Cancer of Unknown Primary (CUP): genetic evidence for a novel nosological entity? A case report.

Cancer of unknown primary (CUP) is an obscure disease characterized by multiple metastases in the absence of a primary tumor. No consensus has been reached whether CUPs are simply generated from cancers that cannot be detected or whether they are the manifestation of a still unknown nosological entity. Here, we report the complete expression and genetic analysis of multiple synchronous metastases harvested at warm autopsy of a patient with CUP. The expression profiles were remarkably similar and astonishingly singular. The whole exome analysis yielded a high number of mutations present in all metastases (fully shared), additional mutations (partially shared) accumulated one after another in a series, and few private mutations were unique to each metastasis. Surprisingly, the phylogenetic trajectory linking CUP metastases was atypical, depicting a common "stream", sprouting a series of linear "brooks", at variance from the extensive branched evolution observed in metastases from most cancers of known origin. The distinctive genetic and evolutionary features depicted suggest that CUP is a novel nosological entity.

[Clinicopathological characterization and autopsybased classification of the Cancer of Unknown Primary origin (CUP) syndrome]. / Az ismeretlen kiindulású malignus daganatok (Cancer of Unknown Primary origin; CUP) klinikopatológiája és boncolásalapú osztályozása.

Cancer of Unknown Primary origin (CUP) is characterized by metastatic tumor spread without identifiable primary tumor. CUP cohort was selected from 6966 autopsy cases (2001-2014). Type-1 ("clinical") CUPs: primary site was not found clinically but identified by autopsy. Type-2 ("clinicopathological") CUPs: no primary site either clinically or by autopsy. Type-3 ("pathological") CUPs: no tumor was suspected clinically whereas autopsy revealed metastatic spread from unidentifiable source. 2160 malignant tumors were found including 80 CUPs (type-1/2/3: 42/29/9). Cumulative incidence declined with time (3.70%; 2001-2007: 4.51%; 2008-2014: 3.19%) due to decreasing incidence of type-1 and -3 CUPs. CUPs were mostly adenocarcinomas and type-1 CUPs usually originate from the lung or pancreas. As a conclusion, type-2 and -3 CUPs may originate from microscopic-sized metastasizing primary tumors. Based on the above classification, improvement of clinical diagnostics may contribute to decreased incidence of type-1 CUPs and transfer of type-3 CUPs into type-2 category.

Integrating a 92-Gene Expression Analysis for the Management of Neuroendocrine Tumors of Unknown Primary

Background: Neuroendocrine tumors (NETs) are rare tumors that can originate from any part of the body. Often, imaging or exploratory surgery can assist in the identification of the tumor primary site, which is critical to the management of the disease. Neuroendocrine tumors (NETs) of unknown primary constitute approximately 10-15% of all NETs. Determining the original site of the tumor is critical to providing appropriate and effective treatment. Methods: We performed a retrospective review of neuroendocrine tumors at our institution between 2012 and 2016 using a 92-gene cancer ID analysis. Results: 56 patients with NETs of unknown primary were identified. Samples for 38 of the 56 underwent the 92-gene cancer ID analysis. The primary site of the tumor was identified with >95% certainty in 35 of the 38 patients. Conclusion: The 92-gene cancer ID analysis identified a primary site in 92% of our NETs study cohort that previously had been unknown. The results have direct implications on management of patients with regard to FDA-approved treatment options.

An integrated tool for determining the primary origin site of metastatic tumours.

AIMS: Cancers of unknown primary sites account for 3%-5% of all malignant neoplasms. Current diagnostic workflows based on immunohistochemistry and imaging tests have low accuracy and are highly subjective. We aim to develop and validate a gene-expression classifier to identify potential primary sites for metastatic cancers more accurately. METHODS: We built the largest Reference Database (RefDB) reported to date, composed of microarray data from 4429 known tumour samples obtained from 100 different sources and divided into 25 cancer superclasses formed by 58 cancer subclass. Based on specific profiles generated by 95 genes, we developed a gene-expression classifier which was first trained and tested by a cross-validation. Then, we performed a double-blinded retrospective validation study using a real-time PCR-based assay on a set of 105 metastatic formalin-fixed, paraffin-embedded (FFPE) samples. A histopathological review performed by two independent pathologists served as a reference diagnosis. RESULTS: The gene-expression classifier correctly identified, by a cross-validation, 86.6% of the expected cancer superclasses of 4429 samples from the RefDB, with a specificity of 99.43%. Next, the performance of the algorithm for classifying the validation set of metastatic FFPE samples was 83.81%, with 99.04% specificity. The overall reproducibility of our gene-expression-classifier system was 97.22% of precision, with a coefficient of variation for inter-assays and intra-assays and intra-lots <4.1%. CONCLUSION: We developed a complete integrated workflow for the classification of metastatic tumour samples which may help on tumour primary site definition.

Epigenetic profiling to classify cancer of unknown primary: a multicentre, retrospective analysis.

BACKGROUND: Cancer of unknown primary ranks in the top ten cancer presentations and has an extremely poor prognosis. Identification of the primary tumour and development of a tailored site-specific therapy could improve the survival of these patients. We examined the feasability of using DNA methylation profiles to determine the occult original cancer in cases of cancer of unknown primary. METHODS: We established a classifier of cancer type based on the microarray DNA methylation signatures (EPICUP) in a training set of 2790 tumour samples of known origin representing 38 tumour types and including 85 metastases. To validate the classifier, we used an independent set of 7691 known tumour samples from the same tumour types that included 534 metastases. We applied the developed diagnostic test to predict the tumour type of 216 well-characterised cases of cancer of unknown primary. We validated the accuracy of the predictions from the EPICUP assay using autopsy examination, follow-up for subsequent clinical detection of the primary sites months after the initial presentation, light microscopy, and comprehensive immunohistochemistry profiling. FINDINGS: The tumour type classifier based on the DNA methylation profiles showed a 99·6% specificity (95% CI 99·5-99·7), 97·7% sensitivity (96·1-99·2), 88·6% positive predictive value (85·8-91·3), and 99·9% negative predictive value (99·9-100·0) in the validation set of 7691 tumours. DNA methylation profiling predicted a primary cancer of origin in 188 (87%) of 216 patients with cancer with unknown primary. Patients with EPICUP diagnoses who received a tumour type-specific therapy showed improved overall survival compared with that in patients who received empiric therapy (hazard ratio [HR] 3·24, p=0·0051 [95% CI 1·42-7·38]; log-rank p=0·0029). INTERPRETATION: We show that the development of a DNA methylation based assay can significantly improve diagnoses of cancer of unknown primary and guide more precise therapies associated with better outcomes. Epigenetic profiling could be a useful approach to unmask the original primary tumour site of cancer of unknown primary cases and a step towards the improvement of the clinical management of these patients. FUNDING: European Research Council (ERC), Cellex Foundation, the Institute of Health Carlos III (ISCIII), Cancer Australia, Victorian Cancer Agency, Samuel Waxman Cancer Research Foundation, the Health and Science Departments of the Generalitat de Catalunya, and Ferrer.

Pan-cancer transcriptome analysis reveals a gene expression signature for the identification of tumor tissue origin.

Carcinoma of unknown primary, wherein metastatic disease is present without an identifiable primary site, accounts for ~3-5% of all cancer diagnoses. Despite the development of multiple diagnostic workups, the success rate of primary site identification remains low. Determining the origin of tumor tissue is, thus, an important clinical application of molecular diagnostics. Previous studies have paved the way for gene expression-based tumor type classification. In this study, we have established a comprehensive database integrating microarray- and sequencing-based gene expression profiles of 16 674 tumor samples covering 22 common human tumor types. From this pan-cancer transcriptome database, we identified a 154-gene expression signature that discriminated the origin of tumor tissue with an overall leave-one-out cross-validation accuracy of 96.5%. The 154-gene expression signature was first validated on an independent test set consisting of 9626 primary tumors, of which 97.1% of cases were correctly classified. Furthermore, we tested the signature on a spectrum of diagnostically challenging tumors. An overall accuracy of 92% was achieved on the 1248 tumor specimens that were poorly differentiated, undifferentiated or from metastatic tumors. Thus, we have identified a 154-gene expression signature that can accurately classify a broad spectrum of tumor types. This gene panel may hold a promise to be a useful additional tool for the determination of the tumor origin.

A multiplexed marker-based algorithm for diagnosis of carcinoma of unknown primary using circulating tumor cells.

Real-time, single-cell multiplex immunophenotyping of circulating tumor cells (CTCs) is hypothesized to inform diagnosis of tissue of origin in patients with carcinoma of unknown primary (CUP). In 20 to 50% of CUP patients, the primary site remains unidentified, presenting a challenge for clinicians in diagnosis and treatment. We developed a post-CellSearch CTC assay using multiplexed Q-dot or DyLight conjugated antibodies with the goal of detecting multiple markers in single cells within a CTC population. We adapted our approach to size-based CTC enrichment protocols for capturing CTCs and subsequent immunofluorescence (IF) using a minimal set of markers to predict the primary sites for common metastatic tumors. The carcinomas are characterized with cytokeratin 7 (CK7), cytokeratin 20 (CK20), thyroid transcription factor 1 (TTF-1), estrogen receptor (ER) or prostate-specific antigen (PSA. IF has been optimized in cultured tumor cells with individual antibodies, then with conjugated antibodies to form a multiplex antibody set. With IF, we evaluated antibodies specific to these 5 markers in lung, breast, colorectal, and prostate cancer cell lines and blood from metastatic prostate and breast cancer patients. This advanced technology provides a noninvasive, diagnostic blood test as an adjunct to routine tissue biopsy. Its further implementation requires prospective clinical testing.

Classification of Cancer Primary Sites Using Machine Learning and Somatic Mutations.

An accurate classification of human cancer, including its primary site, is important for better understanding of cancer and effective therapeutic strategies development. The available big data of somatic mutations provides us a great opportunity to investigate cancer classification using machine learning. Here, we explored the patterns of 1,760,846 somatic mutations identified from 230,255 cancer patients along with gene function information using support vector machine. Specifically, we performed a multiclass classification experiment over the 17 tumor sites using the gene symbol, somatic mutation, chromosome, and gene functional pathway as predictors for 6,751 subjects. The performance of the baseline using only gene features is 0.57 in accuracy. It was improved to 0.62 when adding the information of mutation and chromosome. Among the predictable primary tumor sites, the prediction of five primary sites (large intestine, liver, skin, pancreas, and lung) could achieve the performance with more than 0.70 in F-measure. The model of the large intestine ranked the first with 0.87 in F-measure. The results demonstrate that the somatic mutation information is useful for prediction of primary tumor sites with machine learning modeling. To our knowledge, this study is the first investigation of the primary sites classification using machine learning and somatic mutation data.

Cancers of unknown primary origin (CUP) are characterized by chromosomal instability (CIN) compared to metastasis of know origin.

BACKGROUND: Cancers of unknown primary (CUPs) constitute ~5% of all cancers. The tumors have an aggressive biological and clinical behavior. The aim of the present study has been to uncover whether CUPs exhibit distinct molecular features compared to metastases of known origin. METHODS: Employing genome wide transcriptome analysis, Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA), we defined the putative origins of a large series of CUP and how closely related a particular CUP was to corresponding metastases of known origin. LDA predictions were subsequently used to define a universal CUP core set of differentially expressed genes, that by means of gene set enrichment analysis was exploited to depict molecular pathways characterizing CUP. RESULTS: The analyses show that CUPs are distinct from metastases of known origin. CUPs exhibit inconsistent expression of conventional cancer biomarkers and QDA derived outlier scores show that CUPs are more distantly related to their primary tumor class than corresponding metastases of known origin. Gene set enrichment analysis showed that CUPs display increased expression of genes involved in DNA damage repair and mRNA signatures of chromosome instability (CIN), indicating that CUPs are chromosome unstable compared to metastases of known origin. CONCLUSIONS: CIN may account for the uncommon clinical presentation, chemoresistance and poor outcome in patients with CUP and warrant selective diagnostic strategies and treatment.

Development and validation of a gene expression tumour classifier for cancer of unknown primary.

Accurate identification of the primary tumour in cancer of unknown primary (CUP) is required for effective treatment selection and improved patient outcomes. The aim of this study was to develop and validate a gene expression tumour classifier and integrate it with histopathology to identify the likely site of origin in CUP.RNA was extracted from 450 formalin fixed, paraffin embedded samples of known origin comprising 18 tumour groups. Whole genome expression analysis was performed using a bead-based array. Classification of the tumours made use of a binary support vector machine, together with recursive feature elimination. A hierarchical tumour classifier was developed and incorporated with conventional histopathology to identify the origins of metastatic tumours.The classifier demonstrated an accuracy of 88% for correctly predicting the tumour type on a validation set of known tumours (n = 94). For CUP samples (n = 49) having a final clinical diagnosis, the classifier improved the accuracy of histology alone for both single and multiple predictions. Furthermore, where histology alone could not suggest any specific diagnosis, the classifier was able to correctly predict the primary site of origin.We demonstrate the integration of gene expression profiling with conventional histopathology to aid the investigation of CUP.

An audit of cancer of unknown primary notifications: A cautionary tale for population health research using cancer registry data.

BACKGROUND: Cancer of unknown primary (CUP) is a common cancer yet little is known about the reliability of incidence data. METHODS: We audited 574 CUP (C80.9) diagnoses (median age 81 years) registered by the New South Wales (NSW) Central Cancer Registry (2004-2007) in a cohort of Australian Government Department of Veterans' Affairs clients. The registry did not clarify diagnoses with notifiers during this period due to interpretation of privacy legislation. For the audit, current registry practice was applied by seeking additional information from CUP notifiers and reclassifying diagnoses as necessary. In addition, clinicopathological characteristics were extracted from notifications. Fisher's exact test and Student's t-test were used to compare the demographic and clinicopathological characteristics of the CUP subgroups. Age/sex-standardised CUP incidence rates and 95% confidence intervals were calculated, standardised to the 2001 Australian population. RESULTS: 172 (30.0%) cases were reclassified to a known primary site, mostly cutaneous, and nine (1.6%) were found to be non-malignant diagnoses. After the audit the age/sex-standardised CUP incidence rates decreased from 26.0 (95% CI 21.2-30.8) to 15.9 (95% CI 12.5-19.3) per 100,000 person-years. Of the 393 remaining CUP cases, 202 (51%) were registered on the basis of a clinical diagnosis (46 by death certificate only) and 191 (49%) by pathological diagnosis (79 by cytology alone). Compared to cases with a pathological diagnosis, cases with a clinical diagnosis were older (85.6 vs. 82.0 years, p<0.001), and the reported number and location of metastases differed (p<0.001); metastatic sites were more likely to be unspecified for clinical diagnoses (36.1% vs. 4.2%). CONCLUSIONS: Cancer registry processes can markedly influence CUP incidence. Future population-based CUP research should take this into account, and consider stratification by basis of diagnosis due to differences in patient and tumour characteristics.

Metabolic phenotypes in primary unknown metastatic carcinoma.

BACKGROUND: The purpose of this study is to evaluate expression of metabolism-related proteins in primary unknown metastatic carcinoma (PUMC) and associated implications for treatment. METHODS: A tissue microarray containing 77 cases of PUMC was constructed and immunohistochemical staining was used to evaluate expression of the following proteins: Glycolysis-related: Glut-1, carbonic anhydrase (CA) IX, and monocarboxylate transporter (MCT) 4; Glutaminolysis-related: glutaminase1 (GLS1), glutamate dehydrogenase (GDH), and amino acid transporter-2 (ASCT2); and Mitochondrial-related: ATP synthase, succinate dehydrogenase (SDH)A, and SDHB. The association between immunohistochemical staining results and clinicopathologic parameters was evaluated. RESULTS: The expression of metabolism-related proteins was different depending on the histologic subtype. Compared to other subtypes, squamous cell carcinomas (SQ) expressed more Glut-1 (p = 0.028), while adenocarcinomas (AD) expressed more SDHB in the stroma (p = 0.025). The expression of metabolism-related proteins was also different depending on the clinical subtypes. Glut-1 was expressed most in the nodal type and the least in carcinomatosis type, when compared to other subtypes (p = 0.021). The metabolic phenotypes also showed other trends: when the stroma showed no glutaminolysis, the tumor mostly invaded lymph node, bone, and brain, while the tumor invaded regions other than lymph node, bone, and brain when the stroma showed glutaminolysis (p = 0.003). When the stroma showed the mitochondrial metabolic type, the histologic subtype was mainly AD, but the non-mitochondrial type was associated more with SQ (P = 0.049). CONCLUSION: For PUMC, the expression of metabolism-related proteins, such as Glut-1 and SDHB, differs in the tumor or stroma depending on the clinical and histologic tumor subtype.

A 92-gene cancer classifier predicts the site of origin for neuroendocrine tumors.

A diagnosis of neuroendocrine carcinoma is often morphologically straight-forward; however, the tumor site of origin may remain elusive in a metastatic presentation. Neuroendocrine tumor subtyping has important implications for staging and patient management. In this study, the novel use and performance of a 92-gene molecular cancer classifier for determination of the site of tumor origin are described in a series of 75 neuroendocrine tumors (44 metastatic, 31 primary; gastrointestinal (n=12), pulmonary (n=22), Merkel cell (n=10), pancreatic (n=10), pheochromocytoma (n=10), and medullary thyroid carcinoma (n=11)). Formalin-fixed, paraffin-embedded samples passing multicenter pathologist adjudication were blinded and tested by a 92-gene molecular assay that predicts tumor type/subtype based upon relative quantitative PCR expression measurements for 87 tumor-related and 5 reference genes. The 92-gene assay demonstrated 99% (74/75; 95% confidence interval (CI) 0.93-0.99) accuracy for classification of neuroendocrine carcinomas and correctly subtyped the tumor site of origin in 95% (71/75; 95% CI 0.87-0.98) of cases. Analysis of gene expression subsignatures within the 92-gene assay panel showed 4 genes with promising discriminatory value for tumor typing and 15 genes for tumor subtyping. The 92-gene classifier demonstrated excellent accuracy for classifying and determining the site of origin in tumors with neuroendocrine differentiation. These results show promise for use of this test to aid in classifying neuroendocrine tumors of indeterminate primary site, particularly in the metastatic setting.

Efficient identification of miRNAs for classification of tumor origin.

Carcinomas of unknown primary origin constitute 3% to 5% of all newly diagnosed metastatic cancers, with the primary source difficult to classify with current histological methods. Effective cancer treatment depends on early and accurate identification of the tumor; patients with metastases of unknown origin have poor prognosis and short survival. Because miRNA expression is highly tissue specific, the miRNA profile of a metastasis may be used to identify its origin. We therefore evaluated the potential of miRNA profiling to identify the primary tumor of known metastases. Two hundred eight formalin-fixed, paraffin-embedded samples, representing 15 different histologies, were profiled on a locked nucleic acid-enhanced microarray platform, which allows for highly sensitive and specific detection of miRNA. On the basis of these data, we developed and cross-validated a novel classification algorithm, least absolute shrinkage and selection operator, which had an overall accuracy of 85% (CI, 79%-89%). When the classifier was applied on an independent test set of 48 metastases, the primary site was correctly identified in 42 cases (88% accuracy; CI, 75%-94%). Our findings suggest that miRNA expression profiling on paraffin tissue can efficiently predict the primary origin of a tumor and may provide pathologists with a molecular diagnostic tool that can improve their capability to correctly identify the origin of hitherto unidentifiable metastatic tumors and, eventually, enable tailored therapy.

Assigning site of origin in metastatic neuroendocrine neoplasms: a clinically significant application of diagnostic immunohistochemistry.

The neuroendocrine epithelial neoplasms (NENs) include well-differentiated neuroendocrine tumors (WDNETs) and poorly differentiated neuroendocrine carcinomas (PDNECs). Whereas PDNECs are highly lethal, with localized Merkel cell carcinoma somewhat of an exception, WDNETs exhibit a range of "indolent" biologic potentials-from benign to widely metastatic and eventually fatal. Within each of these 2 groups there is substantial morphologic overlap. In the metastatic setting, the site of origin of a WDNET has significant prognostic and therapeutic implications. In the skin, Merkel cell carcinoma must be distinguished from spread of a visceral PDNEC. This review intends to prove the thesis that determining the site of origin of a NEN is clinically vital and that diagnostic immunohistochemistry is well suited to the task. It will begin by reviewing current World Health Organization terminology for the NENs, as well as an embryologic and histologic pattern-based classification. It will present population-based data on the relative frequency and biology of WDNETs arising at various anatomic sites, including the frequency of metastases of unknown primary, and comment on limitations of contemporary imaging techniques, as a means of defining the scope of the problem. It will go on to discuss the therapeutic significance of site of origin. The heart of this review is a synthesis of data compiled from >100 manuscripts on the expression of individual markers in WDNETs and PDNECs, as regards site of origin. These include proteins that are considered "key markers" and others that are either useful "secondary markers," potentially very useful markers that need to be further vetted, or ones that are widely applied despite a lack of efficacy. It will conclude with my approach to the metastatic NEN of unknown origin.

A second-generation microRNA-based assay for diagnosing tumor tissue origin.

BACKGROUND: Cancers of unknown primary origin (CUP) constitute 3%-5% (50,000 to 70,000 cases) of all newly diagnosed cancers per year in the United States. Including cancers of uncertain primary origin, the total number increases to 12%-15% (180,000 to 220,000 cases) of all newly diagnosed cancers per year in the United States. Cancers of unknown/uncertain primary origins present major diagnostic and clinical challenges because the tumor tissue of origin is crucial for selecting optimal treatment. MicroRNAs are a family of noncoding, regulatory RNA genes involved in carcinogenesis. MicroRNAs that are highly stable in clinical samples and tissue specific serve as ideal biomarkers for cancer diagnosis. Our first-generation assay identified the tumor of origin based on 48 microRNAs measured on a quantitative real-time polymerase chain reaction platform and differentiated 25 tumor types. METHODS: We present here the development and validation of a second-generation assay that identifies 42 tumor types using a custom microarray. A combination of a binary decision-tree and a k-nearest-neighbor classifier was developed to identify the tumor of origin based on the expression of 64 microRNAs. RESULTS: Overall assay sensitivity (positive agreement), measured blindly on a validation set of 509 independent samples, was 85%. The sensitivity reached 90% for cases in which the assay reported a single answer (>80% of cases). A clinical validation study on 52 true CUP patients showed 88% concordance with the clinicopathological evaluation of the patients. CONCLUSION: The abilities of the assay to identify 42 tumor types with high accuracy and to maintain the same performance in samples from patients clinically diagnosed with CUP promise improved utility in the diagnosis of cancers of unknown/uncertain primary origins.

Long-term survivors among patients with cancer of unknown primary.

There is a general conception among oncologists that CUP patients behave poorly to treatment and carry a dismal survival. In this paper we are trying to elucidate the different groups of CUP patients and to describe in details the diagnostic and therapeutic management of the prognostically favorable patients. Clinicians should be aware that the favorable CUP cases must be treated according to recent guidelines with either specific locoregional and/or systemic therapy and that they commonly enjoy a long survival. Survival data of 219 CUP patients treated at Ioannina University Hospital from 1995 until 2011 are also presented.

Clinical presentation of carcinoma of unknown primary: 14 years of experience.

A carcinoma of unknown primary (CUP) is a histologically confirmed metastatic cancer without a definitive primary site after performing a detailed medical examination. The purpose of the study was to classify unfavorable CUPs into more reliable disease entities, which reflect the clinical course. We reviewed the medical records of patients diagnosed with a CUP between January 1995 and March 2008. Patients were classified into a conventional favorable-risk group and a newly proposed unfavorable-risk group according to the clinicopathologic features. Five hundred eighty-six patients were diagnosed with CUPs. Fifty-six (9.6%) patients were classified in the conventional favorable-risk group, and 486 (82.9%) patients were classified in the unfavorable-risk group. We further classified the 486 patients into six subgroups with an unfavorable risk, while excluding 29 patients (5.0%) who were not classifiable. The overall survival of the conventional favorable-risk group was 47.0 months (95% CI, 11.1~82.9 months), which was significantly longer than that of any subgroup of the newly proposed unfavorable-risk group (P < 0.001). Patients with squamous cell carcinoma in the abdominopelvic cavity showed similar overall survival with unfavorable-risk group (P = 0.484). Women with non-papillary malignant ascites had a survival in between the favorable and unfavorable groups (P < 0.001). The newly proposed unfavorable-risk group may assist in classifying CUP patients with an unfavorable risk in a clinically more meaningful way. Squamous cell carcinoma in the abdominopelvic cavity should be considered in the unfavorable-risk group and women with non-papillary malignant ascites in an intermediate-risk group. Further studies with molecular profiling would help in classifying and treating patients with CUPs and an unfavorable risk.

[Classification and regression tree analysis of 154 patients with cancer of unknown primary].

OBJECTIVE: To explore the prognostic factors and their impact on survival of patients with cancer of unknown primary (CUP). METHODS: The clinical and follow up data of 154 CUP patients referred to the Cancer Hospital & Institute, Chinese Academy of Medical Sciences from January 1, 2003 to December 31, 2007 were analyzed. Multivariate analysis of survival was performed using recursive partitioning referred to as classification and regression tree (CART) analysis. RESULTS: The median survival for 154 eligible consecutive CUP patients was 18.2 months, and the 5-year survival rate was 1.3%. CART was performed with an initial split on age of 34, and 5 terminal subgroups were formed. The median survival of the 5 subsets ranged from 5.5 months (younger than 34 years old subgroup) to 61.8 months for patients at age of 34 to 60, with one or two organ sites involved, and non-adenocarcinoma histology subsets. CONCLUSIONS: CART can be used to identify previously unappreciated patient subsets and is a useful method for dissecting complex clinical situations and identifying homogeneous patient populations in clinical practice and future clinical trials.