Biology Reference
In-Depth Information
of the hospital bed for the volunteer during the study, and payments
to the volunteer. In addition, occasional experimental mishaps, such as
failure to draw blood at a specified time, invalid assay readings, or
contaminated samples, may occur. It would be unreasonable to assume
that such mishaps require discarding all previously collected data
points, especially for expensive data-collection protocols. Instead,
statistical analyses that allow for missing values in the time-series data
should be considered.
As noted, with only two or three replicates, the SEM cannot be used as
an estimate of the precision of the hormone concentrations. As a rule of
thumb, with less than 15 replicates a variance model must be created
based upon the performance characteristics of the clinical laboratory
assays. We present one such model next.
The minimal detectable concentration (MDC) is the lowest concentration
that can be measured accurately. It is experimentally calculated as
twice the SD of about 15 or 20 samples containing a hormone
concentration of zero. Clinical laboratories will commonly report
hormone levels that are less than the MDC as being too low to measure
accurately. This creates serious problems for the proper analysis of
hormone concentration time series. The algorithms require a numerical
value for the concentration, not a ''too low to measure.'' If these
''too low to measure'' values are replaced with 0.00, the analysis
procedures are forced to find that basal (nonpulsatile) secretion does not
exist. Yet, if these values are replaced with the MDC, the analysis
procedures will incorrectly find a basal secretion yielding a
concentration equal to the MDC. Also, if these values are treated as
missing values, then valuable information (that the value is ''too low to
measure'') is being neglected. Thus, the best treatment of ''too low to
measure'' values, is to force the clinical laboratories to report the actual
small value with a large experimental uncertainty.
To estimate the experimental uncertainties in such cases, a variance
model must be created that accounts for certain performance
characteristics of the clinical laboratory assays. One way to build
such a model is to describe the way the experimental uncertainties of
the assays change as the hormone concentration changes. Figure 9-7
depicts a typical variance model for hormone concentration assays
expressed in terms of the coefficient of variation (CV). The CV is defined
as the SD of the measurements divided by their mean value. A large
value for the CV would indicate that the measured value might not
represent the actual value accurately. This is more likely to happen at
very low hormone concentrations below the MDC or at high hormone
concentrations outside of the optimal range of the laboratory assay.
In Figure 9-7, the CV increases to infinity at hormone concentrations
approaching zero, decreases to a plateau in the optimal hormone
Hormone Concentration
FIGURE 9-7.
A typical variance model, the coefficient of
variation (CV), as a function of the hormone
concentration. The CV is the standard error of
the mean divided by the mean hormone
concentration.
