Removing the Unknown From the Carcinoma of Unknown Primary

Approximately 3% to 5% of all cancers have no identifiable primary site. The management of carcinoma of unknown primary (CUP) has evolved over the past two decades from a wide-ranging work-up into a more algorithm-defined approach. Many subsets (approximately 20%) of CUP fall into recognizable syndromes; these include midline germ cell cancer in young men and isolated axillary lymph nodes in women. These better-prognosis subsets (these include papillary serous cancer in women, colorectal marker positive adenocarcinoma, and prostate-specific antigen [PSA] –expressing adenocarcinoma) can be treated with specific regimens, and their survival is much better than that of other CUPs. However, even after these subsets are excluded, a large category of CUP remains that cannot be further characterized on clinical and/or pathological grounds. Based on the assumption that CUPs are similar in their response to treatment if not in their tissue of origin, chemotherapy regimens have been explored in over 35 prospective clinical trials. Most of these are phase II trials and while comparisons across trials are flawed, some general themes emerge. Combination doublets, (usually platinum and taxane containing) appear to be superior to single-agent regimens while triplet combinations do not appear to be better than doublets. High-dose or dose-dense regimens do not appear to be superior to so-called regular regimens. One of the best outcomes from a phase II trial is from a combination of a platinum/taxane doublet with bevacizumab and erlotinib; median overall survival (OS) for this regimen was 12.6 months (n 60). The only phase III trial performed in CUP, comparing paclitaxel,carboplatin, and etoposide with gemcitabine and irinotecan (n 198), showed no statistically significant difference in outcome between either arm (median OS 7.4 v 8.5 months). In summary, median OS from chemotherapy combinations of CUP have yielded survival rates of 9 to 13 months which parallel the median survival rates of thoracic, GI, and urothelial malignancies that comprise most of these cancers. This so-called lumper approach is intrinsically limited however, and it is increasingly appreciated that the complex molecular underpinnings of cancer need to be precisely targeted to obtain the best outcome for our patients. While imaging studies such as testicular ultrasound, breast magnetic resonance imaging, and [F]fluorodeoxyglucose–positron emission tomography can point to the primary site in oligometastatic disease, once a cancer is widely disseminated, imaging is often not useful in discerning the origin. Some metastatic tumors, such as papillary serous adenocarcinoma of the ovary retain distinctive architectural patterns and can aid in identification of the primary. Many cancers retain distinctive patterns of protein expression, hence obtaining tumor specific markers can help to determine the primary tissue of origin. Immunohistochemistry using antibodies to proteins such as cytokeratins and differentiation markers such as hormone receptors, thyroid transcription factor 1, PSA, carcinoembryonic antigen, and cancer antigen–125 is now standard. Newer protein markers and better algorithms have improved primary site attributions to more than 80%. Nonetheless, some tumors are difficult to classify with protein immunochemistry alone. The use of reverse transcriptasepolymerase chain reaction (RT-PCR) or oligonucleotide microarray providesaquantitativeestimateofgeneexpression withinatissue.Thisdata can help to classify even the more undifferentiated tumors. MicroRNA, short snippets of regulatory RNA that have tissue specific patterns of expression, can also be used to determine primary site. Several commercial tests utilizing this methodology are now available, however, since no gold standard test exists, it is hard to verify the accuracy of this categorization. In the current study, Hainsworth et al use a 92-gene RT-PCR based assay (CancerTYPE ID; BioTheranostics, San Diego, CA) to classify CUP into constituent cancers and then treat each with a current regimen and compare the outcomes to historical standards. The limitations of this approach are well-recognized: comparison of a highly selected, recent phase II trial population with historical controls can lead to an overestimate of benefit. The median OS of 12.5 months, while higher than some of the larger CUP trials, is within the range of several smaller trials. In addition, the selection imposed by waiting for the biopsy to be processed weeds out rapid progressors and tends to improve median survival. As the patient flow diagram shows, 12.8% of patients had insufficient tissue, which is likely to remain a problem in the future. A further 10% of patients were not treatment candidates by the time results became available. An additional 10% received empiric CUP therapy based on their oncologists’ recommendations so approximately 40% of patients enrolled did not receive protocol-specified therapy, and clearly this imposes a major selection bias on the remaining 60% of patients who received therapy. Also, the trial used two versions of the CancerTYPE ID RT-PCR assay, somewhat complicating analysis. Despite these limitations, this trial is a major step forward. The RT-PCR gene expression assay successfully classified 98% of patients. When using the assay for classification, the median OS was 12.5 months, which was better than reported in most large trials of CUP using empirical one-size-fits-all combination therapy. While the trial JOURNAL OF CLINICAL ONCOLOGY E D I T O R I A L VOLUME 31 NUMBER 2 JANUARY 1