Correspondence
Five Genetic Variants Associated with Prostate Cancer
N Engl J Med 2008; 358:2738-2741June 19, 2008
- Article
To the Editor:
We agree with Zheng et al. (Feb. 28 issue)1 that additional research is needed to assess the value of their finding of genetic variants associated with the risk of prostate cancer. Unfortunately, the planned marketing of a test based on this study2 is premature and may cause more harm than good. Finding a genetic association is only the first step in the continuum of translating research into practice.3 The results have not been independently confirmed, and adding the genetic test results to age, region, and family history only marginally improved risk prediction (the area under the curve [AUC] increased from 0.61 to 0.63). The clinical utility of the test is questionable because it cannot be used to reduce risk, since there are no known modifiable risk factors4; to encourage screening, since the balance of benefits and harms is unknown5; or to predict the clinical course of the disease, since the variants were associated equally with aggressive and nonaggressive cancers.1 In the absence of evidence of improved outcomes, this test may lead to unnecessary or potentially harmful procedures.
5 ReferencesRalph J. Coates, Ph.D.
Muin J. Khoury, M.D.
Marta Gwinn, M.D.
Centers for Disease Control and Prevention, Atlanta, GA 30341
rjc5@cdc.gov 1
Zheng SL ,Sun J ,Wiklund F , et al. Cumlative association of five genetic variants with prostate cancer. N Engl J Med 2008;358:910-919
Full Text | Web of Science | Medline2
Kolata G. $300 to learn risk of cancer of the prostate. New York Times. January 17, 2008.
3
Khoury MJ ,Gwinn M ,Yoon PW ,Dowling N ,Moore CA ,Bradley L . The continuum of translation research in genomic medicine: how can we accelerate the appropriate integration of human genome discoveries into health care and disease prevention? Genet Med 2007;9:665-674
CrossRef | Web of Science | Medline4
Nelen S . Epidemiology of prostate cancer. Recent Results Cancer Res 2007;175:1-8
CrossRef | Medline5
US. Preventive Services Task Force. Screening for prostate cancer: recommendations and rationale. Ann Intern Med 2002;137:915-916
Web of Science | Medline
To the Editor:
In his accompanying editorial, Gelmann states that the five polymorphisms reported by Zheng et al. do not yet constitute a viable screening test.1 We think they never will. The use of genetic polymorphisms with modest odds ratios (1.22 to 1.53 in the study by Zheng et al.) to screen for a polygenic disease such as prostate cancer is unlikely to be practical because most men, whether or not they have prostate cancer, will be at average genetic risk.2 In the study by Zheng et al., most case subjects and control subjects had one to three risk factors (85.5% of case subjects and 86.2% of control subjects), and less than 10% of the case subjects were at high risk. A test based on these polymorphisms cannot distinguish adequately between case subjects and control subjects and will miss most cases or have false positive results for most controls (Table 1Table 1
Performance of a Screening Test for Prostate Cancer Based on Six Risk Factors (Five Genetic Polymorphisms plus Family History).). This is reflected in the very poor AUC of the receiver-operating-characteristic curve even when these genes are combined with other risk factors (AUC, 0.63), which is not much better than that which is expected by chance (AUC, 0.50).2 ReferencesCoral E. Gartner, Ph.D.
Jan J. Barendregt, Ph.D.
Wayne D. Hall, Ph.D.
University of Queensland, Brisbane, QLD 4006, Australia
c.gartner@sph. uq. edu. au 1
Gelmann EP . Complexities of prostate-cancer risk. N Engl J Med 2008;358:961-963
Full Text | Web of Science | Medline2
Wald NJ ,Hackshaw AK ,Frost CD . When can a risk factor be used as a worthwhile screening test? BMJ 1999;319:1562-1565
CrossRef | Web of Science | Medline
To the Editor:
Zheng et al. report that the combined effect of family history and five single-nucleotide polymorphisms (SNPs) on the risk of prostate cancer was increased by a factor of 9.46 for men who had at least five factors as compared with those who had none. This contradicts the low discriminative accuracy they observed (AUC, 0.63). For the calculation of this odds ratio, they compared the relatively small highest-risk category with the lowest-risk category. Selection of the lowest-risk category as the reference is a frequently used and powerful approach to demonstrating an association,1,2 but it does not give a realistic impression of the clinical usefulness of the findings. To evaluate the increase or decrease in the risk of disease as compared with the pretest or overall risk, we have recalculated the odds ratios, shown in Table 4 of the article, for number of associated factors as follows: no associated factors, 0.49 (observed in 10% of the controls); one factor, 0.80 (34%); two factors, 1.01 (36%); three factors, 1.34 (17%); four factors, 2.35 (3%); and five or more factors, 4.78 (<1%). Although the findings of Zheng et al. are of great interest, their AUC analysis and a simple recalculation of their data show that the clinical implications are limited.
2 ReferencesA. Cecile J.W. Janssens, Ph.D.
Cornelia M. van Duijn, Ph.D.
Erasmus University Medical Center, 3000 DR Rotterdam, the Netherlands
a.janssens@erasmusmc. nl 1
Lyssenko V ,Almgren P ,Anevski D , et al. Genetic prediction of future type 2 diabetes. PLoS Med 2005;2:e345-e345
CrossRef | Web of Science | Medline2
Maller J ,George S ,Purcell S , et al. Common variation in three genes, including a noncoding variant in CFH, strongly influences risk of age-related macular degeneration. Nat Genet 2006;38:1055-1059
CrossRef | Web of Science | Medline
To the Editor:
The claim by Zheng et al. that five SNPs plus a family history “account for” 46% of cases of prostate cancer is misleading. They invoked the concept of population attributable fraction (PAF), not population attributable risk, as they incorrectly called it. PAF approximates the proportion of disease prevented by eliminating a risk factor.1 Its application to genetics is questionable.
None of the variants are known to be causal; elimination of the variants, even if possible, would not necessarily prevent prostate cancer. In any case, PAF is highly dependent on the reference group.1 A broader definition than the authors' choice of men with no risk alleles (10% of controls) would lead to a lower PAF.
A better gauge of the effect of measured genes on a disease is the extent to which they explain the increased risk associated with having an affected first-degree relative.2 With adjustment for the SNPs, the family-history effect went from 2.26 Table 1 of the article by Zheng et al.) to 2.22 (Table 4 of the article), so they explained just 2% of familial aggregation.
2 ReferencesGianluca Severi, Ph.D.
Cancer Council Victoria, Melbourne, VIC 3053, Australia
gianluca.severi@cancervic. org. au Graham B. Byrnes, Ph.D.
International Agency for Research on Cancer, 69372 Lyon, FranceJohn L. Hopper, Ph.D.
University of Melbourne, Melbourne, VIC 3053, Australia1
Rockhill B ,Newman B ,Weinberg C . Use and misuse of population attributable fractions. Am J Public Health 1998;88:15-19
CrossRef | Web of Science | Medline2
Hopper JL ,Carlin JB . Familial aggregation of a disease consequent upon correlation between relatives in a risk factor measured on a continuous scale. Am J Epidemiol 1992;136:1138-1147
Web of Science | Medline
To the Editor:
Large-scale genomewide studies1 have identified risk alleles with low odds ratios that can have little clinical utility for risk prediction. Zheng et al. combined genotypes to see whether the odds ratio can be increased to clinically useful levels for estimating the risk of prostate cancer. However, the authors did not compare the performance of the combined genotype with the current standard, prostate-specific antigen (PSA). Approximately 20% of their case subjects had a family history of prostate cancer, and 90% had a PSA level of more than 4 ng per milliliter; combining the two would probably have given a similar, if not better, population attributable risk. The addition of the combined genotype to age, region, and family history improved the AUC by a mere 3%, indicating lack of meaningful improvement over current methods. Finally, as with CHEK2 in breast cancer,2 the odds ratio is too low to be useful at such a low prevalence of the combined genotype. An increase in the odds ratio obtained by combining genotypes comes at the price of a decline in the prevalence of the combined genotype.
2 ReferencesMangesh A. Thorat, M.D., D.N.B.
Indiana University School of Medicine, Indianapolis, IN 46202
thoratmangesh@gmail.com 1
Wellcome Trust Case Control ConsiortiumGenome-wide association study of 14,000 cases of seven common diseases and 3,000 shared controls. Nature 2007;447:661-678
CrossRef | Web of Science | Medline2
Offit K ,Garber JE . Time to check CHEK2 in families with breast cancer? J Clin Oncol 2008;26:519-520
CrossRef | Web of Science | Medline
To the Editor:
The model of Zheng et al. has limited predictive ability: using the data in Table 4 of their article, we estimate an AUC of 0.57 for the SNPs. This is dwarfed by the AUC for a single PSA test at 44 to 50 years of age: in a large study of a representative, unscreened population, we reported an AUC of 0.76 for the occurrence of prostate cancer up to 25 years subsequently.1 Moreover, we have shown2 that a single PSA test can accurately predict advanced prostate cancer (AUC, 0.79) and that screening decisions based on a single PSA test before the age of 50 years are likely to lead to better outcomes for patients than the competing alternative strategies of screening all or no men.3 As such, the proven technology of PSA testing currently offers by far the best method of stratifying men according to risk for prostate-cancer screening.
Andrew Vickers, Ph.D.
Hans Lilja, M.D., Ph.D.
Peter Scardino, M.D.
Memorial Sloan-Kettering Cancer Center, New York, NY 10021Drs. Vickers, Lilja, and Scardino report holding patents for assays of free PSA and human glandular kallikrein 2. No other potential conflict of interest relevant to this letter was reported.
3 References1
Lilja H ,Ulmert D ,Bjork T , et al. Long-term prediction of prostate cancer up to 25 years before diagnosis of prostate cancer using prostate kallikreins measured at age 44 to 50 years. J Clin Oncol 2007;25:431-436
CrossRef | Web of Science | Medline2
Ulmert D ,Cronin AM ,Bjork T , et al. Prostate-specific antigen at or before age 50 as a predictor of advanced prostate cancer diagnosed up to 25 years later: a case-control study. BMC Med 2008;6:6-6
CrossRef | Web of Science | Medline3
Ulmert D ,Serio AM ,O'Brien MF , et al. Long-term prediction of prostate cancer: prostate-specific antigen (PSA) velocity is predictive but does not improve the predictive accuracy of a single PSA measurement 15 years or more before cancer diagnosis in a large, representative, unscreened population. J Clin Oncol 2008;26:835-841
CrossRef | Web of Science | Medline
To the Editor:
According to the data of Zheng et al., it could be expected that if very few persons have none of the six risk factors they report (about 4%), then about 19% of the population will have only one risk factor. On the basis of the hypothetical mean lifetime risk of prostate cancer of 10%, men with no risk factors might have less than a 4% lifelong risk, and men with only one risk factor less than a 6% lifelong risk. For these subgroups, the benefit of screening in relation to the risk is even more doubtful than it is in the general population.
François Eisinger, M.D., Ph.D.
Institut Paoli-Calmettes, 13009 Marseille, France
eisinger@marseille.inserm. fr Author/Editor Response
Our finding of a cumulative effect of five genetic variants and family history on the risk of prostate cancer in a population-based case–control study in Sweden was recently confirmed in two U.S. populations.1 The consistent finding of a strong cumulative effect on prostate-cancer risk, together with a relatively high frequency in the general population, calls for a new perspective in the interpretation and use of genetic information.
Until a perfect biomarker for predicting prostate-cancer risk is available, there is a need to search for biomarkers that individually or collectively offer some utility for risk prediction. We emphasize that our approach represents an initial but important step toward this goal. We found that a cutoff of three of these six risk factors had a specificity of 80% and a sensitivity of 32% for discriminating between case subjects and control subjects in this Swedish population. Although its specificity is lower than that of PSA (94%) with the use of a cutoff of 4.1 ng per milliliter, its sensitivity is higher than that of PSA (21%).2 With 10 additional risk variants reported since February 2008, it is expected that the sensitivity and specificity will be further improved.
We do not advocate replacing the PSA test with genetic tests; we envision combining these tests to improve their predictive ability. Although we could not assess joint discriminatory performance because of our study design, we found that the cumulative effect of genetic risk factors is independent of PSA levels and thus can provide complementary information.
Just as men with a family history of prostate cancer are encouraged to start undergoing screening at an earlier age, men with multiple genetic risk factors might similarly be encouraged to begin screening earlier. Unlike family history, which is subject to the limitations of family size, structure, current age of male relatives, and reliability of reporting, genetic markers can be accurately measured anytime.
Although it is difficult to provide risk information with confidence for most men (who carry one, two, or three of the five genetic risk factors) by means of genetic markers alone, men who are at the two extremes in terms of the number of inherited factors may benefit substantially from categorization of their risk for prostate cancer.
Additional studies, especially prospective ones, are needed to further improve prediction of prostate-cancer risk by including additional risk variants, combining them with PSA and other detection methods, and incorporating genetic markers that distinguish aggressive from nonaggressive prostate cancer in order to predict disease progression.
Jianfeng Xu, M.D., Dr.P.H.
Wake Forest University School of Medicine, Winston-Salem, NC 27157
jxu@wfubmc.edu William B. Isaacs, Ph.D.
Johns Hopkins Medical Institutions, Baltimore, MD 21287Henrik Grönberg, M.D., Ph.D.
Karolinska Institutet, SE-171 77 Stockholm, SwedenSince publication of the article, Dr. Xu reports founding Proactive Genomics, a company that will offer genetic testing for predicting individual prostate-cancer risk. No other potential conflict of interest relevant to this letter was reported.
2 References1
Sun J, Chang B-L, Isaacs SD, et al. Cumulative effect of five genetic variants on prostate cancer risk in multiple study populations. Prostate (in press).
2
Thompson IM ,Ankerst DP ,Chi C , et al. Operating characteristics of prostate-specific antigen in men with an initial PSA level of 3.0 ng/ml or lower. JAMA 2005;294:66-70
CrossRef | Web of Science | Medline
- Citing Articles (4)
Citing Articles
1
Wayne Hall, Coral Gartner. (2009) Direct-to-Consumer Genome-Wide Scans: Astrologicogenomics or Simple Scams?. The American Journal of Bioethics 9:6-7, 54-56
CrossRef2
Peter Kraft, Sholom Wacholder, Marilyn C. Cornelis, Frank B. Hu, Richard B. Hayes, Gilles Thomas, Robert Hoover, David J. Hunter, Stephen Chanock. (2009) Beyond odds ratios — communicating disease risk based on genetic profiles. Nature Reviews Genetics 10:4, 264-269
CrossRef3
Coral E. Gartner, Jan J. Barendregt, Wayne D. Hall. (2009) Multiple genetic tests for susceptibility to smoking do not outperform simple family history. Addiction 104:1, 118-126
CrossRef4
Sharon A Savage. (2008) Cancer genetic association studies in the genome-wide age. Personalized Medicine 5:6, 589-597
CrossRef
