Prediction of Optimal vs Suboptimal Cytoreduction of Advanced-Stage Serous Ovarian Cancer with the Use of Microarrays

Abstract & Commentary

Synopsis: These data support the hypothesis that favorable survival that is associated with optimal debulking of advanced ovarian cancers is due to, at least in part, the underlying biologic characteristics of these cancers.

Source: Berchuck A, et al. Am J Obstet Gynecol. 2004; 190(4):910-925.

The ability to characterize the expression of thousands of genes simultaneously has provided new insight into the underlying biology of disease for many cancers. Berchuck and colleagues adapt the technology to evaluate its ability to predict primary surgical outcome in patients with newly diagnosed ovarian cancer. RNA from 44 preselected advanced (21 with survival less than 3 years and 23 with survival more than 7 years) and 5 early stage ovarian cancers were evaluated with the Affymetric Gene Chip containing 22,283 genes for their relative expression among patients achieving optimal vs suboptimal cytoreduction. The top 120 differentially expressed genes were then used to generate a prediction model, which was validated in an out-of-sample process. Specific probability models were generated for gene set in order to optimize the ability to distinguish cytoreductive outcome (19 optimal vs 25 suboptimal and 5 early stage). Berchuck and colleagues identified 32 differentially expressed genes in the optimized prediction model. Patients’ cytoreductive outcome was correctly classified by the model in 72.7% of cases. Five of 25 (20%) suboptimal cases, 7 of 19 (37%) optimal cases as well as 2 of 5 (40%) stage I cases were misclassified. There was no relationship between misclassified cases and clinical or pathologic features. Evaluation of gene clusters in this set did not improve the predictive power. Berchuck et al conclude that these data do support the hypothesis that optimal cytoreduction is associated with prolonged survival and that at least in part, optimal status is due to underlying biological characteristics.

Comment by Robert L. Coleman, MD

There are many factors which, by retrospective and prospective evaluation, have shown to be important in estimating survival in newly diagnosed or suspected ovarian cancer patients. The most recognizable, perhaps, is cytoreductive status following primary surgery; that is, the amount of residual disease following a maximal effort at removing it. In the last 30 years since the relationship was first well documented, many authors have correlated preoperative findings such as bulky radiographic disease, CA-125 levels and distribution of disease with the ability to render a patient "optimal."1-5 While some of these factors have proven to be useful in certain circumstances (eg, patients with poor performance status) relying on them exclusively to triage patients for surgery would exclude a significant fraction (up to 30%) of debulkable patients, potentially lowering their survival by withholding an important part of their treatment package.

Most gynecologic oncologists appreciate that some tumors are just not "debulkable" and some patients rendered "optimal" have shorter than expected survivorship. Conversely, some "suboptimal" patients survive for extended periods of time—a measure of their chemosensitivity. The most frequently cited reason for these clinical observations is tumor biology—some underlying, tumor-specific feature or features that define the clinical behavior of a tumor. In the current article, Berchuck et al tackle this conundrum with state-of-the-art molecular profiling using a gene chip array. Since all human cancer appears to result from accumulating genetic mutation, studying patterns of thousands of genes simultaneously allows one to gain an insight, at the moment of tissue harvest, of the RNA being either over or under produced relative to "maintenance" standards. The technology has already proven beneficial in producing risk classifications for patients with prostate and breast cancer. There are currently a handful of similar array studies being conducted in ovarian cancer specimens evaluating risk analysis, survival and chemosensitivity.

Berchuck et al address the biology question (via the surrogate of debulking status) by evaluating a small cohort of patients (n = 44) dichotomized by their survival (less than 3 years vs greater than 7 years) collected from a previous study by which the gene chip technology was used to investigate patterns predicting survival. The primary end point of the current trial was to evaluate the expression profile of those rendered surgically optimal against those left with greater than 1 cm of residual disease. One hundred twenty genes were differentially expressed in these 2 cohorts and made up the sample from which a prediction model was constructed. Using novel statistical and probability methodology, 32 genes were subsequently teased out, optimizing the model predicting surgical outcome. In all, the accuracy in distinguishing optimal from suboptimal was 72.7%. Although the data support proof-of-concept—that is, a genetic expression profile underlies the clinical outcome found at surgery and suggest a biological component may render tumors less debulkable, the predictive power of the model is similar to that achieved with fairly unsophisticated tools such as serum CA-125 and radiographic assessment. Unfortunately, it did not completely segregate the early stage (and therefore, optimal by nature) cases nor control for optimal as a result of stage (eg, Stage IIIA and IIIB) or surgical effort. In addition, fresh tissue cores are needed making the technique less palatable as a preoperative tool. It is also unknown whether the current model or expression profile is generalizable to the population at large given the polarized profile of the sampled cases. Nonetheless, review of individually under- or over-expressed genes reveals important clues as to what biological processes are ongoing in dysregulated growth and metastases.

It is clear we have just scratched the surface of understanding the genomic-wide events that lead up to and characterize phenotypic behavior of individual tumors. And this is just the genomic level! Since their products, (ie, proteins) drive the real cellular machinery, similar profiling will ultimately provide the level of detail needed to ferret out individual characterizations of clinical behavior. This type of proteomic analysis is now being validated in ovarian cancer screening trials. It is hoped new technologies will make detailed analysis available to patients diagnosed with ovarian cancer enabling a tailored therapeutic program, truly maximizing tumor cytotoxicity while minimizing the effects of treatment.


1. Bristow RE, et al. J Clin Oncol. 2002;20:1248-1259.

2. Chi DS, et al. Gynecol Oncol. 2000;77:227-231.

3. Collins Y, et al. Int J Mol Med. 2004;14:43-53.

4. Lancaster JM, et al. J Soc Gynecol Investig. 2004;11: 51-59.

5. Petricoin EF, et al. Lancet. 2002;359:572-577.

Dr. Coleman is Professor, Department of Gynecologic Oncology, University of Texas Southwestern Medical Center, Dallas, TX.