Biology Reference
In-Depth Information
measuring efficacy. First, it implies that some patients are biologically non-
responsive to erythropoietic therapy and that defining who will benefit (pre-
dicting response) would be an important advance in cost-effectiveness. The
available data for both rHuEPO and darbepoetin alfa suggest that the higher the
dose, until very high doses are achieved,the higher the observed response rate.
Second, response rate is a relatively insensitive measure of the erythropoietic
behavior of a cohort of patients. Increases of <2 g/dL are not captured,and
even substantial decreases in hemoglobin concentration are not distinguished
from modest increases. The need for a more sensitive measure of center for the
red cell production response to a given dose, schedule or drug has led to the
inclusion of mean change in hemoglobin concentration for the each group as
an important approach to comparison. Although this endpoint can be compli-
cated by transfusions and missing data point (see below), when properly ana-
lyzed, it is an important addition to clinical trial methodology in this area.
With the introduction of new erythropoietic agents and emerging initiatives
to improve established doses and schedules of existing agents, it is becoming
necessary to compare the hematopoietic effects observed in cohorts of
patients. Two vexing problems complicate these efforts. First, patients receiv-
ing erythropoietic agents still may require red cell transfusions, especially dur-
ing the first four weeks of erythropoietic therapy. This requirement presents
the investigator with mean hemoglobin values that are contaminated by trans-
fused red blood cells. Three approaches can be used to handle these values.
First, the measured hemoglobin values can be accepted as valid, with the high-
ly suspect assumption being that, with a sufficiently large sample size, trans-
fusions will contaminate all cohorts equally. This approach will increase the
apparent change in hemoglobin concentration attributed to the agent, especial-
ly early in therapy when transfusions are most frequent. Second, the values
from transfused patients can be censored for a period of time, and the measure
of center (mean or median) calculated for the remaining patients.
Unfortunately,this means excluding the least responsive and most anemic
patients at that time point, which will tend to artificially inflate the effective-
ness of the drug. Finally, hemoglobin values for transfused patients can be
excluded for a period of time (usually four weeks) after transfusion, with the
pre-transfusion value imputed in its place. This approach makes the conserva-
tive assumption that all of any increase in hemoglobin concentration observed
for four weeks after transfusion is due to that transfusion, and probably tends
to underestimate the hematopoietic effectiveness of the agent at that dose.
The second methodological issue has to do with the handling of patients
who drop out of study, deaths, and other causes of missing data points. Studies
of treatment of anemia during cancer chemotherapy, even when diligently exe-
cuted, typically will have 20% to 30% of the hemoglobin data unavailable
when the study is complete. The problem can be handled one of two ways.
First, the analysis can be limited to those patients who completed therapy, usu-
ally a relatively select subset of the patients initially registered on study. This
approach usually inflates the apparent effectiveness of the agent, because non-
