| ||||||||||||||||||||||||||||||||||||||||||||||||||||||
© Oxford University Press 2008.
IN THIS ISSUE
False-Positive Results in Cancer EpidemiologyFalse-positive results, or results that cannot be confirmed in subsequent investigations, are an unfortunate yet inherent outcome of observational epidemiologic studies. Nevertheless, these results are often assumed to be scientifically valid and applicable to the general public and are often sensationally portrayed by the media and overinterpreted by the study authors themselves. In a Commentary, Boffetta et al. (p. 988) highlight some examples of likely false-positive findings in cancer epidemiology and discuss the sources of these types of results. They suggest guidelines and principles for analyzing and reporting results of epidemiologic studies in an effort to provide an epistemic and methodological context for authors, readers, and the press to aid in interpreting such results, with the goal of reducing the negative scientific and societal impact of false-positive reports in the field of epidemiology.
Addition of SNPs to Models of Breast Cancer Risk
Adding information on common alleles that are associated with disease could theoretically improve models for projecting individualized disease risk. Gail (p. 1037) tested this possibility by use of data from two genome-wide association studies and from a candidate gene association study that had identified seven common single-nucleotide polymorphisms (SNPs) that were associated with breast cancer risk. He calculated discriminatory accuracy, as measured by area under the receiver operating characteristic curve (AUC), of a model with the seven SNPs and a hypothetical model with 14 such SNPs. Both models had lower discriminatory accuracy than National Cancer Institute's Breast Cancer Risk Assessment Tool (BCRAT), and adding the seven SNPs to BCRAT improved discriminatory accuracy only modestly. The author concludes that to find enough SNPs to achieve high discriminatory accuracy, very large studies would be needed.
In an editorial, Pepe and Janes (p. 978) point out several problems with the AUC to compare prediction models, including the fact that the scale can be somewhat deceptive and that the AUC hides the values of risk calculated for subjects in the population. They also note that information on absolute risk distributions would have made it possible to compare age-specific proportions of high-risk women and to determine the accuracy of risk classification. The editorialists believe that such calculations would complement Gail's calculations.
Reduction of Iron Stores and Cancer Risk
Having excess amounts of iron in the body has been associated with an increased risk for cancer. To determine whether reducing iron levels by phlebotomy (blood collection) at 6-month intervals would change the risk of cancer among patients with peripheral arterial disease who were participants in a randomized clinical trial of iron reduction and vascular outcomes, Zacharski et al. (p. 996) compared iron levels and cancer incidence in the iron reduction and control groups at baseline and during follow-up. Iron levels were similar in both groups at baseline but were lower in iron reduction patients at each 6-month visit. After a mean follow-up of 4.5 years, fewer patients in the iron reduction group had visceral cancers (38 of 636, compared with 60 of 641 in the control group), and cancer patients in the iron reduction group had better cancer-specific and overall survival than cancer patients in the control group. Mean iron levels at follow-up visits were similar for patients in both groups who developed cancer, but levels were lower in patients who did not develop cancer than in those who did.
In an editorial, Edgren et al. (p. 976) discuss how the findings of Zacharski et al. compare with those of previous studies on this topic. They commend the randomized trial design but note the existing potential limitations, such as the fact that the trial was not designed to study cancer outcomes. They caution against immediate clinical application of the results.
Cancer Mortality and Educational Attainment
Rates of death from the four major cancers in the United States—lung, colorectal, female breast, and prostate—have decreased since the early 1990s. However, few studies have examined whether the reduction in death rates varies with level of educational attainment. Kinsey et al. (p. 1003) used data from death certificates covering 86% of the US population to calculate death rates for 25- to 64-year-old non-Hispanic white and black men and women from the four major cancers by educational attainment during 1993 through 2001. Among those with the highest level of education (ie, college degree or more), rates of death from each cancer decreased from 1993 to 2001 for nearly every race and sex stratum. By contrast, for those with the least education (ie, those who did not finish high school), death rates for most cancers increased or were stable for most race and sex strata. The authors conclude that among persons aged 25 to 64 years the recent overall declines in death rates from the four major cancers in the United States mainly reflect declines in more highly educated individuals, suggesting that those with lower levels of educational attainment have not benefited as much from improvements in prevention, early detection, and treatment.
Medicare Managed Care and a Diagnosis of Cancer
Although Medicare managed care may offer enrollees lower out-of-pocket costs and provide benefits that are not available in the traditional fee-for-service Medicare program, it may also restrict choices to control costs. Elkin et al. (p. 1013) compared rates of voluntary disenrollment from Medicare managed care into traditional fee-for-service Medicare among Medicare managed care enrollees with and without a cancer diagnosis between 1995 and 2002. They found that, in the 2 years after a diagnosis of breast, colorectal, prostate, or lung cancer, patients were less likely to disenroll from Medicare managed care than their matched cancer-free peers. The results were consistent across strata of age, sex, race, SEER registry, and cancer stage. The authors conclude that a new diagnosis of cancer does not appear to precipitate disenrollment from Medicare managed care to traditional fee-for-service Medicare.
Cell Cycle Regulator in Prostate Cancer Invasion and Metastasis
Recent work suggests that the cell cycle regulatory factor cyclin A1, which is highly expressed in various cancers, including prostate cancer, has a cell cycle–independent role in carcinogenesis. To elucidate this role, Persson et al. (p. 1022) engineered human prostate cancer cells to overexpress cyclin A1. Overexpression of cyclin A1 increased the invasiveness of the cells in vitro and their metastatic activity in mouse models. The cyclin A1–mediated increase in invasiveness was dependent on expression of two proteins with specific roles in metastasis, metalloproteinase-2 (MMP2) and urokinase plasminogen activator. Cyclin A1 interacted in a sequence-specific manner with the promoters of the MMP2 and vascular endothelial growth factor genes and increased their activity in concert with the androgen receptor (AR). Treatment of cells with an agonist of the of the AR led to translocation of cyclin A1 to the nucleus, and cyclin A1 expression enhanced AR-mediated activation of the prostate specific antigen promoter. Thus, through its effects on the expression of genes with roles in metastasis, cyclin A1 may play a role in the spread of prostate cancer.
Related Articles in JNCI
CiteULike 
Connotea 
Del.icio.us What's this?
J Natl Cancer Inst 2008 100: 1037-1041.
J Natl Cancer Inst 2008 100: 988-995.
J Natl Cancer Inst 2008 100: 1003-1012.
J Natl Cancer Inst 2008 100: 1013-1021.
J Natl Cancer Inst 2008 100: 996-1002.
J Natl Cancer Inst 2008 100: 1022-1036.
J Natl Cancer Inst 2008 100: 978-979.
J Natl Cancer Inst 2008 100: 976-977.
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||