BETHESDA, MdAt the request of the Food and Drug Administration,
a special subcommittee of the Oncologic Drugs Advisory Committee
(ODAC) has begun assessing issues related to the use of
quality-of-life (QOL) endpoints in the approval process for new
We would like quality-of-life endpoints to be hypothesis
driven, and discussed prospectively in designing the protocol,
said Richard Pazdur, MD, director of the FDAs Division of
Oncology Drug Products.
At its first meeting, the subcommittee explored three key
topicsthe definition of QOL measures, the clinical significance
of interpretations of QOL data, and issues relating to QOL data
The 11-member subcommittee, three of whom are ODAC members, is
chaired by David Cella, PhD, director of the Center on Outcomes
Research and Education, Northwestern University Medical School.
It is the goal of this subcommittee to make clear and concrete
recommendations that are as specific as is reasonable, Dr.
In 1985, ODAC took the position that QOL endpoints could serve as the
basis for approving new cancer medications. FDA has interpreted that
view to mean that for drugs that do not have an impact on
survival, demonstration of a favorable effect on quality of life
would be considered more compelling than improvements in other
measures, such as objective tumor response rate.
In seeking updated guidance from ODAC, FDA noted that pharmaceutical
companies have increasingly incorporated quality-of-life endpoints
into randomized, controlled trials. Yet most oncology drugs continue
to be approved based on traditional efficacy endpoints, such as
survival or tumor response.
Indeed, ODAC members themselves often criticize the
quality of QOL data submitted in support of a drug and
seldom seem to regard them as important in deciding whether to
An issue I find particularly troubling is that it is relatively
uncommon that the quality-of-life data presented as part of the drug
application are actually hypothesis driven, said Richard L.
Schilsky, MD, ODAC chair and associate dean for clinical research,
University of Chicago Medical Center.
Trial design is often based on the usual efficacy endpoints of
survival and time to progression, and those endpoints then drive the
sample size for the study, he said. Frequently, what happens
then is that there are descriptive quality-of-life analyses added in,
often without any hypothesis being proposed by the investigator as to
what quality-of-life changes might be expected to occur, he
said. As a result, the quality-of-life analysis is grossly
underpowered because the sample size is really not adequate.
On the issue of defining QOL, the subcommittee reached consensus on
three general points:
Patients are the experts in determining the status of their quality
of life, and their input is needed in formulating the instruments
used to assess quality of life.
No single measure is now capable of assessing quality of life, nor
will one emerge.
Changes in symptoms are important but not sufficient in themselves
for assessing quality of life as the basis for approving a cancer drug.
Four major issues influence the defining of quality of life, said
Carol Moinpour, PhD, of the Division of Public Health Sciences of the
Southwest Oncology Group (SWOG): (1) Whether quality of life is an
objective construct in which the patients perception is
critical, (2) whether quality of life is just health-related or
involves a broader construct, (3) how broad must quality-of-life
impacts be and which domains are relevant, and (4) to what extent are
psychological or social theories important in explaining the impact
of therapy on quality of life.
In general, in cancer clinical trials, we have talked about
quality of life as really measuring health status, Dr. Moinpour said.
Donald Patrick, PhD, professor of public health and community
medicine, University of Washington, said that part of the problem
is the use of this umbrella term of quality of life. It is
often used as a synonym for a patients self-report.
He called patient perceptions the bedrock of quality-of-life
research in the last 5 to 10 years, but posed the question of
which patient perceptions are individual, which are universal, and
how do the two influence an accurate assessment of quality of life.
Finally, Jody L. Pelusi, PhD, cancer program coordinator, Maryvale
Hospital, Phoenix, interjected another factorthe role cultural
differences may play in affecting the concept of quality of life for
different people. This may become an issue as we recruit more
minorities and ethnic groups into clinical trials, she said.
In turning to the question of the clinical significance of QOL data,
Jeff A. Sloan, PhD, called it probably the most unnerving
question about QOL measurements. If you ask, how are you
doing, we can always say how were doing, said Dr. Sloan,
lead statistician at the Mayo Clinics cancer center. But
how do you quantify that?
Clearly, clinical significance does not equate with statistical
significance, he added. Just because you have a P value of less
than 0.05 does not mean you have a clinically significant outcome.
Three Basic Approaches
Today, researchers use three basic approaches to assess changes in
quality of life. One is searching for meaningful changes in group
comparisons, which is usually done by looking at differences between
means or medians. Thats probably the gold standard.
Whether its an acceptable gold standard is open to discussion
right now, Dr. Sloan said.
Another approach is to look at significant shifts in category scores
on an individual item, such as depression. A third approach involves
statistically adjusting group-comparison data to provide insight into
the individual items measured.
Weve been looking at clinical significance, bringing
together all of the literature, and what is both interesting and
puzzling, but very satisfying and almost comforting, is that all of
the methods tend to say the same thing, Dr. Sloan said. If a
set of scores changes by half a standard deviation, people will
say we have somethingindependent of sample size, independent of
the tool or the measure being looked at.
Stacy R. Nerenstone, MD, urged obtaining more information about the
importance of the placebo effect and investigator bias in influencing
patients perceptions of their quality of life.
Most people would agree that symptoms can be influenced by
taking a pill or being involved in a trial, said Dr.
Nerenstone, an oncologist at the Helen & Harry Gray Cancer
Center, Hartford, Connecticut. Investigator bias is
extraordinarily difficult to quantify.
She called understanding these influences very important when
you are going to have drugs that potentially could be licensed
because of their effectiveness on symptom control.
Dr. Schilsky called for greater interaction between clinical
researchers and those carrying out QOL studies. It is
critically important that as trials are designed, the two
investigators get together and think carefully about what are the
clinically important parameters that should be measured and how to
best measure them, he said.
Several subcommittee members questioned whether cancer studies whose
primary endpoints are such traditional measures as survival or
response duration can reliably assess quality of life, in part
because their sample sizes may be too small.
Dr. Pazdur said that one approach under consideration is to conduct
two studies, one focused on traditional clinical endpoints and the
other on QOL endpoints. This option is starting to evolve in
our discussions with the companies. Many quality-of-life tools are
added on without serious discussion with clinical investigators,
Analyzing data from QOL studies is difficult because the issue is
multidimensional and measurements are done over time, said Diane
Fairclough, DrPH, an invited presenter at the meeting. One important
issue is how much missing data can be allowed.
Unfortunately, there is no magic rule, said Dr.
Fairclough, of the AMC Cancer Research Center, Lakewood, Colorado.
What would be acceptable in an adjuvant breast cancer study
would be very different from what would be acceptable in a pancreatic
cancer study because of the real difference in the mortality and
morbidity associated with these cancers.
However, Nan Laird, MD, chair of biostatistics, Harvard School of
Public Health, said the literature on QOL studies suggests that
5% or less of missing data is not a bad rule.
Good QOL analysis begins with good planning of a trial. I
cant give you a nice, simple formula for handling an analysis
of quality of life, Dr. Fairclough said. But I can say
this. Careful planning in the design phase is critical.