April 2013, Vol 2, No 2

← Back to Issue

Physician-Reported Clinical Utility of the 92-Gene Molecular Classifier in Tumors With Uncertain Diagnosis Following Standard Clinicopathologic Evaluation

Benjamin Kim, MD, MPhil


Advances in understanding the genetic basis of many cancers and the development of molecularly targeted therapies are increasing the need for diagnostic resolution. Patient outcomes have improved with the use of predictive biomarker testing, targeted therapies, and site-specific chemotherapy regimens.1,2 However, tumor classification remains unknown or uncertain in a quarter to a third of newly diagnosed metastatic cancer,3,4 particularly in cases that are poorly differentiated or undifferentiated, contain limited biopsy tissue for immunophenotypic analysis, and/or are associated with atypical clinical presentation. Despite innovations in imaging and pathologic techniques, challenges with difficult-to-diagnose metastatic cancers are often compounded by the nonstandardized approaches traditionally employed by clinicians to solve diagnostic dilemmas using iterative immunohistochemical (IHC) staining. Furthermore, this approach may delay diagnosis and incur considerable costs,5,6 as well as deplete tissue that may be essential for downstream predictive biomarker analysis. In recent years, molecular assays have been developed and are currently utilized as diagnostic complements to standard clinicopathologic evaluation in difficult-to-diagnose cases. Cancers with unknown or uncertain primary sites of origin is an area in oncology that may significantly benefit from the use of molecular-based assays, in terms of improving both the accuracy and efficiency of diagnostic classification, as well as more tailored treatment recommendations.

Molecular classifiers have been clinically validated, with overall sensitivities ranging from 83% to 89%.7-10 The accuracy and potential utility of molecular classification in patients diagnosed with cancer of unknown primary (CUP) have been examined in several studies,11,12 and molecular classification has been incorporated into diagnostic algorithms for CUP.3,13 Studies examining patient outcomes have demonstrated favorable outcomes in patients treated based on their molecular diagnosis. In a retrospective study, patients predicted to have a colorectal site of origin by molecular classification and treated with site-specific regimens had a median survival of 27 months, similar to patients with known metastatic colorectal cancer.14 In a study combining retrospective cases from Sarah Cannon Research Institute and prospective cases from The University of Texas MD Anderson Cancer Center, patients with a colon cancer molecular profile had better responses to colon cancer–specific treatment regimens than those who were treated with empiric regimens (ie, taxane/platinum).15 Finally, a prospective trial assessing outcomes in patients with CUP treated based on molecular classification with the 92-gene assay reported interim overall survival results that compared favorably with previous trials of empiric treatment regimens for CUP.16

An analysis of real-world clinical testing demonstrated broader uptake of molecular classification in recent years to resolve differential and confirm suspected diagnoses in challenging cases.7 However, insight into how molecular classification is being utilized in clinical care and how it impacts oncologists’ clinical assessments and decision making have not been evaluated. This study was designed to assess the characteristics and clinical decision outcomes of medical oncologists who have ordered the 92-gene molecular classifier (CancerTYPE ID) and determine how they view the clinical value of the test in their practice.

Patients and Methods

This is a retrospective, survey-based study of medical oncologists who ordered the 92-gene assay as part of routine clinical care. A 61-question survey was developed by the investigators for use in this study. The primary objectives of the study were to examine whether the results of the 92-gene assay aided in: 1) the determination of a clinical diagnosis of the primary site of origin for their patient’s tumor, and 2) the therapeutic decision making for their patient. Additional assessments included physician and patient characteristics; pre-assay diagnostic tests, diagnosis, and treatments (if any); factors considered when ordering the 92-gene assay; circumstances under which physicians considered ordering the 92-gene assay; and how the 92-gene assay helped in the diagnosis of tumors and treatment planning for patients, or why it was not helpful. The survey was uploaded to a secure Web-based portal, pretested by multiple reviewers and on several Internet browsers, and activated for use.

To obtain the study sample population, a database of clinical cases maintained by bioTheranostics was queried to generate a file of 1105 medical oncologists based on the following inclusion criteria: 1) the oncologist must have had a tumor sample analyzed using the second generation of the 92-gene assay,7 2) the patient’s tumor sample must have had sufficient quantity and quality of RNA to pass quality control standards, 3) the 92-gene assay must have provided a prediction for the primary site of origin (ie, not unclassifiable by assay), and 4) the ordering physician’s name and contact information were available. Cases analyzed as part of clinical trials were excluded.

Physician participation was solicited via direct mail communication from investigators (BK and JLM), and physicians were directed to a secure Web site to complete the survey. To minimize selection and recall biases, participating physicians were asked to complete the survey questions as they pertained to the first patient for whom they ordered the 92-gene assay after a specified date (March 15, 2010) and to review the patient’s medical record prior to initiation of the survey. Thus, each physician only completed the survey for 1 patient. The survey was estimated to take 30 to 45 minutes, and physicians were offered $250 to complete the survey. Second and third contacts were made via direct mail or telephone to nonresponders. A third-party contract research organization (Percolation) provided operational support for the study.

The protocol was reviewed and approved by an independent institutional review board (Western Institutional Review Board); informed consent was not required for the study. Physicians were instructed that the survey was to be completed based only on materials (ie, data, documents, records) that had already been collected. No new patient-level data or patient identifying information were collected for the purposes of the survey. All statistical analyses were descriptive in nature, summarizing frequency and percentage

Physician and Patient/Tumor Baseline

Of 1105 physicians receiving invitations, 103 (9.4%) completed the Web-based survey. Physician and baseline patient/tumor characteristics are reported in the Table. The majority of respondents (82%) worked in community-based practices. Respondents had a median of 10 years of practice experience (range, 1-39 years), and most (77%) had utilized the 92-gene assay multiple times in the previous 12 months (median, 3; range, 0-20). The majority of patients had multiple sites of disease, most frequently involving lung (43%), liver (38%), lymph nodes (34%), and bone (10%). Carcinomas (26%) and adenocarcinomas (55%) were the most frequent histologic types based on pathologic evaluation prior to the 92-gene assay. The majority of tumors were poorly differentiated or undifferentiated; only 1 case involved a well-differentiated tumor.

Pre–92-Gene Assay
Diagnostic Testing and Treatment Planning

Patients underwent extensive diagnostic testing prior to physicians ordering the 92-gene assay. The mean number of IHC stains performed was 6.2 (median, 6; range, 1-17). Most patients (87%) had undergone a chest, abdomen, and/or pelvis CT scan, 61% a PET scan, 21% an MRI, 21% a colonoscopy, and 21% an esophagogastroduodenoscopy. In addition, 45% had serum tumor marker tests ordered.

Physicians were asked to report all clinically suspected primary sites of origin prior to ordering the 92-gene assay. The mean number of clinically suspected sites of origin was 2.6; however, there were varying degrees of diagnostic certainty pre-assay: 34% of respondents reported a single suspected site, 66% had 2 or more suspected sites, and 21% suspected 4 or more sites (Figure 1A). The most frequent clinically suspected primary sites of origin prior to ordering the 92-gene assay were lung (46%), gallbladder/biliary tree (26%), pancreas (26%), colon (25%), breast (18%), and liver (17%).


Figure 1

To assess the circumstances under which physicians considered ordering the 92-gene assay, physicians were asked about their confidence in selecting the most appropriate treatment before ordering the assay and where in the treatment course they ordered it. A minority of physicians (18%) indicated high or very high confidence in being able to select the most appropriate treatment regimen prior to ordering the 92-gene assay (35% reported very low or low confidence and 47% moderate confidence). For therapeutic decision making, 31% of respondents ordered the assay after the patient had received at least 1 line of therapy, while the majority of physicians did not begin treatment until after the 92-gene assay results were received: 24% reported that they had a specific regimen in mind but did not begin treatment, while 45% did not have a single regimen in mind before ordering the 92-gene classifier.

Post–92-Gene Assay
Impact on Clinical Diagnosis

The 92-gene assay predicted 26 different tumor types, including both common (eg, lung, breast, colorectal) and rare tumor types (eg, cholangiocarcinoma, islet cell carcinoma; Figure 2A). The impact of the 92-gene assay on diagnostic certainty was assessed by comparing suspected tumor origin sites before and after the assay. Compared with before the assay, the number of clinically suspected sites of origin following the 92-gene assay decreased in the majority of cases (63% decreased, 33% had the same number, and 4% increased), resulting in an overall decrease in the mean number of clinically suspected sites of origin from 2.6 to 1.2. Following incorporation of the 92-gene assay results, most physicians (84%) reported that they were able to elucidate a single tumor site of origin (Figure 1A).

The clinical utility of the 92-gene assay was also assessed by evaluating physician incorporation of the 92-gene assay prediction into their final clinical diagnosis. The 92-gene assay prediction was concordant with the final clinical diagnosis in 84% of cases (Figure 1B). Notably, in one-third of these cases, the molecular diagnosis provided by the 92-gene assay had not been clinically considered prior to molecular testing (Figure 1B). In most cases (80%), no additional diagnostic evaluations were performed following the 92-gene assay. In cases with additional post-assay evaluations, the most commonly conducted test was a PET scan.

Most physicians indicated that the 92-gene assay was helpful in determining a final clinical diagnosis for the primary site of tumor origin (Figure 2B), with 78% of respondents answering “strongly agree” or “agree.” Of these, 45% indicated that the test result increased their confidence in an existing diagnosis, 30% reported that it aided in making an initial diagnosis, 20% said it helped to change the diagnosis, and 6% provided other responses. Among respondents who answered “disagree” or “strongly disagree,” the most common reason was that the test result was inconsistent with the clinical presentation or the oncologist’s impression of the primary tumor origin site.

Figure 2

Impact on Treatment Decision Making
Respondents indicated that the 92-gene assay provided clinical utility for therapeutic decision making, with 81% of physicians answering “strongly agree” or “agree” that the assay was helpful in the treatment decision-making process (Figure 2B). Among these respondents, 52% indicated that the assay guided the selection of an initial treatment regimen, 29% reported that it confirmed the appropriateness of a previously determined regimen, 10% said that it informed a change in treatment, and 6% stated that the test result allowed them to exclude a treatment regimen. Among respondents who did not agree that the 92-gene assay helped in the treatment regimen decision-making process, the most common reason was that the test result was inconsistent with clinical presentation or impression.


There has been a rapid introduction of molecular diagnostic tests in oncology in recent years; however, to date, the value of molecular classification tests on diagnostic certainty and treatment decision making in routine clinical care have not been established. In the study reported here, the 92-gene molecular classification assay was generally ordered to characterize high-grade metastatic tumors that had undergone extensive clinical and pathologic evaluation prior to submission. Physicians reported using the assay to address a broad range of issues surrounding diagnostic uncertainty, including independent confirmation of a suspected diagnosis prior to treatment initiation, reassessment of an initial diagnosis following treatment failure, resolution of narrow differential diagnoses, and identification of primary tumor site for patients diagnosed with CUP. Physicians reported strong diagnostic impact of the 92-gene assay. The final clinical diagnosis was narrowed (in cases with a differential diagnosis pre-assay) or changed (in cases with a single suspected diagnosis pre-assay) in the vast majority of cases examined, and there was high concordance between the assay prediction and the final clinical diagnosis. Notably, in 41% of the cases, the molecular diagnosis was not clinically considered in the initial diagnosis prior to the assay; however, the physician’s final diagnosis was changed to reflect the molecular diagnosis in two-thirds of these cases. In addition, few additional diagnostic tests were ordered after the molecular test, suggesting that most oncologist respondents viewed the results of the assay as being definitive. These findings highlight the difficulty in obtaining a definitive diagnosis for many high-grade metastatic cancers and the potential role of using an objective molecular diagnostic test to resolve a differential diagnosis or identify a primary site that was not considered following typical clinicopathologic characterization.

Difficult-to-diagnose tumors present a challenge to oncologists in terms of therapeutic decision making, as knowledge of tumor type and primary site of origin inform treatment selection from both an efficacy and safety standpoint. In the absence of a confirmed diagnosis of tumor type, patients are traditionally treated with empiric, broadly acting cytotoxic chemotherapy regimens that are associated with modest response, toxicity, and median survival times of only 6 to 10 months.3,17 The data reported in this study highlight the clinical impact of this challenge, as the 6 most common final clinical diagnoses (lung, pancreaticobiliary, ovary, kidney, breast, and melanoma; data not shown) require different treatment strategies, and several have approved, targeted therapies available.18 While not directly assessed in this study, multiple studies have provided evidence that optimizing treatment plans based on accurate tumor site of origin and targeting activated molecular pathways can improve patient outcomes compared with an empiric therapeutic approach.15,19-23 In particular, in a recently published prospective clinical trial, patients with metastatic CUP who were treated with site-specific chemotherapy based on the 92-gene assay prediction had prolonged overall survival compared with historical controls from prior CUP trials from the same clinical trial network (median overall survival, 12.5 vs 9.1 months).16 Finally, the data reported here complement previous studies evaluating the clinical utility of genomic testing in the real-world setting. For example, studies investigating the clinical utility of the 21-gene recurrence score assay have shown that between 21% and 51% of physicians changed their recommendation for adjuvant breast cancer treatment following receipt of the assay results.24-27 In the present study, over half of respondents reported that the 92-gene assay either guided initial treatment choice, changed the treatment, or excluded a treatment.

The heterogeneity of different tumor types presents a challenge to prospectively evaluate the impact of incorporating molecular assays into diagnostic algorithms for metastatic cancer. Novel research designs are needed to help quantify the impact of these technologies. Pragmatic trial designs, which have been proposed for comparing and evaluating the effectiveness of medical interventions in real-world settings,28-30 may have utility in assessing the associated impact of molecular diagnostic testing on clinical outcomes. For instance, patients could be randomized to receive usual diagnostic care or incorporation of a multigene molecular assay. Diagnostic and therapeutic decisions would be at the discretion of the treating physician. While this design would not provide data regarding the efficacy of treatment based on specific genomic characteristics, such an approach would provide an assessment of the impact of a novel strategy (vs current standard of care) on a broad range of health outcomes, including survival, time to treatment, biopsy tissue preservation, total cost of care, and quality of life. Such studies are urgently needed so policies can be developed to ensure the optimal uptake of genomic medicine in clinical practice.

Several limitations should be considered when evaluating the results of this study. First, as a voluntary survey, there is the potential for a systematically different impression of clinical utility in survey responders versus nonresponders. Thus, the results of the study may not be generalizable to all test users. We attempted to mitigate for this possibility by collecting survey results from a large number of physicians (>100 individual oncologists) and only allowing 1 case per physician. Second, survey answers could be affected by recall error. To address this issue, physicians were asked to review their patient’s medical record prior to completing the survey. Finally, we attempted to address the possibility of patient selection bias by instructing physicians to complete the survey for the first patient after a defined date, rather than allowing the physician to choose a particular case to review.


Defining and evaluating the real-world clinical utility of molecular diagnostic assays is an evolving field. More research into how oncologists incorporate the use of these testing modalities, make clinical assessments and treatment recommendations, and impact patient outcomes through their use will be important for assessing their place in standard clinical practice. Additional prospective studies incorporating molecular diagnostic assays will further help to define their clinical utility. Moreover, given the complexity of genomic medicine, decision support tools may be needed to help physicians determine when testing is indicated and how to interpret results. This study was designed to assess the clinical indications and utility of molecular classification in real-world clinical practice. The 92-gene assay was utilized across an array of diagnostically challenging tumors, and medical oncologists reported that use of the assay resulted in more specific diagnoses for primary site of tumor origin and helped them make both diagnostic and treatment decisions.


This study was funded by bioTheranostics, Inc. Study funding included support for the third-party contract research organization, as well as research support provided to BK and JLM. BES, CAS, and MGE are employees and stockholders of bioTheranostics, Inc.

1. Mok TS, Wu YL, Thongprasert S, et al. Gefitinib or carboplatin-pac­litaxel in pulmonary adenocarcinoma. N Engl J Med. 2009;361:947-957.
2. Bokemeyer C, Bondarenko I, Makhson A, et al. Fluorouracil, leucovorin, and oxaliplatin with and without cetuximab in the first-line treatment of metastatic colorectal cancer. J Clin Oncol. 2009;27:663-671.
3. Greco FA, Hainsworth JD. Cancer of unknown primary site. In: DeVita VT Jr, Lawrence TS, Rosenberg SA, eds. Cancer: Principles and Practice of Oncology. 9th ed. Philadelphia, PA: Lippincott Williams & Wilkins; 2011:2033-2051.
4. Hillen HF. Unknown primary tumours. Postgrad Med J. 2000;76:690-693.
5. Schapira DV, Jarrett AR. The need to consider survival, outcome, and expense when evaluating and treating patients with unknown primary carcinoma. Arch Intern Med. 1995;155:2050-2054.
6. Schroeder BE, Laouri M, Chen E, et al. Pathological diagnoses in cases of indeterminate or unknown primary submitted for molecular tumor profiling. Mod Pathol. 2012;25(suppl 2). Abstract 429.
7. Erlander MG, Ma XJ, Kesty NC, et al. Performance and clinical evaluation of the 92-gene real-time PCR assay for tumor classification. J Mol Diagn. 2011;13:493-503.
8. Kerr SE, Schnabel CA, Sullivan PS, et al. Multisite validation study to determine performance characteristics of a 92-gene molecular cancer classifier. Clin Cancer Res. 2012;18:3952-3960.
9. Pillai R, Deeter R, Rigl CT, et al. Validation and reproducibility of a microarray-based gene expression test for tumor identification in formalin-fixed, paraffin-embedded specimens. J Mol Diagn. 2011;13:48-56.
10. Rosenwald S, Gilad S, Benjamin S, et al. Validation of a micro- RNA-based qRT-PCR test for accurate identification of tumor tissue origin. Mod Pathol. 2010;23:814-823.
11. Greco FA, Spigel DR, Yardley DA, et al. Molecular profiling in unknown primary cancer: accuracy of tissue of origin prediction. Oncologist. 2010;15:500-506.
12. Greco FA. Evolving understanding and current management of patients with cancer of unknown primary site. Commun Oncol. 2010;7:183-188.
13. Greco FA, Oien K, Erlander M, et al. Cancer of unknown primary: progress in the search for improved and rapid diagnosis leading toward superior patient outcomes. Ann Oncol. 2012;23:298-304.
14. Hainsworth JD, Schnabel CA, Erlander MG, et al. A retrospective study of treatment outcomes in patients with carcinoma of unknown primary site and a colorectal cancer molecular profile. Clin Colorectal Cancer. 2012;11:112-118.
15. Varadhachary GR, Talantov D, Raber MN, et al. Molecular profiling of carcinoma of unknown primary and correlation with clinical evaluation. J Clin Oncol. 2008;26:4442-4448.
16. Hainsworth JD, Rubin MS, Spigel DR, et al. Molecular gene expression profiling to predict the tissue of origin and direct site-specific therapy in patients with carcinoma of unknown primary site: a prospective trial of the Sarah Cannon Research Institute. J Clin Oncol. 2013;31:217-223.
17. Pavlidis N, Pentheroudakis G. Cancer of unknown primary site.
Lancet. 2012;379:1428-1435.
18. National Comprehensive Cancer Network. NCCN Clinical Practice Guidelines in Oncology (NCCN Guidelines). www.nccn.org/professionals/physician_gls/f_guidelines.asp#site. Accessed August 15, 2012.
19. Abbruzzese JL, Abbruzzese MC, Lenzi R, et al. Analysis of a diagnostic strategy for patients with suspected tumors of unknown origin. J Clin Oncol. 1995;13:2094-2103.
20. Varadhachary GR, Raber MN, Matamoros A, et al. Carcinoma of unknown primary with a colon-cancer profile-changing paradigm and emerging definitions. Lancet Oncol. 2008;9:596-599.
21. Pentheroudakis G, Lazaridis G, Pavlidis N. Axillary nodal metastases from carcinoma of unknown primary (CUPAx): a systematic review of published evidence. Breast Cancer Res Treat. 2010;119:1-11.
22. Pentheroudakis G, Pavlidis N. Serous papillary peritoneal carcinoma: unknown primary tumour, ovarian cancer counterpart or a distinct entity? A systematic review. Crit Rev Oncol Hematol. 2010;75:27-42.
23. Hainsworth JD, Fizazi K. Treatment for patients with unknown primary cancer and favorable prognostic factors. Semin Oncol. 2009;36:44-51.
24. Oratz R, Kim B, Chao C, et al. Physician survey of the effect of the 21-gene recurrence score assay results on treatment recommendations for patients with lymph node-positive, estrogen receptor-positive breast cancer. J Oncol Pract. 2011;7:94-99.
25. Oratz R, Paul D, Cohn AL, et al. Impact of a commercial reference laboratory test recurrence score on decision making in early-stage breast cancer. J Oncol Pract. 2007;3:182-186.
26. Asad J, Jacobson AF, Estabrook A, et al. Does oncotype DX recurrence score affect the management of patients with early-stage breast cancer? Am J Surg. 2008;196:527-529.
27. Lo SS, Mumby PB, Norton J, et al. Prospective multicenter study of the impact of the 21-gene recurrence score assay on medical oncologist and patient adjuvant breast cancer treatment selection. J Clin Oncol. 2010;28:1671-1676.
28. Tunis SR, Stryer DB, Clancy CM. Practical clinical trials: increasing the value of clinical research for decision making in clinical and health policy. JAMA. 2003;290:1624-1632.
29. Luce BR, Kramer JM, Goodman SN, et al. Rethinking randomized clinical trials for comparative effectiveness research: the need for transformational change. Ann Intern Med. 2009;151:206-209.
30. Roland M, Torgerson DJ. What are pragmatic trials? BMJ. 1998;316:285.

Uncategorized - April 22, 2013

The Doctor Won’t See You Now. He’s Clocked Out.

Big government likes big providers. That’s why ObamaCare is gradually making the local doctor-owned medical practice a relic. In the not too distant future, most physicians will be hourly wage earners, likely employed by a hospital chain. Why? Because when doctors practice in small offices, it is hard for Washington [ Read More ]

Uncategorized - April 22, 2013

The RomneyCare Bill Comes Due

The health reform that Mitt Romney passed in 2006 in Massachusetts presaged President Obama’s, and its results are showing what we can expect nationwide. The latest warning comes in a huge new tax increase proposed by Governor Deval Patrick. The second-term Democrat followed his party’s recent habit and proposed an [ Read More ]