Big Data vs Breast Cancer Trials Raises Big Issues

Kathy D. Miller, MD


April 27, 2016

This feature requires the newest version of Flash. You can download it here.

Hi. It's Dr Kathy Miller from Indiana University. I want to think with you today about the power of big data.

We've answered many important clinical questions in oncology with large randomized clinical trials. But we recognize that those clinical trials are expensive, labor intensive, and have some inherent biases because only a small subset of our adult patients enroll in clinical trials.

Some have suggested that maybe clinical trials are a thing of the past.

As bioinformatics capabilities have expanded dramatically over the past decade, some have suggested that maybe clinical trials are a thing of the past. Maybe big data can help us address some of these questions much faster, much less expensively, and maybe with better results, because all of our patients, or a much greater subset of our patients, would contribute to those efforts. We see that when people use the SEER (Surveillance, Epidemiology, and End Results) database or when they use individual claims databases.

That is part of the goal of the American Society of Clinical Oncology's (ASCO) CancerLinQ initiative— combined wisdom in taking data and experiences from all of our patients in the community and putting them together to address these important questions. But will we get good answers or will we just shift from one source of bias to a different source of bias?

With that in mind, I want you to take a look at a fascinating report in one of the most recent issues of the Journal of Clinical Oncology. This report, by Katherine Henson and colleagues,[1] addresses this question. It looks at two specific issues in breast cancer treatment that have been extensively studied: the role and benefits of radiation therapy in women who had breast-conserving surgery and in women who had a mastectomy. Those questions have been addressed in individual clinical trials, and those individual trials have been put together in our every-5-year meta-analysis.[2,3] We think we understand the benefits of radiation in those two settings. Katherine and her colleagues took the SEER database—which reflects real-world patients—and looked at women who had radiation or not, women who had breast-conserving surgery, and women who had mastectomy. They then compared the estimate of benefit that you would come to on the basis of the randomized trials vs the SEER database.

For women who had breast-conserving surgery, both ways of addressing this question found significant improvement, but the benefit seemed much greater in women in the SEER database than in the randomized clinical trials. The story got even more complicated when [Henson and colleagues] looked at post-mastectomy radiation. The clinical trials have shown us that there is a small but real benefit to post-mastectomy radiation. In the SEER database, mastectomy patients who had radiation did worse.

As best as they could with the available data, the authors tried to control for covariants and comorbidities. Controlling for those things slightly narrowed the differences but didn't resolve them. It really had minimal impact. You could come up with explanations for why this might be. Perhaps biases actually informed clinical opinion, and the patients in the real world who had mastectomy and got radiation had worse disease, and that's why they did worse.

Perhaps the same is true in the breast-conserving therapy setting. Perhaps the patients who didn't get radiation had a lot more comorbidities. They did worse—maybe not from their breast cancer, but just worse in general, and that's why the benefit seemed greater.

If we're going to abandon clinical trials in favor of big data, we need to do so very carefully.

We can't know for sure. What we do know is that the results are very different, and if we're going to abandon clinical trials in favor of big data, we need to do so very carefully.

There are biases on both sides, and we need to be ever mindful that there is no study that is free of bias. Our job should be to understand the bias, try to understand its impact, and use those sources where they're best. Large databases might be the only way to address questions of rare toxicities, rare cancers, or rare late events where we'll never have the numbers in a randomized trial. But for common diseases and common important questions, I think we're still going to be doing randomized trials as the only way to try to isolate the important question. Take a look at the article and see what you think.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.