How can it be useful for a population but not for individuals? Populations don't get prescribed interventions, individuals do. Are we going to tell New York to exercise more, but Nevada to exercise less?
Reading this, the Rohrschach test that occured was "Simpson paradox". So certainly a hierarchal logistic regression or something else Bayesian would clarify the results. It's worth noting the trees were invented by Breiman et al as a technique for determining when a patient with MI was in serious danger for a repeat event. Maybe that is what should have been used.
Given the recent study from Stanford on the mechanism in which mRNA triggers heart cell damage and the dates of collected data from this study - I think there is much more to be discussed here. Please start talking more on the elephant in the room… every day that goes by and highly intelligent people/news/“gold standard” journals/professional organizations do not talk about this topic of mRNA technology credibility is lost.
It is obvious that the risk calculators are not looking at the "right" information, or at "enough" information to properly score the risk. The question is what else needs to be included. Our hearts are not just impacted by cholesterol and high BP.
There is no mention of stress levels (my cholesterol increased when I was under a large amount of stress), sleep, diet (even if on a statin, diet matters), support systems, menopause. All these things play into heart health, but very few doctors look at them and/or address them.
When will medicine start looking at the whole person? For too long, physicians have only looked at their particular organ - the cardiologist only looks at the heart, the nephrologist at the kidneys, the pulmonologist at the lungs, the gastroenterologist at the GI tract. Even the internist or family doctor tends to look only at the person in regards to parts. We are finally starting to learn that what goes on in the GI tract can affect other systems. Disease in the mouth can affect the heart. Etc. We need to look at the whole person. Until we take into account the whole person, risk scores will not be helpful.
It is a shame this is "only" a report/rather "small" work. I would have been interested in more details regarding the populations (like established vs. undiagnosed diabetes, statin/PCSK9 use yes/no and onset, ...).
465 people had first MI. 33% of the low risk group and 17% in the high risk group. Seems like it's safest to be high risk. LOL. Maybe the paper has figures for the total sample space: % low risk in total sample space, etc. You said there are so many low risk people in the total sample space - are those figures in the paper? Maybe it's obvious and implicit in the "risk" figures of the calculator, but given the state of information corruption, it needs to be explicit.
Then there's the absolute risk reduction of interventions. 12% risk for a person not on a statin. Reduces to 9% on a statin - 3%. Per calculator. But do all 12 percenters put on a statin actually have fewer MI? Fewer all cause mortality?
There is an interesting paper by Allan D. Sniderman at McGill University that has a meaningful proposed solution to the flaws inherent in the Pooled Cohort Equation and the so-called "MI Risk Paradox."
One of the main arguments in the the Sniderman paper is that the Pooled Cohort Equation (PCE) is heavily weighted towards the age co-variate and on a population level, although older patients have higher incidence of MI per 100,000 patients, thus leading to heavier risk weighting in the PCE, younger cohorts of patients actually have a much higher denominator compared to older cohorts of patients so the sheer numbers (numerator) of incident MIs is actually comparable in younger and older cohorts. Thus, the PCE misses a lot of these patients.
In addition, the paper makes an interesting assertion that it is not just the LDL or ApoB level that affects risk, but the DURATION of exposure to LDL or ApoB is also a huge factor in incident cardiovascular events. Thus, the PCE causes us to begin preventive therapy in the late stages of disease, where the impacts of statin therapy and LDL/ApoB lowering may actually be quite marginal. On a population level, it may make more sense to initiate preventive therapy in the younger cohort with high levels of established risk factors, since will have a much greater impact by lowering the LDL/ApoB concentration AND lowering the LDL/ApoB exposure DURATION. The term they use for this using a "Causal-Benefit Model" to determine suitability for preventive treatment rather than "Risk Prediction/Estimation"
I agree with Dr. Mandrola that performing more imaging is very UNLIKELY to be the answer here and that something closer to what Dr. Sniderman et al. are proposing is the much more likely to have beneficial impact on a population level for lowering cardiovascular event incidence. Unfortunately I don't think we will every be able to have a definitive randomized trial over a 20 or 30 year time horizon to test this hypothesis for a medication that costs almost nothing so it will be difficult to really prove this assertion. Epidemiologic/observational data and biological plausibility may be the best we can do here to better calibrate preventive treatment decisions on a population level.
God keeps reminding us that we are not He; that we can never know the future as we are not omniscient. We have never been good at predicting, and my sense is that will continue. We need to, like the good Dr. Mandrola wrote, do the best we can to help our patients live their best lives possible.
We know what we know today, and that is all we have to work with. When we learn more we can look back and realize how poorly we understood things in 2025 and how ineffective were all those silly things we insisted people do back then, just like has been true since before Hippocrates. I read the other day that while there certainly is cholesterol in arterial plaque, and higher blood cholesterol levels correlate to a degree with poorer outcomes, we really have no idea why cholesterol is deposited in some endothelial cells and not others! I believe we just lack the fundamental understanding of the disease like the physicians of previous ages, not that long ago, lacked fundamental understanding of the plagues of their age. So we muddle on with the small fraction of the truth that we possess today, as our forefathers did in their day, as its all we can do in our ignorance.
One of my biggest complaints about these risk prediction equations is the sort of preposterous precision. So I have a 13.1% risk of cardiac event? Really? Not 13.0%, not 13.2%, but 13.1%?? Why the third digit? I understand that if you take a study population of people who share some (but not all!) of my demographics and risk factors, that the percentage of them who will get the event is 13.1%. But to translate that weirdly precise number to an individual creates a veneer of scientific precision that is not warranted.
For example, check out the ARISCAT score for postoperative risk pulmonary complications:
Imagine a 55-year-old man, normal oxygen sat, no recent illness, no anemia, going for an abdominal surgery. If you check mark the box "< 2 hours" then his risk of complication is "1.6% Low." Change that surgery duration to "2-3 hours," and now it is "13.3% Intermediate." Finally you realize that the whole score in the end filters down to just three options:
Prediction is like looking for lost car keys under the lamp post in the parking lot because there you have light, not because you are sure that you dropped them there…
We can’t even “diagnose” plaque rupture yet (hence the endless parade of consults for positive troponin).
So I suspect the ability to “predict” plaque rupture is still a long
long way away (tough to predict something you can’t even yet detect).
This mirrors my sentiments (and reservations) about therapies. If a treatment produces an ARR of 10%, with an NNT of 10….it is a blockbuster no-brainer (as it should be). And yet we are knowingly asking 9 out of 10 people to waste their time/money and to tolerate side effects, without any idea of who that lucky 10th person will be, who will actually benefit from it. It is forever the limitation of applying average effects from trials onto unique patients (who do not have 10% less chance of an MI…they either have one, or they don’t).
As vonEye pointed out years ago with a beautiful mathematical proof, if you know everything about every individual, you know everything about the population. Conversely, if you know everything about the population, you know NOT ONE THING about any individual. This is the foundational failing of "population health" (there is no such thing) -- it is an interesting set of numbers of almost no value...but the numbers are seldom treated that way.
How can it be useful for a population but not for individuals? Populations don't get prescribed interventions, individuals do. Are we going to tell New York to exercise more, but Nevada to exercise less?
Reading this, the Rohrschach test that occured was "Simpson paradox". So certainly a hierarchal logistic regression or something else Bayesian would clarify the results. It's worth noting the trees were invented by Breiman et al as a technique for determining when a patient with MI was in serious danger for a repeat event. Maybe that is what should have been used.
Given the recent study from Stanford on the mechanism in which mRNA triggers heart cell damage and the dates of collected data from this study - I think there is much more to be discussed here. Please start talking more on the elephant in the room… every day that goes by and highly intelligent people/news/“gold standard” journals/professional organizations do not talk about this topic of mRNA technology credibility is lost.
So true predicting the future is never easy but thoughtful approaches like ‘Sensible Medicine’ give us a solid compass 💡📚
Risk ≠ destiny
Calculators are decent tools for population level decisions, likely with a better positive predictive value than a NPV.
It is obvious that the risk calculators are not looking at the "right" information, or at "enough" information to properly score the risk. The question is what else needs to be included. Our hearts are not just impacted by cholesterol and high BP.
There is no mention of stress levels (my cholesterol increased when I was under a large amount of stress), sleep, diet (even if on a statin, diet matters), support systems, menopause. All these things play into heart health, but very few doctors look at them and/or address them.
When will medicine start looking at the whole person? For too long, physicians have only looked at their particular organ - the cardiologist only looks at the heart, the nephrologist at the kidneys, the pulmonologist at the lungs, the gastroenterologist at the GI tract. Even the internist or family doctor tends to look only at the person in regards to parts. We are finally starting to learn that what goes on in the GI tract can affect other systems. Disease in the mouth can affect the heart. Etc. We need to look at the whole person. Until we take into account the whole person, risk scores will not be helpful.
It is a shame this is "only" a report/rather "small" work. I would have been interested in more details regarding the populations (like established vs. undiagnosed diabetes, statin/PCSK9 use yes/no and onset, ...).
465 people had first MI. 33% of the low risk group and 17% in the high risk group. Seems like it's safest to be high risk. LOL. Maybe the paper has figures for the total sample space: % low risk in total sample space, etc. You said there are so many low risk people in the total sample space - are those figures in the paper? Maybe it's obvious and implicit in the "risk" figures of the calculator, but given the state of information corruption, it needs to be explicit.
Then there's the absolute risk reduction of interventions. 12% risk for a person not on a statin. Reduces to 9% on a statin - 3%. Per calculator. But do all 12 percenters put on a statin actually have fewer MI? Fewer all cause mortality?
There is an interesting paper by Allan D. Sniderman at McGill University that has a meaningful proposed solution to the flaws inherent in the Pooled Cohort Equation and the so-called "MI Risk Paradox."
https://www.jacc.org/doi/10.1016/j.jacadv.2023.100825.
One of the main arguments in the the Sniderman paper is that the Pooled Cohort Equation (PCE) is heavily weighted towards the age co-variate and on a population level, although older patients have higher incidence of MI per 100,000 patients, thus leading to heavier risk weighting in the PCE, younger cohorts of patients actually have a much higher denominator compared to older cohorts of patients so the sheer numbers (numerator) of incident MIs is actually comparable in younger and older cohorts. Thus, the PCE misses a lot of these patients.
In addition, the paper makes an interesting assertion that it is not just the LDL or ApoB level that affects risk, but the DURATION of exposure to LDL or ApoB is also a huge factor in incident cardiovascular events. Thus, the PCE causes us to begin preventive therapy in the late stages of disease, where the impacts of statin therapy and LDL/ApoB lowering may actually be quite marginal. On a population level, it may make more sense to initiate preventive therapy in the younger cohort with high levels of established risk factors, since will have a much greater impact by lowering the LDL/ApoB concentration AND lowering the LDL/ApoB exposure DURATION. The term they use for this using a "Causal-Benefit Model" to determine suitability for preventive treatment rather than "Risk Prediction/Estimation"
I agree with Dr. Mandrola that performing more imaging is very UNLIKELY to be the answer here and that something closer to what Dr. Sniderman et al. are proposing is the much more likely to have beneficial impact on a population level for lowering cardiovascular event incidence. Unfortunately I don't think we will every be able to have a definitive randomized trial over a 20 or 30 year time horizon to test this hypothesis for a medication that costs almost nothing so it will be difficult to really prove this assertion. Epidemiologic/observational data and biological plausibility may be the best we can do here to better calibrate preventive treatment decisions on a population level.
God keeps reminding us that we are not He; that we can never know the future as we are not omniscient. We have never been good at predicting, and my sense is that will continue. We need to, like the good Dr. Mandrola wrote, do the best we can to help our patients live their best lives possible.
We know what we know today, and that is all we have to work with. When we learn more we can look back and realize how poorly we understood things in 2025 and how ineffective were all those silly things we insisted people do back then, just like has been true since before Hippocrates. I read the other day that while there certainly is cholesterol in arterial plaque, and higher blood cholesterol levels correlate to a degree with poorer outcomes, we really have no idea why cholesterol is deposited in some endothelial cells and not others! I believe we just lack the fundamental understanding of the disease like the physicians of previous ages, not that long ago, lacked fundamental understanding of the plagues of their age. So we muddle on with the small fraction of the truth that we possess today, as our forefathers did in their day, as its all we can do in our ignorance.
One of my biggest complaints about these risk prediction equations is the sort of preposterous precision. So I have a 13.1% risk of cardiac event? Really? Not 13.0%, not 13.2%, but 13.1%?? Why the third digit? I understand that if you take a study population of people who share some (but not all!) of my demographics and risk factors, that the percentage of them who will get the event is 13.1%. But to translate that weirdly precise number to an individual creates a veneer of scientific precision that is not warranted.
For example, check out the ARISCAT score for postoperative risk pulmonary complications:
https://www.mdcalc.com/calc/10022/ariscat-score-postoperative-pulmonary-complications
Imagine a 55-year-old man, normal oxygen sat, no recent illness, no anemia, going for an abdominal surgery. If you check mark the box "< 2 hours" then his risk of complication is "1.6% Low." Change that surgery duration to "2-3 hours," and now it is "13.3% Intermediate." Finally you realize that the whole score in the end filters down to just three options:
1.6% Low (<26 points)
13.3% Intermediate (26-44 points)
42.1% High (>45 points)
What even is the point of this?
chat GPTs of future will dominate your conversation with your patient much more to the detriment of medical society risk category predictions
Prediction is like looking for lost car keys under the lamp post in the parking lot because there you have light, not because you are sure that you dropped them there…
Love the Yogi Berra reference.
We can’t even “diagnose” plaque rupture yet (hence the endless parade of consults for positive troponin).
So I suspect the ability to “predict” plaque rupture is still a long
long way away (tough to predict something you can’t even yet detect).
This mirrors my sentiments (and reservations) about therapies. If a treatment produces an ARR of 10%, with an NNT of 10….it is a blockbuster no-brainer (as it should be). And yet we are knowingly asking 9 out of 10 people to waste their time/money and to tolerate side effects, without any idea of who that lucky 10th person will be, who will actually benefit from it. It is forever the limitation of applying average effects from trials onto unique patients (who do not have 10% less chance of an MI…they either have one, or they don’t).
As vonEye pointed out years ago with a beautiful mathematical proof, if you know everything about every individual, you know everything about the population. Conversely, if you know everything about the population, you know NOT ONE THING about any individual. This is the foundational failing of "population health" (there is no such thing) -- it is an interesting set of numbers of almost no value...but the numbers are seldom treated that way.