Introduction
As medical costs continue to rise alarmingly, government and other third party payors have sought ways to
control them. A fundamental requirement for the control of
inpatient hospital costs is a means of classifying patients
admitted to a hospital by a standard which accurately
reflects patients' use of health care resources. The conventional approach to this requirement has been classification of
patients into diagnosis-related groups (DRGs) such as acute
myocardial infarction (AMI), congestive heart failure (CHF)
with and without a surgical procedure, etc.'
Such conventional classification systems, which are
based primarily on diagnostic criteria, fail to take into
account the great range of severity of illness that exists
within each diagnostic category, hence they may not accurately reflect a patient's use of expensive care modalities.
This problem might be expected to be particularly troublesome when a diagnosis-based classification system is used to
compare outputs from different hospitals. For example, an
academic teaching hospital in a large city may be expected to
Methods
25
HORN
TABLE 1-Patient Severity Index
Levels
Characteristic
Moderate
Asymptomatic Manifestations
Stage of
Principal
Diagnosis
Major Manifestation
Catastrophic
None
Low
Moderate
Major
Rate
Prompt
Moderate Delay
Serious Delay
No Response
Residual
None
Minor
Moderate
Major
Complications
None or
very minor
Moderate
(less important
than principal
diagnosis)
Catastrophic
Dependency
Low
Moderate
Major
Extreme
Procedures
(Non O.R.)
Noninvasive
Diagnostic
Therapeutic
or Invasive
Nonemergency
Life Sustaining
Life Sustaining
Interactions
Response
to
Therapy
Major
Emergency
Diagnostic
Severity rating
(circle one):
ity levels which are not treated in all hospitals and also
applies a common distribution of procedure-adjusted severity for each disease condition in each hospital. In this way,
we obtain one statistic (expected average trimmed charge or
LOS) for a disease condition in a hospital which adjusts for
differences in severity distributions and permits meaningful
comparisons across hospitals. For each disease condition, a
hospital's expected average trimmed charge (or cost or LOS)
is computed as follows: For each of the four severity levels,
take the product of a particular hospital's average charge (or
cost or LOS) and the total number of such patients in all
hospitals; sum these products over the four severity levels;
then divide by the total number of patients in all hospitals in
all the severity levels in which each hospital has at least one
patient. *
If all the patients in a disease condition from all the
study hospitals are grouped together, and if those patients
from the severity index levels not represented in all hospitals
are deleted from the group, one obtains a (trimmed) reference population of patients which can be used to compare
outputs among the hospitals. The fraction of patients in each
of the four severity levels in the reference population can be
used as weights to compute a weighted average charge for a
hospital, using its own average charges for each of the four
severity levels. This weighted average is exactly the expected average trimmed charge for the disease condition for the
hospital. Each hospital's own charge structure is used with
the same reference population to give the average charge
which would be expected if they had all treated the same
severity distribution of patients.
If a hospital's charges are monotonically increasing with
increasing severity level (as is usually the case), and if its
own patient population is more severely ill than the reference population, then its overall average charges will be
higher than its expected average trimmed charges. If its own
patient population is less severely ill than the reference
population, then its overall average charges will be less than
its expected average trimmed charges.
Results
Distribution of Severity of Illness
In Table 2 we present the severity distributions for each
of our four study hospitals and for each of the six disease
conditions. For acute myocardial infarction (AMI) and cerebrovascular disease (CVD), there is no statistically significant difference in distributions of severity of illness among
the four hospitals. For the four remaining disease conditions,
we found statistically significant differences in severity distributions. For CHF, the community teaching hospital (B)
had significantly fewer severely ill patients than the other
three hospitals. For COPD, the community non-teaching
hospital (D) had a more severe distribution than the community teaching hospital (B). For GBD and PROS, the academ*For example, in Table 3, the expected average trimmed charge
for AMI patients in the academic teaching hospital is ($2028 x 13 +
$4784 x 202 + $9538 x 43)/258 = $5437.
AJPH January 1983, Vol. 73, No. 1
HORN
A
Academic
B
Community
Disease
Teaching
Teaching
C
Community
Teaching
n=
1
2
3
4
CVD n=
1
2
3
4
CHF n=
1
2
3
4
COPD n=
1
2
3
4
GBD n=
1
2
3
4
PROS n=
1
2
3
93
4%
59
22
15
91
19%
53
14
14
88
6%
64
22
8
85
47%
46
6
1
63
49%
49
2
0
45
40%
58
2
0
96
2%
71
8
19
91
21%
46
22
84
9%
64
15
12
91
25%
46
18
11
11
86
17%
67
8
6
85
54%
37
80
AMI
7
2
95
72%
26
1
1
93
75%
21
4
0
5%
74
20
1
84
39%
48
13
0
93
71%
28
1
0
96
71%
29
0
0
Community
Nonteaching
Statistical
Significance
97
7%
57
16
20
96
24%
51
20
5
98
4%
61
24
11
96
38%
51
4
7
99
73%
24
3
0
24
75%
25
0
0
x2= 13.37
(p=.15)
x2= 7.25
(p > .5)
x2=
26.45
(p=.002)
x2= 20.95
(p=.02)
x2= 17.37
(p=.05)
x2= 25.57
(p=.003)
Myocardial
Infarction
Severity
without
Procedure
Severity
with
Procedure
1
2
3
4
1
2
3
4
*2028 ( 1)
4784 (50)
9538 (14)
*5135( 1)
*1361 ( 1)
4835 (59)
8580( 7)
*2942 (2)
2445 ( 6)
4247 (50)
7570 (11)
3124 ( 5)
4824 (43)
8224 (11)
*2992 (1)
*6512 (1)
-
*8338 ( 2)
*14468 ( 3)
*21784( 1)
Combined
(p value)
2590 ( 13)
4675 (202)
8542 ( 43)
2.43 ( .15
.62( .60
Overall
Mean
Expected
Trimmed Mean
6358
5110
4635
5413
5437
5284
4710
5305
2024 (12)
4961 (27)
7862 (14)
1799 (20)
3526 (35)
6268 (11)
1458 (17)
2603 (45)
7589 (17)
1668 ( 59)
3741 (139)
7516 ( 46)
2.31 ( .18
7.98 ( .001)
.43 (>.5
2451 ( 4)
6260 ( 7)
13096 ( 4)
-
*3076 ( 3)
3982 ( 5)
*10846 ( 2)
*4128 ( 3)
*6138 ( 3)
_
3424( 15)
6362 ( 24)
.98 (>.5
4.13 ( .025)
3.01 ( .20
Cerebrovascular Disease
Severity
without
Procedure
Severity
with
Procedure
1
2
3
4
1
2
3
4
Overall
Mean
Expected
Trimmed Mean
1337 (10)
4550 (32)
9430 ( 4)
*3095 (3)
3990( 5)
7838 ( 9)
27733 ( 6)
*57982(1)
7441
5505
3673
3550
4923
4797
3626
3555
Severity
without
Procedure
Severity
with
Procedure
Overall
Mean
Expected
Trimmed Mean
1
2
3
4
800 ( 6)
2688 (53)
4431 (16)
*16560 1)
1382 (15)
3378 (54)
12544( 5)
1
2
3
4
*20391 ( 3)
*9675 ( 2)
*10174( 2)
( 4)
(56)
(11)
(4)
1260 ( 3)
2694 (53)
5980 (19)
1229 ( 28)
3007 (216)
6323( 51)
2.26 ( .20
2.63 ( .10
4.95( .01
*5314 ( 2)
6220( 4)
_
*6246 ( 3)
*7960( 2)
11120( 11)
3.39( .18
1276
3248
6839
2068
3745
3917
3844
3686
3442
4967
3773
3300
29
HORN
Table 3.-Continued
Chronic Obstructive Pulmonary
Disease
Severity
without
Procedure
Severity
with
Procedure
2
3
4
1
2
3
4
Overall
Mean
Expected
Trimmed Mean
F-Test
1149 (37)
1954 (38)
7585 ( 5)
999 (45)
3449 (30)
5139 ( 4)
*1781 ( 1)
*3322 (1)
*26133 (1)
1169 (33)
2939 (35)
6762 ( 9)
-
Combined
(p value)
1358 (32)
3131 (44)
7884 ( 4)
1153 (147)
2846 (147)
6858 ( 22)
*2472( 1)
*1781 ( 1)
5248 ( 4)
*26133 ( 1)
2.83 ( .08
6.49 ( .001)
.49 ( .50
*2472 (1)
*5889 ( 3)
1934
2442
2749
2657
1972
2427
2382
2637
1582 ( 7)
3050 ( 6)
2104 (16)
4048 ( 4)
1283 ( 8)
2313 ( 8)
1309 (10)
*3178 ( 1)
1661 ( 41)
2944 ( 18)
2.46 ( .16
1.53 ( .5
3229 (23)
6027 (22)
*11495 ( 1)
2723 (52)
7292 (21)
*11376( 1)
*15599 (1)
2688 (57)
5668 (18)
*27125 ( 1)
2805 (51)
5789 (22)
2798 (183)
6206 ( 83)
2.11 ( .20)
1.61 ( .4
Severity
without
Procedure
Severity
with
Procedure
1
2
3
4
1
2
3
4
Overall
Mean
Expected
Trimmed Mean
4199
3911
3382
3413
3766
3876
3306
3412
2323 ( 5)
4452 (12)
*1352 ( 1)
*828 ( 1)
1910 (18)
5001 ( 7)
1969 ( 4)
*2364 ( 3)
1996 ( 27)
4342 ( 22)
.08 (>.5
2.58 ( .18
2738 (13)
5061 (13)
*9119( 1)
2282 (69)
3939 (18)
4594( 4)
2024 (50)
4434 (21)
-
2308 (14)
*7162 ( 3)
-
2231 (147)
4502 ( 56)
5499( 5)
4.47 ( .01
4.39 ( .02
Prostate
Severity
without
Procedure
Severity
with
Procedure
1
2
3
4
1
2
3
4
Overall
Mean
3990
2677
2747
2865
Expected
Trimmed Mean
3359
2424
2807
3355
Discussion
We have described a new severity of illness index which
is generic and has been used to classify medical and surgical
inpatients. We have found that, when we control for severity
of illness and adjust for major operating room procedure, the
differences in total charges, total costs, and LOS across
various types of hospitals frequently disappear.
Both federal and state governments are now beginning
to use discharge abstract data to form case mix groups to
compare resource consumption across hospitals. Major
teaching hospitals are often found to have the highest
resource consumption using these techniques and the opinion has been expressed that the health care industry cannot
afford the luxury of the expensive care offered in teaching
hospitals. Our results show that when resource consumption
is adjusted for severity of illness, the wide differences among
major teaching hospitals and other community hospitals in
our study disappear or become much smaller.
If severity of illness is not taken into account when
developing prospective reimbursement policies, detrimental
effects may occur which could impact the whole medical
care system in the United States. For example, it has been
proposed that medical insurance be adjusted to cover only
the average charge for a principal diagnosis or DRG in the
least expensive hospital in a community. Under such a
policy, as our data in Table 3 illustrate, a patient who went to
the academic teaching hospital for an acute myocardial
infarction would be reimbursed $4,635 (lowest overall average charge for AMI) but would have to pay out-of-pocket the
difference, $1,723 ($6,358-$4,635), or the academic teaching
hospital would have to write off the difference or not take the
patient. However, when the distribution of severity of illness
is adjusted for by using the expected average trimmed
charge, the difference is only $727 ($5,437-$4,710), almost
$1,000 less.
With increasing scrutiny of hospital costs, it is important to be able to analyze the output of a hospital so that an
institution which provides care to severely ill patients can be
differentiated from an institution which provides care to less
severely ill patients; intrinsically higher costs for the former
institution may not be unreasonable. Attempts have been
made in the past to differentiate institutions using case mix
methods which many hospital administrators and physicians
REFERENCES
1. Fetter RB, Shin Y, Freeman JL, et al: Case mix definitions by
diagnosis-related groups. Med Care 1980; 18:1-53.
2. Horn SD, Sharkey PD, Bertram DA: Measuring severity of
illness: homogeneous case mix groups. Med Care (in press).
3. A Prospective Reimbursement System Based on Patient CaseMix for New Jersey Hospitals. 1976-1983. Trenton: New Jersey
State Dept of Health, September 30, 1980.
4. Garg, ML, Louis DZ, Gliebe WA, et al: Evaluating inpatient
costs: the staging mechanism. Med Care 1978; 16:191-201.
5. Young WW: Measuring the cost of care using generalized
patient management paths. Year I Final Report. Pittsburgh:
Blue Cross of Western Pennsylvania, November 1979.
6. Horn SD: The role of severity-adjusted case mix in hospital
management. Hospitals (in press).
7. Draper E, Wagner D, Knaus W: The use of intensive care: a
comparison of a university and community hospital. Health
Care Financing Review 1981; 3:49-64.
8. Horn SD, Chan C, Chachich B, Clopton C: Measuring severity
of illness: a reliability study. Med Care (in press).
9. Zar JH: Biostatistical Analysis. Englewood Cliffs, NJ: PrenticeHall Inc., 1974.
10. Bertram DA, Schumacher DN, Horn SD, Clopton CJ, Lord JG,
Chan C: Hospital case mix groupings and generic algorithms.
QRB 1982; 8:24-30.
ACKNOWLEDGMENTS
It is a pleasure to acknowledge Gregory B. Bulkley, MD for his
valuable comments, Phoebe D. Sharkey for computer analyses,
Kathleen O'Boyle as research assistant, and Susan Stern for technical administration of this study. We also acknowledge the help of
Richard Gross, MD, ScD, Gregory B. Bulkley, MD, George Roveti,
MD, FACP, Sharon Kreitzer, RN, Dale N. Schumacher, MD, MEd,
and Dennis Bertram, MD, MPH, ScD, in conceptualizing the
severity of illness index. This work was funded in part by the Health
Care Financing Administration, Grant 18P-97045.
ADDENDUM
The current version of the Severity of Illness Index is now being used by many institutions across
the United States. These institutions are collecting Severity of Illness data for purposes such as internal
hospital management, examination of differences in physician practice patterns, strategic planning, and
cross hospital comparisons. In California and Maryland there are plans to use the Severity of Illness
Index for prospective reimbursement purposes. Instead of using a case mix grouping system such as
DRGs, 19 diagnostic codes, or the CPHA-PAS A list to define the group into which a patient is placed
to set a reimbursement level, the Severity of Illness Index is used within groups of diagnostic codes
related to an organ system or to a service such as obstetrics, gynecology, or pediatrics; reimbursement
will be based on the severity level within that group.
AJPH January 1983, Vol. 73, No. 1
31