The Emperor’s New Gene Profile

The role of genomic testing for predicting response to chemotherapy

I read — with great interest — the recent study by Von Hoff, et al. in the Nov. 20, 2010 issue of the Journal of Clinical Oncology, as well as the associated editorial in the same issue. The manuscript, titled Pilot Study Using Molecular Profiling of Patient Tumors to Find Potential Targets and Select Treatments for the Refractory Cancers, reported the results obtained in 106 patients who consented to study using immunohistochemistry and microarray analysis for the identification of treatment process.

Of the original 106 patients:

  • There were 86 who underwent profiling and were considered for therapy
  • Of those, 68 were treated
  • Of whom, 66 received the recommended treatment.

The objective of the trial was to improve progression-free survival over that associated with the most recent prior therapy; to determine the percentage of time a target was identified; and finally, to gauge objective response rates by RECIST criteria.

The patients in the study:

  • Were a mixture of solid tumors, including breast, colon, ovary and others.
  • Had failed prior therapy.
  • The median age was 60 years and the majority of patients were female
  • Only the breast cancer patient population was defined in terms of the number of prior therapies — five.

At first blush, this paper would suggest that the era of molecular profiling has arrived. We need only obtain a small biopsy of tissue to identify the “targets” most likely to respond to available or investigational agents. At a closer look, however, we find that the investigators on the trial invented a criterion of response, namely a 1.3 fold improvement in time to progression. What that means is that patients who received an ineffective therapy and showed disease progression, need only improve upon that short response by a mere 30 percent to be counted among the “responders.”

Thus, a patient who failed a therapy after 10 days could theoretically be counted among the successes if their subsequent response to directed therapy was a meager 13 days in duration.

Even using this soft-boiled endpoint, only 18 of the 66 patients (27 percent) met criteria for response. However, these 18 responders should really be calculated against the total 88 patients approved for study, providing an even lower 20 percent result. Indeed, the most rigorous investigators would demand that these 18 be measured against the total 106 patient cohort, which would provide a response rate of a mere 16.9 percent.

Since most investigators don’t have the luxury of inventing their own criteria for response, we might examine this manuscript in the context of more widely used criteria like RECIST. In this context, the objective response rate of six out of 66 was 10 percent, with an additional 14 patients (21 percent) revealing stable disease for four months. However, again using the intention-to-treat analysis (the criteria other investigators must live by) the objective response rate falls to more like 6.8 percent (6/88) or most rigorously 5.7 percent (6/106).

Furthermore, four of the six (66 percent) objective responders, by RECIST criteria, and nine of the 18 (50 percent) were found in breast and ovarian cancers (mostly breast) known to be among the most chemo-responsive of all epithelial neoplasms. By these standards, the capacity of molecular profiling to identify responders begins to seem a bit underwhelming.

The design of the trail also raises some questions:

  • First, the principle end point is IHC (immunohistochemistry), followed by microarray
  • Yet, the specific predictive validity of the micro-array analysis is not addressed

While the authors note that IHC is a well-established and widely used methodology, they largely skirt the second issue noting only that “For MA (microarray), excellent reviews and commentary have been written on the subject of gene arrays and their potential and actual use for predicting clinical response for chemotherapy.”

In essence, we are left with a report that provides a very low objective response rate and succeeds only by meeting its own invented criteria to support the predictive validity for what appears to be mostly an established use of IHC. Should we consider this the birth of molecular profiling? By comparison, our functional platform in similarly heavily pretreated patients has consistently provided significantly higher response rates than those reported in the current analysis. Is it not time for the molecular profiles to match our results?