The Surprising Effect of Mammogram Recommendations

What are the costs and benefits of mammograms? How do you weigh, for example, the potential savings and healthcare cost against the potential lives saved, against the increase in psychic costs -- these are all very, very difficult issues.

My collaborators and I have recently been looking into the highly controversial area of recommendations for whether and when to screen for breast cancer using mammograms. There's concerns that not enough women are getting screened for mammograms.

This is a hugely relevant issue for many women. One in eight women get a breast cancer diagnosis in their lifetimes, so there's this recommendation to get annual testings at age 40. There's also concerns that many women are getting screened and having false positives... 

...which is when you detect a tumor, and you treat it, but if you had left it alone, it wouldn't have been a problem. Overdiagnosis is a problem because you're incurring a lot of costs that you really shouldn't have had to. It causes a lot of anxiety to people if they're diagnosed with cancer, so we really want to reduce overdiagnosis.

So we're asking how does a person who responds to a recommendation -- so, in this case, for mammograms -- differ from people that don't get screened or from the average person in the population. We got data on people who were screened, so we could see the rates of screening by age.

Before age 40, about 10% of people were getting screened. After age 40, that jumped up to about 35% of people. The share of people getting mammograms went up drastically at 40, but the share of people who got mammograms who tested positive for cancer was going down. But that just tells us how the people who respond to the recommendation at 41 differ from the people who get mammograms without a recommendation at 39. It doesn't tell us what we really wanted to know, which is, how do the women who are getting mammograms at 41 -- when it's recommended -- differ from the women who aren't getting mammograms? 

That's a very hard problem to answer because what you need is data on the underlying cancer of people who, by definition, aren't being screened. That's where the biologists and the clinicians came in -- they developed models of the underlying incidence of breast cancer in, say, a random 25-year-old in the population as well as, most importantly, for our purposes, how it progresses in the absence of treatment.

This is, I think, a wonderful case where the medical community has so much to add here, and then the economics community can take that and build onto that. This model gave us the underlying level of cancer, and, using that, we could back out what the cancer level was for people who never got screened. It's like supposing you have a roomful of people. There's ten people, and you know that half of them have cancer, but you don't know who. So you pick five, and you screen them. and only one of them has cancer -- then you know that, of the other five, four of them have cancer, even though you didn't screen them.

So we find that the people who follow the recommendation actually are healthier than the people who don't follow it -- they're less likely to have cancer, and if they're diagnosed with cancer, they're more likely to have an earlier stage cancer or a smaller cancer that's less dangerous. We wanted to study what types of women get mammograms to see whether they're the types of women who would benefit most. 

Right now, it seems like the recommendations are targeting people who are most healthy. Maybe they engage in other preventive health behaviors, and maybe they're doing really well, but we're not reaching the people who have a higher burden of cancer. So we'd like to look into ways that we could target those people better. 

So, in the end, to be clear, we don't resolve or even attempt to resolve the question that, in some sense, motivates the whole literature, which is, should we recommend screening at 40, 45, 50, 35? Instead, all we do is add another piece to the puzzle. 

So our paper brings an additional dimension that should be considered when sort of designing these policies. You have to worry about who the people are that you're reaching, and if they're the people who are more or less at risk for having cancer than a randomly chosen person in the population.