Younger age at cancer diagnosis may be driven by age structure of the HCV population

We read with interest the recent article, “Increased incidence of cancer and cancer-related mortality among persons with chronic hepatitis C infection, 2006–2010,” which addressed important questions about excess cancer risk among people infected with hepatitis C virus (HCV) [1]. In particular, we were interested in the authors’ analysis that compared ages at cancer diagnosis and cancer death between people with HCV and the general population. The authors concluded that the age at diagnosis was significantly younger for most of the evaluated cancers.

Unfortunately, we believe that the authors’ findings regarding age at diagnosis are biased, because of differences in the underlying population age distributions from which the cancer cases and cancer deaths arose. Among the Chronic Hepatitis Cohort Study (CHeCS) population in this analysis, 77% were aged 43–63 years in 2008 (i.e., born during 1945–1965), compared to only 42% of the reference population (general population in SEER [Surveillance, Epidemiology and End Results]-13 cancer registry areas [2]). Of particular concern is the smaller fraction of people aged ≥64 years old (10% in CHeCS vs. 19% in SEER-13), as this age group has the highest cancer incidence. Because the age distribution is shifted toward middle age in the CHeCS, the cancer cases that arise in this cohort will on average be younger than cancer cases in the general population. This same bias is present in the cancer mortality analysis, as the U.S. Census population has a very similar age distribution to the SEER-13 population [2, 3].

As a hypothetical example, an extreme form of this bias would occur if the cohort only followed people from ages 40 to 50 years. By design, the mean age at diagnosis for every cancer type would then be between 40 and 50 years, which would be younger than the mean age in the general population which also includes cancers among older people. This example highlights that bias arises when analyzing only cases without accounting for the underlying source population.

To illustrate the impact of the age distribution of the CHeCS source population on age at cancer diagnosis, we show the total number of cancer cases (all types) in SEER-13 in 2008 by age group (Figure 1) [2]. We then applied age-specific incidence rates from the same SEER data to a population of identical size, but with the age distribution of the CHeCS (based on a prior publication describing the cohort) [4]. As demonstrated in Figure 1, with the same incidence rates, changing the age distribution of the underlying population results in a shift in the age distribution of cancer cases. Using the SEER-13 population age distribution, the median age at cancer diagnosis was in the 64–73 year-old age group, whereas shifting the population age distribution to that of CHeCS gave a median age in the 54–63 year-old age group. This demonstrates that the median age at diagnosis is highly sensitive to the age composition of the underlying population.

Figure 1.

Number of total cancer cases occurring in SEER-13 in 2008 by age group (black bars), and estimated to occur in SEER-13 if the age distribution of the population had been the same as the Chronic Hepatitis Cohort Study (gray bars). Asterisks indicate age groups containing the median age at cancer diagnosis under each scenario.

This same bias arises when comparing the age at cancer diagnosis between other populations with different age distributions. For many cancer types, we previously showed that large 10–20 year age-at-diagnosis differences between people with acquired immune deficiency syndrome (AIDS) and the general population were mainly driven by the much younger age distribution of the U.S. AIDS population [5]. After correcting for differences in the age of the underlying population, few significant differences in age at diagnosis remained. Similarly, we showed that many apparent differences in age at cancer diagnosis between whites and blacks in SEER were driven by the younger age distribution of blacks in the U.S. [6].

There are statistical approaches that can address this bias and allow for an evaluation of earlier onset of cancer. We previously presented median and mean ages at diagnosis after adjusting for underlying age differences through indirect standardization and weighted linear regression [5, 6]. In addition, variation in age-specific incidence rates across groups [6] or in relative risks across ages [5] can provide evidence of differences in the age at cancer diagnosis. Assessing the role of HCV in the development of earlier onset cancers may have implications for cancer etiology and screening, and we appreciate the authors’ attention to this topic. We hope that they consider utilizing these tools to re-analyze their data and report whether the age differences in incident cancers and cancer deaths remain.

Abbreviations

HCV

hepatitis C virus

CHeCS

Chronic Hepatitis Cohort Study

SEER

Surveillance, Epidemiology and End Results

AIDS

acquired immune deficiency syndrome