Poor outcomes in a specific disease in which treatment is often successful may be a sign of health systems issues. It may signal that a significant proportion of the population is unable to get adequate health care.— Otis W. Brawley, MD, MACP
A recent study by Mokdad and colleagues, reviewed in this issue of The ASCO Post, looks at cancer demographic data for 28 cancers and compares mortality rates in 1980 to results in 2014.1 Publishing mortality rates by geographic area and the observation of significant differences is not new. The Centers for Disease Control and Prevention (CDC) regularly publishes mortality data by state. The American Cancer Society publishes mortality data by congressional district.2 The National Cancer Institute (NCI) publishes incidence, mortality, and survival data from 19 different population-based registries. Publishing by county, however, is unusual. Study of such data can spur action with regard to cancer control.
The CDC’s National Center for Health Statistics (NCHS) collects and maintains mortality data for the entire country, which the NCI’s Surveillance, Epidemiology, and End Results (SEER) Program publishes.3 Mokdad and colleagues used this database for their study. Their paper notes a 20.1% overall decline in age-adjusted mortality rate as a whole from 1980 to 2014. Overall, U.S. cancer mortality rates rose from the early 1900s and peaked in 1991. There has been a 25% decline in age-adjusted mortality from the peak through 2014, the last year for which data are available.
The published SEER/NCHS data are age-adjusted to the U.S. year 2000 standard. As a result, published SEER/NCHS numbers cannot be precisely compared to the numbers reported by Mokdad et al; however, trends and patterns can be compared.
In SEER/NCHS data, colorectal cancer death rates have been declining since at least the early 1970s. Generally, the northeastern states show the largest decreases, whereas southern states show the smallest decreases. Indeed, the highest colorectal cancer mortality rates shifted from the northeastern states in the early 1990s to the southern states along the Appalachian corridor during the period 2003 to 2007.4
For breast cancer, there have been significant declines in mortality rates in 36 states and the District of Columbia from the peak in breast cancer mortality in the late 1980s to 2010, but rates for 14 states remained almost level.5 Analyses by county-level poverty rates showed that the decrease in mortality rates began later and was slower among women residing in low-income areas.
There are some caveats in terms of assessment of data in the Mokdad paper. As was mentioned above, the paper uses death rates standardized to the 2010 U.S. population. Age standardization is a common way to compare populations and remove the different age distributions. That is, the mortality rate of each county is modified as if the population had a standard age distribution. For a common disease, such as lung cancer, age standardization truly translates into risk of disease-specific death. A county in Florida with a large older population would be judged on an equal scale compared to a county in Colorado that had a smaller proportion of older people in its population.
County-specific comparisons of cancers with few deaths should be viewed with skepticism, however. This is an important limitation. Age adjustment can be affected by small numbers and even become useless. In 2017, for example, there were approximately 410 deaths due to testicular cancer in the United States.6 There are 3,144 counties in the United States. Many have no testicular cancer deaths in a given year or even a decade. Of those that had testicular cancer deaths, most had few. County-by-county comparisons for such a disease mean little.
If there is variation in the relative risk of mortality in different age groups between the area-specific population and the standard population, age standardization will produce a distorting effect. Indeed, the NCI SEER program publishes age-adjusted incidence and mortality rates for specific population-based registries. All of these registries are larger than all but a few U.S. counties. SEER tests the statistical significance of the difference between each age-adjusted registry rate and the U.S. age-adjusted rate.
All studies using registry data have the weakness of some data being inaccurate—ie, “garbage data.” Mortality data are only as good as the death certificate information. Health-care providers do have biases in attributing cause of death. Moreover, changing diagnostic and laboratory technologies have affected the interpretation of cause of death. For example, significant literature can be found on the development and availability of the serum prostate-specific antigen test and related changes in the attribution of prostate cancer as a cause of death.7
What Can Be Learned
Even with these caveats, much can be learned or at least verified about cancer control from this study. Areas of the United States with a traditionally high smoking prevalence have higher death rates from lung cancer. Indeed, the death rate for males varies by county by as much as a factor of 30. Areas where the smoking prevalence declined in the 1960s and 1970s have lower rates of tobacco-related cancer.8 It is well documented that certain states, such as West Virginia and Kentucky, have had very high smoking prevalence (nearly 30% of adults currently smoke tobacco products in these states). On the other hand, smoking prevalence in Utah and California has declined substantially (today, 10% to 12% of residents smoke). This example also demonstrates the lead time necessary for the effect of a cancer control intervention. It takes 30-plus years for a successful smoking intervention to result in a measureable decline in cancer incidence or mortality.
Geographic variation in cancer incidence and mortality can be driven by socioeconomic status differences among populations. The combination of poor diet, obesity, and lack of physical activity is the second-leading cause of cancer in the United States. These factors are affected by socioeconomic parameters such as education and income and vary by geography.
In the United States, a college education is associated with a 25% lower risk of cancer death. While high socioeconomic status is associated with a reduction in some cancer burden, it is associated with increases in others. In the 1990s, it was noted that women living on Long Island east of New York City had a very high breast cancer incidence rate. Epidemiologic study ultimately correlated the high incidence with a high prevalence of college-educated career women, who for career reasons did not have children or delayed childbearing.9 First complete pregnancy after age 30 or never having a child are well-documented risk factors for breast cancer.
Poor outcomes in a specific disease in which treatment is often successful may be a sign of health systems issues. It may signal that a significant proportion of the population is unable to get adequate health care. An assessment of outcomes speaks to access to care, which can be influenced by factors other than the easy-to-assess availability of health insurance, availability of adequate health-care facilities, or adequate medical expertise. Studies like those by Mokdad et al do help provide evidence to support cancer control planning. ■
Disclosure: Dr. Brawley reported no potential conflicts of interest.
3. Howlader N, Noone AM, Krapcho M, et al (eds): SEER Cancer Statistics Review, 1975-2014. National Cancer Institute. Based on November 2016 SEER data submission, posted to the SEER website April 2017. Available at https://seer.cancer.gov/csr/1975_2014. Accessed April 17, 2017.
Cancer mortality declined overall in the United States between 1980 and 2014. Over this same period, there were important changes in trends, patterns, and differences in cancer mortality among U.S. counties. These patterns may inform further research into improving prevention and...!-->!-->