Primum Non Nocere
The Pitfalls of Early Detection.
Medicine loves a good maxim. One of the most famous, which every medical student learns early on, is primum non nocere: first do no harm.
The phrase has ancient origins, dating back at least to the time of Hippocrates, when archaic methods based on scholastic theories often made the cure worse than the disease.
More than two thousand years later, even after the scientific revolution, not much had changed. When it came to late 19th century medicine, expectations remained low. As Tolstoy put it in War and Peace, “Though the doctors treated him, let his blood, and gave him medications to drink, he nevertheless recovered.”
Since then we have come a long way. Surely advanced technology and evidence-based medicine have made it time to retire the phrase. By now we must at least have figured out how to not do harm.
I’m afraid not. In the age of modern medicine, primum non nocere remains more relevant than ever.
The other day a patient came into my office carrying a news clipping. I want to get this, he said, handing it over. What do you think?
The article described a new kind of test, made by a company called GRAIL, that claims to be able to detect dozens of early stage cancers by finding abnormal DNA and proteins in the blood.
Of course, a positive test is just the beginning. Next comes the workup, with more tests, scans, and biopsies. Then comes the treatment, usually some combination of medication, radiation, and surgery. But in the end, at least in theory, you’ve removed the cancer before it had a chance to spread.
Seems like a no-brainer.
After all, who can argue with a strategy of early detection? It’s the basis for the screening tests that doctors recommend as part of routine preventive care: mammograms, colonoscopies, pap smears and skin checks. It’s why we check blood pressure, sugar and cholesterol at annual physicals – to catch a problem early and nip it in the bud.
But the allure of early detection is also the basis for expensive and unproven marketing schemes such as full-body MRI, microbiome stool testing, and genomic panels like 23andMe.
And let’s not forget Elizabeth Holmes, the dark princess of early detection, whose fraudulent company, Theranos, peddled a magic box that claimed to detect – and thus to prevent – a panoply of diseases in a mere drop of blood.
Even if you believe that one bad apple like Holmes shouldn’t spoil the bunch, grant me at least that any idea seductive enough to attract billions of dollars yet turns out to be a con, deserves a closer look.
In reality, there are at least a dozen ways in which early detection can go wrong. In the interest of time and space, here are three.
The most obvious way that any test can go wrong is if it is wrong, with false positive or false negative results. Contrary to popular belief, false results have less to do with the intrinsic accuracy of a test than with the prevalence of disease in a given population.
This concept is known as Bayes’ theorem. Certainly one of the most useful rules in medicine, some call it the most powerful law in all of probability and statistics. It holds true for tests of anything but is particularly relevant to screening asymptomatic individuals.
Here’s an example of how it works.
Let’s say a given test is 97% sensitive (correctly catches positives) and 95% specific (correctly rejects negatives), which is about as good as it gets. Nevertheless, in a population where the prevalence of disease is low – let’s say 0.5% – the chance that someone with a positive test actually has the disease is only 9%, which is a false positive rate of 91%!
In the case of GRAIL, Bayes’ theorem tells us that even if the accuracy is so good that if you have cancer the test will surely pick it up, nevertheless a positive test does not mean that you have cancer.
Even if it’s right, early detection can still go wrong. This is because the closer you look the more you will find, including low risk disease which, despite the fact that it is cancer, does not pose a threat.
Prostate cancer is a classic example of overtreatment. The second most common cause of death from cancer in men, prostate cancer is unquestionably a grave disease. The problem is, over 90% of prostate cancers occur in older men, with autopsy studies showing that almost half of all men over the age of 70 have evidence of microscopic disease.
The majority of these men die with prostate cancer, not from prostate cancer. PSA testing, which does not distinguish between high risk and low risk disease, picks it all up. And once you know you have cancer, you tend to want it taken out, which means surgery or radiation, both of which carry a high risk of complications.
It got to the point where overtreatment of prostate cancer was causing more than 40 cases of impotence or incontinence for every life saved. That’s a lot of unnecessary pain and suffering.
Breast cancer is another example where detection and treatment has greatly increased without a proportional decrease in mortality, suggesting overtreatment of low risk disease.
A test like GRAIL, designed to detect cancers indiscriminately, has the potential to lead to massive overtreatment, with all the attendant physical and psychological complications that can upend lives for no good reason.
Let’s imagine that we’ve cleared the above two hurdles. The positive test is truly positive, and the cancer that it detects is a high risk tumor that everyone agrees has to come out.
Still, as easy as it is to imagine everything going right, that’s how hard it is to imagine all the ways that things can go wrong.
For me this lesson is personal.
Eighteen years ago, my grandfather-in- law – Rachel’s paternal grandfather – was admitted to the hospital for surgery. Six months earlier he had lost his wife after a prolonged and difficult battle with an autoimmune disease called scleroderma.
He was just beginning to come out of this traumatic experience and starting to live again, seeing friends, going out to dinner or a show. I remember a dance concert he attended with us and noting that he looked happy.
What happened next was classic.
He went to his cardiologist for a checkup. An abnormality was found on his EKG. He was sent for an angiogram. It showed diffuse coronary artery disease that couldn’t be fixed with a stent. Bypass surgery was recommended.
At first it seemed like the surgery went well and he was transferred to the ICU as planned. He never came out. One complication led to another, and after a three month medical drama that would read as a list of mostly incomprehensible medical jargon if I detailed it all here, he finally passed away.
The amazing thing is, nothing was done wrong! Every decision in his case, from beginning to end, both at the time and in retrospect, made sense. I’m sure most doctors would do it all over again.
And yet, here is a man who was doing fine, who most likely had at least a few more good years ahead of him, whose life was cut short by “appropriate” medical interventions that he ultimately would have been better off without.
So is there really no such thing as too much information? Do I really want to take a test that will tell me what lurks in my body or my blood or my genes?
Not unless you can prove to me that in a representative population such a test – including all the intervention that follows – has a clear benefit on morbidity or mortality or both. And even then, it could easily end up being a big mistake.
In the case of GRAIL, that kind of information is not available, and won’t be for many years, if ever.
While I may have strong opinions on the subject, don’t think that means I mind being asked. On the contrary.
I’m grateful that my patient came to me instead of falling for some commercial pitch, that he gave me the opportunity to provide good medical advice: GRAIL is an unproven test, being sold to you for profit, that could easily end up doing more harm than good.
Primum non nocere – still relevant, and harder than it looks.