Full text of DOI:10.1053/ajkd.2002.30552

Comorbidity and Other Factors Associated With Modality Selection
in Incident Dialysis Patients: The CHOICE Study
Dana C. Miskulin, MD, MS, Klemens B. Meyer, MD, Nicolaos V. Athienites, MD,
Alice A. Martin, RN, Norma Terrin, PhD, Jane V. Marsh, MS, Nancy E. Fink, MPH,
Josef Coresh, MD, PhD, Neil R. Powe, MD, MPH, MBA, Mike J. Klag, MD, MPH,
and Andrew S. Levey, MD, for the Choices for Healthy Outcomes in Caring
for End-Stage Renal Disease (CHOICE) Study
● Case-mix factors influence both the selection of dialysis modality and outcomes in end-stage renal disease
(ESRD). A detailed characterization of the case-mix differences between peritoneal dialysis (PD) and hemodialysis
(HD) patients at the onset of dialysis therapy has not been performed, despite the importance of accounting for
baseline differences in future comparisons of outcomes across modality groups. We compared baseline character-
istics of 279 PD and 759 HD patients enrolled in the Choices for Healthy Outcomes in Caring for End-Stage Renal
Disease (CHOICE) Cohort Study, a prospective study of incident dialysis patients. Comorbidity was assessed using
the Index of Coexistent Diseases (ICED), consisting of a medical record review of 19 medical conditions and an
observer-based assessment of 11 physical functions. ICED scores range from 0 to 3, with higher levels reflecting
more severe comorbidity. Comorbidity was less severe in PD patients than in HD patients: the proportions of
patients with ICED 0-1, ICED 2, and ICED 3 were 52%, 26%, and 22%, respectively, among the PD patients and 30%,
39%, and 31%, respectively, among the HD patients (P < 0.001). After controlling for all other factors, the differences
in comorbidity remained significant. As compared with patients with ICED 0-1, the odds of being treated with PD for
patients with ICED 2 and ICED 3 were less (odds ratio [OR] and 95% confidence intervals) 0.31 (0.17 to 0.56) and 0.50
(0.28 to 0.90), respectively. The number and severity of comorbid conditions at the onset of ESRD is significantly
lower in patients choosing PD, independent of other factors influencing modality selection. The increased survival
of PD patients reported in recent studies may simply reflect the self- or physician-directed selection of healthier
patients to PD. Adjustment for case-mix differences in patients treated with PD versus HD is essential to the
assessment of the independent effect of the dialysis modality on outcomes.
© 2002 by the National Kidney Foundation, Inc.
INDEX WORDS: Case-mix; comorbidity; risk adjustment.
HE CHOICE OF dialysis modality is influ-
enced by patients’ social, cultural, eco-
nomic, and medical circumstances, in addition to
patient and physician preferences. As a result,
there are fundamental differences between perito-
neal dialysis (PD) and hemodialysis (HD) pa-
tients even at the onset of dialysis treatment.1
Some of these factors are well-known predictors
of survival and other outcomes.2-9
T
Studies comparing survival of PD and HD
patients have reported conflicting results.10-16 Dif-
ferences in analytic methods have been postu-
lated to account for some of the differences in
results.11-16 A common limitation of all recent US
survival comparisons is the fact that analyses
were adjusted for age, diabetes, and the cause of
end-stage renal disease (ESRD), but not other
comorbid conditions.11,12,16 Comorbidity data has
been collected in Canadian studies,13-15 but be-
cause of well-known differences in PD utiliza-
tion across countries, these results may not be
generalizable to the US. Furthermore, some stud-
ies included prevalent populations.12,16 It is cru-
cial to distinguish between baseline comorbid
conditions that influence modality selection and
comorbid conditions that develop or worsen due
to the dialysis treatment itself. Hence, it is neces-
sary to study these factors at the onset of dialysis
therapy, which requires assessment of an inci-
From the Divisions of Nephrology and Clinical Care
Research, New England Medical Center, Boston, MA; and
the Johns Hopkins Medical Institutions, Baltimore, MD.
Received May 25, 2001; accepted in revised form Septem-
ber 7, 2001.
Presented in part at the 30th Annual Meeting of the
American Society of Nephrology, November 2-5, 1997, San
Antonio, TX, and the 32nd Annual Meeting of the American
Society of Nephrology, November 1-8, 1999, Miami Beach,
FL.
Supported in part by a grant from the Agency for Health-
care Research and Quality (AHRQ) (formerly Agency for
Health Care Policy Research [AHCPR] HS08365 to N.R.P.)
and a Clinical Research Fellowship from the Alberta Heri-
tage Foundation for Medical Research (to D.C.M.).
Address reprint requests to Andrew S. Levey, MD, New
England Medical Center, Division of Nephrology, Box 391,
Boston, MA 02111. E-mail: alevey@lifespan.org
© 2002 by the National Kidney Foundation, Inc.
0272-6386/02/3902-0011$35.00/0
doi:10.1053/ajkd.2002.30552
324
American Journal of Kidney Diseases, Vol 39, No 2 (February), 2002: pp 324-336

COMORBIDITY OF INCIDENT PD VERSUS HD PATIENTS
325
was indexed as the percent from the lower limit of normal
for each laboratory according to the following formula: %
deviation from lower limit normal ϭ (serum albumin –
laboratory lower limit of normal)/laboratory lower limit of
normal ϫ 100. To standardize for variation between all
laboratories, the standardized serum albumin for each pa-
tient was computed as follows: standardized albumin ϭ %
deviation ϫ mean lower limit normal (all laboratories) ϩ
mean lower limit normal (all laboratories). Glomerular filtra-
tion rate (GFR) at the start of dialysis was calculated using a
validated Modification of Diet in Renal Disease Study
(MDRD) formula that uses the serum creatinine, patient age,
race, and gender.18
dent population. Due to the limited case-mix data
collected in prior survival comparison studies, it
is uncertain whether survival differences from
past studies reflect real treatment effects or base-
line prognostic differences between PD and HD
groups.
The Choices for Healthy Outcomes in Caring
for End-Stage Renal Disease (CHOICE) Study is
a Patient Outcomes Research Team (PORT)
funded by the Agency for Healthcare Research
and Quality.17 One of the projects, the CHOICE
Cohort Study, is a prospective study of the rela-
tionship of dialysis modality and dose to subse-
quent outcomes in incident dialysis patients. It
includes a comprehensive cross-sectional and
longitudinal assessment of comorbid factors in
279 PD and 759 HD patients between 1995
through 1999. The objective of the present report
is to compare the baseline characteristics of the
PD and HD patients, with special emphasis on
comorbid conditions, and to assess the relation-
ship of patient characteristics to modality selec-
tion.
Comorbidity Assessment
Comorbidity, referring to medical conditions other than
the primary disease itself,19 was assessed using the Index of
Coexistent Disease (ICED), an instrument that has been
used in another multicenter trial of HD patients20 and has
been validated as a predictor of death in smaller dialysis
populations.21,22 The ICED, ranging from 0-3, with 3 as the
highest severity level, is a measure of both the presence and
severity of different comorbid conditions. It is derived from
the peak scores of the Index of Disease Severity (IDS) and
the Index of Physical Impairment (IPI) using an algorithm
specific to the ICED.22
The IDS consists of 19 categories of medical conditions,
with 4 levels of severity for each condition. Information for
the IDS was abstracted from dialysis unit records, hospital
discharge summaries, medication lists, consultation notes,
diagnostic imaging, and cardiac imaging reports. These data
were collected at each dialysis unit, photocopied, and sent to
New England Medical Center for abstraction and scoring.
Two dialysis nurses, with prior training and experience in
using the ICED, reviewed and scored all charts. The reliabil-
ity of data abstraction and severity scoring was assessed by
comparing peak IDS scores assigned by the nurse reviewer
with a physician reviewer (N.V.A.) in 42 charts. Inter-rater
reliability, assessed by the ␬ statistic, was high (␬ ϭ 0.72).22
The IPI is an observer-based assessment of 11 functional
domains, each with 3 severity levels. The IPI was completed
by a local dialysis nurse familiar with the patient’s level of
functioning, with input from a family member or caregiver,
if necessary.
METHODS
Study Design and Patient Eligibility
We conducted a cross-sectional analysis of baseline data
on incident dialysis patients recruited from 81 dialysis units
in 19 states of the US participating in the CHOICE Cohort
Study. Patients were eligible if they had started chronic
outpatient dialysis during the 3 months before enrollment,
were 18 years or older, spoke English or Spanish, and gave
informed consent. Home HD and hospice patients were
excluded.
Patients were recruited from 79 dialysis units associated
with Dialysis Clinic, Inc (DCI), who were willing to partici-
pate in the study, and from 2 dialysis units in New Haven,
CT associated with Beth Israel Medical Systems that were
invited to participate because of their large number of PD
patients.
Statistical Analysis
Data Collection
Demographic, socioeconomic, and laboratory factors are
described as the mean (SD) for continuous variables and
frequency for dichotomous variables. Statistical significance
of the differences between modalities was tested using
Dialysis modality at baseline was defined as the modality
at 4 weeks after enrollment in the study. All forms of PD
(continuous ambulatory PD, continuous cycling PD, intermit-
tent cycling PD) were combined as a single category.
2
2-sample t tests for continuous variables and ␹ tests for
Data regarding health behaviors, work history, medical
history, preparation for dialysis, social supports, and dis-
tance to dialysis unit was collected from the Baseline Form,
a self-report questionnaire. Laboratory values for the predi-
alysis serum albumin, creatinine and hematocrit levels,
height, and weight were obtained from the Health Care
Financing Administration Medical Evidence Report (HCFA
Form 2728). To account for differences between laboratories
in the measurement of serum albumin, the serum albumin
categorical variables. Highly skewed variables were summa-
rized as the median and range and compared using nonpara-
metric methods. Differences in the proportions of PD versus
HD patients across severity levels of individual IDS and IPI
categories were assessed using ␹2 tests for ordered variables
(with 1 degree of freedom).
We created a multivariable regression model to identify
factors that were associated with the use of PD versus HD at
the onset of chronic outpatient dialysis therapy. The CHOICE

326
MISKULIN ET AL
Study participants were enrolled from 81 independent clin-
ics across the US. Physicians’ preferences, regional and
local economic considerations, and other factors at the
dialysis clinic level likely contribute to the dialysis modality
selection; however, these data were not collected in this
study. A hierarchical model, with patients analyzed within
clinic clusters, was used to account for the fact that these
clinic level factors are more likely to be similar for patients
within a clinic than among different clinics. The general
form of the random intercept model is as follows: [yij ϭ aj ϩ
Bxij] in which regression models for each clinic have the
same slope (B1), but the intercepts (aj) vary for each clinic.
Thus, the differences between clinics are accounted for in
the intercept terms (random effects), and the slope represents
the regression coefficients of the patient level variables,
which are the same across clinics (fixed effects). These
analyses were performed using the GLIMMIX Macro avail-
able through SAS Statistical Software, release 8.1 (Cary,
NC).
Results are reported as multivariable odds ratios (OR)
with 95% confidence intervals for PD versus HD. P values
are reported for the contribution of the categorical variables
to the model and for each level of categorical variables.
Variables of interest included demographic factors (age,
gender, and race), cause of ESRD, socioeconomic factors
(marital status, employment, level of education, social sup-
port, insurance, distance from the dialysis unit), health-
related behaviors (tobacco use, use of “street” drugs), clinical/
laboratory factors (interval from referral to a nephrologist
until start of dialysis, body mass index (BMI), serum albu-
min, serum creatinine, and hematocrit), and comorbidity
(ICED level).
tested first in multivariate analysis including only the main
effect variables of the interaction terms and were subse-
quently added to the fully adjusted model. A P value of 0.05
was considered significant for the interaction term.
A total of 625 patients (155 PD and 470 HD patients) had
complete baseline data. As a sensitivity analysis, we used 2
methods to ensure that the 625 patients, from whom the
model was derived, were not systematically different from
the rest of the cohort. The covariate with the greatest
proportion of missing data was serum albumin, missing for
288 patients: 216 HD patients (28%) and 72 PD patients
(26%). It was the sole missing covariate for 236 of the 288
patients. We re-estimated the model by categorizing serum
albumin and including the 236 patients with missing values
for serum albumin in a separate category. The category
corresponding to patients with missing values for serum
albumin was not statistically significant, suggesting these
patients were not different than the referent group. Also, the
estimates of regression coefficients did not change substan-
tially, again supporting the robustness of the model derived
from 625 patients. Secondly, we used multiple imputations
(SAS version 8.1) to impute values for patients with missing
covariates, a method described elsewhere.23 The model was
re-estimated using 5 sets of imputed values for missing data.
The resultant regression coefficients and standard errors
from the models were averaged to produce a single OR and
95% confidence interval for each covariate. There were no
significant differences between the model of 625 patients
and the model of 1,038 patients with imputed values, the
latter not shown. This further supports the robustness of the
model derived from 625 patients with full baseline data.
Finally, we also re-estimated the model after exclusion of the
2 New Haven dialysis units with a disproportionate number
of PD patients to ensure results were consistent without
these dialysis units.
In constructing the multivariable model, each variable
with a significance level of P Ͼ 0.05 was removed one at a
time, and the model was re-estimated. A likelihood ratio test
statistic of Յ 0.05 indicated that the removed variable
contributed significantly to the model, in which case it was
retained in the final model. Gender was forced into the final
model because of its clinical importance. To control for
potential biases inherent in stepwise variable selection, we
also fit a model with all variables, and there were no large
changes in the individual regression coefficients or the
overall fit of the model.
The ICED level is an aggregate index of comorbidity. To
determine whether some IDS and IPI categories were inde-
pendently associated with modality use, we developed 2
additional multivariable models, in which variables for all
19 IDS categories or all 11 IPI categories, respectively, were
substituted for the ICED level. These models were fully
adjusted for all variables from the final model.
We created 2 interaction terms. We hypothesized that
modality use might differ across age groups according to
comorbidity. It is plausible that older patients with more
severe comorbidity would be less able to perform PD than
younger patients with the same level of comorbidity. Sec-
ondly, we hypothesized that modality use would differ
across ICED groups in patients with diabetes versus other
causes of ESRD. In general, as patients become sicker with
comorbidity, they are less able to perform self-care, and
thus, we would expect that there would be relatively fewer
patients treated with PD as comorbidity increased. However,
we hypothesized that diabetics with ICED 3 scores would be
more likely than nondiabetics with ICED 3 scores to receive
PD due to problems of obtaining vascular access, as much of
the comorbidity in diabetics relates to atherosclerotic vascu-
lar disease. We classified the cause of ESRD as a binary
variable (diabetes v other causes) and created interaction
terms with each of the ICED levels. Interaction terms were
RESULTS
Recruitment
Approximately two-thirds of eligible patients
were enrolled from the participating dialysis units.
Eligible patients enrolled were similar to eligible
patients not enrolled with regard to gender and
age. A total of 1,041 incident dialysis outpatients
were enrolled from 81 dialysis centers; 279 PD
dialysis (27%) and 762 HD patients (73%). Three
HD patients did not have a completed ICED at
baseline and were excluded from analyses in this
report. Nine hundred and twenty-one patients
(89%) were from 79 dialysis units affiliated with
DCI (188 PD and 733 HD patients). The remain-

COMORBIDITY OF INCIDENT PD VERSUS HD PATIENTS
327
ing 117 patients (11%) were from 2 dialysis units
located in New Haven (91 PD and 26 HD).
Patients were enrolled a median of 45 days from
initiation of chronic dialysis (98% within 4
months).
Baseline Characteristics Associated With
Modality Use
Table 1 shows baseline characteristics of pa-
tients treated with PD and HD, without adjust-
ment. PD patients were on average 5 years
Table 1. Comparison of Baseline Characteristics by Dialysis Modality
Overall (n ϭ 1,038)
PD (n ϭ 279)
HD (n ϭ 759)
P Value
Demographic factors
Age (yr)
Female (%)
57.9
46
53.8
44
59.4
46
Ͻ 0.001
0.54
Race (%)
Ͻ 0.001
Caucasian
African American
Other
67
28
5
77
19
4
63
32
5
Cause of ESRD (%)
Glomerular disease
Hypertension
Diabetes
Polycystic kidney disease
Other
Ͻ 0.0001
16
17
47
4
20
9
47
6
15
20
47
3
16
18
15
Socioeconomic factors (%)
Married*
Working†
Education Ն high school‡
Living alone
Insured§
Distance from dialysis unit (Ͼ30 miles)
Health-related behaviors (%)
Smoking status
56
14
70
36
94
13
68
28
82
36
95
28
53
9
66
36
93
8
0.001
0.001
0.001
0.86
0.43
Ͻ 0.001
0.09
Never smoked
Quit
Current
39
46
15
19
38
51
12
25
40
44
16
17
Prior or current use of street drugs
Clinical/laboratory factors
Median interval from nephrology referral
to start of dialysis (mo)
BMI (kg/m2)࿣
0.004
10.2
27.0
(6.7)
3.51
(0.69)
8.4
(3.3)
7.68
(3.34)
28.3
(6.0)
15.9
26.5
(5.7)
3.79
(0.59)
8.3
(2.9)
7.53
(2.92)
29.7
(5.9)
8.3
27.2
(7.0)
3.40
(0.69)
8.4
(3.4)
7.74
(3.49)
27.8
(6.0)
0.005¶
0.14
Serum albumin (g/dL)#
Serum creatinine (mg/dL)
GFR (mL/min/1.73 m2)**
Hematocrit (%)
Ͻ 0.001
0.72
0.39
Ͻ 0.001
NOTE. Sample size for demographic factors ranged from 949 to 1,038; for cause of ESRD, 1,031; for socioeconomic
factors, 963 to 1,005; for health-related behaviors, 927 to 974; for clinical and laboratory factors, 502 to 962.
*Married v single or divorced or widowed.
†Working part-time or full-time v retired/disabled/unemployed.
‡Greater than or equal to a high school level of education v less than a high school education.
§Insured with Medicare, Medicaid, or private insurance.
࿣BMI based on formula using data from pre-ESRD setting (Form 2728): BMI ϭ weight (kg)/height 2 (m2).
¶T test after log transformation of the data.
#Standardized serum albumin (see text).
**Calculated using the MDRD formula. See text for details.

328
MISKULIN ET AL
younger than HD patients. There was no gender
difference between modalities. A higher propor-
tion of PD patients were Caucasian and a lower
proportion were African American.
The cause of renal disease differed by modal-
ity. Ahigher proportion of patients with glomeru-
lar disease received PD, while a significantly
lower proportion of patients with hypertensive
renal disease received PD. The proportion of
patients with diabetes as the cause of ESRD did
not differ among modalities.
A large number of socioeconomic factors dif-
fered by modality. A larger proportion of PD
patients were married, employed, had attained at
least a high school level of education, or lived far
(Ͼ 30 miles) from the nearest dialysis unit.
There was no difference in the proportion of
patients who were insured or lived alone.
Health-related behaviors differed between mo-
dalities. A higher proportion of ex-smokers were
treated with PD, while current smokers more
often received HD. More PD patients admitted to
having used “street” drugs.
visit to a nephrologist until starting dialysis was
longer for PD patients. The level of renal func-
tion (GFR calculated using MDRD formula) did
not differ between PD and HD patients at the
start of dialysis. PD patients had higher predialy-
sis values of standardized serum albumin and
hematocrit. Mean serum creatinine and BMI
prior to the start of dialysis did not differ be-
tween modalities.
Prevalence and Severity of Comorbid
Conditions
The distribution of IDS and IPI categories and
ICED levels in PD and HD patients are shown in
Figs 1 through 3. PD patients had significantly
(P Ͻ 0.05) lower IDS scores for 5 of 19 IDS
categories (congestive heart failure, arrhythmias,
other heart disease, respiratory diseases, and gas-
trointestinal diseases), as shown in Fig 1. Also,
the mean number of IDS conditions reported per
patient was significantly less for PD patients (5.0
v 6.0, P Ͻ 0.001).
PD patients had significantly lower IPI scores
than HD patients (P Ͻ 0.05) for 7 of 11 IPI
categories (circulatory, respiratory, neurological,
mental status, feeding, ambulatory, and visual
Clinical and laboratory factors that may partly
reflect predialysis care also differed between the
modalities. The median interval from the first
Fig 1. Distribution of IDS categories in PD and HD patients at baseline in the CHOICE Cohort Study (left bars, PD;
right bars, HD). Each bar represents the percent of patients with disease, with shading corresponding to the 3
severity levels (IDS 1, 2, or 3). The remaining percent are those without disease in that category (IDS 0). *The
distribution of IDS scores (0-3) was significantly (P < 0.05 by ␹2 tests) lower in PD v HD patients for the following IDS
categories: congestive heart failure, arrhythmias, other heart disease, respiratory diseases, and gastrointestinal
diseases. IHD, ischemic heart disease; CHF, congestive heart failure; arrhythmia; other, other heart disease; htn,
hypertension; CVD, cerebrovascular disease; PVD, peripheral vascular disease; DM, diabetes mellitus; RESP,
respiratory; msk, musculoskeletal; neuro, nonvascular nervous system; GI, gastrointestinal disease; hep, hepato-
biliary disease; GU, urogenital disease; cancer, malignancy; ophth, ophthalmologic; HIV, HIV and AIDS; Heme,
hematologic; coag, anticoagulation.

COMORBIDITY OF INCIDENT PD VERSUS HD PATIENTS
329
Fig 2. Distribution of IPI categories in PD and HD patients at baseline in the CHOICE Cohort Study (left bars, PD;
right bars, HD). Each bar represents the percent of patients with impairment, with shading corresponding to the 2
severity levels (IPI 1 or 2). The remaining percent are those without impairment in that category (IPI 0). *The
distribution of IPI scores (0-2) was significantly (P < 0.05) lower among PD patients in the following IPI categories:
circulatory, respiratory, neurologic, mental status, feeding, ambulatory, and visual impairments.
impairments), as shown in Fig 2. Overall, PD
patients had a lower mean number of IPI condi-
tions per patient (1.0 v 1.9, P Ͻ 0.001).
Because of their lower IDS and IPI scores, PD
patients had lower ICED scores (P Ͻ 0.001) (Fig
3). Only 48% of PD patients had ICED levels of
2 or 3, whereas 70% of HD patients had ICED
levels of 2 or 3.
older, more often Caucasian, more often had
diabetes as the cause of ESRD, less often had
glomerular disease as the cause of ESRD, and
less often worked, had a shorter interval from
nephrologist referral to the start of dialysis, had
lower predialysis serum albumin and serum cre-
atinine levels, and started dialysis at higher lev-
els of renal function (calculated GFR). Even
though the number of IDS and IPI categories per
patient is not part of the ICED scoring algorithm,
patients with higher ICED levels had more IDS
and IPI categories per patient. Again, the HD
patients had more IDS and IPI categories re-
ported per patient than PD patients across each
ICED subgroup.
Relationship of Comorbidity to Other Baseline
Characteristics
We compared patient factors across ICED
severity levels in PD and HD groups separately
(Table 2). Associations and trends were generally
in the same direction for both modalities for the
following: patients with higher ICED levels were
Fig 3. Distribution of co-
morbidity (ICED Levels) in
PD v HD patients at baseline
in the CHOICE Cohort Study.
The proportion of PD v HD
patients within ICED levels
(0-3) was significantly differ-
ent (P < 0.001).

330
MISKULIN ET AL
Table 2. Relationship of Comorbidity to Other Baseline Factors
PD
HD
Level
0-1
Level
2
Level
3
Level
0-1
P
Level 2
Level 3
P
n ϭ 146
n ϭ 71
n ϭ 62
Value
n ϭ 229
n ϭ 296
n ϭ 234
Value
Demographic factors
Age (yr)
Female (%)
51
43
57
49
58
42
Ͻ 0.001
0.60
56
45
62
45
59
49
Ͻ 0.001
0.60
Race (%)
0.15
0.15
Caucasian
African American
Other
72
21
7
82
17
1
81
18
1
58
36
6
62
33
5
70
26
4
Cause of ESRD (%)
Glomerular disease
Hypertension
Diabetes
Polycystic kidney disease
Other
Ͻ 0.001
Ͻ 0.001
30
9
32
9
14
10
63
3
3
10
61
3
22
23
34
6
16
21
47
2
7
16
59
3
20
10
23
15
14
15
Socioeconomic factors (%)
Married
Working
Education Ͼ high school
Living alone
Insured
72
37
82
39
90
26
60
19
80
44
94
27
67
18
85
31
95
35
0.23
0.005
0.74
0.40
0.06
0.48
50
13
71
34
95
9
54
9
64
40
96
7
53
5
65
33
94
9
0.60
0.02
0.25
0.21
0.80
0.80
Distance to dialysis unit Ͼ 30
miles
Health-related behaviors (%)
Smoking Status
Never Smoked
Quit
Current
Prior or current use of street
drugs
0.19
0.80
44
43
13
26
31
58
11
25
30
60
9
41
39
41
18
41
47
43
15
18
15
16
16
23
0.90
0.70
Clinical/laboratory factors
Median time from nephrology
referral to start of dialysis
(mo)
21.8
13.9
12.6
0.03*
12.9
7.9
4.9
0.03*
BMI (kg/m2)
26.8
(5.8)
3.7
(0.60)
8.6
(2.8)
30.1
(5.9)
7.09
(2.10)
26.6
(6.3)
3.6
(0.46)
8.3
(3.3)
28.4
(4.5)
7.53
(3.18)
25.4
(5.0)
3.4
(0.61)
7.6
(2.7)
30.0
(7.1)
8.60
(3.94)
0.30
26.9
(6.6)
3.3
(0.6)
8.9
(3.2)
27.3
(7.3)
7.32
(3.28)
27.6
(7.6)
3.3
(0.6)
8.6
(3.6)
28.0
(5.1)
7.51
(3.43)
27.0
(6.6)
3.1
(0.6)
7.6
(3.1)
28.1
(5.9)
8.43
(3.67)
0.45
0.004
Ͻ 0.001
0.40
Serum albumin (g/dL)
Serum creatinine (mg/dL)
Hematocrit (%)
0.007
0.08
0.14
GFR mL/min/1.73 m2
0.001
0.005
Median no. of comorbid conditions
(per patient)†
IDS categories
IPI categories
4.0
0‡
6.0
1.0
7.1
2.0
Ͻ 0.001
Ͻ 0.001
5.0
0‡
6.0
2.0
7.0
3.0
Ͻ 0.001
Ͻ 0.001
NOTE. Data are presented as proportions (%) or mean (SD) for each ICED level for PD or HD patients, separately.
*Statistical tests of the differences across ICED subgroups were performed using analysis of variance of the data after
natural log transformation.
†Medians are used because the IPI was not normally distributed. Significance was tested across ICED levels using
nonparametric methods (Kruskal-Wallis test).
‡By definition, the IPI score of a patient with ICED 0-1 is 0.

COMORBIDITY OF INCIDENT PD VERSUS HD PATIENTS
331
Factors Associated With Modality Selection in
the Multivariable Model
were unemployed, disabled, or retired, those
who were employed had significantly higher
odds of being treated with PD (OR 4.4 [2.3 to
8.7]). Patients living more than 30 miles from the
nearest dialysis unit also had significantly higher
odds of receiving PD (OR of 5.3 [2.8 to 10.2]).
Marital status and level of education did not
remain significant predictors of PD use after
adjustment for other factors.
Higher serum albumin concentration and lower
BMI were independently associated with treat-
ment with PD. For a 0.1 g/dL increase in serum
albumin, the odds of PD were higher (OR 1.17
[1.12 to 1.23]) and for each 1.0 kg/m2 increase in
BMI, the odds of PD were lower (OR 0.94 [0.91
to 0.98]).
Table 3 shows factors associated with the
modality selection after multivariable adjust-
ment, as derived from 625 patients (155 PD and
470 HD patients) with complete baseline data. In
comparison to patients aged 45 to 65, the odds of
an older patient (Ͼ 65) receiving PD were signifi-
cantly less (OR of 0.53 [0.30 to 0.93]). There
was no difference in modality use between gen-
ders. In comparison to Caucasians, the odds of
PD in African Americans was also significantly
less (OR 0.35 [0.18 to 0.68]). In comparison to
patients with glomerular diseases, the odds of PD
in patients with hypertensive nephrosclerosis were
significantly less (OR 0.30 [0.12 to 0.80]), but
were greater for patients with diabetic renal
disease (OR 2.6 [1.3 to 5.0]).
After controlling for other factors, higher ICED
level was associated with a lesser relative prefer-
ence for PD versus HD. Compared with those
with no to mild comorbidity (ICED 0-1), the
odds of PD in patients with moderate comorbid-
Certain socioeconomic factors remained asso-
ciated with modality selection after multivari-
able adjustment. As compared with patients who
Table 3. Multivariable Analysis of Factors Associated With the Use of PD v HD at the Start of Dialysis
Variable
Odds Ratio (PD v HD)
95% Confidence Interval
P Value
Age (yr)
Ͻ45 yr
45-65
Ͼ65 yr
Female v male
Race
0.08*
0.54
—
0.03
0.92
0.002*
0.81
1.0
0.53
1.03
0.42–1.6
Reference
0.30–0.93
0.64–1.63
Caucasian
African American
Other
Reference
0.35
0.18–0.68
0.65–9.5
0.002
0.19
2.5
Cause of ESRD
Glomerular diseases
Hypertension
Diabetes mellitus
Polycystic kidney disease
Other
Employment (workers)
Distance from clinic (Ͼ30 miles)
Serum albumin (per 0.1 g/dL)
BMI (per 1 kg/m2)
ICED level
Ͻ 0.0001*
—
0.02
0.005
0.97
Reference
0.30
2.6
1.0
0.12–0.80
1.3–5.0
0.03–3.2
0.76–3.9
2.3–8.7
2.8–10.2
1.12–1.23
0.91–0.98
0.98
1.7
4.4
5.3
1.17
0.94
0.20
Ͻ 0.0001
Ͻ 0.0001
Ͻ 0.0001
0.003
0.0005*
—
ICED 0-1
ICED 2
ICED 3
1.0
0.31
0.50
Reference
0.17–0.56
0.28–0.90
0.0001
0.021
NOTE. The model was estimated from 625 patients (155 PD and 470 HD) with complete data. The final model includes
only those variables that remained significant (P Ͻ 0.05) after adjustment. The variables that were tested, but did not remain
significant in the adjusted model, were marital status (married v other), education, prior street drug use, smoking status, living
alone, time from first nephrologist visit to first dialysis, predialysis hematocrit, and serum creatinine. Odds ratios of Ͻ1.0
indicate a lesser use of PD than the reference group. For examples, the odds of PD in patients greater than 65 years old are
0.53 times the odds of PD in patients between the ages 45 to 65 (referent group).
*The contribution of the variable as a whole to the model was tested for statistical significance.

332
MISKULIN ET AL
ity (ICED 2) were significantly lower (OR 0.31
[0.17 to 0.56]) and were also lower for those with
severe comorbidity (ICED 3) (OR 0.50 [0.28 to
0.90]). The difference in OR for PD between
patients with ICED 3 versus ICED 2 was nonsig-
nificant (P ϭ 0.14).
malignancy (more PD) and gastrointestinal con-
ditions (less PD) were significantly and indepen-
dently related to modality. Of the IPI categories,
only one category (respiratory impairment) was
significantly related to modality (less PD use).
These results suggest that, in general, the pres-
ence of an individual disease or impairment does
not influence modality use, but the combination
of diseases and impairments, weighted according
to severity (ie, ICED score), does influence mo-
dality use.
The use of PD did not differ among older
versus younger patients as comorbidity severity
increased. The interaction term (AGExICED)
was not significant in the model adjusted for
main effects (P value 0.17), nor in the fully
adjusted model (P value 0.89). We also tested for
a differential use of PD with increasing ICED
levels in diabetics versus nondiabetics. We used
a binary variable in place of the 5-leveled cause
of ESRD variable, and thus the referent group
changed from glomerulonephritis to nondiabetic
causes of ESRD grouped together. The interac-
tion term (DMxICED) was significant (P Ͻ
0.001); however, the fit of the model was un-
changed from the final adjusted model that used
a 5-level categorical variable for the cause of
ESRD. The significance of the interaction term
likely reflects the loss of information resulting
from the use of a binary variable for cause of
ESRD, and we thus chose not to include it in the
final model.
The final model was also adjusted for clinic.
We found 6 clinics in which the odds of receiv-
ing PD were significantly higher than in other
clinics. This result suggests that there was varia-
tion in the use of PD at clinics in the CHOICE
Study, although it may not reflect true clinical
practice because of the deliberate oversampling
of PD patients from all clinics. Although the
hierarchical modeling attempts to control for the
differential use of PD at various clinics, we
re-estimated the full model after excluding the 2
dialysis units from New Haven that had a higher
proportion of PD patients than the DCI dialysis
units. The magnitude and statistical significance
of the regression coefficients were the same as
the model that includes the New Haven dialysis
units except for hypertensive nephrosclerosis,
which was no longer significant (data not shown).
DISCUSSION
Within the US there is significant variation in
PD utilization. Factors responsible for variation
are not completely understood, but may include
patients’ clinical characteristics, medical judge-
ment, physicians’ and patients’ preferences, and
differences in physician and facility reimburse-
ment.24 A detailed characterization of differences
in PD versus HD patients in the present study
shows the extent by which these patient groups
differ at the onset of chronic dialysis therapy.
We observed fewer and less severe comorbid
conditions in the PD group (P Ͻ 0.001). Overall,
48% of PD patients had moderate or severe
comorbid conditions (ICED levels 2-3) versus
70% of HD patients. In addition, fewer PD
patients were affected with congestive heart fail-
ure, arrhythmias, other heart diseases, respira-
tory diseases, and gastrointestinal diseases, and
the severity of disease was also lower for those
with disease in these categories as compared
with HD patients. PD patients also had fewer
physical impairments, which may serve as an
even stronger predictor of adverse outcomes than
medical diagnoses.25,26
After controlling for other significant factors,
comorbidity remained a strong independent pre-
dictor of modality use. The OR for selecting PD
over HD for ICED 2 versus ICED 0-1 was 0.31
(0.17 to 0.56) and for ICED 3 versus ICED 0-1, it
was 0.50 (0.28 to 0.90). Overall, we interpret the
relationship between more severe comorbidity
and the lesser preference for PD as a reflection of
the greater physical and mental demands re-
quired of patients who perform PD and of the
greater acceptance of the more dependent life-
style of HD for patients who are more con-
strained by comorbid illness.
Relationship of IDS and IPI Categories and
Modality Use
We assessed the contribution of individual
IDS and IPI categories to modality use in sepa-
rate multivariable models. Only 2 categories,
It was interesting to find that PD use did not

COMORBIDITY OF INCIDENT PD VERSUS HD PATIENTS
333
decrease incrementally as comorbidity sever-
ity increased. Possibly, this may reflect a “co-
morbidity threshold” in modality selection,
whereby HD is recommended preferentially for
patients with either moderate or severe comorbid-
ity. Alternatively, there may be some types of
patients with severe comorbidity for whom PD is
preferable. For example, anticipated difficulty in
obtaining vascular access in patients with periph-
eral vascular disease or difficulty with hemody-
namic shifts associated with HD in patients with
cardiac disease may favor the selection of PD in
some patients with diabetes or severe comorbid-
ity. We did not directly survey patients and
physicians to determine which factors contrib-
uted to modality selection in this study; thus, it is
difficult to tell which of these alternative explana-
tions is correct.
The association of serum albumin to modality
selection is strong, even after adjustment for
comorbidity, and this has many possible mean-
ings.27 A low serum albumin may reflect the
presence of comorbid conditions, malnutrition,
or an underlying inflammatory and catabolic
state, which has not yet resulted in clinical dis-
ease that is captured by the ICED, but which
leads physicians and patients to select HD versus
PD. The fact that both factors (ICED and serum
albumin concentration) were independently asso-
ciated with modality selection suggests both fac-
tors provide a better assessment of the full extent
of comorbidity than either factor alone.
clearance in obese patients. Even after adjust-
ment for comorbidity and other factors, patients
with diabetic renal disease were most likely to
use PD. As described earlier, this may reflect a
relative preference for PD in patients with more
severe comorbidity, which was not accounted for
by the ICED score (for example, anticipated
difficulty with HD due to peripheral vascular
disease or cardiac disease). On the other hand,
patients with hypertensive nephrosclerosis were
least likely to choose PD, although this finding
did not remain significant when re-estimated
without the 2 dialysis units from New Haven.
Nonetheless, the direction of the association still
suggests less PD use in those with hypertensive
nephrosclerosis, which may be explained by the
higher incidence of this diagnosis among African
Americans (who were less likely to use PD).
Practical considerations, such as longer distance
from the dialysis unit and employment, were
independently associated with a greater relative
preference for PD. As reported in the Dialysis
Morbidity and Mortality Study (DMMS) Study,29
we find longer pre-ESRD care was associated
with the use of PD, and the level of renal func-
tion at first dialysis was higher for PD patients,
likely a result of the lower frequency of late
referrals in the PD group. This relationship did
not remain significant in the multivariable model,
probably because of confounding by the other
covariates that also described patients more likely
to be referred early to their nephrologist.
In addition to differences in comorbidity, we
also identified important demographic and socio-
economic differences between PD and HD pa-
tients at the start of dialysis. As in other studies,
PD patients were younger than HD patients.28
However, after adjusting for other factors in the
multivariable model, age did not contribute sig-
nificantly to modality use (P value 0.08 for the
contribution of age to the model). This would
suggest that physiologic age and not chronologic
age is more important in the selection of dialysis
modality in the US. African Americans were less
than half as likely to choose PD, which suggests
unmeasured social or cultural differences, as the
relationship remained significant even after ad-
justment for all other factors. As expected, PD
was less frequently selected for patients with
higher BMI, likely due to the physicians’ con-
cerns of achieving adequate catheter function or
To our knowledge, the Case-Mix Severity
Study of 198928 (a special study of the 1989 US
Renal Data System [USRDS]) is the only other
study in which detailed comorbidity data was
reported from a large sample of incident dialysis
patients. In that study, trained chart reviewers
also abstracted data from the medical record. As
in the CHOICE Cohort Study, the Case-Mix
Severity Study showed that PD patients had
fewer comorbid conditions. However, there are
some important differences to consider between
the method of comorbidity assessment in the
CHOICE Cohort Study and in the USRDS Case-
Mix Severity Study. The ICED is a more compre-
hensive list of medical conditions and impair-
ments and records severity levels for each
category of disease and impairment. Measuring
the severity or extent of a comorbid illness is
superior to simply identifying the presence of a

334
MISKULIN ET AL
disease when assessing patient risk.30 In addi-
tion, items of the ICED are aggregated into a
final score based on severity, thus enabling the
effects of more than one comorbid condition to
be considered, while conserving statistical power
to adjust for the many other factors that differ
between PD and HD.
A recent report emphasizes the need for com-
prehensive chart–review-based comorbidity data
collection to provide accurate and complete re-
porting of comorbid conditions. Longenecker et
al31 compared the validity of HCFA Form 2728
data for 17 medical conditions with data col-
lected from medical record review using the
ICED in the CHOICE Study. Overall, the sensi-
tivity of Form 2728 averaged across all 17 condi-
tions was low (58%), suggesting significant un-
derreporting of comorbid conditions on Form
2728. The DMMS Wave 2 reported similar re-
sults.32
We have shown that PD patients are less sick
at the onset of ESRD; consequently, if case-mix
factors are not measured and controlled for, sur-
vival estimates will be biased in favor of the PD
group. When incident patients are studied and
the nonproportionality of the hazard ratio of PD
and HD over time is accounted for, the most
recent analyses of USRDS11 suggest that PD is
associated with superior survival in the first 2
years after dialysis initiation in all subgroups
except female and male diabetics over the age of
55, in whom the risk of mortality is higher and
equal, respectively, to that of HD patients. After
the first year, the survival between groups be-
comes equivalent, except for older female dia-
betics treated with PD who remain at a higher
mortality risk than HD patients. Whether this
changing mortality ratio of PD:HD is the effect
of changes in peritoneal membrane transport
characteristics and loss of residual renal function
in PD patients over time or simply represents a
higher initial death rate of sicker HD patients
with a lag in death rate of the more healthy PD
patients is unknown,11 because these analyses
did not measure baseline comorbidity in the PD
and HD groups. Because results are likely to
overestimate a benefit of PD, it is entirely pos-
sible that PD is equivalent or even inferior to
HD. In fact, a recent Canadian study15 reported
similar findings as in the US, with an early
survival advantage of PD; however, after account-
ing for differences in comorbidity between the 2
groups (HD patients were sicker), this apparent
survival advantage of PD was no longer present.
Analyses of the CHOICE Study, adjusted for the
case-mix factors that differ between these groups
at baseline, will enable an unbiased comparison
of survival in PD- versus HD-treated patients
over time.
There are strengths and limitations of these
analyses. To our knowledge, this is the most
comprehensive comparison of comorbidity and
other factors in PD and HD patients at the onset
of dialysis. The CHOICE cohort appears repre-
sentative of the 1997 US incident dialysis popu-
lation,33 apart from the deliberate oversampling
of PD patients and a slightly younger mean age.
In addition, despite the deliberate oversampling
for PD patients in this study, this cohort of PD
patients is representative of the US incident PD
population.11 Detailed comorbidity data was col-
lected from chart review and weighted according
to clinically defined severity scales. Moreover,
all laboratory data consisted of predialysis values
and baseline demographic data, and records for
comorbidity review were restricted to those re-
corded prior to or within 4 months of starting
dialysis. Thus, the observed relationships truly
reflect the association of modality with baseline
factors rather than outcomes of dialysis care.
There are some limitations of our analysis.
First, baseline factors were assessed at the onset
of dialysis, which for many patients was after the
time of modality selection. Possibly, comorbid-
ity might have differed in the several months
before the onset of ESRD, when patients and
providers were deciding on the modality. Studies
of the evolution of comorbid conditions in pa-
tients with earlier stages of renal disease (chronic
renal insufficiency) will be necessary to answer
this question. Second, although the CHOICE
Cohort consists of over 1,000 incident dialysis
patients, the number of patients within some
subcategories was small, because the ICED
records a large number of IDS and IPI catego-
ries, further divided into 3 or 4 severity levels.
Statistical power may not have been adequate to
show differences that truly exist. Also, it is
possible that certain severe comorbid conditions
and/or patterns of comorbid diseases may influ-
ence physicians or patients to prefer PD, which
may not have been captured in the ICED score

COMORBIDITY OF INCIDENT PD VERSUS HD PATIENTS
335
MD and Alan Kliger, MD) for their support of and involve-
ment in the CHOICE Study. We also recognize and thank the
CHOICE ScientificAdvisory Committee (Paul Eggers, PhD,
Sheldon Greenfield, MD, H. Keith Johnson, MD, Nathan
Levin, MD, Robin Luke, MD, Richard Rettig, PhD, and Paul
Whelton, MD, MSc) for their scientific guidance and over-
sight of the CHOICE Study. We are also grateful to Patti
Pingree, BS, RN, at New England Medical Center who
abstracted comorbidity data from chart review, and Sheldon
Greenfield for his advice and expertise with the ICED.
itself. Third, we examined an extensive list of
sociodemographic and clinical factors that influ-
ence the decision to use PD or HD, but it is likely
that there are some cultural, social, and other
unmeasured factors at the patient level that are
not represented. Also, the physician’s prefer-
ences relating to his or her familiarity of one
modality over the other, patient education pro-
grams prior to the start of ESRD, physician
reimbursement, and other economic consider-
ations were not captured in the model, but have
been shown to be influential in PD utilization
trends.24,34 We did not collect this information
and, thus, cannot estimate the relative influence
of these factors over factors related more directly
to the patient’s medical and social/cultural cir-
cumstances. Fifth, PD utilization rates and the
factors motivating PD use in the US differ from
those in other countries; hence, these results are
likely only generalizable to this country.
REFERENCES
1. Held PJ, Port FK, Turenne MN, Gaylin DS, Ham-
burger RJ, Wolfe WA: Continuous ambulatory peritoneal
dialysis and hemodialysis: Comparison of patient mortality
with adjustment for comorbid conditions. Kidney Int 45:
1163-1169, 1994
2. Lowrie EG, Lew NL: Death risk in hemodialysis
patients: The predictive value of commonly measured vari-
ables and an evaluation of death rate differences between
facilities. Am J Kidney Dis 15:458-482, 1990
3. Mailloux LU, Napolitano B, Alessandro G, Bellucci G,
Mossey RT, Vernace MA, Wilkes BM: The impact of co-
morbid risk factors at the start of dialysis upon the survival
of ESRD patients. ASAIO J 42:164-169, 1996
In summary, we compared baseline factors in
a large, representative cohort of incident HD and
PD patients participating in the CHOICE Cohort
Study. We used the ICED to provide a detailed
description of the presence and severity of comor-
bid conditions at the onset of ESRD. Using
multivariable analysis, we found that PD patients
had less severe comorbidity, and that this relation-
ship persisted after controlling for all other fac-
tors associated with modality selection. How-
ever, the absence of a graded relationship, in
which PD use decreased incrementally as comor-
bidity severity increased, suggests that there may
be a subgroup of patients with severe comorbid-
ity who are preferentially treated with PD. Fol-
low-up results of the CHOICE Cohort Study,
comparing outcomes of incident HD and PD
patients, with adjustment for baseline comorbid-
ity and other factors, should clarify the relative
contribution of baseline and treatment factors to
dialysis outcomes.
4. Miskulin DC, Meyer KB, Athienites NV, Martin AA,
Marsh JV, Fink NF, Coresh J, Powe NR, Klag M, Levey AS:
The Index of Coexistent Disease (ICED) is an independent
predictor of death in a prospective ESRD cohort. J Am Soc
Nephrol 10:175A, 1999 (abstr)
5. Owen WF, Lew NL, Liu Y, Lowrie EG, Lazarus MJ:
The urea reduction ratio and serum albumin concentration as
predictors of mortality in patients undergoing hemodialysis.
N Engl J Med 329:1001-1006, 1993
6. Rocco MV, Soucie JM, Reboussin DM, McClellan
WM: Risk factors for hospital utilization in chronic dialysis
patients. J Am Soc Nephrol 7:889-896, 1996
7. US Renal Data System: Excerpts From the USRDS
1999 Annual Data Report. Am J Kidney Dis 34:S74-S86,
1999 (suppl 1)
8. Keane WF, Collins AJ: Influence of co-morbidity on
mortality and morbidity in patients treated with hemodialy-
sis. Am J Kidney Dis 24:1010-1018, 1994
9. Collins AJ, Ma JZ, Umen A, Keshaviah P: Urea index
and other predictors of hemodialysis patient survival. Am J
Kidney Dis 23:272-282, 1994
10. Held PJ, Port FK, Wolfe RA, Stannard DC, Carroll
CE, Daugirdas JT, Bloembergen WE, Greer JW, Hakim RM:
The dose of hemodialysis and patient mortality. Kidney Int
50:550-556, 1996
11. Collins A, Hao W, Xia H, Ebben J, Everson S,
Constantini M, Ma J: Mortality risks of peritoneal dialysis
and hemodialysis. Am J Kidney Dis 34:1065-1074, 1999
12. Bloembergen WE, Port FK, Mauger EA, Wolfe RA:
A comparison of mortality between patients treated with
hemodialysis and peritoneal dialysis. J Am Soc Nephrol
6:177-183, 1995
13. Fenton SA, Schaubel DE, Desmeules M, Morrison
HI, Mao Y, Copleston P, Jeffrey JR, Kjellstrand CM: Hemo-
dialysis versus peritoneal dialysis: A comparison of adjusted
mortality rates. Am J Kidney Dis 30:334-342, 1997
ACKNOWLEDGMENT
We thank the patients, staff, and physicians who partici-
pated in the CHOICE Study at Dialysis Clinic, Inc, St.
Raphael’s Hospital and New Haven CAPD. We also acknowl-
edge and thank the CHOICE-DCI Clinical Liaison Commit-
tee (Thomas Depner, MD, H. Keith Johnson, MD, K Shashi
Kant, MD, Klemens Meyer, MD, Richard Sherman, MD,
Edward Schroeder, MD, Pradip Teredesai, MD, John Van
Stone, MD, and Philip Zager, MD) and the New Haven
CAPD Clinical Liaison Committee (Frederic Finkelstein,

336
14. Foley RN, Parfrey PS, Harnett JD, Kent GM, O’Dea
R, Murray DC, Barre PE: Mode of dialysis therapy and
mortality in end-stage renal disease. J Am Soc Nephrol
9:267-276, 1998
15. Murphy SW, Foley RN, Barrett BJ, Kent GM, Mor-
gan J, Barre P, Campbell P, Fine A, Goldstein MB, Handa
SP, Jindal KK, Levin A, Mandin H, Muirhead N, Richardson
R, Parfrey P: Comparative mortality of hemodialysis and
peritoneal dialysis in Canada. Kidney Int 57:1720-1726,
2000
16. Vonesh EF, Moran J: Mortality in end-stage renal
disease: A reassessment of differences between patients
treated with hemodialysis and peritoneal dialysis. J Am Soc
Nephrol 10:354-365, 1999
MISKULIN ET AL
JH: Non-medical factors that impact on ESRD modality
selection. Kidney Int 43:S120-S127, 1993
25. McClellan W, Anson D, Birkeli K, Tuttle E: Func-
tional status and quality of life: Predictors of early mortality
among patients entering treatment for end-stage renal dis-
ease. J Clin Epidemiol 44:83-89, 1991
26. Miskulin DC, Meyer KB, Athienites NV, Martin AA,
Marsh JV, Fink NF, Coresh J, Powe NR, Klag M, Levey AS:
The contribution of medical conditions and physical impair-
ments to a comorbidity index: The Index of Coexistent
Disease (ICED). J Am Soc Nephrol 10:175A, 1999 (abstr)
27. Kaysen GA, Rathore V, Shearer GC, Depner TA:
Mechanisms of hypoalbuminemia in hemodialysis patients.
Kidney Int 48:510-516, 1995
17. Powe NR, Klag MJ, Sadler JH, Anderson GF, Bass
EB, Briggs WA, Fink NE, Levey AS, Levin NW, Meyer KB,
Rubin HR, Wu AW: Choices for healthy outcomes in caring
for end stage renal disease. Semin Dial 9:9-11, 1996
18. Levey A, Greene T, Kusek J, Beck G: A simplified
equation to predict glomerular filtration rate from serum
creatinine. J Am Soc Nephrol 11:155A, 2000 (abstr)
19. Feinstein A: The pre-therapeutic classification of co-
morbidity in chronic disease. J Chron Dis 23:455-468, 1970
20. Miskulin DC, Athienites NV, Yan G, Martin AA, Ornt
DB, Kusek J, Meyer K, Levey AS: Comorbidity assessment
using the Index of Coexistent Diseases (ICED) in a multi-
center clinical trial: The Hemodialysis (HEMO) Study. Kid-
ney Int 60:1498-1510, 2001
21. Nicolucci A, Cubasso D, Labbrozzi D, Mari E, Imic-
ciatore P, Procaccini DA, Forcella M, Stella I, Querques M,
Pappani A, Passione A, Strippolis P: Effect of coexistent
diseases on survival of patients undergoing dialysis. ASAIO
J 41:M291-M295, 1995
22. Athienites NV, Miskulin DC, Fernandez GF, Bun-
napradist S, Simon G, Landa M, Schmid C, Greenfield S,
Levey AS, Meyer KB: Comorbidity assessment in hemodi-
alysis and peritoneal dialysis using the Index of Coexistent
Disease (ICED). Semin Dial 13:320-326, 2000
28. Anonymous: U.S. Renal Data System U.S.R.D.S.
1991 Annual Data Report: Patient selection to peritoneal
dialysis versus hemodialysis according to comorbid condi-
tions. Am J Kidney Dis 20:20-26, 1992
29. US Renal Data System: USRDS 1997 Annual Data
Report, National Institutes of Health, National Institute of
Diabetes and Digestive and Kidney Diseases, Bethesda,
MD, 1998
30. Iezzoni LI: Dimensions of Risk, in Iezzoni LI (ed):
Risk Adjustment for Measuring Health Outcomes, chap 2.
Ann Arbor, MI, Health Administration Press, 1994, pp 1-29
31. Longenecker JC, Klag MJ, Coresh J, Levey AS,
Martin AA, Fink NE, Powe NR: Validation of comorbid
conditions on the End-Stage Renal Disease Medical Evi-
dence Report: The CHOICE Study. J Am Soc Nephrol
11:520-529, 2000
32. Roys EC, Port FK, Agodoa LYC, Meyers-Purkiss A,
Brown PL, Jones JT, Daugirdas JT, Pereira BJA, Golper TA,
Wolfe RA: Validation of medical evidence form with DMMS
data showing relative importance as predictor of mortality.
J Am Soc Nephrol, 10:255A, 1999 (abstr)
33. US Renal Data System: Excerpts From the USRDS
1991 Annual Data Report: Incidence and Prevalence of
ESRD. Am J Kidney Dis 34:S40-S50, 1992 (suppl 2)
34. Golper T, Vonesh E, Wolfson M, Baudoin M, Schrei-
ber M: The impact of pre-ESRD education on dialysis
modality selection. J Am Soc Nephrol 11:231A, 2000 (abstr)
23. Schafer JL: Multiple Imputations: A primer. Stat
Methods Med Res 8:3-15, 1999
24. Nissenson AR, Prichard SS, Cheng IKP, Gokal R,
Kubota M, Maiorca R, Riella MC, Rottembourg J, Stewart