I mentioned cancer incidence versus cancer mortality the other day, and I wanted to highlight this NEJM paper, which is a recent and comprehensive look at the topic. You can see several different effects in the data. Hodgkin’s lymphoma, for example, has shown a pretty steady incidence rate over the past 40 years, but steadily declining mortality, which indicates a steady improvement in overall treatment. Pancreatic cancer (mentioned but not graphed in the article) has both lines holding steady: it occurs in the population at the same rate it always did, and over the 40-year span of the data we were sadly unable to do anything to improve the live spans of those diagnosed with it.
For chronic myeloid leukemia (CML), on the other hand, the incidence rate has also been steady, but there’s an abrupt improvement in the mortality figures that kicks in the during the late 1990s, shown at right. The scale is relative to the 1975 figures, and the decline is the introduction of Gleevec (imatinib). Anyone who worked on that one can look at that curve and know exactly what they accomplished. CML, as it turns out, is less genetically complex than most other types of cancer, and was thus well suited to be targeted by specific kinase inhibitors. Getting this to happen for more heterogeneous and mutation-prone tumors is a struggle indeed.
Interpretation of the data becomes trickier when the incidence rate itself shows a change, though. As the authors note, lung cancer is a good example: the incidence rate and the mortality rates moved in almost exact parallel from 1975-2015, which argues that both these effects are real and are tied together. These declines clearly seem tied to the overall decline in smoking: the cancer rates are just about thirty years behind that decline (lung cancer takes time) and they move right along with it. The post earlier this week about cancer news was prompted, in fact, by an apparent change in that tight correlation, because new therapies seem to be making lung cancer mortality decrease even more than the incidence rates would have predicted.
Stomach cancer is a case where the incidence rates and the mortality rates seem to have been steadily falling over the entire 40-year period studied (an effect seen across the industrialized world). There have been no great breakthroughs in overall diagnosis or treatment; it’s just happening less often. About half of that decline is attributed to decreasing H. pylori infection, and the rest to general lifestyle and quality-of-life changes. Cervical and colorectal cancer are in roughly the same category (incidence and mortality both going down), the latter starting its decline in the early 1980s.
And those last two bring up the issue of screening. Colorectal cancer incidence was actually declining before widespread screening was introduced, though – that one seems to be similar to stomach cancer and reflect a real decline in the underlying rate – and since both of these are associated with the digestive tract, they might both be broadly related to improvements in the food supply. Cervical cancer screening seems to have definitely had an effect on the data, and the fact that precancerous lesions are discovered and treated makes the decline on incidence rate especially strong. But screening in general does not make incidence rates go down – the opposite, in fact, which is what one might expect.
That shows up in the data for thyroid cancer, kidney cancer, and melanoma. Incidence rates for these have been climbing steadily (thyroid since the mid-1990s, melanoma across the whole 40 years), but overall mortality has stayed the same. And while treatment has improved, it has not improved nearly enough to cancel out such dramatic incidence increases – if those incidence increases were real. As the authors point out, these would seem to be clear examples of overdiagnosis. We are telling too many people that they have these cancers when they don’t – or at least, when they don’t have any form of them that they really have to do anything about. We are treating people who should not be treated and we are terrifying people who should not be terrified.
The messiest categories (you know these were coming) are breast and prostate cancer. Both showed large increases in incidence rate on the introduction of widespread mammography (1980-1990) and PSA screening (1987-early 1990s), respectively. But the prostate incidence rate has subsequently fallen, while the breast cancer incidence rate has remained at what may be a new, higher plateau. Meanwhile, mortality for both has declined, breast cancer since about 1990 (and more steeply in recent years, it seems) and prostate since about 1995. The resulting curves in both cases would appear to be a mixture of true incidence rates, overdiagnosis due to screening, and advances in therapy, and it’s not easy to untangle the relative effects of all these. Arguing about these matters has been intense, as many will have noted, but there seems little room to argue that both breast cancer and prostate cancer are indeed overdiagnosed.
Putting all of these together into overall cancer rate figures obscures as much as it reveals, to be honest. That’s because one thing I haven’t been mentioning is the relative differences between all these incidence rates. The overall cancer incidence rate for men looks pretty variable over the years, for example, largely because of those big changes in prostate incidence numbers (prostate having taken over as largest single category). For women, an earlier rise in incidence rate seems to have been driven by rises in lung cancer (mostly very real) and breast cancer (a significant amount of which may not have been). Thyroid cancer (over)diagnosis is hurting the incidence rate among women, who are three times as likely to get the diagnosis, although mortality rates are the same for both sexes. And for both men and women, overdiagnosis in melanoma and kidney cancer have made the overall cancer incidence rates look worse than they surely are.
Cancer mortality, though, has been declining across the board since 1990. And that’s hard to argue with, since death is a definitive endpoint for the data. There are an awful lot of people in the business trying to make sure that it goes down even further!