When a census crisis flares up at a hospital, its sources are

fairly easy to identify or guess at. Such developments as prospective

payment, insurance companies’ copayment requirements, and

ambulatory care alternatives contribute to current declines in inpatient population.

It’s harder to understand why lab workload may not drop at the

same time. That’s what puzzled administrators at our 400-bed

university medical center. During the first half of fiscal 1984, census

was off by 6 per cent–not a critical slide, but seemingly significant

enough to slow down activity in the lab. Yet we were as busy as ever.

We decided to take a close look at what affects workload.

Administrators believed our laboratory should be only as busy as the

census is high. But what about seasonal effects? What about our status

as a teaching institution with a house staff that ranges from first-year

medical school graduates to sixth-year residents? Did they

indiscriminately order laboratory tests? What were the real influences

on workload?

First, I needed an adequate measurement of workload. The CAP

workload recording method wouldn’t do for this purpose. We have

employed the method for 10 years, but it contains discrepancies that

limit its usefulness. With each year’s edition of the CAP manual,

time values assigned to a number of procedures change. The values

usually decrease, often not because of new techniques or instruments but

because of new time study data. As a consequence, a laboratory section

may appear to suffer a drop in workload even though the amount of work

is the same.

I also decided against using our records of total number of tests

performed. Ways of counting vary among sections and are subject to

change as new instruments are purchased. A prime example occurred a

year and a half ago when we acquired a new chemistry analyzer.

Electrolytes, calcium, creatinine, BUN, and glucose were combined to

form a profile that now counted as one procedure.

I finally turned to the hospital’s accounting department for

help. They supplied monthly totals of individual tests that patients are

billed for. After all, this was the bottom line: The data translated

directly into money.

Using the number of billed procedures per month as an index of our

workload and as the dependent variable, I tested how closely the figures

paralleled various census categories–by comparing total tests with

total patient days and with total admissions, for example, and pediatric tests with pediatric patient days. Census data came from monthly

medical records on patient days, admissions, visits, and discharges by

service. My statistical tools were correlation and regression analyses.

I reviewed 40 months of data covering October 1980 to February 1984

(data for October 1981) were unavailable). Figure I shows the

correlation coefficients for 11 independent census variables. Nothing

tied in very well with billed procedures. The coefficients ranged from

-0.066 for ER visits per patient to 0.406 for admissions per month. By

service, obstetrics and pediatric patient days correlated best, but

their coefficients were low–0.348 and 0.317, respectively.

Then I applied stepwise regression, combining data for several of

the variables through the hospital’s statistical computer program

(Music/Statpack, McGill University, Montreal). The highest correlation

achieved, joining nine variables, was 0.67. That was an improvement,

though still not high enough. Besides, the formula was complex and

unwieldy.

The discovery that no aspect of hospital census correlates well

with our workload in the lab was interesting and useful. For one thing,

we learned that the medical staff at our teaching hospital does not

order lab tests just because the patient is in the hospital. If that

were the case, average length of stay would correlate better with billed

procedures than 0.119.

The laboratory also could rebut administration’s claim that

census controls workload. This made my work easier.

But what does control workload? I noticed when looking at the raw

data that December was unusual. Census drops, but the number of billed

procedures increases. Why? Patients who can be discharged are sent

home to enjoy the holidays. Those who remain hospitalized are usually

very ill, and they require laboratory tests. Perhaps it was not the

number of patients but the kind of patients that determines how busy the

laboratory is. I now knew the data I wanted to look at.

Our nursing service bases its staffing on a computerized work

index. The index is derived from a system of classifying patients into

categories, depending on how much care they need. The software was

prepared by Medicus Systems (Evanston, Ill.).

There are four categories of patients: type 1 requires 0-2 nursing

hours per 24 hours; type 2, 2-4 nursing hours; type 3,4-10 hours; and

type 4, 10-24 hours. The categories cover the time spent on taking

vital signs, giving medications, assessment and development of care

plans, assessment and evaluation of plans, and certain other nursing

functions. The sicker the patient, the more nursing time that is

required. In a nursing unit or for nursing as a whole, the average

severity of illness in terms of workload is called patient acuity.

To calculate a work index, each patient category is given a

weighted factor. The factor for type 1 is 0.5; type 2, 1.0; type 3,

2.5; and type 4, 5.0. The work index is a simple total of all

patients’ weighted factors. It isn’t the number of FTEs that

will be needed in a nursing unit, but it leads to a staffing estimate.

Acuity equals the work index divided by census or patient days.

Each nursing unit does these calculations. A total work index and

average acuity for the hospital are computed monthly. I decided to

perform correlation and regression studies using the monthly nursing

work index and acuity levels as the independent variables. Nineteen

months of data were available for the study, from September 1982 to

April 1984 (data for June 1983 were unobtainable).

The results were exciting. The monthly work indexes correlated

0.796 with laboratory billings. Acuity levels correlated 0.734. Again

using the Statpak program on the hospital computer, I applied stepwise

regression with the index and acuity in combination. The adjusted

multiple regression coefficient was 0.805. Wonderful! This meant that

more than 64 per cent (0.805 squared) of the variance from perfect

correlation could be explained.

Since the nursing work index and acuity variables correlate well

with the number of billed procedures, they are much better predictors of

laboratory workload than is the patient census. Census is bound to

affect the laboratory to some extent, of course.

Figure II shows the regression model used by the computer to

compare the work index and acuity level with the number of billed lab

procedures. From this model, we get the numerical constants in the

following formula for predicting our laboratory workload: Y = 944 +

37(X.sub.1.) + 7437(X.sub.2.)

Y is the total number of laboratory procedures. The constants

correlate work index and acuity with known laboratory billings. X.sub.1

is the work index and X.sub.2 is the acuity level, supplied to us by the

nursing service. As you can see in Figure III, the equation estimates

laboratory volumes for May, June, and July 1984 that are very close to

actual billed procedures.

I believe this approach could easily be adapted to laboratories in

other institutions as well as to other departments in a hospital. Our

hospital’s administration, looking at the data in Figure III, no

understands why the lab has remained busy despite a decline in patient

census. As a result, I have been asked to apply the work index and

acuity system to the pharmacy and respiratory therapy departments. Both

departments have experienced a revenue decrease this past year without a

corresponding decrease in workload.

Work index and acuity data, along with such other yardsticks as the

CAP workload recording method, are helping us to monitor the impact of

health care changes and to manage our workload. Sometimes you have to

develop your own tools.