then the source of SSr is pure experimental error.” The proper source of SSr
is always the sum of squares due to pure
experimental
uncertainty
(pure experimental error) plus the sum of squares
due to lack of fit.
Stephens
also suggests that we have
put forth the “illusion.
. . that
the lack
of fit F-ratio
would increase
with im-
proving
precision
in the measurement
Stanley
N. Deming
nal standard
in our current
procedure
(2) for the past two years, and have not
observed
any interferences
in phenobarbital quantitation.
L. Morgan
References
Dept. of Chemistry
University
of Houston
Houston,
TX 77004
Stephen
I. Schier, G. M., and Gan, I. E. T., lnterfer.
ence of theophylline and caffeine in the gaschromatographic
estimation of phenobarbital. Clin. Chem. 25, 1191 (1979). Letter.
2. Least, C. J., Johnson, G. F., and Solomon,
H. M., Micro-scale anticonvulsant assay with
Dept. of Chemistry
Univ. of South Carolina
Columbia,
SC 29208
system.” This is no illusion. The expectation of 4e(the
denominator
in the
F-test for lack of fit) gets smaller with
use
process (see equation 3); the expectation
of s ? (the numerator
in the F-test for
lack of fit) also gets smaller with improved precision
in the measurement
process, but not as rapidly
because it
contains
the additive
term associated
with lack of fit (see equation
4). Thus,
the F-test for lack of fit must increase
with improved
precision
in the measurement
process. The only situation
in
which it will not increase
is the very
special case of absolutely
no lack of fit,
in which case the F-ratio would be expected to have a value of unity, independent of the degree of precision in the
measurement
process (3).
Conversely,
the significance
of the
lack of fit will be decreased
if a lessprecise measurement
process is used.
But there is a danger in this knowledge:
mediocre
models can be shown to have
very little lack of fit if a sufficiently
imprecise
measurement
process
is used
(4).
After having
said all this, we are
nonetheless
in agreement
with Stephens
on several points: extrapolation
outside
the region of experimentation
is always
uncertain, and the F-ratio for lack of fit
is useful for anticipating
extrapolative
discrepancies;
the power of the F-ratio
for lack of fit is strongly dependent
upon
the quantity
(and quality) of the replicates; and the F-ratio for lack of fit is
especially
important
to avoid the inad-
vertent confounding
of effects that are
not accounted for in the model when the
primary purpose of the experimentation
is to determine
model parameters.
References
1. Deming, S. N., and Morgan, S. L., The use
of linear models and matrix least squares in
clinical
chemistry.
Clin. Chem. 25, 840
(1979).
2. Mendenhall, W., Introduction
to Linear
Models and the Design and Analysis of Experiments,
Duxbury Press, Belmont, CA,
1968.
3. Neter, J., and Wasserman, W., Applied
Linear Statistical
Models.
Regression,
Analysis of Variance, and Experimental
Designs, Richard D. Irwin, Inc., Homewood,
IL, 1974.
4. Youden, W. J., Experimental
design and
ASTM committees.
Mater. Res. Stand. 1,
862 (1961).
TheophyllineInterferenceIn
PhenobarbitalQuantitatlon
To the Editor:
indicated that no patients received any
medication known to interfere with the
accuracy of the results. Incorporating
a
commercial
control, assayed for phenobarbital
by various methods,
into both
procedures
confirmed
the positive bias
in phenobarbital
quantitation
with the
on-column
methylation
technique.
Commercial
controls and several sera
patients undergoing
detector
and
(1977).
In a recent report, Schier and Gan (1)
indicated that xanthines
interfered
with
the accurate quantitation
of phenobarbital by gas-liquid
chromatography
involving on-column
methylation.
When
quantitating
phenobarbital
concentrations
by on-column
methylation (2) and simultaneous
phenobarhital and theophylline
concentrations
by
off-column
pentylation
(3), we noted
different
phenobarbital
values for all
samples containing
theophylline.
The off-column
method,
with 3-isobutyl-1-methylxanthine
as the internal
standard,
gave results
typically
50%
higher than the on-column
method,
in
which 5-ethyl-5-p-tolylbarbituric
acid
was the internal standard.
Examination
of patients’ medical files
from
of nitrogen/phosphorus
on-column methylation
compared with a
macro-scale procedure involving flame-ioni’.ationdetection. Clin. Chem. 3, 593-595
improved precision in the measurement
concurrent
theophylline
and phenobarbital
therapy
were processed
according to the on-column methylation
technique
but without
the internal
standard.
These
chromatograms
indicated
that theophylline
co-eluted
with the internal
standard,
5-ethyl-5-p-tolylbarbituric
acid.
We circumvented
this problem
by
using allyl-phenyl-barbituric
acid (Alphenal; Gane’s Chemical
Works, Inc.,
Carlstadt,
NJ 07072), previously
used
(4) as the internal
standard
in the oncolumn methylation
procedure.
Forty serum samples,
from patients
receiving
phenobarbital
but not theophylline,
were analyzed
for phenobarbital by on-column
methylation.
Comparison
of phenobarbital
values
obtained
by substituting
allyl-phenyl
barbituric
acid (y) for 5-ethyl-5-p-tolylbarbituric
acid (x) yielded
a nonweighted least-squares
regression line of
v = l.029x + 0.28 and r = 0.996.
We have used Alphenal
as the inter-
:1. Least, C. J., Johnson, G. F., and Solomon,
I-I. M., Gas-chromatographic
micro-scale
Procedure for theophylline, with use of nitrogen-sensitive
detector.
Clin. Chem. 22,
7t-iF--768 (1976).
.1. .Johnson, G. F., Dechtiaruk,
W. A., and
Solomon,
H. M., Gas-chromatographic
determination of theophyllinein human serum
and saliva. Clin. Chem. 21, 144-147 (1975).
Walter Dechtiaruk
Robert Crawford
Roger Frye
Division
of Clinical Chemistry
Therapeutic
Drug Monitoring
Laboratory
The Johns Hopkins
Hospital
Baltimore,
MD 21205
High-Density Lipoprotein
CholesterolAnalysis
To the Editor:
We read with interest the recent Scientific Note by Ash and Hentschel
on the
estimation
of high-density
lipoprotein
(HI)L)
cholesterol
(I), and were disturbed
by the implications
of some of
the procedures
adopted.
We, like Goldberg
(2), are concerned
that HDL analysis
be as precise and
accurate
as is possible, and are pleased
that Ash and Hentschel
have attempted
to improve the Lipid Research
Clinics
(LRC)
method
(3) by combining
a
higher MnClo concentration
HDL precipitation
procedure
(4) with an enzymic
determination
of cholesterol
in the supernatant
fluid. If, however,
Ash and
Hentschel
wished to compare
their results with those obtained
in earlier
studies,
they should have established
the relationship
between
the results
obtained
by their method and those by
the LRC method.
In this way, methodology may be advanced
without necessarily invalidating
clinical correlations
obtained
by older methods.
Instead of this, the authors attempted
to ensure comparability
with the LRC
method by using an unacceptable
pro-
CLINICAL CHEMISTRY, Vol. 25, No. 12, 1979
2055