May 08 2026 This Week in Cardiology

Last week I received feedback from the danger shock investigators. And before I tell you the story, may I say what an honor it is that such prominent and accomplished academics engaged with the This Week in Cardiology podcast. Thank you. About a month ago, I discussed the CHIP BCIS3 trial of Impella support versus no impelling high-risk PCI. Chip BCIS, published in a New England Journal in March, randomized 300 patients with severe left ventricular dysfunction, extensive coronary artery disease.

to Impella versus no Impella and the trial reported uh thirty-seven percent of pairwise comparisons favored the flow pump versus forty-three percent favored standard care. That win ratio was zero point eight five. So basically that was no difference in major adverse cardiac outcomes. However, Death was 54% higher in the impella arm hazard ratio 1.54. So a 54% increase in death in those conference intervals.

were tantalizingly close to significant zero point nine nine to two point four one. There was also a absolute increase in cardiovascular deaths by twelve percent, A number needed to kill of only eight patients in the Impella arm.

Now at the time I emphasized the specific and larger messages of CHIP BCIS and of course specifically is don't use impella for high risk PCI like those patients in chip BCIS. And when you look at that trial And you look at the inclusion criteria and the actual baseline characteristics, I mean, these look like high risk PCI patients.

But I think the larger message of Chip BCIS goes back to my feelings about medical conservatism, and that is that this device had been on the market and used for this indication for 17 years. And now, after nearly 100,000 patients have been treated, we learn that it is probably harmful to be used. So the larger message is come on, let's have evidence before we adopt especially invasive devices like the Impella.

Well I then made the comment that the Impella proponents would come back at me and say, Well John, in danger shock there was a benefit of Impella in patients with uh AMI and cardiogenic shock. Recall that the danger shock trial randomized 360 patients with MI-related cardiogenic shock. Now they screened 1200 to find 360, and they took 10 years to do it. But there was a twenty six percent reduction in death with the device. The P value equals zero point zero four.

And while there is no argument that Danger Shock found a benefit for Impella, there were caveats of that trial, including a it was statistically fragile, B it took ten years to recruit three hundred sixty patients suitable for the trial, Screening more than twelve hundred to get this three hundred sixty suggested the trial included a highly selective group.

And at the time I discussed this, I went out on a limb to suggest another caveat in Danger Shock that I hadn't thought of before until David Brown, Dr. David Brown, uh remind me of it. And that is that in Danger Shock, in the as treated analysis, where they only count patients who receive their assigned treatment, which included nine patients in the Impella arm who did not receive the device, and three patients in the standard arm who did receive the impella,

Well now the hazard ratio was 0.77. Conference intervals went from 0.57 to 1.03. This is a notable finding because If a device is conferring benefit, you would expect more benefit in the analysis where you only count the patients who receive the device. And that was the point of contention of doctor Besk and Professors Hussagar and Moeller, who wrote to me to clarify the as treated analysis.

and they write, we understand the concern. If the device is beneficial, one might expect an even stronger signal among patients who actually received it. However, That expectation is unlikely to hold in this acute clinical context of danger shock. They explain, once clinicians deviate from the randomized strategy, treatment is no longer assigned by chance, but by bedside judgment.

and in cardiogenic shock, this means that patients who appear too unstable are more likely to receive the flow pump often later in the course and as a rescue therapy, while patients who stabilize, particularly in the uh impella arm uh may not receive it at all. And as a result, they go on to explain.

the as treated groups are no longer comparable. The sickest patients are preferentially treated, while more stable patients are not. This introduces a strong confounding by indication and timing bias. Under these conditions, even an effective therapy will appear less beneficial or even neutral because it is disproportionately used in patients with the highest baseline risk.

For that reason, they go on, the expectation that as treated analyses should show a stronger effect is not correct in this specific setting. The observed attenuation is not surprising, it is predictable. In short, the as treated analysis does not estimate the device's effect alone. It also reflects how clinicians select patients for treatment under pressure. The randomized intention to treat analysis remains the only unbiased estimate of the true causal effect.

Now I'm grateful to hear such thoughtful responses. I think that reasoning is highly plausible. And I reached out to David Brown, whose idea it was that the attenuation of effect in the as treated arm was a worrisome caveat, and he responded with this comment. He wrote back to me and he said, I could make the opposite exact opposite argument that the sickest, futile patients

would be the ones least likely to get the impella. That would be expected to yield more benefit for the impella arm in the ash treated as analysis if the device was helpful. doctor Brown also hopes that the authors publish a future paper describing the protocol violations in both arms, as this would clarify the issue. Finally, my final conclusions on danger shock remain unchanged, and that is

The trial shows what my interventional colleagues at Baptist Health Louisville tell me. They always say, John, there are some patients with cardiogenic shock who do not make it out of the lab alive without support. And to me, that's what the tr the trial shows. If you spend 10 years carefully selecting cardiogenic shock patients, you will find a cohort where the device is beneficial. But it does seem like a much smaller group that proponents in the company believe it is.

All right, next topic is immediate complete versus staged revascularization in STEMI patients. Circuit Invention has published a new meta-analysis of trials comparing immediate complete revascarization ICR to staged complete revascarization or SCR. Now the translation is during a stemie, after PCI of the culprit vessel.

You do the other serious lesions at the setting while the patient is in the lab, or you bring them back for a stage procedure that may be a day or so later or even a week or two later. Now, of course, this is a different question from one I have asked previously multiple times on that on this podcast, and that is doing complete revascarization during a STEMI or deferring it altogether and using medical therapy for the non-culprit lesion.

I had to stop talking about this because I believe the correct interpretation of the data is that medical therapy is the proper approach to the non culprit lesions, but interventional cardiologist and certainly their guidelines disagree, and I don't want to rehash that whole debate because it gives me a rash, but I would note that while the complete trials, which is the largest, four thousand patients,

Complete revascarization versus culprit only trial. It was positive for complete revascarization in a mace endpoint. But but it was driven only by MI reduction without a difference in cardiovascular death. But the problem with everything hanging on MI reduction

is that incomplete, you can't really count periprocedural MIs because it's hard to count an MI a periprocedural MI in the setting of a STEMI because enzymes are already high. And if you don't count periprocedural MIs, You're gonna bias in favor of the complete versus deferred strategy. Now then, of course, that's not the only problem, right? We have complete that's a problem because it's only MI reduction, but now we have two other problems with complete revascarization.

And that is that the I modern trial, which included eleven hundred patients, and that was IFR guided complete revascarization. versus deferred PCI for positive stress MRI lesions only. In that trial there was no reduction in MACE. And of course, we have the full revasque trial, where 1,500 mostly STEMI patients were randomized to FFR-guided complete revascarization or no revascarization other than the culprit vessel.

Now, Full Revasque planned to enroll fout four thousand patients like complete, but it was stopped enrolling when the complete trial released its positive results, which was a mistake in my opinion. So full revasque only enrolled fifteen hundred patients, but again, they found no difference in MACE outcomes. Which was remarkable I think because that composite endpoint included unplanned revascarization procedures, which is a pre soft endpoint, but yet it was still completely null.

Now, I'm gonna keep the complete revasation versus culprit only discussion limited to that, but remind listeners that if they discount iModern and full revas, and if you only believe complete trial, you go against a mountain of evidence finding that PCI does not reduce outcomes in stable CAD.

So PCI of non infarct related or stable lesions during STEMI would have to be an exception to the rule that PCI of stable disease does not reduce outcomes over medical therapy. And to me That's a hard rule to break.

Now anyways, on to the question is if you're doing complete revascularization, should you do it during the index procedure, right when the patient's on the table, during the STEMI, or should you do it in stage procedures? Now the trade-offs are obvious, right? Doing it initially is more efficient. It's one versus two procedures. You get to avoid a second arterial access, and it may reduce the outcomes in that short term in the next couple of days.

However, the downside is the patient is acutely ill with myocardial injury happening at that time. And of course, if you're doing multiple lesions, you're giving more contrast and there's just more risk. So in the meta-analysis, the authors found nine RCTs with about forty two hundred patients randomized. The main points of interest of that meta-analysis were short-term and long-term all-cause mortality.

For short term mortality, it was two percent for the immediate complete versus one point two percent for the stage complete. And that's a relative risk of 1.66, so 66% higher if you did it in one procedure. And those conference intervals were 0.99 at 2.78, p-value 0.053. And at one year mortality, the numbers were four point seven percent for immediate complete versus three point five percent for stage complete. And that rate ratio is 1.40. And uh here the p-value is 0.07.

Now, one issue with the meta analysis, and I wonder if it did not come up in peer review, is the author's choice to include the IModern trial, which I just told you was not really immediate complete versus stage complete, but rather It was IFR guided complete revascarization versus uh later stress MRI guided PCI, which is closer to the complete revascarization versus culprit only than it is to immediate versus staged complete revasque.

So in a leave one out sensitivity analysis where they leave out the the really null I modern trial, the all-cause mortality with immediate complete compared with stage complete. is now significantly worse for short term and long term deaths. Short term, the relative risk was two point one times higher and those confidence intervals were one point one six to three point seven seven.

Long-term mortality at one year it was one point five six, and again, still significant at one point zero one to two point four one. And so the authors hedge a bit in their conclusions and they they write in hemodynamically stable patients with STEMI and multivessal disease. An immediate complete revascular strategy may increase short-term cardiac death compared with stage complete revascarization, and the possibility of increased early and late all-cause mortality cannot be excluded.

And quote, despite some uncertainty, these results currently favor a staged complete revasculation strategy for most patients with STEMI and multivessal disease. I think this is a good conclusion, actually. The short term and long term mortality numerically favored staged procedures for complete revascarization and the p-value hovers close to significant.

I'm drawn to the upper bound of the near doubling of the risk of death. Also, leaving out the I modern trial, which really does not belong, pushes the signal even more strongly against immediate complete revascarization. Staging complete revascarization seems wisest if you're going to do complete revascarization. However, let me now tell you the biggest caveat of this data, and that is the translation to the actual patient.

Now, I don't do PCI, but I know enough to know that there is a feel to things in the lab, right? What do I mean by feel? Well Let's say the culprit vessel PCI went easily. It went well. There was little contrast needed, uh a good result and the patient is is rock stable. Now let's say there is a discrete 95% mid-vessel disease, say right corner artery. It's another low-risk PCI, hardly seems dangerous, so you do it.

Contrast that with a difficult or somewhat late culprit PCI. Maybe there's ectopy or soft blood pressure, not shock, just rocky stuff. The patient is rocky. And now the the non culprit vessel is a branch vessel. In this case it seems wise to do the stage procedure. The trials that are included in this meta-analysis, they offer general guidance, but you do have to consider the patients not randomized in these trials, right?

It seems to me that this meta-analysis would tell me that unless an i an immediate, complete revascarization would be easy and the patient is stable, it's probably best to do a stage procedure. Of course, again, if you're going to do a stage procedure, it might even be better to just start medical therapy and wait and see.

All right, next topic polygenic risk scores for prediction. Now Jack has published a 6,000-word paper looking at polygenic risk scores as prediction for eight different cardiac conditions, and the authors are prominent Harvard researchers. So I am a podcaster and an electrophysiologist.

so this is a bit like an HVAC person commenting on aerospace engineering. But I want to say something about this paper because now the authors tell us in in the c discussion They are commercializing the polygenic risk score so that for two hundred twenty-five dollars, people can use their genes to help predict their future health or ill health.

That's moved the seriousness of this from a curious academic paper to taking money from people. Now here is how and I would frame the issue using CAD as a model. Unlike long QT syndrome, coronary disease does not come from one gene. It comes from an amalgam of many genes. Sort of like a family history. You inherit a group of genes from your mom and dad, and if you are unlucky, these genes interact with your environment to make you more likely to have coronary artery disease.

Now modern technology has allowed research teams to create these poly as in many genic genes risk scores. The idea is it's really just a more accurate family history. Then the paper describes the development and invalidation of the polygenic risk score for the eight conditions. They show they show a framework for clinical reporting at the end. Now here is the challenge though, right? A prediction is hard, especially about the future. That's what Yogi Barra says.

We already have simple scores like the PCE or PEVENT score. So these things use simple uh markers like blood pressure, LDL, and AIDS. And these simple scores get people into risk boxes risk boxes, low, intermediate or high. We also even have coronary calcium, which proponents say further risk stratify, though that is debatable.

The challenge for these soup or fancy polygenic risk scores is how much more incremental benefit they add for the average patient, say the HVAC guy or the plumber or whatever, in changing their decisions about his or her life. And I guess if someone is spending two hundred and twenty five dollars, what is the evidence that that it would reduce MI or death over not spending the two hundred and twenty five dollars?

The paper lays out the polygenic risk scores development and then it tells us results and it's complicated and I'm gonna try to simplify and I hope I don't get it completely wrong. Basically, they used three large biobanks, the All of Us or All of US Data Bank, the Mass General Brigham Biobank, and a UK Biobank. For coronary artery disease, there are dozens of public risk scores, and the insight of the research team was to combine all the scores using a statistical technique.

They then trained this combined model in eighty percent of the all of US data, held back twenty percent for internal testing, then applied the final scores to a separate biobank population. They used electronic health records of course to define w who had each condition. and their performance measures were really three different types. The C statistics

Okay, if you randomly pick one person with a disease and one without, what's the probability that the PRS correctly gives the higher score in the sick person? Uh 0.5 is random chance, 1.0 is perfect. Their scores range from about 0.61 to 0.86 across traits. Lifoprotein Little A was the 0.86 and it's the highest one, but of course that makes uh sense because it's entirely genetic determined. For coronary disease, the PRS was 0.67. For AF it was 0.64. And these are very, very modest.

They also use odds ratio per standard deviation. So for every one standard deviation increase in the polygenic risk score, how much disease odds increase? CAD it was for it was 2.07, meaning moving one standard deviation higher in a score roughly doubles your odds. This sounds impressive, but of course it's a population level statistic.

And the third metric they used was percentile based odds ratios. So this divides people into risk bins, high, top 10%, above average, 76 to 90th, average 26 to 75th. And below average, bottom 25%. They then ask, compared to the average group, how much higher is risk in the high group? For CAD, this uh was 3.7%, meaning the top 10% had 3.7 times the odds of CAD compared to the middle 50%.

And overall, the scores sort of worked. People in the high-risk bins genuinely had more disease than people in the average bins, consistently across all data sets. The graded relationship was real. High was greater than above average. Was it greater than average? Was it greater than below average for every trait?

And this held up in prospective analyses too. People who were high-risk enrollment and didn't yet have the disease were significantly more likely to develop over the uh uh oncoming eight years. The most dramatic result was lipoprotein little A. The top 10% had 41 times the odds of elevated lipopro lipoprotein little A compared to the middle group, but

This is almost tautological, right? Because lip lipoprotein A levels are ninety percent heritable. So of course a genetic score is gonna prov predict lipoprotein little A. They then did something called the net reclassification index analysis or NRI analysis. And this tests whether adding the polygenic risk or improved risk classification in people already

scored by the standard clinical tools like the pooled cohort equation. And here they said it showed improvements of 0.14 to 0.18. And I don't know what that means. It's a difficult thing to know what it means. And here's the problem with this paper. Because they don't present the raw present the raw numbers, you don't know whether there are more people correctly or incorrectly classified.

Like it is with coronary calcium, namely that because many more people who are reclassified higher don't have events, the NRI actually misclassifies more than it correctly classifies. Now, I also want to emphasize the high odds ratios, which look impressive. They sit there and jack and you're like 3.0 odds ratio. But the problem with that is most of the wrist curve, the the bell curve of these patients overlap.

So when you cut off the extreme ten percent, the highest risk and compare it to the middle of the risk curve, it looks impressive. But how does that help an individual person? For instance, a CAD odds ratio of 3.7 sounds clinically meaningful, but it tells you nothing about how well the score actually sorts individuals.

The C statistic is in the range of what it is for the Chads fast score, about 0.65 with 0.5 as a coin toss. Now would you pay two hundred twenty five dollars for your Chads fast score? So in the end, we can know if you sorted the entire population into deciles by their CAD polygenic risk score, the top decile is going to have substantially more events than the bottom decile. That's real and reproducible.

But the C statistic of point six whatever tells you for any individual patient the score makes the wrong ranking about a third of the time. And what is more, we have no idea how much better it is than the simple PCE. Finally, and this is most important, we will have no idea from this paper whether using the polygenic risk score will reduce MI or death relative to not using it, which is the same argument I make for coronary artery calcium. It's one thing

that i if it improves risk prediction. We don't even know if it does, but if it does, risk prediction isn't the same as reducing MI or living longer.

All right, final topic today is brief. Practice improvement policies undergo the proper test, which is randomization. Circ Outcome has published a really fun trial called the Quell Trial QUEL, which brashly tested whether or not implementation of a well meaning series of practice improvement policies actually improved outcomes. Now the background here is also obvious, right? We know that atherosclerotic cardiovascular disease is largely modifiable.

Secondary prevention works, but uptake is a challenge. I mean you can't benefit from statins if you don't take them. Well persuasion from clinicians who have time and proper tools may help people take their secondary prevention therapies, and if they do, they'll do better. And policymakers have these great ideas, right? On paper, many of these practice improvement policies make decent sense. But well meaning policies don't always translate to benefit when used in the real world.

So the quality improvement approach used in this trial comes from something called the Breakthrough Series Collaborative, developed in the nineteen nineties. by the Institute for Healthcare Improvement, IHI, in the United States. And the core idea has been borrowed from manufacturing and industrial process improvement. If organizations share what works and learn from each other systematically, they can all improve faster than working alone. The IHI adapted this into healthcare.

And here's how it works. The basic structure brings together multiple clinical teams, in this case twenty-six general practices in Australia around a shared problem. They then cycle through uh A learning workshops. Teams gather to hear from experts and they share their experiences. And then there's PDSA cycles, plan, do, study, X cycles, small rapid test of change at each site.

And then of course they use benchmarking practices, see how they compare to each other using their own data. And finally, there's facilitated support where there's ongoing coaching between workshops. And all that sounds great, doesn't it? So in the Quell trial specifically, practices also receive monthly automated reports pulled from their EHR showing that how their CHD patients were dueling against twelve key performance indicators, and that was the data driven part of this classic model.

The trial used a cluster randomized uh uh process and they recruited Australian primary care practices from twenty nineteen to two thousand twenty two. Importantly, Practices here were the unit of randomization and adults with uh coronary heart disease were the units of analysis. The primary end point was great. It was the proportion of participants who had unplanned cardiovascular disease hospitalization at two years.

And the results? A total of fifty one primary care practices participated, and it resulted in a patient cohort of about seven thousand eight hundred patients. And at twenty four months, get this, sit down for this, there were no significant differences between the groups for unplanned cardiovascular hospitalization, no differences in MACE, no differences in prescription of antiplatelet, statins, ACE inhibitors, Uh there was no difference in uh LDL levels, systolic blood pressure, or anything.

So the authors concluded that a primary care data driven quality improvement program did not improve any important outcome. Not even surrogates like LDL or blood pressure could be affected. They then add this zinger sentence, quote, robust evidence for the use of data-driven, collaborative approach to improving care for people with coronary heart disease in primary care remains elusive, end quote. So I have two things to say. One is that I congratulate the Australian teams for doing this study.

As I have said many, many, many times, that policy that affects patient care should always be tested in RCTs. They should be tested before implementation. because well-meaning does not equate to effective. And policy is surely as important to study as drugs or devices because policy may affect more people than specific drugs or devices.

The second thing to say is that evidence for quality improvement benefit may be elusive because quality improvement tools in factories does not easily transfer to health care. I am skeptical that policy changes will ever show benefit over basic health care delivery, though I do hold out some hope that an AI driven model and an AI used wisely Um, may be a positive in the future.

So that's it for this week in cardiology. As always, I'm grateful that you listened. Thank you. And if you like this podcast, please take the time. to give us a rating, write us a review, and please if you disagree with something I say or have something to clarify, like the danger shock investigators did, please send me an email. I learn from these and they make great teaching points. Until next week. This is John Mandrola from the Heart.org Medscape Cardiology.