Content uploaded by jean-pierre Michel
Author content
All content in this area was uploaded by jean-pierre Michel on Mar 24, 2019
Content may be subject to copyright.
Geriatrics index of comorbidity was the most accurate predictor of death
in geriatric hospital among six comorbidity scores
Dina Zekry
a,
*, Bernardo Hermont Loures Valle
b
, Claudia Lardi
a
, Christoph Graf
a
,
Jean-Pierre Michel
a
, Gabriel Gold
a
, Karl-Heinz Krause
c
, Franc¸ois R. Herrmann
a
a
Rehabilitation and Geriatrics Department, Geneva University, 3, chemin du Pont-Bochet, 1226, Th^onex, Switzerland
b
Clementino Fraga Filho Hospital, Rio de Janeiro Federal University, Rio de Janeiro, Brazil
c
Pathology and Immunology Department, Geneva University, Geneva, Switzerland
Accepted 18 November 2009
Abstract
Objectives: To compare the abilities of six validated comorbidity indices (Charlson index, cumulative illness rating scale [CIRS], index
of coexistent diseases, Kaplan scale, geriatrics index of comorbidity [GIC], and chronic disease score) to predict adverse hospitalization
outcomes (death during hospitalization, length of stay, and institutionalization).
Study Design and Setting: Prospective cohort of 444 elderly inpatients (mean age 85.3) was randomly selected from Geneva geriatric
hospital.
Results: In univariate analyses, GIC was the best predictor for all outcomes. The risk of death was 30 times higher and the risk of
prolonged hospitalization and being institutionalized was eight to nine times higher in patients with scores of class 3 or 4. In adjusted
logistic regression models, GIC remained the best predictor of death during hospitalization. Higher GIC scores accounted for 25% of
the variance of this outcome, with mortality rates differing by a factor of four between the highest and the lowest scores. CIRS was a strong
predictor of a prolonged hospital stay and institutionalization, accounting for 10% of the variance of these outcomes.
Conclusion: GIC was the most accurate predictor of death during hospitalization. CIRS could be used to select elderly patients at
admission as an indicator of improvement at discharge. Ó2010 Elsevier Inc. All rights reserved.
Keywords: Comorbidity scores; Aged; Elderly; Death; Length of stay; Institutionalization
1. Introduction
Elderly patients often suffer from multiple chronic con-
ditions that individually and jointly affect their quality of
life, use of health services, morbidity, and mortality [1].
Several indices have been proposed to quantify comorbidity
in adults. However, only some of them are valid and reli-
able for use as a measure of comorbidity in applied clinical
research [2] or in elderly patients [3,4]: (1) The Charlson
comorbidity index (CCI) is the most extensively studied
comorbidity index (CI) for predicting mortality. It is
a weighted index that takes into account the number and se-
verity of comorbid conditions [5]. This index was created to
enhance the prediction of 1-year mortality in a cohort of
medical young patients, but it has been used to predict other
health outcomes, such as functional status. It gives a highest
weight for conditions that are not frequent (i.e., AIDS) in
the elderly; and for other conditions, so frequent in elderly
patients (i.e., dementia) the weight is lower, (2) the cumu-
lative illness rating scale (CIRS) addresses all relevant
physiological systems rather than being based on specific
diagnoses and consists of two parts: the CI and the severity
index [6]. The advantage of this scale built for geriatrics pa-
tients is that it assesses the severity of diseases according to
their impact of disability, (3) The index of coexisting dis-
ease (ICED) was developed to predict in-hospital postoper-
ative complications and 1-year health-related quality of life
of patients who underwent total hip replacement surgery.
This index has a 2-dimensional structure, measuring dis-
ease severity and disability, which can be useful when con-
sidering mortality and disability as the outcomes of interest
[7]. A major limitation of the ICED is that it requires med-
ical records and highly trained reviewers who must follow
complex decision rules in creating the index, (4) The Ka-
plan index was developed specifically for use in diabetes re-
search [8], (5) the geriatrics index of comorbidity (GIC)
takes into account the number and severity of diseases,
No author received any consultancy fees or has any company holdings
or patents. There are no conflicts of interest.
* Corresponding author. Tel.: þ41-22-3056355; fax: þ41-22-3056115.
E-mail address:dina.zekry@hcuge.ch (D. Zekry).
0895-4356/$ - see front matter Ó2010 Elsevier Inc. All rights reserved.
doi: 10.1016/j.jclinepi.2009.11.013
Journal of Clinical Epidemiology 63 (2010) 1036e1044
but although it was built for geriatric patients, it has the pe-
culiarity of not including disability [9], and (6) the chronic
disease score (CDS) is an alternative CI based on the drugs
taken by the patient rather than clinical diagnoses [10].
These tools were initially validated in institutionalized
elderly patients in a retrospective manner. A previous study
examined the prognostic value of the CCI in predicting a 3-
year mortality and functional decline in patients receiving
long-term care from 88 residential care facilities in Quebec,
Canada (291 dependent elderly adults with a mean age of
83.3 years). The CCI performed well in predicting both out-
comes [11]. The CIRS is significantly associated with mor-
tality, acute hospitalization, medication usage, laboratory
test results, and functional disability among frail elderly in-
stitutionalized patients [6]. Recently, Di Bari et al. [12]
showed that these measures of comorbidity (CCI, ICED,
GIC, and CDS) predicted death and disability in basic ac-
tivities of daily life in 688 Italian community dwellers with
a mean age of 74 years. However, the value, relevance, and
pertinence of these CIs as predictors of hospitalization ad-
verse outcomes in the very elderly remain unknown.
In this prospective study, we compared the performance
of these six validated and widely used CIs in predicting ad-
verse hospitalization outcomes in the elderly, including
death during the hospitalization period, a prolonged hospi-
tal stay, and institutionalization. The study population was
derived from a study cohort of very elderly, acutely ill
geriatric inpatients.
2. Methods
2.1. Patients and data collection
We carried out a prospective study in a 300-bed geriatric
hospital (HOGER) of the University Hospitals of Geneva,
Switzerland, for acute illness. Patients and data collection
have been described elsewhere [13]. Briefly, patients were
recruited by clinically trained staff. All patients older than
75 years and consecutively admitted on selected days be-
tween January 2004 and December 2005 were included.
We selected a random sample of patients for each day, us-
ing a computer-generated randomization table. The local
ethics committee approved the protocol, and the patients
or their families or legal representatives gave signed written
informed consent. Demographic data for the patients stud-
ied did not significantly differ from data for all patients ad-
mitted to the HOGER during 2004e2005. Our sample was
therefore representative of all patients admitted to this hos-
pital, demonstrating the reliability of the randomization
procedure used in this study.
Medical history was recorded on a standardized form
and the same geriatrician carried out physical examinations
on all patients. Annual follow-up over a 4-year period, with
the same assessment carried out each year, was planned in
the study protocol.
2.2. Sociodemographic data
The data recorded included age, sex, native language,
marital status, living arrangement, and educational level.
2.3. Cognitive diagnosis
The same neuropsychologist assessed all subjects for
clinical dementia, at least 1 week after admission, to avoid
the effects of concomitant delirium. The mini-mental state
examination scores (0e30) [14] and the short cognitive
evaluation battery [15,16] were used. Based on screening
results, the same neuropsychologist then carried out a com-
prehensive standardized neuropsychological assessment to
determine the etiology and severity of clinical dementia,
as previously described [13].
2.4. Assessment of comorbidity
The same geriatrician calculated all six scores for each
patient by extensive review of the patient’s medical records
and administrative data for diagnoses established at or be-
fore enrollment in this study.
1. Charlson comorbidity index [5]
The CCI is a list of 19 conditions; each is assigned
a weighting (1e6). Weightings reflect the ability of
each condition to predict 1-year mortality, as origi-
nally reported for cancer patients. They are fixed
for each diagnosis and range from 1 (for conditions,
such as myocardial infarction or mild liver disease,
with a relative risk >1.2 and !1.5) to 6 (assigned
to metastatic cancer, with a relative risk >6). The
CCI is the sum of the weightings for all conditions
observed in a patientdhigher scores indicated
greater comorbidity.
2. Cumulative illness rating scale [6]
The CIRS identifies 14 items, corresponding to dif-
ferent systems. Each system is scored as follows: 1
(none)dno impairment to that organ or system; 2
(mild)dimpairment does not interfere with normal
activity, treatment may or may not be required, prog-
nosis is excellent; 3 (moderate)dimpairment inter-
feres with normal activity, treatment is needed,
prognosis is good; 4 (severe)dimpairment is dis-
abling, treatment is urgently needed, prognosis is
guarded; 5 (extremely severe)dimpairment is life
threatening, treatment is urgent or of no avail, poor
prognosis. The illness severity index (summary score
based on the average of all CIRS items, excluding
psychiatric or behavioral factors) and the CI (sum-
mary score based on a count of organ system with
moderate or greater impairment, excluding psychiat-
ric or behavioral factors) can then be calculated using
these scores.
1037D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
3. Index of coexistent diseases [7]
The ICED is based on the presence and severity of 19
medical conditions and 11 physical impairments, us-
ing two scales: the index of disease severity (IDS)
and the index of physical impairment (IPI). The final
ICED score is determined by an algorithm combin-
ing the peak scores for the IDS and IPI. The ICED
score ranges from zero to three (four classes), reflect-
ing increasing severity.
4. Kaplan scale [8]
This index uses two forms of classification: focusing
on the type of comorbidity and the pathophysiologic
severity of the comorbid conditions present, respec-
tively. The type of comorbidity can be classified as
vascular (hypertension, cardiac disorders, peripheral
vascular disease, retinopathy, and cerebrovascular
disease) or nonvascular (lung, liver, bone, and nondi-
abetic renal diseases). Pathophysiologic severity is
rated on a 4-point scale, ranging from zero (comor-
bidity is absent or easy to control) to three (recent
full decompensation of comorbid disease). The rating
of the most severe condition determines the overall
comorbidity score. Scores for vascular and nonvascu-
lar comorbidity can be calculated, based on the most
severe condition in each subscale.
5. Geriatric index of comorbidity [9]
In computing the GIC, each of the 15 more prevalent
clinical conditions (ischemic or organic heart dis-
eases, primary arrhythmias, heart diseases with a non-
ischemic or nonorganic origin, hypertension, stroke,
peripheral vascular diseases, diabetes mellitus, ane-
mia, gastrointestinal diseases, hepatobiliary diseases,
renal diseases, respiratory diseases, parkinsonism
and nonvascular neurologic diseases, musculoskele-
tal disorders, and malignancies) is graded on a 0e4
disease severity scale on the basis of the following
general framework: 0 5absence of disease,
15asymptomatic disease, 2 5symptomatic disease
requiring medication but under satisfactory control,
35symptomatic disease uncontrolled by therapy,
and 4 5life-threatening or the most severe form of
the disease. The GIC classifies patients into four clas-
ses of increasing somatic comorbidity. Class 1 in-
cludes patients who have one or more conditions
with a disease severity grade equal to or lower than
1. Class 2 includes patients who have one or more
conditions with a disease severity grade of 2. Class
3 includes patients who have one condition with a dis-
ease severity of 3, other conditions having a disease
severity equal to or lower than 2. Class 4 includes pa-
tients who have two or more conditions with a disease
severity of 3 or one or more conditions with disease
severity of 4.
6. Chronic disease score [10]
This is a measure of comorbidity obtained from
a weighted sum of scores based on the use of 30 dif-
ferent classes of medication. An integer weight be-
tween one and five is given to each of the selected
classes of medication; the overall score is then the
sum of the weightings.
2.5. Adverse outcomes of hospitalization
The adverse outcomes considered include hospital stays
greater than the median value, death during the hospitaliza-
tion period, and changes in living arrangements at dis-
charge (institutionalization).
2.6. Statistical methods
We checked for the normal distribution of data for contin-
uous scores (CCI, CIRS, Kaplan scale, and CDS) using
skewness and kurtosis tests and carried out standard trans-
formations to normalize non-Gaussian variables. As it was
not possible to normalize these scores, they were catego-
rized into quartiles to facilitate comparison with the four
classes of the other two indices, ICED and GIC. Colinearity
among the six indices was checked using Spearman rank
correlation coefficient. Multiple logistic regression analysis
was then carried out using age, sex, and the six comorbidity
scores as independent variables and each outcome as depen-
dent variable to identify the best predicting score for each
outcome, whereas adjusting for all the others. Outcomes
were considered as dichotomous data (death during hospital-
ization, a prolonged hospital stay [longer than the median
duration], admission to long-term care). Odds ratios and
95% confidence intervals were calculated. Statistical analy-
ses were performed with Stata software version 10.1 (Stata-
Corp LP, College Station, TX, USA).
3. Results
We included 444 patients in this study (mean age
85.3 66.7, 74% women). Table 1 summarizes frequency dis-
tribution of patients according to each comorbidity score.
As there were no patients in the ICED classes 1 and 2,
we considered only classes 3 and 4, providing binary data
for the analyses. Likewise, only 2% of the patients were
classified as class 1 by the GIC, allowing us to combine
classes 1 and 2 for the analysis.
For the other four indices, the distribution was almost
equal among the four quartile ranges, with approximately
25% of the patients per range.
Table 2 shows the patient’s destination after hospitaliza-
tion, comparing living arrangement before and after.
3.1. Univariate and multiple logistic regression analysis
Spearman rho values among the six indices ranged be-
tween 0.038 and 0.548, which does not meet the criteria for
1038 D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
colinearity usually set at O0.900. We carried out univariate
logistic regression analyses including age, sex, and the six
CIs tested predicting the three adverse hospitalization out-
comes (Table 3). We then tested full multiple logistic regres-
sion models containing all the variables. No new differences
were observed; thus, results are presented only with variables
that were positive in the univariate models.
3.1.1. Length of stay (median 532 days)
In univariate analysis, age, quartiles or class 3 or 4
scores were found to be independent predictors of pro-
longed hospitalization. GIC class 4 scores were the stron-
gest predictors of a prolonged stay in hospital, with
a difference of a factor of nine in adverse outcome rate be-
tween patients with the highest and lowest scores.
This association was not observed when all variables
were introduced into the analysis, with only the third and
fourth quartiles of CIRS scores remaining statistically sig-
nificant and accounting for 10% of the variability of this
outcome. Higher classes of the ICED also remained weakly
significant, with P50.045.
3.1.2. Death during hospitalization
Of the 444 patients, 27 died during the hospitalization
period (6%). In univariate analysis, mortality was signifi-
cantly associated with age (not with sex) and with the high-
est score of the CCI, CIRS, ICED, Kaplan scale, and GIC
but not with the CDS. GIC class 4 scores were the strongest
predictors of death during hospitalization, with a difference
of a factor of 37 in adverse outcome rates between patients
with the highest and lowest scores.
When all variables were included in the model, only the
GIC classes 3 and 4 remained statistically significant. High-
er GIC comorbidity scores accounted for 24% of the vari-
ance of this outcome. Higher classes of the ICED score
also remained weakly significant, with P50.045.
3.1.3. Institutionalization
Table 2 summarizes the destinations of patients after
hospitalization. Sixty-one (14.3%) patients were institu-
tionalized and 10% of the initial cohort was transferred to
another hospital (surgery, intensive care).
Univariate analysis revealed that institutionalization was
significantly associated with the highest score of the CIRS,
ICED, Kaplan scale, and the GIC but not with the CCI or
CDS. GIC class 4 and CIRS fourth quartile scores were
the strongest predictors of this outcome, with the rate of in-
stitutionalization differing by factors of nine and five, re-
spectively, between patients with the highest and lowest
scores.
When all variables were included in the model, only the
CIRS classes 3 and 4 remained statistically significant.
Higher CIRS comorbidity scores accounted for 10% of
the variance of this outcome.
3.1.4. Summary of results
Of the six indices, the GIC explained the largest percent-
age of variation in the frequency of these three outcomes in
Table 1
Quartile range and frequency of six comorbidity scores
Level/classes
a
CCI CIRS ICED
a
Kaplan GIC
a
CDS
Quartile
range score N(%)
Quartile
range score N(%) N(%)
Quartile
range score N(%) N(%)
Quartile
range score N(%)
10e3 165 (37) 0e11 121 (27) 0 0e2 128 (29) 9 (2) 0e3 122 (28)
2 4 91 (20) 12e14 107 (24) 0 3e4 156 (35) 34 (8) 4e6 117 (26)
35e6 91 (20) 15e18 119 (27) 93 (21) 5 55 (12) 310 (70) 7e8 109 (24)
47e14 97 (23) 19e30 97 (22) 351 (79) 6e16 105 (24) 91 (20) 9e15 96 (22)
Data are expressed as number of cases (%).
Abbreviations: CCI, Charlson comorbid index; CIRS, cumulative illness rating scale; ICED, index of coexistent diseases; Kaplan, Kaplan scale; GIC,
geriatrics index of comorbidity; CDS, chronic disease score.
a
Quartile ranges do not apply to ICED and GIC, because continuous scores were not calculated using these tools and patients were assigned directly to
four classes.
Table 2
Destination after hospitalization (n5444)
Living arrangements
Total N(%)
After hospitalization
Before hospitalization Alone Partner Family Protected residence Nursing home Died in hospital Transfer
Alone 258 (58) 179 0 0 0 36 16 27
Partner 105 (27) 0 70 0 0 15 7 13
Family 36 (8) 0 0 27 0 3 1 5
Protected residence 27 (6) 0 0 0 16 7 3 1
Nursing home 18 (4) 0 0 0 0 17 0 1
Total N(%) 179 (40) 70 (16) 27 (6) 16 (4) 78 (18) 27 (6) 47 (10)
Data are expressed as number of cases (%).
1039D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
Table 3
Univariate and multivariate logistic regression including all variables for predictors of the three adverse hospitalization outcomes (length of stay greater
than the median, death during hospitalization, institutionalization) (n5444)
Outcomes Independent variables
Univariate logistic regression Multiple logistic regression
Crude OR 95% CI Adjusted OR 95% CI
Length of stay
Age 1.03 1.00e1.06* 1.02 0.99e1.05
Male vs. female 0.91 0.60e1.39
CCI
Quartile
1 1.00 d
2 1.27 0.76e2.12
3 1.77 1.07e2.94* 1.24 0.66e2.32
4 1.89 1.12e3.17* 1.45 0.79e2.65
CIRS
Quartile
1 1.00 d
2 1.82 1.07e3.09* 1.36 0.72e2.54
3 3.51 2.03e6.07*** 3.00 1.64e5.46***
4 5.07 2.84e9.04*** 4.08 1.91e8.7***
ICED
Class
1þ2þ3 1.00 d
4 2.00 1.24e3.2* 1.73 1.01e2.96*
Kaplan
Quartile
1 1.00 d
2 1.38 0.73e2.6 0.59 0.27e1.30
3 2.10 1.30e3.39** 1.10 0.53e2.27
4 2.40 1.42e4.08*** 1.32 0.75e2.29
GIC
Class
1þ2 1.00 d
3 8.22 3.46e19.5*** 0.88 0.33e2.33
4 9.03 4.08e20.0*** 1.56 0.69e3.52
CDS
Quartile
11.00 d
2 1.88 1.11e3.17* 1.18 0.65e2.14
3 2.03 1.18e3.50** 1.57 0.83e2.94
4 2.06 1.23e3.46** 1.61 0.92e2.85
Death in hospital
Age 1.07 1.00e1.15* 1.06 0.98e1.15
Male vs. female 0.99 0.38e2.6
CCI
Quartile
1 1.00 d
2 1.68 0.88e3.20
3 1.74 0.92e3.27
4 2.49 1.34e4.60** 1.15 0.96e1.37
CIRS
Quartile
1 1.00 d
2 1.72 1.07e3.09
3 4.29 2.03e6.07
4 6.84 2.84e9.04* 1.21 0.20e7.14
ICED
Class
1þ2þ3 1.00 d
4
a
* 1.36 1.01e1.83*
(Continued )
1040 D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
Table 3
Continued
Outcomes Independent variables
Univariate logistic regression Multiple logistic regression
Crude OR 95% CI Adjusted OR 95% CI
Kaplan
Quartile
1 1.00 d
2 1.23 0.20e7.50
3 4.94 0.88e27.82
4 9.70 2.14e43.69** 1.71 0.28e10.50
GIC
Class
1þ2 1.00 d
3 34.30 13.75e87.82*** 3.68 3.01e6.26***
4 37.14 14.75e93.53*** 4.34 3.92e9.52***
CDS
Quartile
1 1.00 d
2 0.62 0.14e2.64
3 1.60 0.49e5.21
4 2.13 0.67e6.70
Institutionalization
Age 1.05 1.00e1.10* 1.03 0.98e1.08
Male vs. female 0.95 0.50e1.80
CCI
Quartile
1 1.00 d
2 1.42 0.66e3.07
3 1.50 0.70e3.20
4 1.69 0.80e3.57
CIRS
Quartile
1 1.00 d
2 1.98 0.77e5.09
3 2.98 1.23e7.21* 2.73 1.10e6.77*
4 5.53 2.31e13.21*** 5.56 2.18e14.22***
ICED
Class
1þ2þ3 1.00 ddd
4 2.31 1.05e5.08* 1.75 0.75e4.03
Kaplan
Quartile
11.00 d
2 1.65 0.69e3.89
3 2.22 0.88e5.55
4 2.27 1.09e4.72* 1.65 0.91e3.00
GIC
Class
1þ2 1.00 d
3 3.25 1.24e11.20*** 1.50 0.39e5.79
4 4.62 3.46e13.20*** 1.53 0.32e7.25
CDS
Quartile
1 1.00 d
2 0.57 0.24e1.43
3 0.94 0.44e2.04
4 1.40 0.70e2.844
Abbreviations: OR, odds ratio; CI, confidence interval; CCI, Charlson comorbid index; CIRS, cumulative illness rating scale; ICED, index of coexistent
diseases; Kaplan, Kaplan scale; GIC, geriatrics index of comorbidity; CDS, chronic disease score.
*P!0.05, **P!0.01, ***P!0.001.
a
ICED class 4 strongly predicts the outcome.
1041D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
univariate analyses. When all scores were compared in a lo-
gistic regression after controlling for age and sex, the GIC re-
mained a strong predictor for death during hospitalization.
However, the CIRS performed better than the other indi-
ces in predicting a prolonged hospital stay and institu-
tionalization. The CDS performed the most poorly for
predicting death during hospitalization and institutionaliza-
tion. The risk of being hospitalized for longer than the me-
dian ranged from 1.88 for the lower scores to 2.06 for the
higher scores, showing poor discrimination between these
groups of patients. CCI scores were not predictive of insti-
tutionalization at all and were less predictive of prolonged
hospitalization or death during hospitalization than the
ICED or Kaplan scale.
4. Discussion
One of the main strengths of this study was the compre-
hensive and detailed assessment of the presence and extent
of comorbidities: the same medical doctor scored the six
CIs for all patients to ensure a high accuracy of scoring.
The prospective collection of comorbidity data allowed bet-
ter control over the quality of the data needed to quantify
comorbidity. We carried out, for the first time, a prospective
study comparing the use of six CIsdthe most widely used
and validated in elderly subjectsdfor the prediction of
three adverse outcomes of hospitalization in elderly pa-
tients with acute disease. Previous studies, as described ear-
lier, have used only one comorbidity score and have mostly
been retrospective.
In our prospective study, introducing all parameters into
the model, having checked for the absence of colinearity
and adjusting for age and sex, the GIC provided a better
measure of comorbidity than the other indices tested, when
death during hospitalization was the outcome of interest.
The CIRS could be used as a method for selecting elderly
patients at admission and as a prognostic predictor for im-
provement at discharge. The results obtained for the CIRS
were similar to previous findings in a retrospective analysis
of patients aged 90e99 years, admitted over a 6-month pe-
riod to a district hospital in Australia. One hundred three
patients were included in the study with an average age
of 92 years and a male-to-female ratio of 1:3. Fifty-five per-
cent of hospitalized patients came from nursing care facil-
ities. Characteristics of patients from nursing homes were
compared with those of patients from the community.
The physical burden of illness was measured by the CIRS.
There was a significant (P!0.05) correlation between
high CIRS scores and duration of the hospital stay. The
death rate for this group of patients was higher (13%) than
the proportion of patients with a prolonged hospitalization
period (10.2%). There were significant differences in the
CIRS scores between patients who died and those who sur-
vived; the CIRS is thus potentially a useful tool in predict-
ing this outcome [17]. In our univariate analysis, high CIRS
scores were associated with death during hospitalization,
with death rate differing by a factor of six between patients
with the highest and lowest scores. These results confirmed
those of Salvi et al. [18] that previously demonstrated the
CIRS’s ability to predict 18-month mortality and rehospi-
talization in a cohort of 387 patients aged 65 and older from
an acute internal medicine ward. One advantage of the
CIRS is its suitability for use in common clinical practice:
it is based on measures of clinically relevant physiological
systems and uses a clear and clinically sound ranking of se-
verity. Given its validity and reliability, the CIRS seems to
provide a very useful measure of comorbidity for clinical
research. This index appears to be sufficiently reliable be-
cause it allows all the comorbid diseases from clinical ex-
aminations and medical files to be taken into account in
a comprehensive manner [19]. The CIRS, however, has
some limitations and improvements are needed, such as
the inclusion of psychiatric disturbances, which are highly
prevalent in the elderly. Such limitations may explain why,
when all variables were included in the model, only the
GIC class 3 and 4 scores remained statistically significant
for the prediction of death during hospitalization.
Similarly, previous studies confirmed the impact of the
GIC index on the prediction of 6-month survival in a popu-
lation of 1,402 hospitalized elderly patients (age 80.1 67.1
years; 68% female) with chronic disability consecutively
admitted to an acute care unit in Italy. As observed in our
study, patients with GIC class 1 and 2 scores were scarce
in this acute geriatric ward. In a Cox regression analysis,
adjusting for factors associated with mortality in univariate
models (low levels of serum albumin and cholesterol,
anemia, dementia, chronic obstructive pulmonary disease,
coronary heart disease, renal diseases, gastrointestinal
diseases, and advanced cancer) and taking class 2 as a refer-
ence, patients with GIC scores in class 4 had a risk of death
three times higher than patients with the lowest scores [9].
The CDS was the poorest predictor for all the adverse out-
comes considered. This is consistent with other previous
studies. The low predictive value of this medication-based
score for short-term outcomes may be because of the use
of preventive treatments or treatment for benign conditions
in healthier patients. For example, elderly women who are
generally healthy and aware of health risks are likely to take
lipid-lowering drugs and hormone replacement therapy.
Such patients are likely to fare better than patients whose pri-
mary diagnosis has a poor short-term prognosis that may de-
ter treatment of secondary conditions. This is consistent with
earlier findings that sicker patients are less likely to be
treated for comorbid conditions [20], particularly if these
conditions are not immediately life threatening; additionally,
medication for treating these conditions has preventive ef-
fects, for example, oral antidiabetic agents [21] or lipid-
lowering drugs [22]. Users are thus often healthier than
would be suggested by their medication-based scores. Al-
though these findings are yet to be confirmed in other popu-
lations, they suggest that medication-based scores should be
1042 D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
used only in situations when the available data on the med-
ication taken by the patients are of much better quality than
the diagnostic data, or are the only source of information.
The CCI was not predictive of institutionalization at all
and performed more poorly than the ICED or Kaplan scale
for predicting prolonged hospitalization and death during
hospitalization. The CCI is the most extensively studied
CI for predicting mortality [2]. It was designed and scaled
to predict mortality rather than functionally relevant comor-
bidity. This index does not take into account the severity of
certain major diseases but only the presence of the disease.
For example, in the case of congestive heart failure, patients
with either a mild or a severe form of the disease will be as-
signed a score of 1. This index may therefore fail to identify
important diseases, or their severity, in the elderly, which
may otherwise act as predictors of adverse outcomes. The
CCI has previously been found to be limited in determining
the full range of diseases in elderly patients [19]. For this
reason, some studies tried to outperform the CCI comparing
the predictive capacity on mortality, readmission, and
length of stay of the original CCI with a new CI regarding
a larger range of diseases. Their results favor the utilization
of newly developed indices [23,24]. On the contrary, Bun-
tinx et al. [25], in a large cohort of 2,624 institutionalized
elderly people, showed that the CCI is a predictor of
short-term mortality and, to a lesser extent, also of hospital-
ization. In addition, the CCI has been shown to predict costs
of chronic disease in primary care patients and in conse-
quence being useful to predict resource utilization [26].
The GIC classifies patients based on increasing somatic
comorbidity and takes into account disease severity. This
probably explains why, when including all variables in the
model, this index remained statistically significant and the
best predictor for death during hospitalization in these el-
derly patients with acute disease. In the logistic regression
model, the ICED also remained statistically, but weakly, sig-
nificant. A distinct advantage of the ICED is that this index
includes information on physical impairment in the assess-
ment of comorbidity. Physical impairment is considered to
be an additional dimension of comorbidity [27], reflecting
symptomatic, uncontrolled, or advanced stages of disease.
The ICED is the only one of these measures studied that
has a 2-dimensional structure, measuring both the severity
and extent of the disability associated with pathophysiologic
disease. This could be particularly useful in studies assess-
ing mortality and disability as outcomes of interest [2].
The Kaplan index performed well in our univariate anal-
yses but lost all significance when all variables were con-
trolled for. This index was specifically developed for use
in diabetes research and contains clinically relevant infor-
mation. It distinguishes between vascular and nonvascular
comorbidity and uses severity rankings based on parame-
ters derived from common clinical practice. The validity
of this test makes the Kaplan scale a useful CI for clinical
diabetes research [2] but probably less useful for assessing
comorbidity in the elderly.
Currently, there is no accepted standardized method for
measuring and quantifying the prognostic value of comor-
bid conditions in hospitalized elderly patients with acute
disease. Our results showed that it is unlikely that any
one particular index can be used to predict a variety of rel-
evant outcomes. According to our results, the choice of
measures will depend on the outcomes of interest as previ-
ously stated by Byles et al. [28]. We can recommend more
usefully the GIC in predicting vital outcomes because of its
link to physiological aspects of diseases, whereas the CIRS
captures more comorbidity information related to the care
because of its link to functional aspects of diseases. These
findings have widespread implications for improved plan-
ning of the hospitalization period through the discharge
of very ill elderly patients with acute disease.
The ways that health researchers have measured comor-
bidity has advanced our understanding in aging population
but an important issue in geriatrics remains the need for
new and better measures of the health status of elderly indi-
viduals that summarize the complex disorders that burdened
them. Studies contrasting multimorbidity, which is defin-
eddfollowing van den Akker et al. study [29,30]das the
co-occurrence of two or more diseases in one person, without
defining an index disease and comorbidity, corresponding to
additional diseases to one index disease are needed. It would
be essential to take into account not only the number of
comorbid conditions and an index weighted by the severity
of the comorbid conditions but also the associations among
diseases.
Acknowledgments
The authors thank the teams of Mrs. O. Baumer, L.
Humblot, and M. Cos for their technical assistance. This
work was supported by grant 3200B0-102069 from the
Swiss National Science Foundation.
References
[1] Gijsen R, Hoeymans N, Schellevis FG, Ruwaard D, Satariano WA,
van den Bos GA. Causes and consequences of comorbidity: a review.
J Clin Epidemiol 2001;54:661e74.
[2] de Groot V, Beckerman H, Lankhorst GJ, Bouter LM. How to mea-
sure comorbidity: a critical review of available methods. J Clin Epi-
demiol 2003;56:221e9.
[3] Karlamangla A, Tinetti M, Guralnik J, Studenski S, Wetle T, Reuben D.
Comorbidity in older adults: nosology of impairment, diseases, and
conditions. J Gerontol A Biol Sci Med Sci 2007;62:296e300.
[4] Rozzini R, Sabatini T, Barbisoni P, Trabucchi M. How to measure co-
morbidity in elderly persons. J Clin Epidemiol 2004;57:321e2.
[5] Charlson ME, Pompei P, Ales KL, Mackenzie CR. A new method of
classifying prognostic comorbidity in longitudinal studies: develop-
ment and validation. J Chronic Dis 1987;40:373e83.
[6] Parmelee PA, Thuras PD, Katz IR, Lawton MP. Validation of the cu-
mulative illness rating scale in a geriatric residential population. J Am
Geriatr Soc 1995;43:130e7.
1043D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044
[7] Greenfield S, Sullivan L, Dukes KA, Silliman R, D’Agostino R,
Kaplan SH. Development and testing of a new measure of case-mix
for use in office practice. Med Care 1995;33:AS47e55.
[8] Kaplan MH, Feinstein A. The importance of classifying initial
co-morbidity in evaluating the outcome of diabetes mellitus.
J Chronic Dis 1974;27:387e404.
[9] Rozzini R, Frisoni GB, Ferrucci L, Barbisoni P, Sabatini T,
Ranieri P, et al. Geriatric index of comorbidity: validation and
comparison with other measures of comorbidity. Age Ageing
2002;31:277e85.
[10] Von Korff M, Wagner EH, Saunders K. A chronic disease score from
automated pharmacy data. J Clin Epidemiol 1992;45:197e203.
[11] Bravo G, Dubois MF, H
ebert R, De Wals P, Messier L. A prospective
evaluation of the Charlson Comorbidity Index for use in long-term
care patients. J Am Geriatr Soc 2002;50:740e5.
[12] Di Bari M, Virgillo A, Matteuzzi D, Inzitari M, Mazzaglia G,
Pozzi C, et al. Predictive validity of measures of comorbidity in older
community dwellers: the Insufficienza Cardiaca negli Anziani Resi-
denti a Dicomano Study. J Am Geriatr Soc 2006;54:210e6.
[13] Zekry D, Herrmann FR, Grandjean R, Vitale AM, De Pinho MF,
Michel JP, et al. Does dementia predict adverse hospitalization
outcomes? A prospective study in aged inpatients. Int J Geriatr
Psychiatry 2008;37:83e9.
[14] Folstein MF, Folstein SE, MacHugh PR. Mini Mental State: a practi-
cal method for grading the cognitive state of patients for the clinician.
J Psychiatr Res 1975;12:189e98.
[15] Solomon PR, Hirschoff A, Kelly B, Relin M, Brush M, DeVeaux RD,
et al. A 7 minute neurocognitive screening battery highly sensitive to
Alzheimer’s disease. Arch Neurol 1998;55:349e55.
[16] Robert PH, Schuck S, Dubois B, Oli
eJP,L
epine JP, Gallarda T, et al.
Screening for Alzheimer’s disease with the short cognitive evaluation
battery. Dement Geriatr Cogn Disord 2003;15:92e8.
[17] Nagaratnam N, Gayagay G. Validation of the cumulative illness
rating scale (CIRS) in hospitalized nonagenarians. Arch Gerontol
Geriatr 2007;44:29e36.
[18] Salvi F, Miller MD, Grilli A, Giorgi R, Towers AL, Morichi V, et al.
A manual of guidelines to score the modified cumulative illness
rating scale and its validation in acute hospitalized elderly patients.
J Am Geriatr Soc 2008;56(10):1926e31.
[19] Harboun M, Ankri J. Comorbidity indexes: review of the literature
and application to studies of elderly population. Rev Epidemiol Sante
Publique 2001;49:287e98.
[20] Redelmeier DA, Tan SH, Booth GL. The treatment of unrelated dis-
orders in patients with chronic medical diseases. N Engl J Med
1998;21:1516e20.
[21] Glynn RJ, Monane M, Gurwitz JH, Choodnovskiy I, Avorn J. Aging,
comorbidity, and reduced rates of drug treatment for diabetes melli-
tus. J Clin Epidemiol 1999;52:781e90.
[22] Glynn RJ, Knight EL, Levin R, Avorn J. Paradoxical relations of drug
treatment with mortality in older persons. Epidemiology 2001;12:682e9.
[23] Martins M, Blais R. Evaluation of comorbidity indices for inpatient
mortality prediction models. J Clin Epidemiol 2006;59:665e9.
[24] Holman CD, Preen DB, Baynham NJ, Finn JC, Semmens JB.
A multipurpose comorbidity scoring system performed better than
the Charlson index. J Clin Epidemiol 2005;58:1006e14.
[25] Buntinx F, Niclaes L, Suetens C, Jans B, Mertens R, Van den
Akker M. Evaluation of Charlson’s comorbidity index in elderly
living in nursing homes. J Clin Epidemiol 2002;55:1144e7.
[26] Charlson ME, Charlson RE, Peterson JC, Marinopoulos SS,
Briggs WM, Hollenberg JP. The Charlson comorbidity index is adap-
ted to predict costs of chronic disease in primary care. J Clin Epide-
miol 2008;61:1234e40.
[27] Iezzoni LI. Dimensions of risk. In: Iezzoni LI, editor. Risk adjustment
for measuring health outcomes. Ann Arbor, MI: Health Administra-
tion Press; 1994. p. 1e29.
[28] Byles JE, D’Este C, Parkinson L, O’Connell R, Treloar C. Single
index of multimorbidity did not predict multiple outcomes. J Clin
Epidemiol 2005;58:997e1005.
[29] Van den Akker M, Buntinx F, Roos S, Knottnerus JA. Problems in
determining occurrence rates of multimorbidity. J Clin Epidemiol
2001;54:675e9.
[30] Van den Akker M, Buntinx F, Knottnerus JA. Comorbidity or multi-
morbidity: what’s in a name? A review of literature. Eur J Gen Pract
1996;2:65e70.
1044 D. Zekry et al. / Journal of Clinical Epidemiology 63 (2010) 1036e1044