After intubation, I struggled with his hypotension all night. The concept of auto-peep, explained to me the next morning, is one I will take to my grave.

Fun

While there were plenty of terrifying and humbling moments, I remember The Brockton fondly because we had a great deal of fun there. There was a bowling alley in the basement that we could use on quiet nights. On call, we had keys to the “widgets”, essentially golf carts, that allowed travel through the tunnels between buildings. Not surprisingly, widget racing was a thing.

As a snacker, I loved the commissary (now called a PatriotStore). I ate the ButterCrunch cookies almost daily. The ICU was stocked with chocolate Ensure pudding. I became a devotee of this rich snack until I realized each cup contained 550 kcal, explaining why it tasted so good.

Uncomfortable memories

There were experiences that make me realize how much things have changed, for the better, in the last 30 years.

• None of the patient rooms were private; one had six beds. It was common for patients to chime in about one another’s condition and plan.

• A patient once confided to me that he didn’t mind having an IJ placed because it meant that my intern, a young woman, would spend 15 minutes in close proximity to him.

• Dr. G, the attending in charge of the sleep lab, would not allow patients to leave if their CPAP could not be properly titrated. Instead, the patients had to wait until the subsequent Thursday, when ENT did their surgeries, to have a tracheostomy tube placed. As these patients had to be staffed by residents, they remained on our service. We referred to them as POGs. Prisoners of G.

• We sometimes commented that our patients seemed indestructible. This was said with respect, awe, really, but it reflected our concern that we were often in over our heads.

Camaraderie

The stress of The Brockton bound me to my fellow residents. Their support was unforgettable. My first resident warned me to beware of the intern shuffle, in which an intern would walk a few steps in one direction, then another, then another, because he was unable to prioritize his tasks. My resident insisted that I call him whenever I caught myself doing the intern shuffle.

This same resident once chased a van I was riding in across the entire VA campus to save me (and likely a patient). In the middle of the night, a cardiac arrest was announced at another building on the campus. Many of these buildings were residences for veterans with chronic conditions. During my orientation, I learned that when one of these codes was called, I was to report to the hospital’s entry, where a van would meet us and take us to the patient.

On this night, 5 days into my internship, I raced downstairs and hopped in the van, which quickly pulled away with a respiratory therapist and me on board. I said to the driver, “Wait, my resident is not here yet.” In a calm and measured tone, he said, “My instructions are to leave as soon as I have the respiratory therapist and a doctor.” Looking out the window, I saw my resident sprinting after us. He did not let up until we got to the code. For the rest of the month, I’d run downstairs and hide inside the lobby until I could jump on the heels of my resident when he passed me. I instructed my interns to do the same in later years.

Whenever someone finished their stint at Brockton, they were gifted a plastic thermal mug – mine was graced with an American Flag and the maxim, “America is Free, Thanks to our Veterans.” The rest of the house staff then gave a Brockton salute. The salute needs some explanation. Our pagers were radio receivers. When yours sounded, it alerted you to activate the broadcast, the page operator telling you whom to call. An unintended benefit of the system was that if your resident saw you reach for your pager, she could also listen in and monitor your activity. If the pager was activated when no message was being recited, it played only static. Hence, the Brockton salute. Everyone would raise their hand, holding their beeper, activating it to honor you with static.

There is much about our experience at The Brockton that would be unacceptable today. There was too little supervision, care often seemed less than patient-centered, and there was some inappropriate behavior. My time there was critical to my training.

Share

The Tweet

I’m 36. I’m a physician. I take a statin—and ezetimibe—every day. No symptoms. No cardiac history. Just an honest read of the evidence. Here’s what I found—and why I stopped waiting for a reason to act.

Read more

I love this essay, mostly because Dr. Raudasoja has really made me think.

Adam Cifu

Modern clinical medicine rests on a quiet pretense: that the visit can read whether a clinical intervention worked. It cannot. The prescriber writes a prescription, the patient takes it, they meet again at three months, and a judgment is rendered. If the patient is better, the drug helped. If the patient is not better, either the dose is too low or the drug is wrong. Sometimes, if several drug treatments fail, the patient’s condition is considered treatment-resistant. This is the framework behind some of the most common conditions we treat — depression, dementia, and acute and chronic pain.

The framework rests on a biased assumption: that treatment effects are observable at the individual patient level. They mostly are not. Average drug effects in conditions like those named above are so small that they are rarely detectable. Set against the natural fluctuations of illness, detecting a drug response at the bedside becomes impossible.

Yet healthcare still leans on this framework. I call it the theatre of treatment. The performance is for the patient, for the chart, for the system that pays the clinician’s time. It is for the clinician’s own sense of doing something. The clinician follows a script: prescribe, wait, adjust the dose, switch the agent, and treat each step as a verdict on the medicine. No one is acting in bad faith; the performance is structural. It is the same performance medicine has staged for two thousand years, in which clinicians believe they are treating, while mostly they are watching the disease take its own course.

The numbers behind the prescription

Statistics support that we are acting in a theatre of treatment. Roughly one in eight patients with major depression responds to an antidepressant. The number is real and traceable to the literature. However, it does not say what it pretends to say. It is largely a statistical artifact of where the responder threshold is drawn and of the amount of variance in the study population, both treatment-related and not.

Consider a study in which all treatment group participants have 51% symptom improvement, and the control group participants 49%. Draw the response line at 50%, and you get a 100% response rate in the treatment group and 0% in the control group, number needed to treat of one. The figure does not tell us how much patients improve or whether some improve more than others; it is a single number that compresses many things together, and it is largely meaningless for treatment decisions.

What we should be looking at is the average benefit on the symptom scale itself. For antidepressants, the mean drug-placebo difference on the Hamilton scale (0–52 points) is about 1.8 points, against a minimal important change — the smallest difference patients themselves perceive as meaningful — of 3-5 points. A 1.8-point difference is unlikely to be noticed by a patient. Whether some subgroups of patients see larger effects, we usually do not know; what we do know is that responder analysis cannot answer the question. Even a patient who is worse at follow-up tells us nothing on their own: they may be even worse without the drug. In fact, because the absolute benefit of a drug tends to scale with symptom severity, the patients who look like the clearest non-responders could be the ones the drug is helping most.

The same logic applies elsewhere in the formulary. The new Alzheimer’s drug donanemab slowed cognitive decline by roughly half a point on the 30-point MMSE over 18 months. This compares with a placebo group whose scores declined by about 2 points. The minimal important difference is about 1.5 points. Compare these results to older acetylcholinesterase inhibitors that improve MMSE at 26 weeks by about 1 point.

Paracetamol improves the pain of osteoarthritis by about 4 points on a 0–100 scale, where the minimal important change is around 10.

The trials are likely well-designed. The benefits are real. They are also, for the average patient, small enough that the patient would likely call them no change.

The follow-up illusion

If the population-level signal is small, why do some patients seem to improve? Symptoms change for reasons that have nothing to do with the drug, and on a much larger scale. Depressive episodes lift on their own; chronic pain waxes and wanes; cognitive decline is non-linear . A Dutch primary-care cohort followed patients with major depression for three years: 43 percent recovered from their first episode, 40 percent had a fluctuating course, and only 17 percent remained continuously depressed. In clinical trials, about 43% of patients in the placebo arm improve by at least 50% at 6 to 8 weeks. That improvement is not the drug — it is the natural course of the illness, for example, regression to the mean (patients enroll when they are at their worst, and worst rarely lasts), the clinical contact itself, structured attention of being in a trial, and whatever else has shifted in their lives between visits.

Clinicians help maintain the illusion. In a 2021 study, 542 American primary-care physicians estimated how much routine prescriptions — warfarin, antihypertensives, bisphosphonates, statins — would help a specific patient. The median estimates ranged from 20% to 50% absolute risk reduction. The actual figures from the trials are around one percentage point. The doctors who overestimated the most were the ones most likely to prescribe.

The clinician walks into the visit, then, with a belief about drug benefit the trials do not support, meets a patient whose course is mostly determined by the illness and other non-treatment-related factors. The visiting doctor asks, “Is it better?” and then treats the answer as a verdict on the efficacy of the medication.

An old mistake in a new costume

Waiting for a response and reading the patient’s improvement as evidence that the treatment worked is the inferential structure medicine has always used. The history of medicine is full of treatments that did nothing, and clinicians and patients still waited for the “response.”

What has changed is that today’s drugs often do something, but the something is still not observable and not large enough to make the structure work. We are still in the same theatre, waiting for patients to “respond” and feeling vindicated when they do. When they do not, the answer is more medicine — a higher dose, a different drug, another chance at a response.

What should change look like

What should change? If we were honest about response, we would prescribe less. That would be a real gain, but probably not the main one. Think about the opportunity costs. A patient waiting for an antidepressant to work may be a patient not investing in the things that might do more for depression — exercise, sleep, the slow rebuilding of a life. A healthcare system organized around three-month follow-ups to read drug response is one whose clinicians spend their scarcest resource — time — on a reading they cannot reliably make. Drop the reading, and that time becomes available for more important conversations. Unlike our predecessors, we already have the evidence to see the theatre for what it is — and to act on something better.

Share

Aleksi Raudasoja, MD, PhD, is an editor at Finnish Current Care Guidelines and is the responsible editor of Finnish Choosing Wisely Recommendations.

Photo Credit Giusi Borrasi