He had a sudden cardiac arrest. How does this change the probability that he has a particular genetic condition?

Megan McArdle writes:

I have a friend with a probability problem I don’t know how to solve. He’s 37 and just keeled over with sudden cardiac arrest, and is trying to figure out how to assess the probability that he has a given condition as his doctors work through his case. He knows I’ve been sharply critical of doctors’ failures to properly assess the Type I/Type II tradeoff, so he reached out to me, but we quickly got into math questions above my pay grade, so I volunteered to ask if you would sketch out the correct statistical approach.

To be clear, he’s an engineer, so he’s not asking you to do the work for him! Just to sketch out in a few words how he might approach information gathering and setting up a problem like “given that you’ve had sudden cardiac arrest, what’s the likelihood that a result on a particular genetic test is a false positive?”

I agree that the conditional probability should change, given the knowledge that he had the cardiac arrest. Unfortunately it’s hard for me to be helpful here because there are too many moving parts: of course the probability of the heart attack conditional on having the condition or not, but also the relevance of the genetic test to his health condition. This is the kind of problem that is addressed in the medical decision making literature, but I don’t think I have anything useful to add here, beyond emphasizing that the calculation of any such probability is an intermediate step in this person’s goal of figuring out what he should do next regarding his heart condition.

I’m posting the question here, in case any of you can point to useful materials on this. In addition to the patient’s immediate steps in staying alive and healthy, this is a general statistical issue that has to be coming up in medical testing all the time, in that tests are often done in the context of something that happened to you, so maybe there is some general resource on this topic?

1. For genetic conditions using PCR or sequencing techniques, the main mechanism of false positive is contamination. False negative arises because of damaged reagents or human error. If you get a positive from two labs and the genetic conditions causes sudden cardiac arrest you’re probably done with the diagnosis.

It’s much harder to diagnose things that are less straightforward than “my genetic code contains known allele X”. That’s a truly binary question

• Rahul says:

How many genetic conditions that definitively cause sudden cardiac arrest? How many of those can we reliably detect via sequencing today? Just curious.

Aren’t most symptoms a complex combination of genetic predisposition and the environment?

• Rick says:

I am the patient in question for the original post.

On September 27th I experienced Sudden Cardiac Death while sitting at my desk at work; my coworkers gave me CPR, EMS arrived promptly, and I was transported to the local hospital where I experienced multiple seizures, ventricular arrhythmias, asystole, and nonischemic cardiomyopathy. After they got my heart going again they cooled me down for 24 hours to prevent brain damage. After being warmed back up I quickly recovered and am currently wearing a temporary Life Vest. My heart catheterization revealed a clear and healthy heart. Unless the cardiac MRI this week turns up something that requires some other treatment, I will probably receive a permanently implanted defibrillator device (ICD) in the next few weeks.

I do not know how many conditions there are that can cause SCD, but after doing some reading on the topic here is a (probably incomplete) list of possible genetic causes:

1. Congenital Long QT Syndrome (this is the one the Drs think is most likely in my case, though I might also have the acquired variety rather than the genetic)
2. ARVD (another candidate for me, but less likely)
3. ARVC (nobody has mentioned this to me as being a candidate)
4. Brugada syndrome (nobody has mentioned this to me as being a candidate)
5. CPVT (nobody has mentioned this to me as being a candidate)
6. Wolff–Parkinson–White syndrome

There are probably others.

• Rahul says:

All the best Rick! Thanks for this list. Do you have any idea for which ones of these six there’s a genetic test for? The sort Daniel alluded to above “PCR or sequencing techniques”. Maybe someone else knows.

• There are things that are congenital which are not genetic (ie. caused by random events during development rather than genetic coding issues), and there are things that are genetic but not determined by a single known gene (for example alcoholism risk increases dramatically if your parents were alcoholics, but there’s no single known gene that causes this). In those cases where causes for the same thing could be genetic or non-genetic, it’s hard to figure out what to do about testing.

If any of these syndromes are the result of a known single gene mutation, doing those tests makes good sense, because the false positive or negative rate in modern testing labs using sequencing or PCR based tests is going to be pretty low. But the key is that there should be a known strong relationship between the mutation and the cardiac syndrome you have. If there’s a 10% increased risk of a very rare disease due to the gene (say goes from 1 per 100k to 1.1 per 100k), this is meaningless, if 98%+ of everyone with the gene has your symptoms by age 50 then this is a good reason to do testing.

Even syndromes that are caused by say a known combination of 2 or 3 genes, you should probably test for. But something like “congenital long qt syndrome, possibly the acquired type” is probably less likely to be actionable. If you don’t have the genetic type, but you do have acquired, does that change your treatment vs if you do have the genetic type? I’m guessing not but if it does… you should do the testing.

I’m not an expert on any of the medical issues here, just someone who has worked on data analysis of sequence and PCR data for biological research and knows something about how those data collection techniques work.

If you have specific quantitative information you’ve put together on the syndromes and want to have someone work out actual calculations for you, post back and we can discuss, there are many people here with skills that can help.

• Rahul says:

” if 98%+ of everyone with the gene has your symptoms by age 50 then this is a good reason to do testing.”

That’s exactly my point. Are there many of these? I don’t know. But I’m skeptical.

• Imaging guy says:

” if 98%+ of everyone with the gene has your symptoms by age 50 then this is a good reason to do testing.”
That’s exactly my point. Are there many of these? I don’t know. But I’m skeptical.

They are called fully penetrant Mendelian diseases although they are quite rare.

• Rahul says:

Thanks! This is the sort of critical detail that’s ignored by all the hype generated around genetic testing.

• Rick says:

Thanks for the well wishes.

My understanding is that the treatment for all of these conditions is pretty much the same: a permanently implanted subcutaneous defibrilator, or ICD.

• In this case the main reason to do testing is if knowledge of the genetics leads you to expect a certain additional set of issues which you might plan for or treat prophylactically.

If not it seems you would best get the implant and ignore the tests. The value of the information is low if it doesn’t lead to any change in action.

• Rick says:

Thanks for the discussion to everyone. It’s really helped me to think critically about what’s ahead of me.

• Rahul says:

“In this case the main reason to do testing is if knowledge of the genetics leads you to expect a certain additional set of issues which you might plan for or treat prophylactically.”

@Dan, do you have any examples? This sort of thing always sounds great in the abstract but I’m still trying to understand the sort of actionable intelligence that doctors can treat prophylactically. Any examples?

• I’m not a doctor, or a biologist, so I asked my wife who’s a biologist. She gave a couple of examples, it’s very common in *genetic* disorders that there are multiple consequences of a particular syndrome and that the one that presents initially isn’t the one that is the worst.

She gave the example of Ehlers-Danlos syndrome. It’s a defect in the production of collagen, so typically you’d have injuries in joints and bruising and arthritis type conditions, early in life. Some people have been accused of abusing their children because the child with this undiagnosed syndrome looks like he or she has been beaten from bruising etc. As time goes on though the structural integrity of the heart and major blood vessels is compromised and they may die suddenly of a burst aorta. So by diagnosing the genetic disorder you know to monitor the heart and major blood vessels and then can intervene surgically to prevent this fatal outcome.

This pattern is pretty typical with genetic disorders because they accumulate problems through time as the genes fail to do their job.

• Charles Carter says:

I’m a general internist/ hospitalist. Your concerns and questions, naturally should be addressed by your EP physician. I’ll offer the following.
In response to Megan McArdle’s original question, I think most of us in practice today are at least qualitatively aware of changes in conditional probabilities after germane findings or events. There are exceptions and I think our heuristic black boxes aren’t great quantitatively.
Regarding your differential diagnoses, Brugada, long QT and WPW are usually evident on an EKG. And long QT usually results in polymorphic VT or “torsades des pointes” VT, distinguishable by tracings during the event.I believe an electrophysiology study can distinguish among these though your EP doc is best to decide if the study has decision-altering utility.
But I’m not convinced you frame the question in the most useful way. When you say your cath was clean, I assume you mean there was no coronary stenosis. You also state you were diagnosed with a cardiomyopathy. The vast majority of cardiomyopathies are associated with diminished LV function, decreased LV ejection fraction. Regardless of etiology the majority of deaths in patients with these illnesses, as well as with hypertrophic cardiomyopathy, are arrhythmogenic. If your EF is below 40% or you show certain other abnormalities by echo, that might be a more fruitful path to explore.
Sounds like you’ve received state of the art care so far. Best of luck.

• Rick says:

Thanks, Charles. My EP seems very knowledgeable and the care I received at Miami Valley here in Dayton, Ohio seemed to be top notch for sure.

It was established early on that I had a couple different ventricular arrythmias occur, including VT and Vfib (there might have been others). I believe the Long QT diagnosis is the favored view because of my EKG results.

I do not fully understand how the cardiomyopathy fits in, but it does appear on my discharge chart. However, the focus of every conversation with the dr so far has been on confirming Long QT. My understanding it the purpose of the cardiac MRI is to get an additional look at how healthy my heart is, as well as rule out other unusual possibilities.

The dr who performed the cath procedure told me as far as he was concerned, I have the healthy heart of a 37 yr old that he would expect to see. It showed nothing.

• Rick says:

Oh yes and when I was admitted, my EF was at 15%. It rose to 55% within about 12 hours of being admitted to the ICU (I am told this helped them rule out infection). They didn’t tell me what it was when I left, but I assume it is back to normal. I feel exactly the same as ever before, except perhaps a little bit more tired; but I assume that’s probably the beta blockers.

• Gordon Fox says:

Comment 1: hypertrophic cardiomyopathy (HCM) is the most common cause of SCA, as well as being the most common of the heritable heart diseases. Truth in advertising: I know a bit about it because I have HCM and, as a quantitative biologist, have worked a fair bit with the Hypertrophic Cardiomyopathy Association to help educate patients about the genetics, physiology, and occasionally, statistics.

Comment 2: HCM is highly heritable, but basic diagnoses don’t require genetic tests. Echocardiogram and/or MRI is what’s normally done. But “highly heritable” means that something between 50 and 60% of cases can definitively be assigned to be caused by known mutations (mostly in cardiac sarcomere proteins). The remainder of genetic “tests” are simply uninformative because much is still unknown about the genetics.

Comment 3: So I’m not sure that a statistical approach is going to be all that useful. Though it might be informative and entertaining. I don’t know about any of these other conditions, but I do know that with HCM, causation is complicated: numerous people with genes that certainly can cause the disease don’t ever develop it. And today genetic testing is mainly used as a way to decide which family members need to be watched closely; it’s really not yet used clinically.

Comment 4: Yep, having an SCA is a pretty good reason to have an ICD implanted, regardless of the cause of your trouble.

Comment 5: Some of the candidate diseases you list aren’t mutually exclusive. As an example, there are people with Long QT as well as HCM.

Comment 6: I don’t know about the other diseases, but we strongly recommend that people who may have HCM be evaluated by docs who specialize in the disease. The closest place to you that has such a specialty clinic also happens to be one of the best, at Cleveland Clinic. I travel a long way to get treated there, and lots of others do. What you might consider is sending them your records and asking their opinion – they do this all the time. They’ll charge you for it, of course, but it’s probably a sensible approach.

• Rick says:

HCM is not one that has not yet been mentioned to me specifically, but it definitely sounds like it COULD fit. I assume they’ll be looking for that when I finally get my cardiac MRI- but perhaps not?

However, I’ll definitely follow up with my doctor on this, and depending what he says, perhaps follow up your HCA. Unfortunately Cleveland Clinic does not accept our healthcare coverage but it might be worth making an exception on this one.

• Rick says:

Update: a heart surgeon friend (what a great friend to have in my situation!) comments that HCM should be able to be identified on my MRI pretty easily. But he did say based, on my echo he’d be very surprised if I have it- though it’s definitely possible.

2. Clyde Schechter says:

Dan Lakeland’s comment is spot on. These tests’ sensitivity and specificity are really both limited only by issues of specimen handling and lab administration. The actual analyses have virtually 100% sensitivity and specificity.

That said, it may or may not be worth while for this person to invest time and energy into the question he poses. Really the first question he should be asking is: how will this alter my treatment? While we have extensive knowledge about the genetics of many diseases, there are, despite the hype, very few situations today where knowledge of your genes provides actionable information. That may change over the next several years or decades, but at present “personalized medicine” remains a distant goal in all but a handful of domains.

• Receiving a diagnosis is not *only* about how it will affect treatment. For many people with complex (or simple, I suppose) medical conditions, the psychological impact of having a diagnosis is important. For many, a diagnosis is helpful, even when treatments are limited or ineffective. Simply knowing what is wrong provides a certain amount of satisfaction. Of course, this isn’t true for everyone…hence the reason some people choose not to have genetic testing done for things like Huntington’s, or BRCA, etc…

In any case, I feel that this figuring out how to rationally and economically deal with these sorts of scenarios is very important. It seems (I don’t have the data) that there has been an explosion in expensive diagnostic testing in healthcare. It also seems that much of this testing is done in a reflexive manner, without proper consideration of what the goal/outcome of the testing will be.

The original question is a perfect prototype for lots of what goes on daily in healthcare.

• Andrew says:

Michael:

I agree, and that was my motivation for writing this post, both for the particular patient in question and for the larger question of clarifying uncertainties and decision making in such problems.

• Rick says:

Another complicating factor: we are members of what is known has a “healthcare cost sharing ministry” (Medishare).

Because of this, we feel a moral responsibility to other members of the sharing group not to “waste” ministry funds on unnecessary tests that provide no actionable information for my treatment. If the test could provide some information that is needed to devise a treatment plan- such as my hearth catheterization- that is a different story. But these genetic tests, with the high false positive rate, seem like a waste of time and resources to me. But I also recognize that I’m pretty ignorant when it comes to medical knowledge.

• Rodney Sparapani says:

Hi Rick:

First of all, good luck to you. We all hope you will live a long healthy life. In regards to personalized medicine, I don’t want to make myself into an expert. But, having read a lot about this stuff recently, most of the individual traits that are known are non-genetic factors like age, gender, weight, smoking history, etc. Where genetic factors are probably best studied (besides BRCA) is for newborns who are born seriously ill where the diagnosis is challenging, and therefore, the treatment uncertain. In these cases, the cost of doing an entire genome sequencing (which is dropping fast) and reading the tea leaves is well worth it since these children are often hospitalized for long periods.

Rodney

• Charles Carter says:

True for whether the gene is present, not necessarily the illness.
Don’t know if the genetic basis for Wolfe-Parkinson-White is known, but it’s a useful example. This is a defect in cardiac conduction with a pathognomic EKG. Most with this never experience arrhythmias related to it. Some develop supraventricular tachycardia which is lethal only very very rarely. A very small minority develop VT, though it’s a known consequence.

3. Larry says:

He might consider studying Pearl’s The Book of Why, which describes causal models/diagrams and how to use them to handle Bayesian problems. Several of his cases discuss epidemiological problems. (He cites/explains the mammogram case in particular.)

4. Jim Dannemiller says:

One could simplify the problem by assuming that sudden cardiac arrest is associated with a mutation in a single gene (e.g., the one being genotyped). Prior to having the cardiac event, the probability that this person would carry the mutation is a function of the proportion (q = 1-p where p is the proportion of the wildtype, unmutated allelle) of that mutated allele in the population/subpopulation of which this person is a member. There are two ways that the person could carry the mutation: 1) they could be homozygous (carry two copies, probability q^2), 2) they could be heterozygous (carry one copy of the mutation, probability 2pq). These probabilities assume that the population is in what is called Hardy-Weinberg equilibrium: basically that mating is random and there is no strong selection pressure on the alleles of that gene among other conditions. After the cardiac event, that probability changes depending on the strength of the association between the mutation and the disorder. That, in turn, can depend on the penetrance of the mutation: high penetrance means that a high proportion of those carrying the mutation will also show the disorder. Finally, it is relevant that genotyping mistakes from reputable labs are fairly infrequent these days, so as a general statement, if the test says that you carry the mutation, you probably carry the mutation.

This is similar to some conditions such as breast cancer. Mutations in the BRCA1 and BRCA2 genes are strongly associated with breast and ovarian cancer. Those mutations appear much more frequently in women with Ashkenazi Jewish ancestry. Prior to being diagnosed with breast cancer, a women with Ashkenazi Jewish ancestry would have a probability of having that mutation that depended (in the absence of any additional family history of breast cancer) on its frequency in the population of Ashkenazi Jewish women. After being diagnosed with breast cancer, the probability that she has the mutation is extremely high. I would point this person toward the literature on the BRCA1 and BRCA2 genes and their association with breast cancer for guidance on how to determine these probabilities. Some genetic counselors also understand these kinds of problems and are trained to be able to engage with these issues in a quantitative way.

5. Clyde Schechter says:

“After being diagnosed with breast cancer, the probability that she has the mutation is extremely high. “

That’s not true. Yes, an Ashkenazi Jewish woman who has been diagnosed with breast cancer has a higher probability of being a BRCA1 or 2 carrier than just the general prevalence of those mutations in the Ashkenazi Jewish population. But a breast cancer diagnosis alone does not make that problem “extremely” high under any reasonable interpretation of “extremely.”

What would make the probability very high, high enough to warrant genetic testing and its sequelae, is if the diagnosis comes at a young age, or if there is a strong family history of breast or ovarian cancer. But if a 70 year old Ashkenazi Jewish women gets a breast cancer diagnosis and has no, or minor, family history of breast cancer, her probability of having a BRCA 1 or 2 mutation is only minimally elevated above Ashkenazi Jewish prevalence.

Indeed, in terms of actionable information, BRCA testing is more valuable if done *before* any cancer diagnosis, triggered by a strong family history of breast or ovarian cancer, particularly family history of such a diagnosis at a young age. You have better options then.

6. Phil says:

Maybe I’m underthinking this, but perhaps all this guy needs is Bayes’ Theorem:

p( condition | heart attack at age 37 ) = p( condition ) * p( heart attack at age 37 | condition ) / p( heart attack at age 37 )

Perhaps “heart attack at age 37” is hard to find in the literature, but maybe there’s something like “heart attack before age 40” or similar, and the equation can be rewritten that way.

In words, this is:

The probability that he has the genetic condition, given that he had a heart attack before age 40, is equal to:
(the probability that he has the condition given no additional information (i.e. what is the fraction of people in the population who have the condition))
times
(the probability of having a heart attack before age 40, given that he has the condition (i.e. what fraction of people with the condition have a heart attack before age 40)
divided by
(the probability of having a heart attack before age 40, given no additional information (i.e. what fraction of people in the population have a heart attack before 40).

• This calculation is problematic in many ways. It’s very likely for somewhat rare syndromes that the base frequency is unknown (p(condition) in your calc) and that the conditional rate of heart attacks is not well known (p(Heart Attack at age 37 | condition) in your calc)

Furthermore if there are several conditions of interest, the calculation gets more complicated, particularly if he could have more than one of them (though if they’re rare then this seems potentially ignorable, but if even one of them is reasonably common, you probably have to consider it in combination)

Finally, there’s the question of how much information you gain by doing the testing, ie. how these probabilities change when tests come back with certain results.

Most likely the quantities needed to plug into the calculation are not well known. You can start to deal with that by explicitly splitting out frequencies and probabilities into distinct quantities, and you can put probability distributions over the frequencies, but this is only likely to help you if you have a largish dataset of tests and heart attack survivors so that you could get a reasonably tight posterior distribution over the frequencies of interest.

• Phil says:

I don’t disagree with anything you say, Daniel. But if you don’t have a half-decent estimate of p(condition) or p(heart attack | condition), nor of course p( condition | heart attack) which is the unknown quantity of interest, then clearly it is impossible to quantify the probability in a useful way. Which is fine, not every question has a known answer. But if one _could_ answer the question, the Bayes’s Theorem approach I outlined would provide the answer.

If it were me, I might still try the exercise: take the estimated rates of the conditions in the population, and the estimated rates of heart attack given the condition, and attempt some sort of Bayes Factor approach to guessing which condition I have (with ‘none of the above’ being a wildcard of course). This would seem to be of mostly academic interest, given that the treatment would be the same in any case.

My general feeling is that everyone is trying to help Rick, whereas I am not trying to help Rick, I am just trying to answer the statistical question posed by the original poster! So I tried to do that. (Of course I would be delighted to help Rick, if I could, but I can’t).

• So, to expand on this calculation though in order to figure out the diagnostic value of testing, we really want

p(Condition | HeartAttack, TestResult, Background)

how do we combine the data on heart attack and test result? Are they statistically independent given our background knowledge? That seems unlikely. In particular we can imagine something like:

p(Condition | HeartAttack, TestResult, Background) = p(HeartAttack | Condition) p(Condition | TestResult,Background) p(TestResult | Background)/ p(HeartAttack,TestResult | Background)

not sure that this helps us, are these quantities easier to figure out, or calculate with? I think the thing this points out is that if

p(HeartAttack|Condition) is very large (near 1, it’s a condition that very commonly causes heart attack) and p(Condition | TestResult) is near 1 (it’s a good test), then for p(Condition | …) to be a probability (between 0 and 1) then almost everyone who gets the result will have had a heart attack. Basically the heart attack adds no additional information compared to the test result.

The same will not necessarily be true without testing for the condition. There are lots of reasons you could have a heart attack other than having this condition. But if the test is specific, those reasons become irrelevant after the test.

• On the other hand, the *type* of heart attack may well be very unusual in the absence of given conditions. So you might have
p(HeartAttackOfThisType | Condition) is negligibly small for Condition=Normal (like 1e-7), and nontrivially large for several conditions Condition[i] for i=1..5 or so. Then, even if the other conditions are unusual, they will wind up being dominant over Condition=Normal

This seems to be the situation Rick is in. The doctors don’t think it’s likely he’s a “normal” person, instead he almost certainly has one of the conditions he listed.

The next question is which condition do you have? and what do you do about it if you have it? Decision theory. The main reason to narrow down the probability of the correct condition is if the main, or ancillary care decisions depend importantly on the condition. As I said above somewhere for example you might have secondary symptoms that could be prevented or pre-treated or the like.

If the “best” treatment decision is the same for all conditions[i] that are in consideration after the knowledge of this unusual cardiac event, then all that you need to know is “it’s very likely one of these, we do the same thing for all of them”. If on the other hand some specific conditions have different additional/secondary treatments, then it makes sense to decide whether those additional secondary treatments are required, and so you’d want to test *only* for those conditions that involve specialized additional treatments. If the test is good, then after a positive test for a given thing requiring ancillary treatment, you’d have actionable information.

• Rahul says:

I think the methodology makes sense. But I do think we sorely lack the data to do anything actionable with this sort of framework.

It all sounds great in principle, but whenever I’ve tried applying it to reality I get stuck.

7. jim says:

Hi Rick,just want to wish you the best. Take care.

8. Rick says:

In case anyone is interested in more details, my wife maintained a Caring Bridge page during the worst of it:

https://www.caringbridge.org/visit/rickteacheyjr

I’m still interested in the original question of how my SCD event affects how I should interpret potential the results future genetic tests. But after much discussion here and elsewhere, at the moment it does seem like there is little point, in my specific case, in going forward with them.

However if I do become interested in tests after discussion with doctors I’ll be sure to post details back here and see what I can learn from all of you. It will take some time to actually get there (weeks at the very least).

Also a big thanks to Megan for reaching out about my case in the first place; you’re a good friend.

• Ross Mohan says:

Hey Rick best of luck to you.

Sounds like your best weapon so far (aside from the fact that you made it through the initial attack) is your desire to explore and improve further. Good luck with that.

A lot of smart people here and I know next to nothing about this but if I were in your boots I’d be looking at some of the epigenetic gateways that might turn off or turn on the manifestation of this genetic predisposition.

Again, very best of luck to you and your family.

9. Prashant Singh says:

Hi Rick,

All the best to you and your family.

Embrace the uncertainty and cherish each day – I was diagnosed with Idiopathic Cardiomayopathy 3 years ago. Ace inhibitors and beta blockers since then, lung surgery to fix a collapsed lung [spontaneous pneumothorax] not related to the cardiomyopathy. Tests are inconclusive on whether its genetic or environmental.

Solution: I got a dog [Great Pyrenees!!]

https://www.ahajournals.org/doi/abs/10.1161/CIR.0b013e31829201e1

I may not have control on the probability of the event occuring due to genetic factors but I can change the probability of the event occuring by tweaking my environmental factors.

All the best mate!

10. Joshua Pritikin says:

Even if you have a “healthy 37 year old heart,” you should consider Caldwell Esselstyn’s diet to reverse heart disease. Even if this does not address your particular root cause, it won’t hurt since it’s a health diet anyway.