Glossary of Pharma-Related Terms
Ascertainment bias:
Study centers carrying out a clinical drug trial may have a specific clientele, or
patients of high compliance may predominantly enroll in study, so that results lack
generalizability. Meta analyses combing several drug trials are then necessary for the
quantification of between-study variation.
Biological ethnicity:
Quantitative genetic similarity between individuals, or groups of individuals, determined
by a multilocus set of molecular-genetic polymorphisms which is represented as a
multidimensional point in a Genetic Vector Space. The process of quantifying genetic
similarity always gives a similarity of 0.5 between first-degree relatives and of 0.25
between second-degree relatives. More generally, for unrelated subjects, genetically
similar subjects form compact clouds ("clusters"), while genetically dissimilar subjects
are located in more distant regions, thus giving rise to the notion of "biological"
ethnicity. Response to antidepressant and antipsychotic drug treatment appears to be
similar across populations, whereas the unwanted side effects of drugs display considerable
ethnic variation.
Cross-sectional analysis:
Drug trials are designed in such a way that all patients enrolled in the study have
to be assessed at prespecified "design days" using standard rating instruments
(e.g., HAMD). Design days typically encompass baseline (day 0) and weekly assessments
over 4-6 weeks (day 7, 14, 21, 28, 35, 42). Repeated measurement analyses are carried
out to compute mean scores as a function of time across patients (thereby implicitly
defining an "average patient"). In clinical practice, however, assessments vary
by typically ±2 days around the prespecified design days. This fact cannot be
modeled in cross-sectional analyses [cf: longitudinal analysis].
Cure model, one-dimensional:
The two central aspects of psychotropic drug response, (1) the proportion of patients
in whom a therapeutic response is induced ("incidence") and (2) the time to onset of
improvement ("latency"), are often combined into a single, 1-dimensional "cure" model by
means of survival-analytical techniques. This cure model incorporates non-responders in
addition to modeling the time to an a priori defined, clinical response. Patients who
drop out of the trial prior to the envisaged observation period are treated as
"censored" data under the assumption of random censoring. The nonparametric generalized
maximum likelihood for the model is obtained from the Kaplan-Meier (KM) product limit
estimator.
Cure model, two-dimensional:
The two central aspects of psychotropic drug response, (1) the proportion of patients
in whom a therapeutic response is induced ("incidence") and (2) the time to onset of
improvement ("latency"), can be disentangled by 2-dimensional "cure" models on the
basis of survival-analytical techniques. When separating incidence and latency through
a 2-dimensional cure model, tests become available [Kolmogorov-Smirnov, Cramer-von Mises]
to explicitly compare the speed of improvement between treatments among improvers.
Double-blind design:
In clinical drug trials following a double-blind design, patients are randomly allocated to
either active substances or placebo, with patients and clinicians not knowing the substance
with which the individual patient is treated. In principle, this minimizes study bias
arising from both the clinician's and the patient's expectations.
Drug trial:
Clinical study involving either two or more active compounds (drug under investigation
plus at least one active comparator), or at least one active compound and placebo. The
study is designed in such a way that differences in efficacy can be tested statistically,
with psychopathology assessed through standard rating instruments.
Improvement:
20% baseline score reduction, irrespective of later deterioration above the
achieved improvement score.
Improvement, sustained:
20% baseline score reduction, without subsequent deterioration (a 10% fluctuation
around achieved improvement score is admissible).
Incidence:
There are two central aspects of psychotropic drug response, (1) the proportion of patients
in whom a therapeutic response is induced ("incidence") and (2) the time to onset of
improvement ("latency"). These two aspects are analyzed by means of 1-dimensional or
2-dimensional cure models.
Latency:
There are two central aspects of psychotropic drug response, (1) the proportion of patients
in whom a therapeutic response is induced ("incidence") and (2) the time to onset of
improvement ("latency"). These two aspects are analyzed by means of 1-dimensional or
2-dimensional cure models.
LOCF method:
(Last Observation Carried Forward) A considerable proportion (20-45%) of patients who
enroll in a typical clinical drug trial withdraw prior to the envisaged study period,
presenting a problem for statistical comparisons of efficacy between drugs. The LOCF
method handles the problem of missing data by "carrying forward" the last observation
of these patients throughout all remaining predefined assessment days, just as if the
patient continued taking the drug without change, thus maintaining sample sizes for
cross-sectional analyses between drugs.
However, the LOCF method introduces bias and its results can be misleading
when the rates and circumstances of premature withdrawal vary between drugs.
Longitudinal analysis:
Longitudinal data analysis involves examining the progression of data over precisely
measured units of time. In contrast, drug trials use a cross-sectional design
which lumps observations into "design days", prescribed time points which typically
encompass baseline (day 0) and weekly assessments over 4-6 weeks (day 7, 14, 21, 28,
35, 42). As assessments vary in clinical practice by typically ±2 days around
the prespecified design days, longitudinal analyses evaluate the individual
—rather than the "average"— patient by means of survival-analytical
methods at the exact calendar date of assessments, so that a substantial loss of
information is avoided.
Meta analysis: Combined
analysis of several similarly designed drug trials under the overall goal of quantifying
ascertainment biases and between-study variation. Differences between drug trials and
between treatment modalities can be assessed through random-effects (GLS, or Generalized
Least Squares) models, which aim at measuring unexplained heterogeneity in cases
where fixed-effects models seriously underestimate the standard errors for estimates
and LS-means, thus biasing inferences.
Onset of action:
Time point at which a particular drug induces a therapeutic effect in the
individual/average patient; estimated through the onset of improvement criteria
which, however, amalgamate drug-induced, placebo-induced and spontaneous change.
Onset of improvement:
Time point at which improvement criterion is met by the individual patient
subsequent to placebo run-in/washout (often used in the sense of a time span: time
to onset of improvement).
Placebo-controlled design:
Antidepressant drug trials reveal that a considerable proportion of patients (25-40%)
respond to placebo treatment. Drug-placebo differences depend essentially on severity at
baseline and hardly ever reaching significance in patients with mild depression. For
this reason, efficacy assessments in clinical drug trials typically require a placebo
control along with a minimum baseline score of, for example, 15 on the 17-item HAMD
depression scale at entry into study (after washout). Conducting a placebo-controlled
antipsychotic drug trial is very difficult, and such trials are only approved by ethics
commitees under stringent conditions.
Predictive factors:
Current knowledge about the mechanisms of action of antidepressants and antipsychotics
is rather limited. In consequence, it is currently impossible to make any predictions
of whether or not a particular patient will respond to a particular treatment
—or develop unwanted side effects. Under antidepressants, there are two known
predictive factors which apply to the "average patient": (1) severity at baseline,
with unwanted side effects outweighing the drugs' (small) beneficial effects in mild
depression, and (2) early onset of improvement, which predicts later response in 70%
of patients, whereas non-improvement within the first two weeks of treatment reduces
the probability of later response to 15%, and to 8% if there is no improvement within
the first three weeks.
Premature withdrawal:
A considerable proportion of patients who enroll in a clinical drug trial
withdraw prior to the envisaged study period for a variety of reasons, among which
"lack of effect" and "unwanted side effects" are prominent. Since the proportion
of premature withdrawals lies in the range of 20-45% in a typical drug trial,
significant biases can be introduced if subsequent data analyses do not compensate
for this problem.
Publication bias:
Negative results tend to remain unpublished, while published studies display a clear bias
in favor of the studies' principal sponsor.
Response:
50% baseline score reduction, irrespective of later deterioration above the
achieved improvement score.
Response, sustained:
50% baseline score reduction without subsequent deterioration (a 10% fluctuation
around achieved improvement score is admissible).
Severity at baseline:
The efficacy of antidepressants in the treatment of major depressive disorder (MDD)
depends essentially on the severity of depression at baseline, with drug-placebo
differences hardly ever reaching significance in mild MDD cases. For this reason,
clinical drug trials typically require a minimum baseline score of 15 on the 17-item
HAMD scale at entry into study (after washout).
Survival analysis:
Method of approach that evaluates the time to some particular event for each individual
patient, for example, the time to onset of improvement, premature withdrawal, or
response. Patients who prematurely withdraw are kept in the analysis until the time
point of drop-out and thereafter treated as "censored data". This approach avoids
biases introduced by the LOCF method of cross-sectional analyses. Caveat: the processes
underlying therapeutic effect and premature withdrawal must be statistically independent.
Time to response:
Time point at which improvement criterion is met by the individual patient
subsequent to placebo run-in/washout.
Unwanted side effects: All
classes of antidepressants and antipsychotics act indirectly by triggering recovery in
patients who otherwise would not show improvement. That is, psychotropic drugs act
unspecificly with respect to the primary target symptoms, thereby giving rise to unwanted
side effects which may compromise compliance with an otherwise beneficial medication
regimen.
Washout:
In order to assess the efficacy of an investigational compound versus one or more
established comparators and/or placebo, it is necessary to "standardize" patients
with respect to their pharmacological state. Thus, patients must (1) meet certain
inclusion/exclusion criteria to enroll in the trial (e.g., clinical diagnosis,
severity of symptoms), and (2) pass a 3-7 day "washout" or "run-in" period before
the baseline assessment (day 0) is carried out. In a subgroup of patients, recovery
already starts during washout, so that these patients may not meet the "minimum
baseline severity" criterion.