On the enormous variability of cancer behavior

ResearchBlogging.orgPerhaps one of the most common misconceptions held about cancer among lay people is that it is one disease. We often hear non-physicians talk about “curing cancer” as though it were a single disease. Sometimes, we even hear physicians, who should know better, using the same sort of fuzzy thinking and language about “curing cancer” as well. But cancer is not a single disease. Indeed, it’s a collection of dozens of different diseases, with different cell types of origin, pathophysiologies, behaviors, and treatments. True, there are a fair number of commonalities between cancers in terms of shared molecular mechanisms of evading the immune system, overcoming the body’s natural checks and balances to proliferate in an uncontrolled manner, induce angiogenesis to supply themselves with a blood supply, and metastasizing, but the differences make talk of “curing cancer” to be wishful thinking. Of course, curing individual cancers is entirely possible. Indeed, we already routinely cure Hodgkin’s lymphoma, for example, as well as breast cancer, colon cancer, and various leukemias, among others. It’s the seductive idea that there is a “magic bullet” or a single treatment that will cure cancer that is the wishful thinking.

Indeed, a recent study out of Norway1 could be taken as evidence that even different cases of the same cancer might reasonably be considered different diseases, so different are their behaviors. This study was recently reported in the media thusly:

LONDON (Reuters) – Five percent of breast cancer tumors appear to double in size in just over a month, Norwegian researchers said on Thursday in a study underscoring the potential benefits of more frequent screening.

The study published in the journal Breast Cancer Research also suggested detection rates of just 26 percent for a 5 mm tumor, and 91 percent for a 10 mm tumor.

The researchers used a computer model fed with national screening and cancer data to calculate how quickly tumors grow and estimate the proportion of breast cancers detected at screening.

This data on nearly 400,000 women aged 50 to 69 helped them estimate that about 5 percent of tumors may double in just over a month, growing from 10 mm to 20 mm. This was mainly among younger women in the study’s age group.

The amazing thing about this study is that it demonstrates just how much variability in the growth of a common tumor like breast cancer. The power of this study comes largely from the nationalized health care of a country like Norway, where there is a high-quality, population-based Cancer Registry and a unique personal identity number for each inhabitant of the country. Because reporting of cancer cases to the registry is mandatory, this centralization allows the linking of disparate sources of health information, including serial mammography results, details of cancer diagnoses and the pathology of tumors, and treatment results, allowing the examination of close to 400,000 women. Through a complex statistical model which, I admit, I don’t fully understand, investigators modeled cancer results of the Norwegian Breast Cancer Screening Program, which took into account the size at which tumors were first detected, age, background incidence of cancer from historical data, tumor sizes at the time of definitive surgery to remove them, and various other sources.

The results of this study can be summarized as follows:

  • Tumors in women 50-59 years of age take a mean 1.4 years to grow from 10 mm in diameter to 20 mm in diameter while tumors in women between 60-69 years of age take a mean of 2.1 years.
  • Overall, the mean doubling time for all women was 1.7 years.
  • The mean “sojourn time,” defined as the time between when a tumor is visible on an imaging study and when it becomes noticeable to a woman as a lump was 2.9 years.
  • Detection rates were 26% for a 5 mm tumor, and 91% for a 10 mm tumor, implying that the vast majority of tumors are detected when they are between 5 adn 10 mm in diameter.
  • 5% of tumors had a doubling time of less than 1.2 months to double in diameter while, at the other end of the scale, 5% of tumors took more than 6.3 years.

This last finding is the one that made my jaw drop. It also made the study’s statistician sit up and take notice:

“The variation was larger than what I was expecting,” said Harald Weedon-Fekjaer, a statistician at the Cancer Registry of Norway, who led the study.

Indeed. The difference between the fastest growing 5% of tumors and the slowest growing 5% of tumors is striking. It tells us that for some women, waiting a month for surgery is a big deal, while for others, waiting a year probably wouldn’t matter. The problem, of course, is that we have no good way to predict accurately which patient will fall into which category, except in a very rough way. Not surprisingly, the patients with the faster-growing tumors tended to be younger and the women with slower-growing tumors tended to be older. Ultimately, a curve was produced that showed the percentiles for different growth rates, and this curve shows extreme variability:


Each line represents a percentile. The point where all the curves intersect is the point where tumors were 1.5 cm in diameter, which was taken as the baseline tumor size. From the way the curves fan out, you can see the variability in growth rates of the various groups of women. It is this extreme variability that inspired me to speculate about how different breast cancers behave almost as different diseases. Of course, this is nothing that physicians taking care of breast cancer patients didn’t already know: That there is a huge variation in the clinical behavior of breast cancer. Anyone who’s ever taken care of breast cancer patients for a significant length of time has seen the patient whose tumor was very indolent and slow growing and thus who did well for years even with stage IV disease. On the other end of the spectrum, we’ve all seen the occasional patient whose tumor was so rapidly progressive that it appeared to have grown in just the 2-4 weeks that is the typical interval between diagnosis and definitive surgery. What this study does that’s helpful is to quantify the magnitude of that difference.

The model used in this study is not perfect. (No model is.) However, it is the first one that I’m aware of that modeled directly so many different parameters for such a large study population size. perhaps the biggest problem I see is the necessity to estimate the background “expected incidence” of breast cancer in the population, which may have been underestimated due to increased use of hormone replacement therapy during some of the years included in the study. There are some other potential weaknesses in the model. For instance, the model assumes that the sensitivity of the test increases towards 100% as tumor size increases, while some tumors may never become visible on mammograms, whose confounding factors could be hard to distinguish from the effects of HRT. Even so, the model has a fair amount of support, and is based on a very large number of women.

About a year ago, I wrote about how the relationship between the early detection of cancer and decreased death rates from cancer is not as simple a relationship as one might think. One area where models like this can help is in providing hard figures to guide us in deciding how often mammography screening is indicated. In other words, what’s the best balance between detecting cancer early and cost, or, phrased differently, what’s the best screening strategy to account for length bias? In the 5% of women with the fastest-growing tumors, it’s not hard to see how, taken together, this data suggests that no screening regimen is likely to detect a cancer before it reaches 2 cm in diameter except maybe 10% of the time. The converse of this is that the 5% of women with slow-growing tumors could do just as well with screening every three years, which is the norm in many European countries. Once again, though, the problem is that we have no way of knowing who will fall into which category. This sort of data is also highly useful in giving us a much better estimate of lead time bias.

Perhaps the most useful potential of this sort of model is that it could lead to strategies that could minimize the phenomenon of overdiagnosis; i.e., women with small tumors that would never have grown to threaten their lives and were only detected because of a sensitive screening program. As the authors put it:

Whereas screening with mammography has been related to reduced mortality in several randomized trials [32,38], so-called overdiagnosis remains a controversial topic. Following the conservative definition of the number of overdiagnosed cases as ‘the number of women who would not had breast cancer in their life time without participating in mammography screening’, our new model can be used to estimate the level of overdiagnosis under different screening designs…The new method presented here provides a toolbox for estimating this and other central issues related to mammography screening.

Designing mass screening programs that strike the best balance between sensitivity (not missing real cancers), minimizing overdiagnosis, and cost is a highly complex and difficult business. Although mammography screening clearly decreases mortality in women over 50, whether it does in younger age groups is not as clear, for instance. Moreover, it’s not clear whether screening yearly, every other year, or every three years is optimal. Models like this can be used to change such variables and model what is likely to happen as a result. This sort of information can thus help clinicians refine and optimize screening regimens.


1. Weedon-Fekjaer, H.H., Lindqvist, B.H., Aalen, O.O., Vatten, L.J., Tretli, S.S. (2008). Breast cancer tumor growth estimated through mammography screening data. Breast Cancer Research, 10(3), R41. DOI: 10.1186/bcr2092