The Health Care Financing Administration (HCFA) plans to use the Unifo
rm Clinical Data Set System (UCDSS) to collect data on hospitalized Me
dicare patients. This study examined the value of UCDSS data for creat
ing severity of illness measures. UCDSS data were obtained from a stud
y hospital and from a national data set for patients with pneumonia (n
= 528) and stroke (n = 565). Models to predict length of stay or an a
dverse event were derived for each condition using HCFA claims data al
one, UCDSS data alone, and UCDSS data supplemented with additional inf
ormation also abstracted from charts. The models were derived from one
set of patients and validated on another. The R2 for predicting lengt
h of stay in the validation data for the UCDSS model was 0.29 for pneu
monia and 0.19 for stroke compared to R2 values from the claims model
of 0.09 for stroke and 0.06 for pneumonia. UCDSS models also were bett
er than claims models for predicting adverse events. The best UCDSS mo
dels included International Classification of Diseases, Ninth Revision
, Clinical Modification (ICD-9-CM) codes and other information requiri
ng clinical judgment, and were improved by adding more information on
patient functional status. Some findings were more strongly associated
with outcome for the study hospital than for the national data. These
results suggest that UCDSS models will predict outcome much better th
an the claims based models currently used by HCFA for the analysis of
hospitalization-related mortality; more functional status information
should be added to UCDSS; and despite an extensive objective database,
the most predictive UCDSS models require clinician-assigned diagnosti
c codes.