RedPath offers the PathFinderTG® molecular test as a way to provide definitive diagnoses, prognostic information and predict responses to chemotherapy. While integrating the molecular information that a test like PathFinderTG® provides is of interest and the subject of research for neoplasms, currently the specific molecular features, associated genetic biomarkers and their relationships with clinical outcomes are not well defined. Accordingly, their role in clinical decision making, including selecting treatment options, has not been defined.

 

Because of the scope of claims by the company for widespread application of PathFinderTG® in multiple organ systems and clinical scenarios, and over 500 papers they reference “supporting the clinical efficacy of PathFinderTG®” (http://www.redpathip.com/publications.asp, accessed 04/05/2008), we chose 2 representative applications to address in this policy—pancreatic cysts and gliomas. Published studies reviewed for this policy include those cited by RedPath as providing “clinical validation” for PathFinderTG®, as well as those representative of the current medical literature describing what is known of the molecular profiles of various tumor types (specifically pancreatic cystic and glial neoplasms) and their potential role in clinical decision-making.

 

The diagnosis of cystic pancreatic lesions is usually performed by endoscopic-ultrasound-guided fine-needle aspiration sampling the fluid and cyst wall for cytologic examination and analysis. Cytologic examination of these lesions can be difficult or indeterminate due to low cellularity, cellular degeneration, procedural difficulties, etc. Ancillary tests are often performed on the cyst fluid in an attempt to aid in the diagnosis (e.g., amylase, lipase, carcinoembryonic antigen levels), but results can still be equivocal. Information provided by additional testing modalities would, therefore, be potentially useful.

 

The PathFinderTG® test is suggested to serve the following purposes based on panels of molecular markers, including KRAS point mutations and loss of heterozygosity (LOH), in pancreatic cyst samplings/fluid:

 

From a convenience sample of tissue specimens archived between 1999 and 2004 (Lapkus et al) described mutations in brushing samples from pancreatic and bile duct lesions. No description of patients or tumor stage was reported. Various molecular findings and mutations from 40 pancreatic duct and 21biliary brushing specimens were described, including LOH for a panel of 15 markers (1p, 3p, 5q, 9p, etc.) in addition to point mutation in KRAS-2 for cells obtained from cytologic preparations that had been diagnosed by conventional cytology as malignant (n=10), atypical (n=20) or benign (n=10).. The authors suggested “[knowing that cancer evolves through a temporal process of mutation acquisition raises the question whether this sequence can be determined in clinical specimens.” Tissue histology was available for 4 samples in the malignant group (all carcinomas); 4 in the benign group (3 were benign); and 9 in the atypical group (4 carcinomas). These results provide little evidence to support clinical value of the mutation patterns. Results from just 17 of 40 samples were compared to tissue diagnoses; measures of accuracy are not calculable. The cross-sectional study design further prevents meaningful conclusion.

 

Khalid et al. (2004) reported results from a retrospectively assembled case series using LOH in the diagnosis of pancreatic malignancy. In an apparent convenience sample of 26 patients with either surgical removal (n=25) or “long term follow-up” (n=1) there were 17 cancers (6 pancreatic and 11 cholangiocarcinomas) and 9 non-cancers. No description of patients or tumor stage was reported. Using “standard” microscopic analysis of the cytology specimens, 8 were judged malignant, 10 indeterminate and 8 benign. Results of various molecular tests (LOH, KRAS, etc.) were evaluated on cytology and tissue specimens. Reported sensitivity was 100% (95% CI: 83–100%) and specificity 100% (95% CI: 70–100%; CIs calculated) for a diagnosis of malignancy by PathFinderTG®. Whether the molecular findings would have had clinical impact was unclear. In addition, results were not presented in a manner that allowed direct comparison between the molecular and standard testing, both for the cytology analysis and the tissue specimens.

 

In a case series, Khalid et al. (2005) prospectively collected fluid from 116 pancreatic cysts. Molecular analyses were restricted to 31 patients who underwent surgery and “5 patients [who] reached a final diagnosis of malignancy based on cytology....” In this subgroup of 36, molecular analyses included KRAS point mutation and a broad panel of LOH markers. Whether those conducting molecular assays were blinded to confirmatory results was not reported. Cutoffs defining significant LOH at loci were established using normal specimens—more than a 2 standard deviation was defined as abnormal. The sensitivity and specificities of KRAS mutation alone distinguishing premalignant from malignant cysts were 93% (95% CI: 62–100%) and 87% (95% CI: 62–96%) (CIs calculated). Subsequent testing for LOH did not alter sensitivity and correctly reclassified one false–positive result increasing specificity to 93% (95% CI: 70–100%). Analysis from this small sample did not demonstrate incremental value for LOH testing added to KRAS.

 

These studies highlight important issues pertaining to test accuracy and whether there is evidence that PathFinderTG® results provide incremental information that informs diagnostic and prognostic decision-making for patients with pancreatic cysts, including the following:

 

First, step-wise accumulation of KRAS mutations and allelic losses (LOH) accompanying cancer progression in pancreatic ductal adenocarcinomas have been found (Furukawa 2006, Koorstra 2008).  However, overlap between molecular findings in inflammatory and neoplastic conditions has also been described—specifically KRAS mutations in chronic pancreatitis (Popovic 2007, Khalid 2006, Muller 2001, Uehara 1999).  Progression of genetic mutations in mucinous pancreatic tumors has been described and although there appear to be similarities with pancreatic ductal adenocarcinomas, the precise pattern and progression of these mutations remains to be elucidated. For example, while suggesting the pattern of KRAS mutation followed by allelic loss predicts malignancy in pancreatic cysts, Khalid et al. also noted a lack of clarity in the “pattern and rate of mutation accumulation….”

 

Second, potential sampling discrepancies between the molecular profile of a pancreatic cyst wall aspirate or fluid and tissue specimen needs to be further addressed. This is particularly important in MCNs of the pancreas, which may show differing histology in different areas of the tumor, and may have focal invasion (Singh 2007). To further complicate this issue, RedPath does not address the genetic heterogeneity seen among the subtypes of MCNs (for example, KRAS mutations have been described as occurring in 20% of mucinous cystic adenomas, 33% of adenomas with moderate dysplasia, and 89% with carcinoma in situ.

 

Third, whether those performing the assay were blinded to surgical results was not stated in the studies reviewed. Clear description of case selection (e.g., consecutive or convenience) was absent.

 

Fourth, sample sizes were small as reflected in the wide confidence intervals calculated for reported sensitivities and specificities.

 

Finally, both adequate follow-up and replicated results (validation) are required to demonstrate whether PathFinderTG® results inform diagnostic and prognostic decision-making in a manner that will benefit patients. We are unaware of long-term studies with defined clinical outcomes using PathFinderTG® to differentiate neoplastic pancreatic cysts from those determined by the test as “definitely benign” implying progression to cancer improbable. RedPath suggests that some cases of MCN of the pancreas need not be resected based on “benign” molecular profiles that predict indolent behavior. However, international guidelines state that “unless there are contraindications for operation, all MCNs should be resected.” (Tanaka 2006) This is due to the uncertainty of the natural history of MCN and the presumed malignant potential of all types (Tanaka 2006, Khalid 2007, Oh 2008).

 

In summary, the evidence reviewed does not demonstrate that PathFinderTG® has incremental clinical value for diagnosis or prognosis of pancreatic cysts and associated cancer.

 

Limitations of pathologic examination for diagnosing and grading gliomas have been described (Aldape 2007, Coons 1997).  This is partly due to the small brain biopsy size as well as to interobserver variability between pathologists. In addition, different glial tumors may respond differently to therapy. Defining prognosis and predicting therapeutic response is therefore of considerable interest. The PathFinderTG® test is proposed to address these issues using panels of LOH markers from stereotactic brain biopsies or fine-needle aspirations to:

 

In one study cited as clinical validation for PathFinderTG®, investigators retrospectively analyzed archived tissue from 197 gliomas, performing polymerase chain reaction (PCR) for a panel of 12 to 16 LOH markers (Mohan 2004).  Loss of 1p and 19q were analyzed in all cases, with a subset (n=93) undergoing an extended panel of 16 markers. They reported, “The qualitative and quantitative aspects of glioma allelic loss were correlated with the clinical impression of treatment response [to procarbazine, CCNU, vinblastine] as determined by imaging studies.” Treatment response was noted to be classified as none, partial (25% decrease in size with stable or decreasing steroid doses), or complete (“disappearance of all contrast enhancing tissue for gliomas and resolution of abnormalities on FLAIR sequences for low-grade gliomas”). However, tables reported “major response” (undefined) rather than “complete response.” Of 96 patients with high-grade gliomas, 49 had follow-up of 6 months or longer and were assessed for therapy response: 11 had “no response,” 17 “partial,” and 21 “major” response; deaths occurring prior to 6 months were not reported. Predictive values (calculated from reported data) of at least partial response for 1p loss were positive predictive value (PPV) 100% (95% CI: 89–100%), negative predictive value (NPV) 65% (95% CI: 41% to 83%); for 19q loss PPV 82% (95% CI: 64–92%), NPV 21% (95% CI: 8–43%). It is not clear whether the molecular data results were blinded for the classification of treatment response. The retrospective design, lack of follow-up information for 47 patients, and no report of any censoring due to mortality are significant study limitations.

 

In a retrospective case series, Finkelstein et al. (2004) described the use of LOH to make the distinction between neoplasm and reactive gliosis. The study included tissue from 15 cases of reactive gliosis, none of which showed allelic loss (LOH), and 54 patients with gliomas of varying histologic type and grade. Within the group of 54 patients, 14 of 19 (74%) of the well-differentiated neoplasms showed allelic loss; 35 of 35 high-grade gliomas showed at least one allelic loss alteration and 33 of 35 showed more than one loss. Within this same study, 16 “diagnostically challenging cases of reactive gliosis vs glioma” were analyzed in the same manner, with the conclusion that LOH analysis in 8 of 11 (73%) were correctly predicted by the analysis, based upon clinical follow-up of 13–16 months. The sample size was small and the confidence intervals wide and it will be necessary to apply the criteria determined in this study to additional validation samples.

 

The only molecular test currently being performed on a relatively frequent basis for gliomas is the 1p/19q chromosomal loss for oligodendrogliomas. Yet optimal use of this test in the clinical care of patients remains unclear (Lassman 2007).  Although LOH involving multiple chromosomes has been observed in gliomas, and in some studies, LOH is predictive of tumor response to particular chemotherapeutic regimens, it is not clear the information is meaningfully incremental to that provided by the 1p/19q data alone (Thiessen 2003).  Also,some studies have shown conflicting results in terms of chemoresponsiveness for tumors with LOH involving chromosomes other than 1p/19q…” (Thiessen).   As noted by Lassman, “Current wisdom suggests that one of the main reasons for the poor responsiveness (of gliomas to therapy) is the existence of multiple genetic subsets of gliomas and the related lack of molecular stratification prior to treatment. Therefore, several strategies have been employed to divide gliomas into molecular subgroups that are either more or less responsive to existing therapies. Whether such stratification will result in effective therapies specific for each glioma subset is not yet clear…. To what extent molecular subdivision of gliomas will be clinically useful remains to be seen.”

 

The evidence reviewed does not demonstrate that PathFinderTG® testing for diagnosis or predicting response to therapy provides a benefit in patients with gliomas.

 

The evidence reviewed has significant limitations as discussed. Demonstrating utility of a test for diagnostic and prognostic purposes, or to predict therapeutic response requires that results accurately inform clinical decision-making in a manner leading to a net health benefit defined by clinical outcomes. Results must also be clearly reproducible as shown by applying the test (with a priori defined cut-offs) to independent samples for validation. The impact of this technology on health outcomes is not known and outcomes with this technology compared with existing alternatives (i.e., incremental value) are not known.