Full text of DOI:10.1016/S0272-6386(00)70039-4

Potential Novel Predictors of Mortality in End-Stage
Renal Disease Patients With Sleep Disorders
Robert L. Benz, MD, Mark R. Pressman, PhD, Edward T. Hovick, MD,
and Donald D. Peterson, MD
● Patients with end-stage renal disease (ESRD) have an annual mortality rate exceeding 20%, although some
survive many years. The ESRD population has a high incidence of sleep disorders, including sleep apnea and
periodic limb movements in sleep (PLMS). Sleep disorders result in sleep deprivation, which can negatively affect
immune function and cardiovascular-related outcomes, common causes of death in patients with ESRD. This study
examined predictors of mortality in patients with ESRD with sleep problems. Twenty-nine consecutive patients with
ESRD reporting disrupted sleep or daytime sleepiness were studied by all-night polysomnography. All patients
were followed up until death, transplantation, or study termination. Among the variables studied, including such
previously reported predictors as serum albumin level, urea reduction ratio, and hematocrit, only the PLMS index
(PLMSI), arousing PLMSI (APLMSI), and total number of arousals per hour of sleep significantly predicted mortality.
The 20-month survival rate with a PLMSI less than 20 was greater than 90% versus 50% for a PLMSI of 20 or greater
(exact log-rank, P ؍ 0.007). For the deceased versus survivor groups, mean PLMSI was 119.1 versus 19.8 (P ؍ 0.01)
and APLMSI was 48.1 versus 7.8 (P ؍ 0.00006), with a mean survival of 10.3 versus greater than 25.5 months,
respectively (P ؍ 0.001). Median survival of patients with a PLMSI greater than 80 was only 6 months. PLMSI,
APLMSI, and total arousals per hour of sleep were strongly associated with mortality in patients with ESRD with
sleep disorders independent of other factors and may be novel predictors of near-term mortality.
ꢀ 2000 by the National Kidney Foundation, Inc.
INDEX WORDS: Mortality; predictors of mortality; end-stage renal disease (ESRD); dialysis; sleep apnea; periodic
limb movements in sleep (PLMS); sleep disorders; sleep; anemia; urea reduction ratio (URR).
problems about their sleep or daytime alert-
Editorial, p. 1221
ness.9-13 When dialysis patients with a sleep
problem were studied objectively in the sleep
laboratory, 53% to 75% were found to have
either sleep apnea, periodic leg movements in
sleep (PLMS), or both.14-17 In a series of 45
patients with ESRD with sleep problems, 71%
were found to have PLMS, and 95%, sleep
apnea.18 The only randomized study of sleep
disorders in dialysis patients to date found a 72%
incidence of sleep apnea, although it did not test
fully for PLMS.19 This finding is of potential
significance because untreated obstructive sleep
apnea is associated with both increased cardiovas-
cular morbidity and increased mortality in the
general population.20-22 Because the incidence of
both sleep complaints and objective findings of
sleep apnea and PLMS are considerably greater
in the ESRD population than in the general
population, the potential effects could also be
greater.23
ATIENTS WITH end-stage renal disease
(ESRD) undergoing maintenance dialysis
P
are reported to have an annual mortality rate in
excess of 20%.1 However, some patients survive
on dialysis treatment for more than 20 years. A
variety of risk factors, including low serum albu-
min levels and inadequate dialysis, are associ-
ated with increased mortality in the dialysis
population.2-7
Sleep disturbance is a common feature of
ESRD.8 Recent questionnaire-based surveys of
dialysis patients reported that 41% to 83% report
From the Division of Nephrology; Department of Medi-
cine, Division of Pulmonary and Critical Care Medicine,
Sleep Medicine Services, Jefferson Health System–Main
Line, The Lankenau Hospital and Medical Research Center,
Wynnewood; and the Department of Medicine, Jefferson
Medical College, Philadelphia, PA.
Received July 19, 1999; accepted in revised form Decem-
ber 17, 1999.
Address reprint requests to Robert L. Benz, MD, Chief,
Division of Nephrology, Jefferson Health System–Main Line,
Suite 130, The Lankenau Hospital, 100 Lancaster Ave,
Wynnewood, PA 19096. E-mail: benz@op.net
A better estimate of the incidence of sleep
disorders in the ESRD population is lacking
because of the absence of a full laboratory-based
study of sleep disorders in a randomly selected
ESRD population. Additionally, many patients
with such sleep disorders as PLMS may be
unaware of their sleep disorder.16
ꢀ 2000 by the National Kidney Foundation, Inc.
0272-6386/00/3506-0005$3.00/0
doi:10.1053/ajkd.2000.7459
The strong link between renal failure and
1052
American Journal of Kidney Diseases, Vol 35, No 6 (June), 2000: pp 1052-1060

POTENTIAL NOVEL PREDICTORS OF MORTALITY
1053
man et al.29 All PLMS were required to consist of at least
four consecutive, repetitive, dorsiflexion movements of the
legs or feet lasting 0.5 to 5 seconds, with an interval of 5 to
90 seconds during sleep.
sleep disorders is highlighted by a limited num-
ber of studies showing that sleep disorders of
patients with ESRD improve significantly or are
completely eliminated after successful renal trans-
plantation.24,25 However, hemodialysis is not re-
ported to have significant beneficial effects on
sleep apnea of ESRD.26
In preparation for a study investigating the
causes of sleep disorders in patients with ESRD,
records of patients who had previously under-
gone polysomnography (sleep studies) were re-
viewed. Agreater than expected number had died
since their sleep study. Retrospective review of
parameters to distinguish deceased patients from
longer term survivors was undertaken. Addi-
tional patients with ESRD with reports of sleep
problems were studied prospectively and fol-
lowed up to determine whether sleep disorders,
along with clinical, dialytic, and biochemical
profiles, could predict mortality.
PLMS were required to meet these criteria and to occur
independently of other events in the sleep recording. Be-
cause legs and feet may move as part of a general arousal
complex that might occur after apneas and hypopneas,
apparent movements of the legs or feet that occurred at the
termination of apneas or hypopneas or were associated with
snores or other sources of sleep disturbance were not in-
cluded in the analysis of PLMS. PLMS were further subdi-
vided as arousing PLMS (APLMS) when arousals as previ-
ously defined occurred secondary to the limb movements.
Obstructive hypopneas were scored if their duration was
10 seconds or longer and there was a visible reduction in
airflow or respiratory effort associated with at least a 3%
decline in SaO and/or arousal from sleep. Obstructive ap-
2
neas were scored in the absence of airflow but with contin-
ued presence of respiratory effort for at least 10 seconds
associated with at least a 3% decline in SaO and/or an
2
arousal from sleep. Central apneas required the complete
absence of airflow and respiratory effort for at least 10
seconds associated with at least a 3% decline in SaO and/or
2
an arousal from sleep.
METHODS
Statistical Methods
Subjects
Analysis. Two general approaches to data analysis were
taken. First, descriptive analyses comparing mean values
between deceased and surviving patients were performed
using independent-group t-tests. Deceased was defined by
death within 14 months after polysomnography. Survivors
lived longer than 14 months after polysomnography. All
patients included in this first analysis were treated by dialy-
sis only. Patients undergoing transplantation were excluded
from this analysis only because they did not continue to be
treated with dialysis. They were included in the Kaplan-
Meier survival analyses noted next. Parameters compared
included sleep and clinical, dialytic, and biochemical param-
eters (Table 1). Parathyroid hormone values were not com-
pared because of a change in laboratory methods during the
study period. The urea reduction ratio (URR) is a surrogate
measurement of urea clearance obtained by dialysis calcu-
lated by the formula:
Twenty-nine consecutive patients with ESRD who re-
ported difficulty sleeping or daytime sleepiness and were
referred for evaluation to the Sleep Disorders Center be-
tween 1990 and 1993 were studied. Mean age was 58.7 Ϯ
12.7 (SD) years (median, 59 years; range, 45 to 75 years),
and there were 18 men and 11 women and 15 white and 14
black patients. As part of clinical evaluation for sleep com-
plaints, all patients underwent all-night polysomnography in
the American Sleep Disorders Association–accredited Lan-
kenau Hospital Sleep Disorders Center (Wynnewood, PA).
Patients were followed up until death, transplantation, or
study termination. Survival time was measured from the
date of polysomnography.
Polysomnography was performed on evenings after dialy-
sis for all patients. The standard electrodes and sensors used
and parameters monitored were electroencephalogram (C3
or C4), electro-oculogram (bilateral outer canthi), electromyo-
gram (bilateral anterior tibialis and mentalis muscles), elec-
trocardiogram, airflow (bilateral nares and mouth), respira-
tory effort (chest and abdomen), and oxygen saturation
(finger pulse oximetry). Patients went to bed at their usual
times and awakened spontaneously. Data were recorded
continuously and stored both on paper and computer using
the Semi-Automatic Sleep Scoring System (Telefactor, Con-
shohocken, PA).
URR ϭ (predialysis BUN Ϫ postdialysis BUN) ϫ 100
/predialysis BUN
where BUN is blood urea nitrogen level.
The second approach characterized survival of patients
with and without high PLMS indices (PLMSI) or APLMS
indices (APLMSI). Kaplan-Meier survival curves were con-
structed to estimate median survival times with appropriate
accounting for censoring. Exact log-rank tests (StatExact
version 3 for Windows; Cytel Software Corp, Cambridge,
MA) were used to test hypotheses of equivalency in mortal-
ity rates, taking into account the small sample sizes.30
Incidence-density rates and incidence-density ratios (IDRs)
were computed instead of risk ratios.
All analyses of the polysomnographic data were per-
formed visually. No automatic computer analysis was per-
formed. All sleep data were analyzed using the standard
method of Rechtshaffen and Kales.27 Additionally, brief
arousals (not included in the manual by Rechtshaffen and
Kales) were analyzed as recommended by the Atlas Task
Force of the American Sleep Disorders Association.28
PLMS were scored according to the technique of Cole-
The mortality incidence density is the average number of
deaths per unit of time over a specified time interval. It can

1054
BENZ ET AL
Table 1. Comparison of Deceased Versus Surviving Groups
Deceased Survivors
Variable
Mean
SD
Mean
SD
P
General
Survival post-PSG (mo)
Age (y)
Duration of dialysis (mo)
Movements
10.3
61.0
44.6
6.3
14.9
29.4
25.5
57.3
32.8
12.2
9.4
24.3
0.001
NS
NS
PLMSI (per h)
APLMSI (per h)
Sleep
119.1
48.1
159.6
32.3
19.8
7.8
27.8
10.6
0.01
0.00006
Total recording time (min)
Total sleep time (min)
Sleep efficiency (%)
Stage 1 (%)
Stage 2 (%)
Stage 3 (%)
393.9
263.5
66.6
26.5
47.8
2.5
69.0
88.5
16.6
13.4
11.2
2.4
454.1
332.1
72.2
22.9
51.8
1.9
108.3
117.6
18.7
15.2
11.2
2.3
NS
NS
NS
NS
NS
NS
Stage 4 (%)
REM (%)
Arousals per hour
Respiratory
9.9
13.2
93.2
8.9
5.9
44.4
9.4
12.9
52.7
11.4
10.2
34.2
NS
NS
0.01
Respiratory disturbance index*
Low SaO2 (%)
44.9
76.2
37.1
14.6
44.8
81.6
37.4
7.3
NS
NS
Laboratory values
Hematocrit (%)
BUN (mg/dL)
Iron saturation (%)
Ferritin (ng/mL)
Calcium (mg/dL)
Phosphorus (mg/dL)
Albumin (g/dL)
29.9
61.9
24.8
61.7
9.3
5.8
3.63
58.5
3.6
17.3
17.0
39.9
1.4
1.4
0.29
8.5
33.4
56.1
26.3
80.0
9.6
5.9
3.81
63.6
4.9
13.0
9.7
61.1
1.2
1.7
0.27
4.2
0.06
NS
NS
NS
NS
NS
NS
NS
Urea reduction ratio
Abbreviations: PSG, polysomnogram; NS, not significant; REM, rapid eye movement.
*Number of apneas per hour of sleep.
be estimated as the total number of deaths divided by the
total accrued population-time expressed in, for example,
person-months. It differs from mortality risk in that it is
inherently a population-level description of mortality risk
rather than a patient-level measure. The computational advan-
tage of incidence risk in contrast to mortality risk is that
patients need not be followed up for equal lengths of time.
Moreover, patient-level risks (ie, probabilities) for particular
time intervals of interest (eg, 2-year survival) may be de-
rived by computation of cumulative incidence.
The IDR is simply the ratio of two incidence densities.
The IDR was used to compare mortality rates between
groups of patients (eg, high versus low PLMSI or APLMSI
rates). For example, a value of 2.0 means that the mortality
rate in one group of patients is two times as large as the
other. Adjusted IDRs were computed from proportional
hazards regression models using exact estimation methods.
The statistical precision of IDR estimates was determined
using confidence intervals. Cox proportional hazards regres-
sion models were used to determine whether controlling for
other clinical variables (Table 2) significantly affected esti-
mates of the mortality rates. Multivariable models were
restricted to one grouping factor and one potential confound-
ing variable at a time in light of the small sample size.
To eliminate bias resulting from post hoc group defini-
tions, categories of variables were defined without regard to
survival outcome, using cumulative distribution functions.
For example, in the definition of categories of PLMSI, data
were plotted in a PLMSI cumulative distribution graph.
Based on the distribution curve, three categories of patients
were determined: those with PLMSI values less than 20
(group I), PLMSI values of 20 or greater but less than 80
(group II), and PLMSI values of 80 or greater (group III).
Before assessing group differences in mortality, these groups
were compared using one-way analysis of variance for
interval variables and chi-square or generalized Fisher’s
exact tests for categorical variables to identify the most
important potential confounders.31
Two patients underwent kidney transplantation before 14
months of follow-up. These patients were treated as cen-
sored at the time of transplantation in the survival analyses
but excluded in the descriptive comparison between de-
ceased and surviving groups.
Survival analyses. The primary null hypothesis tested
with Kaplan-Meier analysis was that survival curves for
patients with PLMS rates less than 20, 20 or greater but less
than 80, and 80 or greater per hour were equal. Testing of
this hypothesis was performed using an exact log-rank

POTENTIAL NOVEL PREDICTORS OF MORTALITY
Table 2. Patient Characteristics by PLMI Group
1055
PLMSI
Group
Comparison*
Ն80/h (n ϭ 7)
20-Ͻ80/h (n ϭ 9)
Ͻ20/h (n ϭ 13)
Variable
Age (y)
RDI (events/h)
Mean sat (%)
Dialysis (mo)
URR
Mean
SD
Mean
SD
Mean
SD
F
P
66.7
42.8
90.1
53.0
65.0
9.3
33.9
3.2
31.5
5.6
54.5
38.8
92.5
37.2
57.6
11.1
57.1
0.5
25.5
7.5
57.1
52.7
90.9
23.3
59.6
14.0
35.9
3.0
15.5
7.1
2.2
0.3
1.6
3.2
1.9
2.5
0.144
0.758
0.213
0.057
0.184
0.119
Albumin
3.52
0.23
3.90
0.36
3.60
0.34
Abbreviation: RDI, respiratory disturbance index; Sat, oxygen saturation.
*F tests from one-way analysis of variance for interval.
statistic (StatExact).32 The exact log-rank test is a widely
cited technique valid in small samples that are based on
exact conditional distributions. It represents an alternative
approach to those techniques relying on maximizing the
unconditional likelihood function for parameter estimation
that usually require large samples for justification of their
validity.32 Exact log-rank is increasingly recognized as a
method of choice for dealing with small sample inference in
the analysis of clinical trial data.33 Survival curves for
patients with PLMSIs less than 20 and greater than 20 were
also constructed and compared with exact log-rank tests.
Survival curves for APLMSI groups of less than 20, 20 to
40, and greater than 40 were also constructed and tested in a
similar manner.
To rule out substantial confounding, Cox regression mod-
els were estimated that included the PLMS or APLMS
factors and one potential covariate at a time. We sought to
show that the difference between unadjusted and adjusted
mortality ratios did not differ by a clinically significant
amount, thus ruling out substantial confounding. If control
for a variable results in no substantial parameter estimate
change, confounding is not regarded as a major problem
(Table 3).34,35
RESULTS
Of the 29 patients studied, 10 died within 14
months of initial polysomnographic testing. Me-
Table 3. Cox Regression Analysis of Mortality Rates
Adjusted Rate Ratio for PLMSI
Ն80/h v Ͻ20/h 20-Ͻ80/h v Ͻ20/h
95% CI 95% CI
PLMSI P From Cox
Regression After
Adjustment for Covariate
(␹2 with df ϭ 2)
Variable
IDR*
IDR*
PLM (unadjusted)
Age (decade)
RDI
Mean sat (%)
Years of dialysis
BUN
MCV
Calcium
Glucose
Phosphorus
URR
Albumin
Male
White
Coronary artery
Diabetes
25.5
59.9
24.7
30.1
26.6
26.2
113.8
29.2
27.7
26.8
59.5
24.2
29.0
26.9
41.1
16.0
3.0-215.7
4.2-846.4
2.8-216.7
3.2-281.7
2.7-266.2
3.1-221.8
8.7-1,496
3.4-253.1
3.3-235.3
3.1-232.7
4.1-872.1
2.8-212.8
3.3-253.0
3.1-232.7
4.0-421.1
1.7-147.9
4.3
4.7
3.2
3.0
4.1
4.4
3.7
3.4
5.7
2.7
6.4
4.7
3.8
4.5
4.6
10.8
0.5-41.7
0.5-45.5
0.3-35.5
0.3-34.3
0.4-40.5
0.5-42.7
0.3-34.1
0.3-34.2
0.6-64.4
0.2-31.0
0.6-64.4
0.5-45.9
0.4-36.8
0.5-44.2
0.4-53.1
0.7-178.3
0.002†
0.007
0.009
0.004
0.006
0.002
0.0008
0.002
0.003
0.002
0.007
0.007
0.002
0.002
0.002
0.049
Abbreviations: RDI, respiratory disturbance index; PLMSI, periodic leg movements in sleep index; MCV, mean corpuscular
volume.
*Incidence density ratio.
†Unadjusted Wald chi-squared test (df ϭ 2) from Cox regression.

1056
BENZ ET AL
dian survival time of patients in the deceased
group was 10.3 months (range, 3 to 14 months)
compared with 25.5 months (range, 15 to 48
months) in the surviving group. Nine of 29
subjects had no PLMS. Eight of these 9 subjects
were survivors. The one patient with no PLMS
who died had severe sleep apnea with 99 apneas
and/or hypopneas per hour of sleep. Virtually all
deaths were attributed to arteriosclerotic-, car-
diovascular-, or infection-related events. Un-
explained sudden deaths were diagnosed as car-
diovascular by assumption. Arteriosclerotic
cardiovascular disease was twice as common as
infection and/or sepsis diagnoses as causes of
death.
Table 1 lists the results of comparisons be-
tween means of variables in deceased and surviv-
ing groups with independent t-test results. All
patients had clinically significant sleep apnea,
PLMS, or both. Differences in means were statis-
tically significant for PLMSI (119.1 versus 19.8;
P ϭ 0.01), APLMSI (48.1 versus 7.8; P ϭ
0.00006), and total arousals per hour of sleep
(93.2 versus 52.7; P ϭ 0.01). Arousals consisted
of brief periods of electroencephalography-
defined wakefulness lasting less than 15 seconds
but more than 2 seconds.
Values in the longer term survivors tended to
drift up over time because of changes in accept-
able target URRs.
The Kaplan-Meier survival curves related to
the primary hypothesis test are shown in Fig 1.
The exact log-rank statistic rejected the null
hypothesis of equivalency in survival distribu-
tions between patients with a PLMSI less than 20
versus a PLMSI between 20 and 79 versus a
PLMSI of 80 or greater at P Ͻ 0.00001. For
patients with a PLMSI greater than 80 (group
III), the estimated median survival time was 6
months (95% confidence interval, 4 to 20).
Controlling for the effects of albumin level,
hematocrit, and duration of dialysis (months)
still resulted in large IDRs for PLMSI, but these
reduced the size of the unadjusted values by one
half or less. Despite the reduction in size of the
statistical effect, controlling for the effects of
hematocrit, albumin level, and duration of dialy-
sis by Cox regression did not result in loss of
statistical significance of PLMSI or APLMSI as
independent predictors of survival time. Clini-
cally significant differences in mortality contin-
ued to exist after controlling for potential con-
founding from these variables. However, these
findings suggest that the greatest PLMSIs are
more likely to be found in patients with the
lowest hematocrit and albumin values. Cox re-
In both groups, these occurred almost exclu-
sively after sleep-disordered breathing and/or
APLMS. Because sleep apnea was equally high
in each group, the increased number of arousals
in the deceased group was secondary to in-
creased APLMS. Sleep apnea occurred fre-
quently in both the deceased and survivor groups.
However, there were no significant differences in
the respiratory disturbance index or low SaO
2
values between the two groups. Compared with
the survivor group, there was a trend for in-
creased stage 1 sleep and decreased sleep effi-
ciency (both indicative of more fragmented sleep)
in the deceased group. The deceased group also
showed a trend for less time in bed and less total
sleep time.
Among non–sleep-based variables, hematocrit
showed a trend toward lower values in the de-
ceased group (29.9% versus 33.4%; P ϭ 0.06).
URR, serum albumin level, and comorbidity
were not different between groups. Note that
URR values listed in Table 1 are less than those
targeted currently because patients in this study
underwent polysomnography from 1990 to 1993.
Fig 1. Kaplan-Meier survival curves comparing pa-
tients with PLMSI of 80 or greater (n ؍ 7) versus 20 to
less than 80 (n ؍ 9) versus less than 20 (n ؍ 13) per
hour. The unadjusted differences among the survival
curves were significantly different (exact log-rank test,
P F 0.0001). Differences remained significant control-
ling for group differences in mean length of time under-
going dialysis, age RDI, URR, albumin, sex, coronary
artery disease, and diabetes. Abbreviation: PSG, poly-
somnography.

POTENTIAL NOVEL PREDICTORS OF MORTALITY
1057
gression–adjusted mortality ratios were esti-
mated for other potentially confounding clinical,
biochemical, dialytic, and sleep-related factors
as well. No appreciable effects on mortality IDRs
or P of the PLMSI predicting survival time were
found.
Survival curves for patients grouped by
PLMSIs less than 20 and greater than 20 were
also found to be significantly different by exact
log-rank test (P ϭ 0.007). The 20-month sur-
vival rate for patients in the group with PLMSI
greater than 20 was 50%, whereas the group with
PLMSI less than 20 showed a 20-month survival
rate of 90% (Fig 2). Survival curves for APLMSI
less than 10 and 10 to 40 were found to be
significantly different from the survival curve for
APLMSI greater than 80 (Fig 3).
A limited number of Spearman correlations
were also performed to better understand the
relationship between duration of dialysis and
other variables. Duration of dialysis was corre-
lated only with PLMSI (r ϭ 0.42; P Ͻ 0.02) and
APLMSI (r ϭ 0.44; P Ͻ 0.01). Duration of
dialysis was not correlated with a measure of
sleep. PLMSI and APLMSI were also correlated
with hematocrit (PLMSI, r ϭ Ϫ0.36; P Ͻ 0.05
and APLMSI, r ϭ Ϫ0.36; P Ͻ 0.05).
Fig 3. Kaplan-Meier survival curves comparing pa-
tients with an APLMSI of 40 or greater (n ؍ 7), 10 to
less than 40 (n ؍ 8), and less than 10 (n ؍ 14) per hour
of sleep. The unadjusted differences among survival
curves were statistically significant (exact log-rank
test, P F 0.0001). Differences remained significant
controlling for group differences in mean length of
time undergoing dialysis, age RDI, URR, albumin, sex,
coronary artery disease, and diabetes. Abbreviation:
PSG, polysomnography.
DISCUSSION
This report presents evidence for potentially
new predictors of mortality in patients with ESRD
who report problems with sleep and daytime
alertness. All statistical procedures performed
found that in this group of patients with ESRD,
only PLMSI, APLMSI, and total arousals per
hour of sleep differed significantly between sur-
viving and deceased groups and significantly
predicted death. The Kaplan-Meier curves sug-
gest that with greater numbers of PLMS, there is
a greater risk for death within a shorter period of
time. PLMSI greater than 20 or APLMSI greater
than 40 predicted mortality. When all patients
were divided into two groups by PLMSI less
than or more than 20, the 20-month survival rate
was 50% for those with a PLMSI of 20 or greater
versus greater than 90% survival for those with a
PLMSI less than 20 (P ϭ 0.0008; Fig 2).APLMSI
was six times greater in the deceased than survi-
vor group (P ϭ 0.00006).
More thoroughly researched risk factors, such
as URR, albumin level, and hematocrit, were not
found to be independent predictors of mortality
in this relatively small, select group of patients.
Controlling for their effects in the estimation of
relative mortality rates in the PLMSI and
APLMSI categories reduced the size of the over-
all effect of PLMSI and APLMSI, but did not
Fig 2. Kaplan-Meier survival curves comparing pa-
tients with PLMSI of 20 or greater per hour (n ؍ 16)
versus patients with PLMSI less than 20 per hour (n ؍
13). The unadjusted difference in survival was statisti-
cally significant (exact log-rank test, P ؍ 0.007). Differ-
ences remained significant controlling for group differ-
ences in mean length of time undergoing dialysis, age
RDI, URR, albumin, sex, coronary artery disease, and
diabetes. Abbreviation: PSG, polysomnography.

1058
BENZ ET AL
prevent them from reaching statistical signifi-
cance. The reduction in the size of the statistical
effect, however, suggests that the relationship
between PLMSI and APLMSI with other estab-
lished risk factors, such as serum albumin level,
URR, and hematocrit, as well as duration of
dialysis (months), may be more complicated
than is evident in this study. High numbers of
PLMS may reflect dialytic inadequacy or persis-
tence of an abnormal uremic milieu, including
anemia. Correction of anemia in such patients
has been reported to reduce PLMS and APLMS
by approximately one half.36 PLMS may be
secondary to or act as a marker of cerebral and/or
peripheral dysfunction in patients with ESRD
induced by other, perhaps multiple, pathological
processes present in renal failure or secondary to
dialysis. This hypothesis is lent support by re-
ports that successful renal transplantation mark-
edly improves ESRD-related sleep disorders, in-
cluding PLMS.24,25
population. Coleman et al29 noted a primary
finding of RLS in only 8 of 53 patients with a
primary diagnosis of PLMS. In our own clinical
practice, only 2% of all patients with a primary
diagnosis of PLMS also report severe RLS (Benz
and Pressman, unpublished data). Thus, PLMS
may occur independently of RLS and result from
factors independent of RLS. Additionally, RLS is
characterized by subjective sensations primarily
in the lower limbs that occur during wakefulness
before sleep and are inferred to be present by
history alone. RLS severity, therefore, cannot be
quantified in the same objective manner as PLMS.
An initial working hypothesis for this study
was that sleep apnea was likely to be the main
risk factor for mortality in this group. Obstruc-
tive sleep apnea is reported to be associated with
increased morbidity and mortality in the general
population.20,21 The failure of sleep apnea in
patients with ESRD to predict mortality may be
explained by several factors.
An alternate hypothesis of the relationship
between PLMS and mortality in this population
considers the possible effects of chronic, severe
sleep fragmentation and sleep deprivation. Sleep
was severely disrupted, particularly in the de-
ceased group. Total arousals from sleep were
significantly greater than in the deceased group.
Chronic sleep fragmentation has a variety of
well-described consequences, including the car-
diovascular sequelae of circadian disruption and
loss of nocturnal dipping of blood pressure, result-
ing in increased hypertensive burden and left
ventricular hypertrophy.37-39 Sleep deprivation
has also been implicated as a source of impaired
immune function.40-42 The trend for patients with
increased duration of dialysis (months) to have a
greater PLMSI (and consequently greater mortal-
ity risk) could represent longer exposure of those
patients to the effects of chronic sleep disruption.
Further support for the relationship between
PLMS, APLMS, and mortality can be found in
the recent report that restless legs syndrome
(RLS) is associated with increased mortality in a
population similar to that described here.43 RLS
and PLMS are hypothesized to have similar
causes and respond to treatment with similar
medication. A primary report of RLS is associ-
ated with a finding of PLMS in 80% of patients
in the general population.44 However, PLMS
may occur in the absence of RLS in the general
First, sleep apnea of ESRD is often reported to
be of the rarer central type, not the obstructive
type found in patients previously reported to
have increased morbidity and mortality. Kimmel
et al16 reported that 7 of 26 patients with ESRD
studied had predominantly central sleep apnea,
and Mendelson et al26 found that 5 of 11 dialysis
patients had between 53% and 90% central ap-
neas. Central sleep apnea is not related to obesity
and is often associated with less nocturnal hypox-
emia.45 Thus, the effect of obesity-related obstruc-
tive sleep apnea on mortality in the general
population may not necessarily be extrapolated
to patients with ESRD.
Second, sleep apnea may require a much longer
time before increased mortality becomes evi-
dent. He et al20 reported that statistically signifi-
cant increased mortality in sleep apnea patients
in the same age range as our patients was not
evident for at least 7 years after the initial sleep
study. Thus, sleep apnea may not be a significant
risk factor for mortality in the majority of pa-
tients with ESRD. Longer studies or larger sam-
ple sizes may be needed to correctly assess the
impact of sleep apnea on patients with ESRD.
One potential limitation of this study is that
patients were not randomly selected. Rather, all
had reported problems regarding sleep or day-
time sleepiness and all had been referred to a
sleep disorder center for evaluation. A majority

POTENTIAL NOVEL PREDICTORS OF MORTALITY
1059
6. 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
7. Owen WF Jr, Lew NL, Liu Y, Lowrie EG, Lazarus JM:
The urea reduction ratio and serum albumin concentration as
predictors of mortality in patients undergoing hemodialysis.
N Engl J Med 329:1001-1069, 1993
8. Papper S: Clinical Nephrology. Boston, MA, Little,
Brown, 1978, p 120
9. Holley JL, Nespor S, Rault R: Characterizing sleep
disorders in chronic hemodialysis patients. Trans Am Soc
Artif Int Organs 38:M456-M457, 1991
of surveyed patients with ESRD have complaints
about their sleep, and asymptomatic patients
have been noted to have PLMS without their
knowledge.9,10,12,13 However, a definitive study
of the incidence of sleep disorders in a large
random sample of patients with ESRD has not
been published. This leaves open the question of
precisely how common sleep disorders are in this
population.
Mortality rates for patients with ESRD remain
unacceptably high. This appears to be especially
true for dialysis patients in the United States
whose survival time is less than that of many
other technologically advanced countries.46 Ef-
forts to reduce the mortality rate have been the
subject of numerous publications, consensus pan-
els, and multicenter studies, including the Na-
tional Institutes of Health HEMO Study. Total
fiscal Medicare, state, and private-payer budgets
for the ESRD population now exceed $13 billion
annually, stimulating demand for improved out-
come.1 Barrett et al3 recently concluded that
predictors of short survival time in ESRD have
largely proven inadequate. Determining predic-
tors of mortality is the first step in ultimately
understanding inadequacy of therapy and focus-
ing on beneficial interventions. PLMSI and
APLMSI, as well as the total number of arousals
per hour of sleep, represent potential novel pre-
dictors of mortality in patients with ESRD. Fur-
ther study is indicated to delineate the relation-
ships between chronic renal failure, sleep
disorders, and mortality to improve the long-
term outcome of this high-risk population.
10. Vegia J, Goncalves N, Gomes F, Santos N, Baptista
A, Paiva T: Sleep disturbances in end-stage renal disease
patients on hemodialysis. Dial Transplant 26:380-383, 1997
11. Walker S, Fine A, Kryger MH: Sleep complaints are
common in a dialysis unit. Am J Kidney Dis 26:751-756, 1995
12. Burman-Urbanek M, Sanner B, Laschewski F, Kruezer
I, Wiemann J, Konermann M, Sturn A: Sleep disorders in
end-stage renal disease. Pneumologie 49:S158-S160, 1995
(suppl)
13. Wendland KL, Greinert I: Sleep disorders in hemodi-
alysis patients. Rehabilitation 28:74-77, 1989
14. Wadhwa NK, Mendelson WB:Acomparison of sleep-
disordered respiration in ESRD patients receiving hemodi-
alysis and peritoneal dialysis. Adv Perit Dial 8:195-198,
1992
15. Millman RP, Kimmel PL, Shore ET, Wasserstein AG:
Sleep apnea in hemodialysis patients: The lack of testoster-
one effect on its pathogenesis. Nephron 40:407-410, 1985
16. Kimmel PL, Miller G, Mendelson NM: Sleep apnea
syndrome in chronic renal disease. Am J Med 86:308-314,
1989
17. Pressman MR, Benz RL, Schleifer CR, Peterson DD:
Sleep disordered breathing in ESRD: Acute beneficial ef-
fects of treatment with nasal continuous positive airway
pressure. Kidney Int 43:1134-1139, 1993
18. Pressman MR, Benz RL, Peterson DD: High inci-
dence of sleep disorders in end-stage renal disease patients.
Sleep Res 24:417A, 1995 (abstr)
19. Hallett M, Burden S, Stewart D, Mahony J, Farrell P:
Sleep apnea in end-stage renal disease patients on hemodialy-
sis and continuous ambulatory peritoneal dialysis. ASAIO J
41:M435-M441, 1995
20. He J, Kryger MH, Zorick FJ, Conway W, Roth T:
Mortality and apnea in obstructive sleep apnea. Chest 94:9-
14, 1988
21. Shepard JW: Cardiopulmonary consequences of ob-
structive sleep apnea. Mayo Clin Proc 65:1250-1259, 1990
22. Kryger M: Monitoring respiratory and cardiac func-
tion, in Kryger M, Roth T, Dement WC (eds): Principles and
Practice of Sleep Medicine. Philadelphia, PA, Saunders,
1989, pp 702-708
23. Young T, Palta M, Dempsey J, Skatrud J, Weber S,
Bad S: The occurrence of sleep-disordered breathing among
middle aged adults. N Engl J Med 328:230-1235, 1993
24. Langevin B, Fouque D, Leger P, Robert D: Sleep
apnea syndrome and end-stage renal disease: Cure after
renal transplantation. Chest 103:1330-1335, 1993
25. Yasuda T, Nishimura A, Katsuki Y, Tusji Y: Restless
REFERENCES
1. US Renal Data System: USRDS 1997 Annual Data
Report. The National Institutes of Health, National Institute
of Diabetes and Digestive and Kidney Diseases, Bethesda,
MD, 1997
2. Hutchinson TA, Thomas DC, MacGibbon B: Predict-
ing survival in adults with end-stage renal disease: An age
equivalence index. Ann Intern Med 96:417-423, 1982
3. Barrett BJ, Parfre PS, Morgan J, Barre P, Fine A,
Goldstein MD, Handa SP, Jindal KK, Kjellstrandt CM,
Levin A, Mandin H, Muirhead N, Richardson RM: Predic-
tion of early death in end-stage renal disease patients starting
dialysis. Am J Kidney Dis 29:214-222, 1997
4. Maillou LU, Bellucci AG, Mossey RT, Napolitano B,
Moore T, Wilkes BM, Bluestone PA: Predictors of survival in
patients undergoing hemodialysis. Am J Med 84:855-862, 1988
5. Collins A, Ma J, Ebben J: Patient survival is associated
with hematocrit (HCT) level. J Am Soc Nephrol 8:190A, 1997
(abstr)

1060
BENZ ET AL
legs syndrome treated successfully by kidney transplanta-
tion: A case report. Clin Transplant 138, 1986
26. Mendelson WB, Wadhwa NK, Greenberg HE, Gu-
jarty K, Bergofsky E: Effects of hemodialysis of sleep apnea
syndrome in end-stage renal disease. Clin Nephrol 33:262-
266, 1990
disorders and daytime sleepiness in hemodialysis patients
(The SLEEPO Study). Am J Kidney Dis 34:1089-1095,
1999
37. Rostand SG, Irk KA, Rutsky EA: Relationship of
coronary risk factors to hemodialysis-associated ischemic
heart disease. Kidney Int 22:304-308, 1982
27. Rechtshaffen A, Kales A: A Manual of Standardized
Terminology, Techniques and Scoring System for Sleep
Stages of Human Subjects. LosAngeles, CA, Brain Informa-
tion Service/Brain Research Institute, 1968, pp 1-60
28. Bonnet M, Carley D, Carskadon M, Easton P, Guille-
minault C, Harper R, Hayes B, Hirshkowitz M, Periklis K,
Keenan S, Pressman M, Roehrs T, Smith J, Walsh J, Weber
S, Westbrook P (Atlas Task Force): EEG arousals: Scoring
rules and examples. Sleep 15:173-184, 1992
29. Coleman RM, Pollak CP, Weitzman ED: Periodic
movements in sleep (nocturnal myoclonus): Relationship to
sleep disorders. Ann Neurol 8:416-421, 1980
30. Cytel Software Corporation: StatExact 3 for Win-
dows. Statistical Software for Exact Nonparametric Infer-
ence. Cambridge, MA, Cytel Software Corporation, 1996
31. Mehta CR, Patel NR: A network algorithm for per-
forming Fisher’s exact test in r ϫ c contingency tables. J Am
Stat Assoc 78:427-434, 1983
38. Rosansky SJ, Menachery SJ, Whittman D, Rosenberg
JC: The relationship between sleep deprivation and the
nocturnal decline of blood pressure. Am J Hypertens 9:1136-
1138, 1996
39. Foley RN, Parfrey PS, Harnett JD, Kent GM, Martin
CJ, Murray DC, Barre PE: Clinical and echocardiographic
disease in patients starting end-stage renal disease therapy.
Kidney Int 47:186-192, 1995
40. Krueger JM, Karnovsky ML: Sleep and the immune
response. Ann NYAcad Sci 496:510-516, 1987
41. Moldofsky H, Lue FA, Davidson JR, Gorczynski R:
Effects of sleep-deprivation on human immune function.
FASEB J 3:1972-1977, 1989
42. Toth LA, Tolley EA, Krueger JM: Sleep as a prognos-
tic indicator during infectious disease in rabbits. Proc Soc
Exp Biol Med 203:179-192, 1993
43. Winkelman JW, Chertow GM, Lazarus JM: Restless
legs syndrome in end-stage renal disease. Am J Kidney Dis
28:372-378, 1996
44. Montplaisir J, Boucher S, Poirer G, Lavigne G,
Leapiere O, Lepserance P: Clinical, polysomnographic and
genetic characteristics of restless legs syndrome: A study of
133 patients diagnosed with new standard criteria. Mov
Disord 12:61-65, 1997
45. White D: Central sleep apnea, in Kryger M, Roth T,
Dement WC (eds): Principles and Practice of Sleep Medi-
cine. Philadelphia, PA, Saunders, 1994, pp 630-641
46. Held PJ, Brunner F, Odaka M, Garcia JR, Port FK,
Gaylin DS: Five-year survival for end-stage renal disease
patients in the United States, Europe, and Japan, 1982 to
1988. Am J Kidney Dis 15:451-457, 1990
32. Mehta CR, Patel NR: Exact logistic regression: Theory
and example. Stat Med 14:2143-2160, 1995
33. Simon R: A decade of progress in statistical method-
ology for clinical trial. Stat Med 10:1789-1817, 1991
34. Kleinbaum DG, Kupper LL, Morgenstern H: Epide-
miologic Research: Principals and Quantitative Methods.
New York, NY, Van Nostrand Reinhold, 1982
35. Mickey RM, Greenland S: The impact of confounder
selection criteria of effect estimation. Am J Epidemiol 129:
125-137, 1989
36. Benz RL, Pressman MS, Hovick EH, Peterson DD: A
preliminary study of the effects of correction of anemia with
recombinant human erythropoietin therapy on sleep, sleep