Last month, a study raised questions about over-treatment of ductal carcinoma in situ, more commonly known as DCIS. It's the earliest form of breast cancer and some doctors don't even think it warrants the label "cancer." Nevertheless about 60,000 women receive a DCIS diagnosis annually in the United States -- more than the number that are diagnosed with all forms of colon cancer put together -- and the vast majority are treated as aggressively, or more aggressively, than those with invasive breast cancer.
Suffice it to say: DCIS is the poster child of the need for a new research agenda in American medicine -- research on when it's safe, even beneficial, to have less medical care.
Research on less is not easy. Imagine designing an experiment to test the safety of moving from the most common current DCIS therapy (lumpectomy and radiation) to a much less aggressive approach (say, taking an oral anti-estrogen drug). The first thing to do is to decide how many patients need to be recruited, which in turn depends on how common the outcome is. Last month's study found that the 10-year risk of breast cancer death among women with DCIS following current therapy is about 1 percent (for context, over 30 percent of women diagnosed with colon cancer will die from it in 10 years). Even though the less aggressive approach to DCIS might be better, you would want to design the experiment to be certain that it was not substantially worse: say, no higher than a 1.5 percent risk of breast cancer death.
That study would need to enroll about 20,000 women. To recruit that many women would require the participation of multiple medical centers. And the experimental process could easily take two decades - by the time the researchers had identified the centers willing to participate, trained study coordinators in each, recruited all the women needed and followed each for the 10 years.
Welcome to the world of studying low probability events. When the risk is high, the study can be small. But when the risk is low, the study needs to be big -- and long. Because study participants will have so few events, investigators need a lot of them -- and a lot of time -- to have enough events to study. So it's hard to know the real effect of attempts to lower already low risks. Doctors are increasingly worried about their patients getting too much medical care. Interventions to deal with low probability events are a prime candidate for research on less -- settings where treatment can be worse than the disease.
Not uncommonly, interventions that can successfully reduce the risk of death from one disease end up increasing the risk from others. While this may be an acceptable trade-off for patients whose death is most likely to be from that one disease, it may not be for those who are at higher risk to die from something else. Women with DCIS, for example, are more than 5 times more likely to die from other causes than they are from breast cancer. They might reasonably ask for an experiment that ensures that they are not trading off one form of death for another. That requires examining the effects of intervention on death for any reason.
There's just one problem: That study would require about 80,000 women.
For many of us, it's not all about avoiding death anyway. Other things matter: like how much we are poked and prodded, how much pain and sickness medical intervention causes, and how vulnerable the entire process makes us feel. While you might be willing to go through hell to move from a 50 percent to 20 percent risk of cancer death, you might less enthusiastic if the move was only from 1.5 percent to 1 percent. Ask yourself this: How much are you willing to go through to increase the chance of not having a breast cancer death from 98.5 percent to 99 percent?
And then there's the question: How much will this all cost me? Patients increasingly have "skin in the game" and can be on the hook for substantial out-of-pocket payments. Medical expenditures can lead to serious financial stress; even among the insured, they are a common cause of bankruptcy. That can't be good for your health.
Yet the medical care industrial complex is gearing up to identify more low probability events. The reason is simple: That's where the money is. Just about everybody and their brother seems to be searching for disease markers: a minor genetic variant, too much of protein X, too little of protein Y, a concerning pattern in your immune function, or something funny in your breath. They are just trying to sell tests that find things -- with little attention given to the question of what to do when things are found.
By the way, that's the way we got into the DCIS mess. We were just trying to find smaller and smaller cancers. No one asked the question what will we do when we find them.
Medical research is increasingly dominated by industry funding. The industry's interests that foster research on more ought to be counterbalanced by an equivalent effort devoted to research on less. American medicine needs a cadre of appropriately skeptical clinician-researchers asking, Does this really serve our patients' interests?
That requires data. And because experiments on low probability events require impracticable numbers of patients, that data will need come from current medical practice. To get data on 100,000 women with DCIS, last month's study used the federal government's cancer registry. The registry has detailed data on a patient's cancer, their first course of treatment and whether or not they are alive today. There's nothing about how their cancer was found or what they went through after their initial therapy and three quarters of the country is not included. And this is from arguably the best cancer registry in the world. Furthermore, cancer registries are light-years ahead of registries for heart disease, diabetes or Alzheimer's disease.
The data are out there -- in the computers of insurance companies. But because there are multiple insurers and because individuals (and their employers) regularly change insurers, it's virtually impossible to get a comprehensive picture of what is actually happening to patients in the United States.
Call it the Balkanization of data. Even if you don't chose to support a single-payer system, you ought to support a single-data system. It's hard to learn whether it would be better to have less medical care in the future, if we don't even know what's happening now.