Separating Facts from Misinterpretations in Cervical Cancer Epidemiology: Why Biostatistics should matter to every physician
A recent article in the journal Cancer that focused on racial disparities in cervical cancer mortality has garnered widespread attention. The article, “Hysterectomy-corrected cervical cancer mortality rates reveal a larger racial disparity in the United States”, was reported on in a New York Times Health Piece, “Wider Racial Gap Found in Cervical Cancer Deaths”, and similar commentaries appeared on CNN, Time, and other news outlets. The article in Time magazine opened with the bleak conclusion that “cervical cancer kills more women in the U.S. than health experts have previously believed”. The January 25th digest from The Skimm, an email newsletter with 3.5 million subscribers, similarly highlighted that “a new study was published that showed more women in the US are dying of cervical cancer than people realized.”
So what’s notable about this coverage?
The interpretations offered by Time.com and The Skimm are incorrect.
It’s not that more women are dying from cervical cancer, but rather that those deaths from cervical cancer are instead occurring at higher rates.
It may help to take a step back to understand the science. Cancer of the cervix, which is the lower part of the uterus in the female reproductive system, is the 21st most common type of cancer and in 2016 is estimated to have caused 12,990 new cases and 4,120 deaths. The most important risk factor for invasive cervical cancer is human papillomavirus (HPV) infection. Screening for cervical cancer is accomplished through Pap smears, which involve collection of cells from the cervix and examination under a microscope to find early cancerous abnormalities.
Notably, the cervix is often taken out during a hysterectomy. This procedure primarily involves the surgical removal of the uterus but can involve the cervix, ovaries, and fallopian tubes depending on the reason for the hysterectomy. Common indications include excessive bleeding and uterine fibroids, and CDC data indicate that 600,000 hysterectomies are performed each year making it the “second most frequently performed major surgical procedure among reproductive-aged women”.
The authors of the paper in Cancer sought to understand how the cervical cancer death rates (calculated by dividing the total deaths by the number of vulnerable women) would change if women post-hysterectomy were not considered to be at risk for cervical cancer. After all, almost all of these women no longer have a cervix from which a cancer could arise and the American College of Obstetricians and Gynecologists, the American Cancer Society, and the United States Preventative Services Task Force all recommend against cervical cancer screening post-hysterectomy if the cervix was removed and the hysterectomy was not performed as treatment for pre-cancer or cancer of the cervix.
The authors found that the overall prevalence of hysterectomy was 20% among women older than 20, and was higher for black women than white women. So, they removed this population from the denominator, and rather than comparing “cervical cancer deaths” to “all women” they compared “cervical cancer deaths” to “all women except those who have had hysterectomies” — such that the mortality rate truly reflected only deaths among those who would be at risk for cervical cancer. The adjusted mortality rates were higher and revealed wider racial gaps, as shown below.
aMortality rates are deaths per 100,000 women
bCalculated as the change in mortality rate divided by the uncorrected rate
Returning to the popular coverage of this paper, the New York Times was correct in stating, “The new rates do not reflect a rise in the number of deaths, which recent estimates put at more than 4,000 a year in the United States. Instead, the figures come from a re-examination of existing numbers, in an adjusted context.” Time.com and The Skimm, which concluded that “cervical cancer kills more women” and “more women in the US are dying of cervical cancer,” respectively, were incorrect. It wasn’t the number of deaths that increased. Rather, our understanding of the consequences or severity of cervical cancer on women vulnerable to it changed.
As an illustration, imagine that four deaths are observed in a population of 100 people. At first glance we might say that the mortality rate was 4%. And, if all four of those deaths were from cervical cancer, it might be reasonable to say that there was a cervical cancer mortality rate of 4%.
Four deaths in a theoretical population of 100 people would yield a mortality rate of 4%.
If we realize that not all 100 people were initially at risk for cervical cancer we could reach a dramatically different conclusion. Imagine that half the population is male and that 10 of the remaining 50 women (20%) had undergone total hysterectomy such that they no longer had a cervix. In this scenario, we might consider only the remaining 40 women with a cervix to be at risk for cervical cancer. And here our cervical cancer mortality would be 4/40 = 10%, which is 2.5 times the original mortality rate — even though there were no additional deaths.
If only 40 individuals were at risk of dying from a given condition, then we could recalculate the mortality rate as 10%. This a dramatic increase in the mortality rate but does not reflect a change in the underlying number of deaths.
This sort of re-framing isn’t a novel concept. We don’t discuss our concerns about appendicitis in an individual who has already had their appendix surgically removed. We wouldn’t worry about the possibility of uterine rupture in a male with abdominal pain (since they don’t have a uterus). And if I wanted to test a new measles vaccine and claimed that it was 100% effective at preventing measles infection in a group of individuals who had all been sick or vaccinated with measles in the past (which confers lifelong immunity), no one would take me seriously. I wouldn’t be lying, but I wouldn’t be saying anything of value either. Only people at risk for a given condition should be considered when we are calculating infection rates, cure rates, or death rates. And changes in rates, since they are calculated from both the number of events and the population at risk for those events, may only reflect a change in one of those factors.
The newsworthy story here is not that more deaths are occurring, but rather that those deaths are afflicting the population in ways that we had not fully appreciated. Black women have much higher hysterectomy rates and so the true rate of cervical cancer among those women who were at risk for the disease had been obscured. Why is this racial discrepancy so stark, and how is it that the oldest black women have a corrected mortality rate on the order of rates in less developed countries? Could it be a result of differences in actual cancer biology, or vaccination, or screening, or access to care? In order to begin to answer these questions and address the consequences of cervical cancer we must first understanding the underlying statistics.
Do you have questions or comments on this article, or feedback for future columns in this epidemiology series? I welcome your thoughts at firstname.lastname@example.org!