NCI has published a 30-point checklist it uses to evaluate proposed
clinical studies that rely on genomics and proteomics to determine the patients’
The institute’s standardized, prospective approach is a direct
consequence of three ill-fated Duke University studies that relied on genomic
predictors to determine therapy for breast and lung cancer patients.
The studies were terminated after this publication reported that a
key investigator at Duke—Anil Potti—had falsified his credentials, claiming,
among other things, to have been a Rhodes scholar (The Cancer Letter,
July 16, 2010).
Subsequent reanalysis of Duke data led to retractions of papers published
in the world’s most significant medical journals and formation of a panel of
the Institute of Medicine to determine when findings generated from
genomic research can be safely applied in clinical studies. Altogether, 19
Potti papers have been retracted, partially retracted or corrected.
NCI officials have been working on the 30-item checklist, which appears on
page 3 of this publication, for over a year.
The criteria are used to evaluate proposals for institute-sponsored studies,
and the checklist will become a formal guideline after it’s published in a peer-
reviewed journal. The institute’s paper accompanying the checklist is close
to being submitted, officials said to The Cancer Letter.
“We’ve made a point of emphasis that we want most of our trials to contain
biomarkers that select specific patient populations to be in these trials,” said
Jeffrey Abrams, associate director of the Cancer Therapy Evaluation Program
and acting director, clinical research, at the NCI Division of Cancer Treatment
and Diagnosis. “We want to leave the old days where we just gave these
drugs to a broad group of patients with a certain histologic type of cancer
and then hope that the drug would work.
“The drug companies have also moved in to this approach,” Abrams said to
The Cancer Letter. “Everybody is trying to develop a drug with a companion
diagnostic if at all possible.”
The checkpoints reflect the fundamentals of omics-based studies, said Daniel
Hayes, the Stuart B. Padnos Professor of Breast Cancer Research and Clinical
Director, Breast Oncology Program at the University of Michigan Comprehensive
“This is, in my opinion, what should be ‘Tumor Markers 101’—and, frankly,
failure to adhere to these minimally rigorous SOPs is what has made the field
so chaotic and led to the Duke fiasco,” said Hayes, who served on the IOM
panel. “These in many ways reflect the roadmap laid out in the IOM report.”
Checklist Will Become NCI Policy
The NCI checklist applies to studies that rely on high-dimensional, mathematically-
modeled “omic” classifiers, as opposed to biological classifiers, which assign
patients to therapy because they may have a particular gene or a particular
mutation targeted by the investigational therapy.
“We are aiming at the complex mathematical models generated from high-dimensional
data, because that’s where we see people get into a lot of trouble,” said Lisa
McShane, a biostatistician at the Biometric Research Branch at the NCI Division of
Cancer Treatment and Diagnosis, and the first author on the paper describing the
“It is not our intent to tread on the territory of strongly biologically driven classifiers.
However, sometimes the lines can blur between those two types of classifiers. Should
the same kind of validations take place with those kinds of studies?
“Validation of the analytical performance is still important for biologically driven classifiers,
but they are not derived as computational black boxes, so different considerations may
apply in assessing the body of evidence.”
Click Here to Subscribe and read the whole story