|
Abstract
It is proposed to apply statistical methods
of industrial process or quality control to the cumulating data in a
statistical monitoring center of a clinical trial. Some specific issues of the application
of these methods over calendar to patient characteristics (at a particular
individual patient time) or to more formal monitoring characteristics (like the
number of errors detected in double data entry) are discussed. Statistical
methods that are used in this context are: breakpoint regression, recursive
residuals with cusum- and V-mask techniques and Shewhart charts. In a brief way
it is discussed how desired information about quality characteristics can be
included in the trials database scheme. SAS-Macros, that allow easy data
management (e.g. importing tables via ODBC interface) and application of the
statistical methods, are presented. Some examples with graphical outputs are
given from an ongoing trial on patients with primary malignant melanoma.
Key Words
Statistical quality control, clinical trials,
quality monitoring, Shewhart charts, Cusum charts, breakpoint regression,
recursive residuals, quality characteristics, relational database
Conclusion
The development of statistical quality and
process control techniques started in the 1920s in relation with the production
industries in the United States. In the 1930s Shewhart developed the control
chart technique. Since then quality control methods have successfully been used
in many fields.
As a clinical trial is a process, too, and as
there are quality characteristics that can be measured, statistical quality
control methods can also be applied to clinical trials. Benefits of the usage
of statistical quality control in clinical trials: It is relatively cheap, it
is not personnel intensive, it can centrally be performed from the monitoring
center, it is not work intensive and it provides a fast and direct access to
information.
However there are some special
characteristics of clinical trials that have to be taken into account when
applying statistical quality control methods. Differently from the production
industry there are no production runs, which would form a natural sample basis.
There is no sampling of observation. The sample size varies from time interval
to time interval. Patient values will change over time, due to therapeutical
effects. Repeated measurements are possible, as patients may have more than one
visit in a given time interval. The process 'clinical trial' cannot be directly
corrected.
Quality characteristics that are analyzed in
clinical trials can be divided into two groups: a) quality characteristics that
base on patient data such as laboratory values, relative compliance or the
number of adverse events; b) quality characteristics as data from data
monitoring like the number of queries on CRF-data, relative frequency of
reminders or the performance of double data entry.
The statistical methods presented in this
context have been selected for the situation of clinical trials. Generally two
types of data are considered: a) measurement data which are modeled using the
normal distribution, b) event data, where the Binomial and Poisson distribution
is used. Basically all quality characteristics are monitored over calendar
time, but only values, which were taken at the same individual patient time in
the trial, are used in analysis because of potential therapeutical effects.
Event data and measurement data for Shewhart charts are aggregated over
calendar time intervals. Here all observations in a time interval have the same
time value, the interval midpoint time. All methods have been adopted for the
characteristics of clinical trials.
Breakpoint analysis is used to find out
whether in a pre-specified timepoint a change in the course of the data has
happened. The one-step-ahead forecast error is used to detect actual deviations
from the long-term trend. Here cusum charts with the V-mask technique are
applied. For measurement data the linear regression, for proportions the
logistic regression, for frequencies and rate the Poisson regression is used.
Control charts are used to view and analyze
the course of data over time. These charts are directly taken from classical
quality control. For measurements,  -, R- and s-charts are drawn. For event data, p-, c- and
u-charts are available.
Two special kinds of plots are used to
visualize special characteristics of the data over time. These are single plots
and cumulative recruitment plots.
If information for certain quality
characteristics, e.g. compliance, scheduling of examination, time intervals in
the workflow, performance of double data entry or queries on CRF-data shall be
available, corresponding entities, attributes and relationships have to be
included in the information model. The additional workload for the modification
of the information model and the software program seems to be low, as compared
to the whole effort of designing and programming the clinical trials database
and the potential advantages for quality monitoring.
The SAS-macros have been implemented for
statistical quality control in clinical trials. They allow easy usage,
facilitate data management, provide a number of general and specific methods of
statistical quality control and perform a number of data transformation and
data preparation steps.
The usage of the macros follows three steps:
first, initialization is provided, second, data management is performed, finally, statistical analysis is
done. The macros have the feature of flexible usage, that means the user is free to write his own SAS-programs and combine
then with the macros.
As far as it is possible, standard
SAS-procedures and their standard output are used for the macros that perform
the analysis. However, in general, the macros are not simply call interfaces
for standard SAS-procedures. The macros perform a large part of data management
and calculation. And in some cases, e.g. for the one-step-ahead forecast error,
nearly the whole functionality is programmed in the macro language. The
SAS-macros can be called in the SAS-language and from a graphical user interface. Various examples are given from an ongoing trial for the treatment
of Melanoma patients.
Finally it should be mentioned that the
application of statistical quality control methods in clinical trials provides
an easy, cheap and direct access to relevant information about the quality of a
clinical trial. The benefit of their usage is the retrieval of information, how
quality characteristics behave over time. It has to be kept in mind that the
additional effort of performing statistical quality control in clinical trials
is comparatively low to other activities, e.g. source data verification.
Back to "Statistical Quality Control in Clinical
Trials"
| 