B. Mozes et al., CASE-MIX ADJUSTMENT USING OBJECTIVE MEASURES OF SEVERITY - THE CASE FOR LABORATORY DATA, Health services research, 28(6), 1994, pp. 689-712
Objective. We evaluate the use of routinely gathered laboratory data t
o subclassify surgical and nonsurgical major diagnostic categories int
o groups homogeneous with respect to length of stay (LOS). Data Source
s and Study Setting. The source of data is the Combined Patient Experi
ence database (COPE), created by merging data from computerized source
s at the University of California San Francisco (UCSF) Medical Center
and Stanford University Medical Center for a total sample size of 73,1
17 patient admissions. Study Design. The study is cross-sectional and
retrospective. All data were extracted from COPE consecutive admission
s, the unit of analysis is an admission. The outcome variable LOS prox
ies hospital resource utilization for an inpatient stay. Nine (candida
te) predictor variables were derived from seven lab tests (WBC, Na, K,
CO2, BUN, ALB, HCT) by recording the whole-stay minimum or maximum te
st result. Data Collection/Extraction Methods. Patient groups were for
med by first assigning to major diagnostic categories (MDCs) all 73,11
7 admissions. Each MDC was then partitioned into medical and surgical
subgroups (sub-MDCs). The 13 sub-MDCs selected for study define a stud
y population of 32,599 patients that represents approximately 45 perce
nt of inpatients. Within each of the 13 sub-MDCs, patients were random
ly assigned to one of two data sets in a ratio of 2:1. The first set w
as used to create, the second to validate, three different LOS predict
ors. Predictive accuracies of individual DRG classes were compared wit
h those of two alternative classification schemes, one formed by recur
sive partitioning (the sub-MDC) using only lab test results, the other
by partitioning with both lab test results and individual DRGs. Princ
ipal Findings. For the eight largest sub-MDCs (81 percent of study pop
ulation), individual DRGs explained 23 percent of the within sub-MDC v
ariance in LOS, laboratory data classes explained 31 percent, and clas
ses derived by considering individual DRGs and laboratory data explain
ed 37 percent. (Each result is a weighted average R2. The average numb
er of LOS classes into which the eight largest sub-MDCs were partition
ed were 20, 10, and 10, respectively. Within six of the eight, partiti
oning on the basis of laboratory data alone explained more within sub-
MDC variance than did partitioning into individual DRGs. Conclusions.
Routine test data improve the accuracy of LOS prediction over that pos
sible using DRG classes. We note that the improvements do not result f
rom overfitting the data, since the numbers of LOS classes we use to p
redict LOS are considerably fewer than the numbers of individual DRGs.