The blood count is one of the most widely performed laboratory
tests, however we define it. Years ago, a complete blood count
consisted of several manual procedures such as a red cell count and
hematocrit, and usually a peripheral blood film evaluation and
differential white count. Today, the CBC usually includes a panel of up
to nine discrete measurements. Thanks to the development of rapid,
precise, and accurate hematology instruments, labs now perform more
blood counts than ever.
Changing reimbursement incentives, however, have pressured the
medical establishment to scrutinize costs and benefits of every aspect
of patient care, blood counts included. In the current setting,
we’re unlikely to find funds for more and better testing; in fact,
we may be able to add new procedures only by deleting others. The blood
count, with its multiple components, offers potential savings without a
reduction in quality of service.
In our hematology labs, a microcomputer helps us identify and
eliminate unnecessary blood film reviews. The computer
“reads” data directly from our automated hematology analyzer,
flags specimens that need review, and cues technologists on possible
causes for the abnormality. Here’s how we developed the system.
We began with a re-examination of the basics: Why do physicians
order blood counts, and what do they hope to learn from the results? Is
a rapid normal or abnormal result sufficient when the test is used as a
screening procedure? We expect the blood count to indicate certain
basic conditions in peripheral blood cells, as shown in Figure 1. If
these expectations are reasonable, then we must determine how our
sophisticated instruments can help examine specimens most efficiently.
Optimal instrument use rests on the principle that a specimen is
“normal” it selected quantitative and qualitative parameters
lie within a prescribed set of ranges, indicating no need for additional
studies. These studies, such as the stained blood film review and
differential leukocyte count, cost considerable time, labor, and money.
By eliminating them when not clinically indicated, we can trim costs and
make better use of technologist time. The decision sapling in Figure
II–the algorithm is too simple to qualify as a decision
tree–illustrates this line of reasoning.
On that basis, we laid the scientific groundwork for our study.
Hematology instruments produce whole sets of quantitative and
qualitative data on a single blood specimen within one minute after
sampling. By qualitative data, we mean the histograms automatically
generated by all larger instrments. Quantitative data analysis used to
employ strict statistical methods, such as determining normal ranges by
the mean, plus or minus 2 standard deviations. As the inadequacies of
this simple method became clear, percentile determinations came into
We have pursued a somewhat different tack in our studies over the
last few years, by attempting to set clinically useful limits for blood
film reviews. We evaluated various arbitrary limits on quantitative CBC
data to insure that no significant findings from the automated analysis
would be missed. By comparing individual determinations to each
patient’s entire set of test results, we arrived at the flagging
limits shown in Figure III. As we’ll see, the personal computer
handles the task of flagging out-of-range results.
Two other parameters have proved quite sensitive to hematologic abnormalities. The first, the multivariate reference range, originated
in clinical chemistry as a statistical technique designed to simplity
analysis of large amounts of data, such as chemistry panels.
This method merges all eight quantitative CBC results and
determines their relative normality by the size of the resulting number.
The multivariate statistic is far more sensitive than histogram analysis
to the presence of circulating normoblasts, for instance, and has
flagged such specimens almost unerringly.
The second parameter is the histogram distance, or HD. So far we
have worked primarily with the white cell histogram because of its
greater importance. We ran 250 normal blood specimens and stored their
white cell histograms on the microcomputer. Next, the computer
determined the mean and variability of all these curves and calculated a
Chi square distance at four carefully selected points–shown in Figure
IV, using the formula: [sigma] (observed — expected).sup.2./expected
To obtain the HD, the computer compares the patient’s
histogram to the reference curve at all four points. Specimens that
differ significantly are flagged for further study.
Our system was designed to maximize the capabilities of laboratory
staff members as well as instruments. In addition to extending our
automation, it frees skilled technologists from the tedium of routine
blood film evaluations, allowing them to concentrate on those specimens
that really need attention. For this reason, we have avoided attempts
to computerize blood evaluation beyond the parameters of the CBC. Now
let’s take a closer look at how our computer puts theory into
We interfaced a multichannel automated hematology analyzer, the
Ortho ELT-8/ds with Data Handler, and an Apple IIe personal computer,
including disk drive and CRT, and equipped with a serial input/output
card. The interfacing process required some real detective work since
the complexity of the instrument’s software made access difficult.
Finally, after much research and effort, we were able to capture all
relevant data and display it on the CRT in real time. Figure V depicts
how the two systems interconnect. With expert advice, other
multichannel instruments and microcomputers can probably be interfaced
and programmed in a similar manner, although our experience is limited
to this system.
When a specimen’s index parameters on the automated analyzer
fall outside any of the predetermined limits, the computer flags the
variant value with a flashing arrow. These single and double flags
produce a footnote-like CRT display of appropriate prompts (Figure VI).
These prompts give the operator useful information for blood film
examination, and can be custom-programmed for any laboratory’s
Of course, few if any laboratory tests are 100 per cent sensitive
and 100 per cent specific. We have deliberately set the flagging limits
to point out all abnormal specimens. As a result, the computer includes
some false-positive specimens for review, but we also lower the risk of
missing a significant abnormality. Our latest review shows a
specificity of 76 per cent, with a predictive value of 86 per cent for
We tabulated the sensitivity of the individual flags–that is, how
often each flag yields an abnormal result (Figure VII). The more
complex statistics, like histogram distance, are abnormal in about half
of all specimens. Out-of-range hemoglobin, on the other hand, is
flagged in a far lower percentage of cases, but this and other less
frequent flags often provide the most direct clues to abnormalities.
In a review of 3,500 cases, we
identified clear flagging patterns associated with various clinical
conditions, also shown in Figure VII. We determined possible causes
based on observation and experience. The next phase of our study
confirmed the correlation of non-flagged ABCs with the clinical
condition on a case-by-case basis.
At this point, it’s difficult to measure how the system has
affected the laboratory’s workload. Orders for differential counts
have decreased by some 50 per cent, but some of this drop may be due to
a change in ordering protocols that allows physicians to order an
automated blood count alone, without differential. In any case, an
increasing proportion of CBCs are being ordered as screening counts
only, with the option for further review left up to the laboratory.
Our system now functions in three sections of the hematology lab
system. We were careful to introduce and develop the new method in an
evolutionary rather than a revolutionary way. Technologists have
accepted it well, once they realize that the computer helps them use
their time and skills most effectively.
Most important, our personal computer allows us to cut unnecessary
testing without compromising patient care. In light of prospective
payment, that’s a strategy with considerable implications for the