Analytical Bias in a Quality Control Scheme - Clinical Chemistry

1040 ALLEN fT AL Clinical Chemistry
Methods
The presently used quality control method was instituted 2 years
ago with assistance from The Ohio State I. lilversity Statistics l)ivisioii
(2-4).
Control samples analyzed in the laboratory are of three types. There
are samples the technologist knows are controls: These are of pooled
serum processed and stored frozen. After the test has beeii standardized,
only the first daily result of the 1)001 for every constituent is
used for statistical analysis. Tilell there are coniinerical control
samples, used at the discretion of the technologist; these results are
not kept on record. Finally, there are control samples unknown to the
technologist, consisting of commercial control samples at two or three
levels of concentration. Oniy the results of the control having a concentration
approximating that of the pool have been tabulated (Tables
land 2).
Control samples, not known as controls to tile technologist, were
submitted at randoni times during the work day with a sillluiated
patient ielentilication. The time was determined by a table of ratidom
numbers. Results of the pool as well as the unknown commercial controls
were plotted daily on conlputer-prepared control charts indicating
the mean and tile permissible deviation. Values outside the prescribed
limits were brought to the attention of the supervisor. The limits used
to evaluate daily performance were usually the 2 standard deviations
calculated over a 3-month period. The number of analyses is about fit)
for each test. No changes iii methodology occurred during this -niontii
period for any one test.
Results
Results of control analysis over a period of 6 months showed differences
between sonic of tile known and unknown controls (Tal)le 1).
rflle mean values of the known calcium pool and tile blind control for
calcium, for instance,differ by about 1% while the standard deviation
differsby a factor of about 2.
1xanumation of time results of 18 tests in Table 1 shows that the
known control samples for 16 of the 18 tests had a smaller standard
deviation and 15 of tile 18 tests had a smaller coefficient of variation
than the unknown controls. The prohalility of this occurring by cilance
alone is less than 0.001 for the standard deviation and less than 0.01
for the coefficient of variation in a one-tailed test. Since the data did
not follow a parametric or normal distribution, a nonparametric statistical
analysis, that is, a binomial distribution, was employed for the
calculations.