FDA's Oncologic Drugs Advisory Committee agreed that retrospective studies can be used to support predictive biomarkers. The agency convened the meeting in hopes of coming away with guidance on how to assess forthcoming applications to restrict therapies to patients, using predictive biomarkers.

At the Dec. 16 meeting, members agreed that if a biomarker is supported by two well-conducted retrospective studies with similar endpoints, large patient-size, replicated outcomes, and high rates of tissue ascertainment, there is no need for additional studies to validate it.

FDA Wary Of Retrospective Studies

Retrospective analyses of clinical trials are frequently limited by deficiencies in study design elements, FDA warned. For one, they include potential for bias in selecting the biomarker and failure to adjust for multiplicity. In addition, the specimens available for analysis might not be representative of specimens from the intended use population.

Furthermore, when specimens were collected during a trial without reference to any special requirements for measuring a particular biomarker, some or all of those specimens might be analytically unfit for testing for it after the fact.

Another caveat is that early testing for the biomarker might use a device that has not yet been analytically validated. If the completed trials are old enough, the clinical and therapeutic context that existed when it was designed and executed might no longer be relevant.

According to FDA, in the ideal case for genomic biomarkers, the development of the assay methodology should be an integral part of the clinical drug development program. In addition, the clinical studies required to establish efficacy of the drug and those needed to establish the prognostic and predictive value of the genomic biomarker are measured by a well-characterized assay and occur in tandem.

However, the agency said it recognizes that there may be legitimate reasons for failure to prospectively consider the impact of genomic biomarkers early in drug development, particularly when advances in scientific knowledge occur during or after a drug's development.

Real-World Examples From ImClone and Amgen

The agency presented the committee with real-world examples in the form of Amgen and ImClone's proposed labeling changes for their colorectal cancer therapies based on retrospective analyses correlating efficacy outcomes with patients' K-ras gene status.

The sponsors are using the studies to argue that patients with wild-type K-ras tumors experience enhanced benefit when undergoing therapy with each drug. Amgen is pushing to restrict monotherapy of Vectibix to patients with the wild-type gene; ImClone proposes to include K-ras data in the Erbitux label.

Both sponsors submitted summary results in subgroups defined by retrospective K-ras testing of available tissue from randomized studies in patients with metastatic colorectal cancer. ImClone submitted four and Amgen submitted one.

FDA gave the retrospective studies a green light as long as they met the following criteria: the trial was adequate, well-conducted and well-controlled; the samples size was sufficiently large; tumor tissue was obtained in more than 95 percent of patients (see ¹ Also see "Tissue Samples Will Play Important Role In Identifying Predictive Biomarkers" - Pink Sheet, 22 December, 2008.); the assay was deemed acceptable by FDA; genetic analysis was performed by individuals masked to treatment assignment; and FDA approves an analytical plan for hypothesis testing.

ImClone's NCIC CO.17 trial showed a progression-free survival of 3.8 months in Erbitux-treated patients with wild-type K-ras compared to 1.8 months in Erbitux-treated patients with a mutated form of the gene. Its CRYSTAL trial showed a progression-free survival of 9.9 months compared to 7.6 months. Data from OPUS indicated a progression-free survival rate of 7.7 months compared to 5.5 months. Similarly, EPIC showed progression-free survival of 4.0 months compared to 2.6 months.

Amgen's Study 20020408 also showed a statistically significant increase in progression-free survival: 12.3 months for patients with wild-type K-ras treated with Vectibix versus 7.4 months for K-ras mutant patients treated with Vectibix.

Committee members did not vote on whether the studies Amgen and ImClone presented validate their request for label additions that note benefit in wild-type K-ras patients, as FDA has not yet reviewed all the data. But many appeared impressed with what they saw at the meeting. "I don't know what more you could ask for," said temporary voting member Richard Simon, National Cancer Institute. "It was a slam dunk as far as I was concerned."

Retrospective Studies OK, But Challenges Remain

The committee did not vote on anything, but rather discussed bulleted topic points provided by the agency.

In answering when it would be appropriate to limit the use of a drug to a subgroup based on retrospective analysis, overall, members agreed that acceptable retrospective studies should include similar endpoints, large patient-size, replicated outcomes, and high rates of tissue ascertainment.

Temporary voting member Ralph D'Agostino, Boston University, said that analyses should be hypothesis-driven and validation should be built in. In terms of securing statistical power and multiplicity control, he stressed the importance of having enough subjects, data and samples.

"I think that if we have two really good retrospective trials and they meet all the specified conditions, I don't think we need prospective trials," said voting member David Harrington, Dana-Farber Cancer Institute, Boston.

He also said that as a committee, they must be careful not to send a message that nothing can be learned from trials that don't meet their endpoints. "We don't want to imply that exploratory analyses are risky, when they are often helpful," he said.

A prospective study would be needed to establish treatment effects in a subgroup when results cannot be replicated. Members were most impressed by analyses that replicate results.

The possibility of false-negatives was also discussed. "We need to be very careful about the patients who appear not to benefit because of their biomarker status and be sure we aren't experiencing a false-negative," said Harrington.

The properties of clinical studies that would be unsuitable for demonstrating efficacy in a biomarker subgroup included ones with controversial biology, controversy in the analytic techniques, small population size, and too many variables.

In assessing when it is acceptable to limit future enrollment to a biomarker-selected subset of an actively accruing clinical trial based on external information, the committee agreed that situations where it makes the most sense to stop a trial include when biomarker data indicating a patient would be harmed by the drug, or when such data showed no chance of benefit.

But committee members also acknowledged that there will be gray areas. "I think the things that are much more difficult are the things where you're saying, 'Well, we think this type of patient may benefit,'" said voting member Jean Grem, University of Nebraska. "There's always going to be a lot of other reasons why a patient may or may not respond to a combination or monotherapy."

The agency also asked the panel to discuss the importance of timing and rigor in determining the analytic performance of the companion diagnostic test.

Temporary voting member Ronald Przygodzki, Department of Veterans Affairs, noted that issues may arise because of the timing of sample accrual; whether it comes from a node or a primary index tumor; and sample size.

He said some sort of standardization is going to be necessary and recommended FDA present guidance, especially to establish a firm cut-off for positivity.

Others disagreed. "I think ideally you want to have an analytically valid test used prospectively in your trial," said temporary voting member Richard Simon, National Cancer Institute. "I think very often that's not going to be the case. If you're going to [have] the mindset that we have to have the perfect test and a test that is going to last for a decade ... I think that would just bring a halt to moving to predictive oncology. I think we need more realistic standards than that."

- Lauren Smith (lauren.smith@elsevier.com)

Committee OKs Retrospective Studies, But With Caveats

More from Archive

More from Pink Sheet