International Journal for Quality in Health Care 1999; Volume 11, Number 5: pp. 375–384
Difference between observed and
predicted length of stay as an indicator of
inpatient care inefficiency
ROSA JIMÉNEZ1, LIBIA LÓPEZ2, DANIEL DOMINGUEZ2 AND HUMBERTO FARIÑAS1
1
Research Section, Hermanos Ameijeiras Hospital, and 2Escuela de Salud Publica, Ministry of Health, Havana City, Cuba
Abstract
Objectives. To evaluate the performance of the difference between observed and predicted length of stay (OLOS–PLOS)
as an inefficiency of care indicator for inpatients.
Setting. The Internal Medicine and the General Surgery departments of Hermanos Ameijeiras Hospital in Havana.
Design and study participants. Two sets of clinical histories were needed for each department: one for deriving the predictive
equation and another to validate it. The equation was a linear multiple regression model which included variables recognized
as affecting length of stay. The validation group of histories was thoroughly examined and separated into two groups: (i)
adequate efficiency or mild problems and (ii) inefficiencies considered to be moderate or severe. This classification was the
gold standard to obtain a receiver operating characteristic (ROC) curve for the indicator.
Results. The function explained 41% of the total variation for Internal Medicine and 70% for General Surgery. The
indicator’s mean difference between the two validation groups of histories was around 10 days for both departments. The
areas under the ROC curve were 0.80 for Internal Medicine and 0.88 for General Surgery. Sensitivity and specificity > 0.7
for detecting inefficiencies of care are achieved with a cut off point of 2 days for Internal Medicine and 1 day for General
Surgery.
Conclusions. The use of predictive equations might be quite useful for detecting efficiency problems in inpatient health
care.
Keywords: access and evaluation, health care quality, length of stay, process assessment
Rigorous control of the time between a patients’ admission
to hospital and their discharge is an imperative of today’s
health systems, as hospital length of stay (LOS) is the major
contributor to health costs from institutional medical care
[1].
Mean LOS is the simplest indicator of stay length in a
hospital, and should measure both bed use and efficiency of
inpatient health departments [2]. In Cuba, mean LOS is
calculated for each department and compared with fixed
LOS norms. However, if the variations due to the patients’
characteristics are not considered in the norms, one runs the
risk of having an erroneous perception of hospital care
efficiency or, what is worse, physicians can be induced to
discharge patients sooner than necessary.
The literature reports several efforts to establish proper
LOS norms, efforts that were made, in general, for reasons
related to the need to control the costs of inpatient care.
Furthermore, efficiency in health care is a necessity that goes
beyond cost containment; it has been demonstrated that
hospitals with higher levels of morbidity and mortality are
also the ones with lower efficiency levels [3] a fact that
supports the assumption that efficient health care is also high
quality care.
Currently, the health services literature is discussing what
are called ‘attention protocols’ in which an attempt is made to
regulate patient management, including the moment ancillary
exams are indicated, the treatment provided and discharge
ordered [4–6]. However, there is plenty of evidence about
Address correspondence to Rosa Jiménez, Sección de Investigaciones, Hospital Hermanos Ameijeiras San Lázaro 701, Centro
Habana, Ciudad de La Habana (Havana City), Cuba. Tel+53 7 576020. Fax:+53 7 335036. E-mail: [email protected]
 1999 International Society for Quality in Health Care and Oxford University Press
375
R. Jiménez et al.
factors that affect LOS. Even recently, many papers continue
to analyse the association between patients’ characteristics
(including their socio-economic and cultural environment)
and the hospital LOS [7–12].
This means that health administrators must face a big
contradiction: how to detect inefficient care when there is
no simple way of knowing how long a patient should remain
in the hospital. Furthermore, even if the proper LOS could
be approximately known it would have to be adjusted to be
appropriate to the local setting because of the huge differences
of care among hospitals, regions and countries.
This led to the idea of using the difference between
predicted LOS and observed LOS as an indicator of care
efficiency. Predicted LOS could be calculated by means of a
statistical tool such as multiple linear regression; however,
development of such a tool has required a long series of
studies. The first studies were meant to evaluate the influence
of different variables, reported in the international literature
and previously mentioned, upon LOS in internal medicine
and surgical wards [13,14]. Results, using multiple linear
regression, indicated a significant effect on LOS for (among
others) severity of disease, place of residence, diagnosis, type
of surgical intervention, surgical complications and the need
for surgical re-intervention. These studies corroborated the
fact that the variables included, despite not being the only
ones that affect LOS, explain an important proportion of it
and are readily available, directly or indirectly, in the patient’s
clinical history. Severity of disease was one of the most
important variables for predicting LOS. The former studies
used an Index of Severity proposed and validated by Horn
[15,16] but, before the new method could be implemented,
a local severity index had to be developed. Two studies were
carried out with this aim [17,18]. Shortly after this, the
performance of the new indicator was evaluated, first in the
Internal Medicine department (as a representative of the
clinical area) and afterwards in the General Surgery department (as a representative of the surgical departments).
The results of these latter evaluations are reported here.
Methods
Data collection
The necessary information was obtained from clinical histories
of patients discharged alive from four Internal Medicine
wards and from all wards of the General Surgery department
of Hermanos Ameijeiras Hospital in Havana. This is a 900bed facility for secondary and tertiary levels of care with all
clinical and surgical specialities for adults.
Two data sets were needed for each department: one for
deriving the predictive equation and another to validate it.
The Internal Medicine wards selected were those without
preference for any particular diagnosis. The histories of 200
patients discharged consecutively from January 1 1997 were
included, 100 hundred for each data set.
In the General Surgery department, in order to attain a
fair representation of all types of surgery procedures, patients
376
were grouped into categories according to a classification
used in a previous study [18] (see Appendix 2 for summary).
Included were 80 consecutive patients in each category; this
produced 466 clinical records of patients discharged since
May 1996, 263 for the derivation group and 203 for the
validation group. Not to do this would have resulted in the
inclusion of a large proportion of patients from the more
frequently performed procedures and very few or none at all
from other less frequent ones.
In both departments patients with any of the following
characteristics were excluded: foreigners, those leaving hospital against medical advice or to another institution, with
principal diagnosis undetermined, not operated (for surgical
wards) and died before discharge. Less than 5% of all
admissions in the period were excluded.
Variables, procedures and statistical analysis
As for two previous studies [13,14], information about the
following variables was obtained from each clinical history.
For Internal Medicine the variables were principal diagnosis,
age, place of residence, and disease severity. Principal diagnosis
was grouped into the following seven categories: (i) cardiovascular system diseases; (ii) digestive system diseases; (iii)
respiratory system diseases; (iv) malignant tumours; (v) urinary
and genital system diseases; (vi) benign tumours, metabolic
and endocrinological diseases, anaemia, skin and subcutaneous cellular tissue diseases, bone and connective tissue
diseases, infectious diseases; and (vii) other diseases. Place of
residence was grouped into two categories: (i) city of Havana
and (ii) rest of the country. Disease severity was measured
by means of a severity index constructed and validated in a
previous study [17]. The index is a weighted sum of points
given by the eight variables that constitute the items (see
Appendix 1).
For General Surgery the variables were age, sex, surgical
intervention, nature of the intervention, complications score,
ward, and disease severity. Surgical intervention was grouped
into five categories according to complexity as proposed by
an expert and it was treated as a nominal variable. The
categories are detailed elsewhere [18]; Appendix 2 shows an
example of each category. The nature of the intervention was
noted as elective or urgent. The complications score was
constructed by assigning to each complication 1–3 points
according to its severity (as defined by an expert) and then
summing all points. Points were assigned as follows. For
surgical wound complications: subcutaneous emphysema
(1 point), haematoma, bleeding (2 points), sepsis, dehiscence
(3 points). For general complications: retention of urine,
phlebitis, headache, cutaneous rash, cough and expectoration,
diminution of haemoglobin, paralytic ileus, emesis and/or
nausea (1 point), urinary tract infection, fever, abdominal
distention, acute laryngitis, diarrhoea (2 points), pneumonia,
jaundice, intestinal occlusion, empyema (3 points).
The ward was considered as nominal: 16A, 16B, 17A, 17B
(wards 1–4, respectively). Place of residence was considered
in three categories (and was treated as ordinal): region directly
served by the hospital (Centre of Havana and Old Havana),
Indicator of care inefficiencies
other municipalities in Havana City, and other provinces. This
variable was categorized differently in the two departments
because of the differences in the way it affects LOS. Presumably surgeons pay more attention than clinicians to the
place of residence of the patient according to the distance
from the hospital. Severity of disease was measured with an
index, described in a previous study [18] (see Appendix 1).
Observed length of stay (OLOS) was calculated for Internal
Medicine as the difference measured in days between admission date and discharge date and for Surgery between
operation date and discharge date.
With the derivation group of clinical histories, the optimum
function for explaining the relationship between LOS and the
described variables was obtained. A linear multiple regression
equation with LOS as the response variable stood for the
model.
Steps derived the final function. To avoid multi-colinearity,
linear correlation between all pairs of quantitative variables
was calculated. Association between qualitative variables was
assessed with the Phi coefficient from the contingency table.
This exploration led to the elimination of age from the
function for the General Surgery department because of its
high correlation with the severity index. A second step
involved residual analysis, which allowed exploration of the
normality assumption and the necessity of quadratic or interaction terms. In the Internal Medicine department no deviations from normality were detected. For the General
Surgery department the residual analysis showed the need
for the inclusion of a quadratic term for the complications
score. Twenty outliers from Internal Medicine and 16 from
General Surgery were excluded from the derivation of the
predictive function. This exclusion, although large, should
help to guarantee the future performance of the equation for
detecting inefficiencies. Outliers are likely to be patients that
are inefficiently (or at least differently) treated. The final
function yielded a Determination Coefficient (R2) of 0.41 for
Internal Medicine and 0.70 for the Surgery wards.
In the validation group, each history was thoroughly examined
for inefficiencies. Emphasis was placed on the following events:
(i) time elapsed between indication and performance of diagnostic procedures; (ii) time elapsed between indication and
performance of therapeutic procedures (e.g. blood transfusions,
removal of catheter); (iii) time elapsed between indication and
consultations with other specialities; (iv) weekend leave or other
absences; (v) time elapsed between admission and diagnostic
discussion (for Internal Medicine only).
The moment (or date) of the indication and the performance of a procedure is written down in the progress
notes by the physician and (on some occasions) the nurse.
The model used to confirm the indications for the procedure,
which usually refers back to the history, also shows both
dates. If another specialist is called for consultation, the
physician providing the care notes the first date and the
consultant physician himself, the date the procedure is performed.
Histories were separated then into two groups: (i) adequate
efficiency or mild problems, and (ii) inefficiencies considered
moderate or severe. A history was classified as group (ii) if
Table 1 Frequency of inefficiency problems found in clinical
histories of the validation group1
Problems
Internal General
Medicine Surgery
............................................................................................................
More than 2 days between indication 20
8
and performance of diagnostic
procedures
More than 2 days between indication
2
6
and performance of therapeutic
procedures
More than 2 days between indication
6
5
and consultation with other
specialities
Weekend leave or other absences
5
7
interfering with diagnosis or treatment
More than 2 days between admission
3
–
and diagnostic discussion
Other delays
–
5
1
Each history may have more than one inefficiency problem.
time considered in points i–v above was more than 2 days
and if weekend leave or other absences (if any) interfered
with diagnostic or therapeutic indications and thus caused an
increase in LOS. Otherwise the history was classified as group
i. Table 1 shows the frequency of the inefficiency problems
encountered. Classification was made by two of the authors
who separately examined each history and discussed discrepancies until they reached agreement.
The predicted LOS (PLOS) for each patient was obtained
according to the linear regression function obtained with the
derivation group of histories. The difference: OLOS – PLOS
(indicator) was obtained for each patient. The reviewers did
not know the PLOS of any patient nor the function used
for its calculation.
The mean and SD of the indicator were calculated for
each group of clinical histories (classified according to type
of care) and the 95% confidence interval calculated for the
difference. A receiver operating characteristic (ROC) curve
was constructed to evaluate the ability of the new indicator
to detect care inefficiencies using the expert classification as
the gold standard. Three histories were eliminated from the
validation group of the Internal Medicine department because
of excluding characteristics missed in the first review.
Results
Tables 2 and 3 show the patients’ characteristics in both
groups and both departments; groups appear to be very
similar in almost all features. Regression coefficients for the
best functions in each department are shown in Tables 4 and
5. The function explains 41% of the total variation for the
Internal Medicine department and 70% for the General
Surgery department. Table 6 shows means and SD of the
377
R. Jiménez et al.
Table 2 Characteristics of the groups included: Internal Medicine department
Derivation group (n=100)
Validation group (n=97)
............................................................ .........................................................
Frequency (%) Mean SD
Frequency (%) Mean SD
.............................................................................................................................................................................................................................
Age
55.60
17.43
53.79
18.18
Severity index
70.28
28.06
74.53
27.02
Length of stay (days)
12.01
5.63
11.54
7.03
Diagnostic groups
Cardiovascular system diseases
33
30.9
Digestive system diseases
15
16.5
Respiratory system diseases
22
16.5
Malignant tumours
3
3.1
Urinary and genital system diseases
3
2.1
Benign tumours, metabolic and endocrine
21
21.6
diseases, anaemia, skin and subcutaneous diseases,
bone and connective tissue diseases, infections
Other
3
9.3
Place of residence
Havana City
76
69.1
Rest of the country
24
30.9
Table 3 Characteristics of the groups included: General Surgery department
Derivation group (n=262)
Validation group (n=203)
.............................................................. ..............................................................
Frequency (%) Mean SD
Frequency (%) Mean SD
.............................................................................................................................................................................................................................
Age
Severity index
Length of stay (days)
Intervention category1
1
2
3
4
5
Place of residence
Centre of Havana and Old Havana
Other municipalities of Havana City
Rest of the country
Sex
Male
Female
Complications index
0 points
1–3 points
4 or more points
51.13
17.45
4.58
16.28
6.13
4.95
49.37
17.72
6.68
15.3
11.8
19.1
36.6
17.2
19.7
20.7
20.2
19.2
20.2
33.2
47.3
19.5
26.6
48.3
25.1
32.4
67.6
37.4
62.6
89.3
8.4
2.3
79.3
14.8
5.9
17.00
6.57
9.25
1
See Appendix 2.
indicator proposed, according to type of attention. There is
a large difference between the two groups. The group in
which inefficiency problems occurred exhibits, in both departments, the highest value for the indicator.
378
Figure 1 shows the ROC curves for the two departments.
For the Internal Medicine department a value of 2 as a cutoff point for the indicator would achieve a sensitivity and
specificity of > 0.7 (Table 7). For the General Surgery
Indicator of care inefficiencies
Table 4 Best function for predicting length of stay in Internal Medicine department
Estimated
Standard
95% Confidence
Variable
coefficient
error
interval1
P
.............................................................................................................................................................................................................................
2
0.087
0.017
0.05, 0.12
0.000
Severity index
Age
−0.117
0.026
−0.17, −0.06
0.000
Place of residence3
−0.722
0.921
−2.56, 1.11
0.436
1.406
2.129
−2.84, 5.65
0.511
Diagnosis 14
Diagnosis 2
3.276
2.220
−1.15, 7.70
0.145
Diagnosis 3
1.506
2.190
−2.86, 5.87
0.494
Diagnosis 4
−1.852
3.231
−8.30, 4.59
0.568
Diagnosis 5
6.004
3.207
−0.39, 12.4
0.065
Diagnosis 6
5.031
2.214
0.61, 9.45
0.026
Constant
9.915
2.663
4.60, 15.23
0.000
R2 =0.412
1
For population coefficients.
See Appendix 1.
3
1, City of Havana; 2, Rest of the country.
4
Dummy variables for principal diagnosis group. The seventh diagnosis group is the reference.
2
Table 5 Best function for predicting length of stay in General Surgery department
Estimated
Standard
95% Confidence
Variable
coefficient
error
interval1
P
.............................................................................................................................................................................................................................
0.044
0.021
0.004, 0.08
0.0376
Severity index2
Intervention 13
5.163
0.381
4.41, 5.91
0.0000
Intervention 2
4.828
0.378
4.08, 5.57
0.0000
Intervention 3
1.785
0.332
1.13, 2.44
0.0000
Intervention 4
0.407
0.294
−0.17, 0.99
0.1680
−0.842
0.271
−1.38, −0.31
0.0022
Ward 14
Ward 2
0.287
0.218
−0.34, 0.91
0.1897
Ward 3
0.200
0.302
−0.40, 0.80
0.5087
0.245
0.147
−0.44, 0.53
0.0974
Place of residence5
Complications index6
0.929
0.284
0.37, 1.49
0.0013
Squared complications
−0.049
0.057
−0.16, 0.06
0.3912
Index
−0.858
0.298
−1.44, −0.27
0.0045
Type of surgery7
Constant
1.581
0.595
0.41, 2.752
0.0086
R2=0.698
1
For population coefficients.
See Appendix 1.
3
Dummy variables for intervention group. Group 5 is the reference.
4
Dummy variables for ward. Ward 4 is the reference.
5
Center of Havana and Old Havana: 1, Other municipalities in Havana City: 2, Other provinces: 3.
6
Sum of points provided each complication is assigned 1 to 3 points according to severity. See text.
7
Elective: 0, Urgent: 1.
2
department a similar situation would be achieved with a cutoff point of 1 (Table 8). A cut-off point of 4 for Internal
Medicine would achieve 0.65 for sensitivity and 94.8 for
specificity. For General Surgery, this cut-off point achieves
0.65 for sensitivity and 94.9 for specificity. The area under
the ROC curve was 0.88 for Internal Medicine and 0.80 for
General Surgery. Both values can be considered acceptable.
The low frequency of inefficiency problems is the reason for
low positive predictive values (+PV) for the cut-off points
mentioned above. However if we focus on +PV, 4 as the
379
R. Jiménez et al.
Table 6 Mean and SD for the indicator in each service and
attention group
Table 7 Sensitivity, specificity and predictive values (PV) for
different cut-off points of the indicator1: Internal Medicine
Number of cases Mean
SD
............................................................................................................
Internal Medicine
Adequate care
77 (79%)
−2.19
4.36
Inefficient care 20 (21%)
8.17
8.88
95% CI for the mean difference:
(6.35, 14.37)
General Surgery
Adequate care
177 (87%)
0.43
5.11
Inefficient care
26 (13%)
11.51
13.88
95% CI for the mean difference:
(4.53, 17.61)
Cut-off
Sensitivity Specificity +PV
−PV
point
............................................................................................................
−9
100.0
5.2
21.5
100.0
−7
100.0
11.7
22.7
100.0
−5
100.0
23.4
25.3
100.0
−3
100.0
44.2
31.7
100.0
0
80.0
72.7
43.2
94.1
2
70.0
84.4
53.8
91.5
4
65.0
94.8
76.5
91.2
6
50.0
94.8
71.4
88.0
8
35.0
98.7
87.5
85.4
12
35.0
100.0
100.0
85.6
14
30.0
100.0
100.0
84.6
16
20.0
100.0
100.0
82.8
20
15.0
100.0
100.0
81.9
22
10.0
100.0
100.0
81.1
24
5.0
100.0
100.0
80.2
>24
0.0
100.0
100.0
79.4
1
Expressed in percentages.
Table 8 Sensitivity, specificity and predictive values for different cut-off points of the indicator1: General Surgery
Cut-off
Sensitivity Specificity +PV
−PV
point
............................................................................................................
−5
−3
−2
−1
0
1
2
3
4
5
6
7
9
11
13
15
20
25
30
40
50
>50
Figure 1 Receiver operating characteristic curves for
(a) Internal Medicine and (b) General Surgery.
380
1
100.0
100.0
96.2
92.3
92.3
76.9
69.2
65.4
65.4
50.0
38.5
38.5
34.6
34.6
30.8
26.9
26.9
23.1
19.2
3.8
0.0
0.0
Expressed in percentages.
0.6
6.8
9.0
28.8
56.5
80.2
88.7
93.2
94.9
96.6
97.2
97.2
97.7
98.9
98.9
98.9
98.9
98.9
98.9
99.4
99.4
100.0
12.9
13.6
13.4
16.0
23.8
36.4
47.4
58.6
65.4
68.4
66.7
66.7
69.2
81.8
80.0
77.8
77.8
75.0
71.4
50.0
0
0
100.0
100.0
94.1
96.2
98.0
95.9
95.2
94.8
94.9
92.9
91.5
91.5
91.1
91.1
90.7
90.2
90.2
89.7
89.3
87.6
87.1
87.2
Indicator of care inefficiencies
cut-off would achieve a +PV of 0.76 for Internal Medicine
and 0.65 for General Surgery. Negative predictive values
(–PV) are in a better position: 4 as cut-off results in a –PV
of 0.91 for Internal Medicine and 94.9 for General Surgery.
Discussion
In the last few years we have witnessed a growing interest
in assessing the efficiency of medical care – particularly for
hospitalized patients. The current trend to design ‘attention
protocols’ or guides that strictly define ways of handling
patients with different diseases might imply a number of
regulations that change usual medical practices and therefore
introduce reluctance among practitioners. The procedure we
propose and validate in the present work can be adapted to
local circumstances and practices and does not include extra
obligations or specific restrictions.
Our results support the hypothesis that differences between
observed and predicted LOS reflect, with high probability,
inefficiencies in health care for hospitalized patients.
The literature reports some other similar approaches. Bernard et al. [4] and Ryan et al. [19] propose the use of the
difference between PLOS and OLOS for similar purposes
but in both cases, the PLOS is the historical LOS mean for
each diagnosis-related group. Best et al. [20] proposed a
quality indicator for health care in non-surgical patients which
is based on the ratio of observed to expected mortality, the
latter obtained through logistic regression. Rosenthal and
Harper [21] report some health care quality indicators that
measure quality and efficiency. They developed predictor
models using logistic regression (for binary events) or linear
regression (for LOS). The discriminatory power of the models
was measured with ROC curves and the coefficient of
determination (R2).
Hartz et al. [22] evaluated three methods for the detection
of unnecessary hospital days, one of them very similar to
ours. The best of the three methods was a clinical algorithm,
which entailed the review of each record by an expert, who
must look for unnecessary hospitalization days. They also
found that a method that does not depend on patients’
characteristics (a comparison with a fixed LOS) was as good
as an equation-based screen. We have two comments on
their methods: a clinical algorithm that requires the review
of all charts is surely the best way to detect excess stay but
is inefficient; and the variables included in the equation that
predicts LOS must not be limited to those available at
admission. The authors also note this latter comment.
Our study has surely some limitations, principally in relation
to external validity: the information needed might not be
readily available in clinical histories from other hospitals; the
complications score and the intervention grouping used for
the surgical areas have not been specifically validated. Applications of our results must be restricted to the method;
individual hospitals, even in Cuba, must construct and validate
their own models.
Another comment on our method, which can also be seen
as a limitation, is that it implies that the hospital is setting
its own standards. This could be a real problem if inefficiency
was the regular way of care and if there were no other control
mechanisms. This is not the case, but it must be taken into
account if an attempt to use such a procedure is made in
another setting. Another type of limitation is that some of
the variables included in the model are process variables
(surgical intervention or need for reintervention). These variables are included as proxies for patients’ characteristics but
they depend on a medical decision.
Nevertheless, we support the idea that the way to assess
and control efficiency of care for inpatients must rely on
individual comparisons between real and ideal stay. This ideal
stay cannot be established theoretically because it depends
on several aspects that concern the patient, the hospital and
the whole environment. Predictive equations, which include
the greatest number of aspects available, represent a valuable
tool for approaching ideal comparisons.
Some advantages can be mentioned for this method: (i)
once the mechanism to collect the necessary information is
set up, the evaluation system should work smoothly; (ii) the
predictive equations can be readjusted periodically incorporating changes expected to occur in medical practices;
and (iii) the calculation of summary measures such as means
and SDs of the proposed indicator (OLOS – PLOS) can also
be carried out for each ward, department or hospital.
This is not to say that utilization review should be entirely
replaced by ‘predictive equations’ but the method we suggest
could be used, like other quantitative indicators, as a shortcut
for detecting problems. Implementation of the indicator is
another issue, but this should not be difficult if the physician
providing the care completes the history, as it is done in
Cuba.
Acknowledgements
The authors thank T. Piazza (Berkeley University) for reading
the early versions of the manuscript, correcting the English
and making many useful comments. J. Nieto (Johns Hopkins
School of Public Health) reviewed the final versions and
made many valuable suggestions.
References
1. Mozes B. Unnecessary hospitalization days. Isr J Med 1989; 25:
360–361.
2. Rios N. Health Indicators Used More Frequently in Health Service
Administration (in Spanish). Habana, Cuba: Facultad de Salud
Pública, Instituto Superior de Ciencias Médicas de La Habana,
1987.
3. Bradbury RC, Golec JH, Steen PM. Relating hospital health
outcomes and resource expenditures. Inquiry 1994; 31: 56–65.
4. Bernard AM, Mynard RAJE, Rosevear JS, McMahon LF. The
integrated inpatient management. Med Care 1995; 7: 663–675.
5. Weingarten S, Riedinger M, Johnson B, Ellrodt AG. Reducing
LOS in the coronary care unit with a practice guideline for
381
R. Jiménez et al.
patients with congestive heart failure. Med Care 1994; 12: 1232–
1243.
characteristics in a General Surgery Unit (in Spanish). Gaceta
Sanitaria 1994; 8: 180–188.
6. Odderson IR, McKena BS. A model for management of patients
with stroke during the acute phase. Stroke 1993; 12: 1823–1827.
15. Horn SD, Sharkey PD, Bertram DA. Measuring severity of
illness: homogeneous case-mix groups. Med Care 1983; 21:
14–30.
7. Lutjens LRJ. Determinants of hospital length of stay. J Nurs
Adm 1993; 23: 14–18.
16. Horn SD, Horn RA. Reliability and validity of the Severity of
Illness Index. Med Care 1986; 24: 159–178.
8. Cleary PD, Greenfield S, Mulley AG et al. Variations in LOS
and outcomes for six medical and surgical conditions in Massachusetts and California. J Am Med Assoc 1991; 266: 73–79.
17. Jimenez RE, Vazquez J, Fariñas H. Construction and validation
of a Severity of Disease Index for inpatients in clinical areas
(in Spanish). Gaceta Sanitaria 1997; 11: 122–130.
9. Horn SD, Sharkey PD, Buckle JM et al. The relationship between
severity of illness and hospital LOS and mortality. Med Care
1991; 29: 305–317.
18. Jimenez RE, Dominguez E, Fariñas H, Fuentes E. Construction
and validation of a Severity of Disease Index for inpatients in
surgical areas (in Spanish). Rev Cub Salud Publica, 1999; 24: (in
press).
10. Morgan M, Beech R. Variations in LOS and rates of day case
surgery: implications for the efficiency of surgical management.
J Epidemiol 1990; 44: 90–105.
11. Burns LR, Wholey DR. The effects of patient, hospital and
physician characteristics on LOS and mortality. Med Care 1991;
29: 251–271.
19. Ryan S, Lowe CS, Hardes G. A quantitative approach to quality
improvement and resource allocation. Qual Clin Prac 1995; 15:
11–16.
20. Best WR, Cowper DC. The ratio of observed-to-expected
mortality as a quality of care indicator in nonsurgical VA patients.
Med Care 1994; 4: 390–400.
12. Rhodes RS, Sharkey DR, Horn SH. Effect of patient factors
on hospital costs for major bowel surgery: implications for
managed health care. Surgery 1995; 117: 443–450.
21. Rosenthal GE, Harper DL. Cleveland health quality choice. A
model for collaborative community-based outcomes assessment.
J Qual Improve 1994; 8: 425–442.
13. Hernandez C. Factors affecting LOS in an Internal Medicine
Department (in Spanish). Thesis for obtaining the degree of
First Grade Specialist in Biostatistics, Public Health Faculty,
Institute of Medical Sciences of Havana, Havana, Cuba, 1990.
22. Hartz AJ, Bade PF, Sigmann P et al. The evaluation of screening
methods to identify medically unnecessary hospital stay for
patients with pneumonia. Int J Qual Health Care 1996; 8: 3–11.
14. Jimenez RE, Gutierrez AR, Fariñas H, Suarez N, Fuentes E.
Variation in length of postoperative stay related to patient
382
Accepted for publication 11 June 1999
Indicator of care inefficiencies
Appendix 1. Summary of the
development and validation of a severity
of illness score for hospitalized patients in
clinical and surgical areas
The purpose of both studies was to develop and validate a
semi-quantitative variable (a linear combination of items) that
could measure severity of illness during a hospitalization
period and would be easy to obtain from the information
contained in a regular clinical chart.
Construction of the variable entailed item selection and
search for item weights. Experts and literature were consulted
and clinical records provided empirical information (74 records for clinical areas and 174 for surgical ones). In both
cases, the result was the proposal of an index with two
alternatives: one quantitative and the other ordinal with three
levels of severity (only the quantitative one was used in this
study). Stepwise regression procedures were used for the final
versions. Validation included four aspects of validity (face,
content, construct, and criterion), general reliability, inter-rater
agreement and internal consistency. Results show satisfactory
validity in all aspects. The reliability coefficient was 0.95 for
the clinical index and 0.97 for the surgical one. Kappa
coefficients were 0.4 and 0.8 respectively for the ordinal
forms of the index.
The following table shows the results for both indexes
according to items and weights used in this work.
Variables
Scales
Weights
.............................................................................................................................................................................................................................
1
Internal Medicine
Age
14–35 years: 1; 36–60 years: 2; More than 60 years: 3
15
Residual effects
None: 0; Some, but did not change life style: 1; Life habits
25
changed: 2; Deceased: 3
Admission
Urgent: 1; Elective: 0
11
Principal diagnosis aetiology
Malignant: 1; Other: 0
11
Invasive diagnostic or therapeutic
Yes: 1; No: 0
12
procedures
Response to treatment
Immediate: 0; Non-immediate: 1; No response: 2
18
Number of organ failures
(Number)
28
Complications
None: 0; Yes, but not life threatening: 1; Life threatening: 2
15
General Surgery2
Residual effects
None: 0; Some, but did not change life style: 1; Life habits
4.1
changed: 2; Deceased: 3
Urgent procedures
Yes: 1; No: 0
2.3
General complications
Yes: 1; No: 0
1.3
Organ failure
Yes: 1; No: 0
0.9
Intervention type
Urgent: 1; Elective: 0
0.9
Stay in UIT
Yes: 1; No: 0
0.8
Invasive diagnostic or therapeutic
Yes: 1; No: 0
0.9
procedures
Aetiology of comorbidities
None: 0; Non-malignant: 1; Malignant: 2
Wound complications
Yes: 1; No: 0
0.5
Age (years)
0.4
1
Data are reproduced from Tables 1 and 2 in Jimenez RE, Vazquez J, Fariñas H. Construction and validation of a Severity of Disease
Index for inpatients in clinical areas (in Spanish). Gaceta Sanitaria 1997; 11: 122–130, by permission of the publisher.  Editorial Garsi,
S.A., España.
2
Data are reproduced from Jimenez RE, Dominguez E, Fariñas H, Fuentes E. Construction and validation of a Severity of Disease Index
for inpatients in surgical areas (in Spanish). Rev Cub Salud Publica (in press), by permission of the publisher.  Editorial Ciencias Médicas,
La Habana.
383
R. Jiménez et al.
Appendix 2. Examples of surgical
interventions included in each category1
Category
1
2
3
4
5
1
Intervention examples
Renal transplantation, pneumectomy, rectum amputation, hepatic resection
Nephrectomy, lobectomy, small bowel resection, hepatic node exeresis
Mastectomy, appendectomy, gastrostomy, splenectomy
Cholecystectomy, jejunostomy, salpingectomy, oophorectomy
Breast node exeresis, thoracic wall fistulectomy, wound granuloma resection,
epigastric herniorrhaphy
Data are reproduced from Jimenez RE, Dominguez E, Fariñas H, Fuentes E. Construction and validation of a Severity of Disease Index
for inpatients in surgical areas (in Spanish). Rev Cub Salud Publica (in press), by permission of the publisher.  Editorial Ciencias Médicas,
La Habana.
384