Originally published as JCO Early Release 10.1200/JCO.2004.06.129 on September 20 2004

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Hensley, M. L. Articles by Spriggs, D. R. Search for Related Content PubMed PubMed Citation Articles by Hensley, M. L. Articles by Spriggs, D. R.

Cancer Screening: How Good Is Good Enough?

Memorial Sloan-Kettering Cancer Center, New York, NY

Ovarian cancer is diagnosed in approximately 25,000 women and accounts for more than 14,000 deaths in the United States each year.¹ Stage for stage, the prognosis for women with ovarian cancer does not differ greatly from that of women with breast cancer. However, the higher mortality associated with ovarian cancer is related to the fact that approximately three fourths of women have stage III or IV disease at diagnosis. Like early-stage breast cancer, early-stage ovarian cancer is potentially curable, with survival rates in excess of 80% for patients with stage I and II disease.² Thus, there is intense interest in the development of a screening test capable of diagnosing ovarian cancer in its presymptomatic, early stages.

The challenges to this effort are several. The prevalence of ovarian cancer is an order of magnitude lower than that of breast cancer. When the sensitivity and specificity of any single screening procedure or assay are relatively low, the application of such tests in a low-prevalence disease results in very low positive predictive value. Finally, in ovarian cancer, positive test results require an invasive procedure—laparoscopy or laparotomy—to rule out malignancy.

It has been suggested that an ovarian cancer screening test should have a positive predictive value of at least 10%; therefore, 10 women would undergo surgery to diagnose one cancer.³ Given the low prevalence of ovarian cancer in the general population, to achieve this positive predictive value, the screening test would require a sensitivity of greater than 75% and a specificity of at least 99.6%. These high test performance thresholds were nearly achieved in a pilot randomized trial of 22,000 postmenopausal women using CA-125 screening compared with no screening, and sonography for screened women with elevated CA-125 values.⁴ The specificity of CA-125 was 98% (thus 2% of all women were referred for sonography) and the sensitivity of the two-step screening strategy was 71% for all ovarian cancers. However, the sensitivity for early-stage cancers was only 40%. Overall survival did not differ between the screened and nonscreened groups, although the study was neither designed nor powered to detect such a difference. In addition, although 31% of the cancers in the screened group were stage I or II, compared with 10% of the cancers in the nonscreened group, there is no clear evidence that this shift was attributable to screening. Historically, even among nonscreened populations, 25% to 30% of women present with early-stage cancers, and it is unlikely that presymptomatic detection of stage III and IV cancers would significantly reduce ovarian cancer mortality.

In this issue of the Journal of Clinical Oncology, Skates et al⁵ report the results from a study that aimed to improve the first-step sensitivity of serum screening for early-stage ovarian cancers, while fixing specificity at 98%. Their approach was to combine the information contributed by four serum markers using logistic regression, classification trees, and mixture discriminate analysis, and then determine which combination yielded the highest sensitivity for early-stage cancers. Using sera collected before surgery, the authors created a training set of 63 ovarian cancer patient cases (27 of which were from women with early stage cancers) and 126 controls, and a validation set of 60 patient cases and 98 controls. The authors determined that the greatest sensitivity for early-stage ovarian cancer was obtained with the combination of CA-125, CA 72-4, and macrophage colony-stimulating factor. In the mixture discriminant analysis model, this combination had a sensitivity of 70% for early-stage cancer when specificity was fixed at 98%. The authors postulate that a first-step specificity of 98%, when combined with a second-step sonography, would yield a combined two-step screening approach specificity of 99.6%. Given this substantial improvement in sensitivity over first-step screening with CA-125 alone, with preserved specificity, why is this three-marker approach not yet ready for general application? The answers relate to some of the greatest challenges in ovarian cancer screening, including our currently limited knowledge of the pathophysiology of tumor progression and the difficulty of screening for a low-prevalence disease.

A strength of this study is the use of an independent validation data set from patients at institutions other than those in the training set, thus decreasing bias in the estimates of sensitivity. However, the validation sera set of early-stage cancers included three patients with stage IIIA epithelial ovarian cancer, 12 with borderline stage I or II tumors, and three with nonepithelial ovarian cancers. Thus, 18 of the 60 patient cases in the validation set were not early-stage, invasive, epithelial ovarian cancers, and were not ideal screening targets. The performance of this same three-marker approach might therefore differ when evaluated for its sensitivity for detection of stage I or II invasive epithelial cancers. Like any case-control study, there is always some trade-off in deciding how different or how alike to create the patient case and control populations. Knowing some detail about the control populations, particularly for factors known to affect CA-125 levels such as menopausal status, might help assess the potential generalizability of these data.

Given the low prevalence of ovarian cancer and the limited availability of preoperative sera, the sera for the patient cases in both the training and the validation set were from women undergoing pelvic surgery, presumably for suspicion of pelvic pathology. Although the training and validation sets may both be enriched for patients with early-stage disease, it is possible that the marker profile of a patient who is about to undergo surgery for possible pelvic pathology may differ markedly from that of asymptomatic women in a true screening population.

This issue points toward a critical missing piece in our understanding of the pathophysiology of epithelial ovarian cancer: we do not know whether advanced-stage ovarian cancer progresses in a stepwise fashion from early stage to late stage, or whether those early-stage cancers are genetically programmed to behave in a more latent fashion. There may be molecular characteristics that predispose certain ovarian cancers to present as either early- or late-stage disease. If this were the case, the early-stage cancers that were in the training and validation sets may be a very select group, such that these early stage cancers are, by definition, diagnosable. There are, however, recent gene expression array analyses suggesting that expression patterns are not different between early- and late-stage poorly differentiated cancers.⁶

One additional issue related to generalizability and applicability of these data is more straightforward. The three-marker approach demonstrates high sensitivity at a set specificity in validation sera in which the disease prevalence was 60 of 158 (38%). The prevalence of ovarian cancer is 30 to 50 per 100,000 women (0.03%). Therefore, even with this improved sensitivity and high two-step specificity, the positive predictive value when applied to a typical screening (normal risk) population may not achieve the target positive predictive value of 10% or greater.

How shall we proceed in the search for a strategy to detect this disease in its potentially curable stages? The effort must be multifaceted. The genetic and molecular profiles of early- and late-stage cancers need to be clarified to answer the question of whether and how disease progresses from the potentially curable to the probably lethal. In conjunction with this approach, if screening is performed, it should be done in carefully monitored cohorts such that sera from true screening populations are available for study. Markers of poor prognosis may be especially valuable in early ovarian cancer screening, given that those proteins may identify the stage I patients with the worst risk.

Novel markers, such as lysophosphatidic acid,⁷ OVX1,⁸ and YKL-40,⁹ need additional study, given that CA-125 is known to be elevated in only approximately 50% of women with stage I and II cancers.¹⁰ As potential markers are identified, their assays must ultimately be both reproducible and available for general use. Approaches that recognize the probable importance of change over time in serum marker levels (such as the Risk of Ovarian Cancer Algorithm,¹¹ also developed by Skates) will almost certainly be required.

As screening approaches are developed, attention must be paid to clinical implementation. In high-risk populations, anxiety about ovarian cancer is high.¹² Approaches that require that multiple biomarkers reach specific trigger thresholds may be difficult for both the physician and the woman undergoing screening. There will be a temptation to act on the results of a single abnormal marker. Similarly, women will be screened serially, and the probability of a false-positive screening test will increase well beyond that estimated by retrospective, one-time test performance evaluation studies. In our own screening experience, the psychologic stress of a positive test is great and may undermine future screening participation.

Finally, the most promising approaches must be subjected to prospective randomized testing. Even this step is fraught with controversy. Will patients even agree to be randomly assigned to a no-screening arm? Should the end point be reduced ovarian cancer mortality or will a change in the ratio of early-stage to late-stage cancers be sufficient? Do we accept the assumptions that early-stage cancer precedes late-stage cancer, and that diagnosing early-stage cancers will decrease mortality? Clearly, a surrogate end point of increasing the percentage of women diagnosed with early-stage ovarian cancer would be more practical, requiring many fewer patients and years, but that approach has bedeviled prostate-specific antigen screening¹³ for prostate cancer. In the final analysis, only a randomized trial with enormous numbers of patients, examining a survival end point, will establish whether screening both finds cancers early and saves lives.

Authors' Disclosures of Potential Conflicts of Interest

The authors indicated no potential conflicts of interest.

REFERENCES

1. American Cancer Society: Cancer Facts and Figures 2003, American Cancer Society. http://www.cancer.org/docroot/STT/content/STT_1x_Cancer_Facts__Figures_2003.asp

2. Young RC, Walton LA, Ellenberg SS, et al: Adjuvant therapy in stage I and II epithelial ovarian cancer: Results of two prospective randomized trials. N Engl J Med 322: 1021-1027, 1990[Abstract]

3. Jacobs I, Bast RC Jr: The CA125 tumor-associated antigen: A review of the literature. Hum Reprod 4: 1-12, 1989

4. Jacobs IJ, Skates SJ, MacDonald N, et al: Screening for ovarian cancer: A pilot randomized trial. Lancet 353: 1207-1210, 1999[CrossRef][Medline]

5. Skates SJ, Horack N, Yinhua Y, et al: Pre-operative sensitivity and specificity for early stage ovarian cancer when combining cancer antigen (CA) -125 II, CA 15-3, CA 72-4, and macrophage colony-stimulating factor using mixture of multivariate normal distributions. J Clin Oncol 22: 4059-4066, 2004[Abstract/Free Full Text]

6. Shridar V, Lee J, Pandita A, et al: Genetic analysis of early verus late stage ovarian tumors. Cancer Res 61: 5895-5904, 2001[Abstract/Free Full Text]

7. Xu Y, Shen Z, Wiper DW, et al: Lysophosphatidic acid as a potential biomarker for ovarian and other gynecologic cancers. JAMA 280: 719-723, 1998[Abstract/Free Full Text]

8. Xu F-J, Yu Y-A, Daly L, et al: OVX1 radioimmunoassay complements CA125 for predicting the presence of residual ovarian cancer at second-look surgical surveillance procedures. J Clin Oncol 11: 1506-1510, 1993[Abstract/Free Full Text]

9. Dupont J, Tanwar MK, Thaler HT, et al: Early detection and prognosis of ovarian cancer using serum YKL-40. J Clin Oncol 22: 3330-3339, 2004[Abstract/Free Full Text]

10. Zurawski V, Orjaster H, Andersen A, et al: Elevated CA125 levels prior to diagnosis of ovarian neoplasia: Relevance for early detection of ovarian cancer. Int J Cancer 42: 677-680, 1988[Medline]

11. Skates SJ, Xu FJ, Yu YH, et al: Toward an optimal algorithm for ovarian cancer screening with longitudinal tumor markers. Cancer 76: 2004-2010, 1995[CrossRef][Medline]

12. Hensley ML, Robson ME, Kauff ND, et al: Pre- and post-menopausal high-risk women undergoing screening for ovarian cancer: Anxiety, risk perceptions, and quality of life. Gynecol Oncol 89: 440-446, 2003[CrossRef][Medline]

13. Schmid H-P, Riesen W, Prikler L: Update on screening for prostate cancer with prostate-specific antigen. Crit Rev Oncol Hematol 50: 71-78, 2004[Medline]

This Article Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Email this article to a colleague Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Save to my personal folders Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Hensley, M. L. Articles by Spriggs, D. R. Search for Related Content PubMed PubMed Citation Articles by Hensley, M. L. Articles by Spriggs, D. R.