Nephrol Dial Transplant (2005) 20: 1192–1200
doi:10.1093/ndt/gfh806
Advance Access publication 12 April 2005
Original Article
A detailed analysis of sodium removal by peritoneal dialysis:
comparison with predictions from the three-pore model of
membrane function
Marissa C. Aanen1, Daniele Venturoli2 and Simon J. Davies3
1
Department of Nephrology, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands,
Department of Nephrology, Lund University Hospital, Lund, Sweden and 3Department of Nephrology, University
Hospital of North Staffordshire and Institute of Science and Technology in Medicine, Keele University, UK
2
Abstract
Background. The development of fluid and salt retention is a potential problem for all peritoneal dialysis
(PD) patients. Sodium removal by the peritoneum is
predominantly determined by convective fluid loss but
influenced by diffusion and sieving due to free water
transport as predicted by the three-pore model (TPM).
The aim of the study was to establish the effect of
transport status, dwell length and glucose concentration on observed ultrafiltration (UF), dialysate sodium
concentration ([Naþ]D) and removal, and compare this
with that predicted by a computer program based on
the principles of the TPM.
Methods. This was a cross-sectional study of UF and
[Naþ]D collected prospectively from dwells classified
by length, glucose concentration and membrane
transport characteristics. Solute transport, converted
to area parameter and UF capacity, was measured on
each occasion by the peritoneal equilibration test.
These parameters, along with plasma [Naþ], were
entered into the computer model. Fixed values for
other parameters, e.g. hydraulic conductance and
lymphatic absorption and sump volume, were used.
Results. A total of 1853 dwells from 182 patients
[10% were on automated PD (APD)] were analysed.
There was a high degree of correlation (r ¼ 0.83–95,
P<0.001) between the observed and predicted values
for UF, [Naþ]D and sodium removal across the full
range of dwell categories. The model overpredicted UF
as the net volume increased with increasing glucose
concentration, independently of solute transport. This
bias was not fully explained by the preferential use
of hypertonic dialysate by patients with reduced UF
capacity. The prediction of [Naþ]D described sodium
Correspondence and offprint requests to: Professor Simon J. Davies,
Department of Nephrology, University Hospital of North
Staffordshire, Princes Road, Hartshill, Stoke-on-Trent, ST4 7LN,
UK. Email: [email protected]
sieving, which was overestimated in a small number of
patients with UF failure. There were no discrepancies
between continous ambulatory PD (CAPD) and APD
patients.
Conclusion. This analysis endorses the TPM as a
description of membrane function, particularly in
relation to sodium sieving and removal. The relationship between dialysate glucose concentration and
achieved UF appears to be more complex; even
accounting for extended time on treatment and
reduction in the osmotic conductance in patients
preferentially using hypertonic exchanges, further
adjustments may be needed to account for the
tendency to overestimate UF.
Keywords: automated peritoneal dialysis (APD);
computer modelling; continuous ambulatory
peritoneal dialysis (CAPD); hydraulic conductance;
peritoneal membrane; sodium sieving
Introduction
Adequate water and sodium balance is crucial in the
management of patients on peritoneal dialysis (PD)
especially when the patient is anuric. In at least one
prospective cohort study, the removal of sodium and
fluid has been identified as a predictor of mortality
in PD patients independent of residual renal function
[1], and failure to achieve >750 ml of daily ultrafiltration (UF) in anuric patients was associated with
increased mortality in the European Automated
Peritoneal Dialysis Outcome Study (EAPOS) [2].
The combination of fluid overload along with hypertension can be a major factor for developing cardiovascular disease, the leading cause of death in PD
patients [3].
ß The Author [2005]. Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved.
For Permissions, please email: [email protected]
A detailed analysis of sodium removel by peritoneal dialysis
The three-pore model (TPM), developed by Rippe,
provides a quantitative description of sodium and
water removal across the peritoneal membrane [4]. In
particular, this approach is able to account for the
phenomenon of solute sieving that results in an early
fall in the dialysate sodium concentration, due to its
dilution by osmotically driven free water transport [5],
occurring via aquaporins [6,7]. This dissociation
between UF of water and sodium transport is, therefore, of greatest clinical relevance to patients using
automated PD (APD) during short dwells [8] that
might have clinical consequences [9]. Prediction of
time-dependent UF profiles by the TPM has been
validated previously in a small number of patients [10].
It also forms the theoretical basis of the personal
dialysis capacity (PDC) test [11], which is rather better
at predicting daily UF volume than PD Adequest [12];
whereas PDC utilizes individual data from dwells
of differing length and glucose concentrations, the
latter relies on a single peritoneal equilibration test
(PET) and overnight fluid volume to determine UF
capacity. The ability of the TPM to describe dialysate
sodium concentration and, in combination with
UF volume, the net sodium removal under differing
conditions of dwell length, peritoneal solute transport
characteristics and strength of the glucose exchange,
has not, so far, been evaluated. The purpose of this
study was to compare observed UF dialysate sodium
concentration, and thus sodium removal, with that
predicted by the TPM, under differing conditions
(glucose concentration and dwell length) using computer simulations individualized for peritoneal transport
status and plasma sodium concentration.
Subjects and methods
Study design
A cross-sectional analysis was performed using prospectively
collected data from dialysis dwells of PD patients undergoing
their routine treatment. For each dwell, the net UF, dialysate
sodium concentration and thus calculated sodium removal
was measured and categorized according to dwell length,
solute transport and glucose concentration. These empirical
data were then compared with that derived from the
TPM using grouped median measures of peritoneal solute
transport and plasma sodium concentrations in the modelling process. Comparisons were also made between the
APD and continuous ambulatory PD (CAPD) patients
where the same length exchanges were performed, again
comparing these with values predicted by the TPM computer
simulations.
Sample collection
Data for the present analysis were collected between January
1998 and May 2001. Dialysate and blood samples were
analysed for sodium content when patients were undergoing
routine tests for dialysis adequacy and peritoneal membrane
function. This procedure was performed every 6 months,
1193
unless the patient developed peritonitis in the preceding
month. Each assessment involved the patient collecting each
dialysate drain for a 24 h period followed the next day by a
standardized PET. In CAPD patients, all the exchanges were
brought to the hospital and analysed separately, the net UF
being corrected for the overfill of dialysis bags, typically
found to be 200 ml, which bypasses the peritoneal cavity
as part of the flush before fill procedure, but is collected in
the drainage bag. This amount was derived from regular
monitoring of overfill performed following completion of this
study, corroborated independently by another study [13], and
from personal communications with senior Baxter personnel
confirming that the overfill volume has been stable for at
least the duration of this study. For APD patients, long day
dwells were treated in a similar fashion, except that there
is no overfill flush volume; exchanges using icodextrin were
excluded, and all the patients performed a 9 h dwell. This was
because use of APD during the study period was confined to
anuric patients who required an additional fill volume. The
overnight exchanges were collected into a single drainage
bag, sampled by the patient to determine the mean sodium
concentration. Net overnight UF was determined from the
APD device and divided by the number of exchanges.
Information, including dialysate fluid type (glucose 1.36,
2.27 and 3.86%) and dwell length, was collected for each
exchange. Dwell lengths were defined and categorized as
follows: short, 90 min, typical for overnight exchange in
APD patients; medium, 300 min, typical for daytime dwells
in CAPD patients; and long, 540 min, being either overnight
dwells in CAPD or daytime dwells in APD patients.
Peritoneal equilibration test (PET)
The PET, performed as the first exchange of the day in all
patients, was utilized to measure solute transport and UF
capacity, and this was performed as described previously [14].
Briefly, a standard 4 h dwell period was used (first exchange
of the day), using a 2.27% glucose concentration 2 l volume
exchange. The patients used their usual overnight dialysis
regime, and both the overnight and test drainage volumes
were measured. The dialysate:plasma ratio of creatinine at the
completion of the 4 h dwell period (D/Pcreat) was used as the
estimate of low molecular weight solute transport. As glucose
interferes with the assay for creatinine in a linear fashion,
concentrations for both these solutes are measured at 4 h
and the true value for creatinine obtained by subtracting
the glucose concentration multiplied by a correction factor
derived locally by our laboratory (0.00047).
Sodium analysis
Plasma and dialysate sodium, glucose and creatinine were
determined on an automated discrete random access
analyser (DAX 72, Bayer Instruments, Basingstoke, UK).
This employs an indirect electrode method to measure
sodium concentration, using pre-dilution of the sample,
thus minimizing the influence of sodium binding to protein
and inorganic ions. Because concerns have been raised as
to the accuracy of direct electrode methodology [15], the
indirect electrode measurements were cross-checked using
the flame photometry technique. In this comparison of
methods, the results were always within 1 mmol/l of each
other.
1194
M. C. Aanen et al.
Modelling
For the modelling of the data, three transport groups (low
average, high average and high) were formed according to
their D/P creatinine ratio obtained from the 4 h 2.27% PET
data. Each transport group was divided further according
to the dwell lengths (90, 240, 300 and 540 min, see above for
definitions) and glucose strength concentrations (1.36, 2.27
and 3.86%). For each of these groups, the median of the
estimate of solute transport made in the PET (D/P creatinine
ratio) was converted into a value for area parameter, having
first established the relationship between these parameters as
defined by the model.
The computer simulations were performed on a personal
computer using a specifically designed ExcelÕ worksheet
(Microsoft, Bellevue, WA), developed by one of us (D.V.),
according to the principles of the TPM of peritoneal
transport [16]. This worksheet enables rapid simulations to
be performed whilst allowing the operator to enter both
empirical data and alter one or more of the model parameters. For the present analysis, the original parameters were
employed as recently published (see Table 1) [16,17]. Notably,
the permeability area product (PS) values for glucose and
sodium were perturbed, as described; the value for glucose
is increased to account for the effects of interstitial and
intracapillary glucose concentrations, whereas that for
sodium is decreased in keeping with observations that its
mass transfer coefficient is much lower than predicted from
its molecular weight. The Gibbs–Donnan effect for sodium
is also accounted for in the model, employing a dissociation
constant of 0.93. In addition, the computer model also
enables the operator to use a fixed value for hydraulic
conductance (LPS) even when the value for the area
parameter is altered, and this is the approach we adopted
for this study.
Three parameters were altered for the analysis of each
dwell, namely the glucose concentration, the median value
of sodium plasma concentration and the area parameter.
Drained dialysate volume, dialysate sodium concentration
and thus derived sodium removal were calculated according
to dwell length in each case.
Statistical analysis
All data are expressed as mean values, with error bars
denoting SDs. The comparison between observed and
predicted data is expressed graphically as a scattergram
and statistically using Pearson linear regression, in which
each point represents one of the 27 subcategories according to
transport category, dwell length and glucose concentration.
The agreement and systematic bias of observed and predicted
values for these categories were analysed using the Bland
and Altman method. Comparisons between solute transport
groups or dialysis modalities were made using non-paired
t-tests or analysis of variance (ANOVA) where appropriate.
Results
Table 1. The parameters used for the computer simulation
three-pore model
Parameter
Value
Small pore radius (rs) Å
Large pore radius (rl) Å
Fractional small pore UF coefficient (as)
Fractional transcellular pore UF
coefficient (ac)
Fractional large pore UF coefficient (al)
UF coefficient (LpS) ml/min/mmHg
Osmotic conductance to glucose
(LpSsg) ml/min/mmHg
‘Unrestricted’ pore area over unit
diffusion distance for small pores
(A0/X) cm
PS (‘MTAC’) for glucose ml/min
(perturbation: increase by 1.7009)
PS (‘MTAC’) for Naþ ml/min
(perturbation: decrease by 0.3542)
Peritoneal lymph flow (L) ml/min
Transperitoneal hydrostatic pressure
gradient (P) mmHg
Transperitoneal oncotic pressure
gradient (pprot) mmHg
Dialysis fluid volume instilled ml
Peritoneal residual volume (Vr) ml
Serum urea concentration mmol/l
Serum sodium (and sodium associated
‘anion’ concentration) mmol/l
Serum glucose concentration mmol/l
Dissociation (Donnan) factor
43
250
0.900
0.020
0.080
0.074
Calculated
Variable
Patient demographics of the different dwell groups
A total of 1853 dwells from 182 patients (55% male)
were analysed. Ten percent of patients were using APD,
and these had been on PD dialysis significantly longer
than CAPD patients for each of the glucose and
transport categories (Table 2). Generally, the use of
higher glucose concentrations was associated with
longer time on treatment, especially in CAPD patients
(ANOVA, P ¼ 0.001) and higher solute transport. The
only exception to this was the APD 3.86% group,
where the low average transporters had been on treatment longer than high average and high transporters.
No consistent significant differences in body surface
area were found between these groups.
Calculated
Calculated
Comparison of observed and predicted data
0.3
8
Ultrafiltration. Figure 1a–c and Table 3 summarize
the comparison between observed and predicted UF
according to each of the 27 potential categories (three
transport, three glucose concentrations and three dwell
lengths). Overall, there is a strong and highly significant correlation, r ¼ 0.89, P<0.001. It can also be seen
that there is a systematic bias, such that as the net
UF increases, there is an increasing tendency for
the modelled data to overpredict the observed data
(Bland and Altman correlation r ¼ 0.84, P ¼ 0.001).
This overprediction is independent of transport category (Figure 1b) and was most marked in the very
small number of patients using 3.86% glucose
exchanges exclusively for short exchanges as part of
22
2000
300
20
Variable
6.5
0.93
The glucose osmotic conductance, glucose MTAC and sodium
MTAC are calculated by the model, while the area parameter
(A0/X) and sodium concentration are inserted for each dwell time
point.
A detailed analysis of sodium removel by peritoneal dialysis
their APD regime (Figure 1c). The possibility that this
overprediction of UF reflects the selective use of 3.86%,
and to a lesser extent 2.27%, glucose exchanges by
patients with worse membrane function was tested
1195
by comparing the UF capacity achieved in the PET
in these different categories. This analysis demonstrated that patients using stronger glucose fluids did
indeed tend to have less good UF capacity: 1.36%,
Table 2. Time on treatment (±SD) of patients in each of the dwell categories, grouped according to glucose concentration (1.36, 2.27 and
3.86%) and membrane transport category (TC)
TC
Glucose concentration
1.36%
Months on treatment
Total
CAPD
APD
2.27%
3.86%
n
Mean±SD
n
Mean±SD
n
Mean±SD
LA
HA
H
544
492
49
22.3±23.5
23.0±25.4
31.8±33.5a
296
432
88
23.0±22.3
26.2±27.0
42.3±43.4b
66
110
43
28.6±27.3
27.0±27.4
47.9±24.4b
LA
HA
H
LA
HA
H
526
469
44
18
23
5
21.6±23.2
21.4±23.9
26.4±27.1
40.8±25.2c
55.1±33.5d
79.5±49.2d
121
160
31
27
56
13
17.1±17.1
19.1±17.8
30.4±24.9
49.9±23.7d
46.1±37.4e
70.7±63.6d
61
96
38
5
14
5
26.8±26.6
23.9±25.1
48.5±25.5c
50.7±28.4f
48.9±33.1
43.4±15.8
n refers to the number of dwells in each subgroup. Data are also broken down by modality, CAPD and APD. Dwells are compared, within
the glucose concentration group, using ANOVA test and post hoc Bonferroni.
LA ¼ low average; HA ¼ high average; H ¼ high.
a
¼ compared to other transport groups, P<0.05.
b
¼ compared to other transport groups, P<0.01.
c
¼ compared to equivalent CAPD group, p<0.001.
d
¼ compared to equivalent CAPD group, p<0.05.
e
¼ compared to equivalent CAPD group, p<0.02.
f
¼ compared to equivalent CAPD group, p<0.01.
(a)
3300
3100
Observed (ml)
2900
2700
2500
2300
2100
1900
1700
1500
1500
1700
1900
2100
2300
2500
2700
2900
3100
3300
Predicted (ml)
Fig. 1. Correlation between observed and predicted net ultrafiltration, r ¼ 0.89, P<0.001. Each point represents one of 27 different
combinations of glucose concentration, dwell length and transport category. (a) Groupings according to glucose concentration: 1.36%
(open triangles), r ¼ 0.85, P<0.001; 2.27% (filled circles), r ¼ 0.96, P<0.001; and 3.86% (open circles), r ¼ 0.83, P<0.001. (b) Data are
grouped according to transport category: low average (open circles), r ¼ 0.85, P<0.001; high average (open triangles) r ¼ 0.92, P<0.001;
and high (filled diamonds) r ¼ 0.94, P<0.001. (c) Data are grouped by dwell length, r ¼ 0.85, P<0.001: short (filled diamonds), r ¼ 0.02,
NS; medium (open circles) r ¼ 0.94, P<0.001; and long (open triangles) r ¼ 0.98, P<0.001.
1196
(b)
M. C. Aanen et al.
3300
3100
Observed (ml)
2900
2700
2500
2300
2100
1900
1700
1500
1500
1700
1900
2100
2300
2500
2700
2900
3100
3300
2700
2900
3100
3300
Predicted (ml)
(c)
3300
3100
Observed (ml)
2900
2700
2500
2300
2100
1900
1700
1500
1500
1700
1900
2100
2300
2500
Predicted (ml)
Fig. 1. Continued.
Table 3. Comparison between observed and predicted ultrafiltration according to category
Category
Low average transport
High average transport
High transport
Glucose 1.36%
Glucose 2.27%
Glucose 3.86%
Short (90 min)
Medium (300 min)
Long (540 min)
All groups
Mean difference (SD)
predicted observed
Net ultrafiltration (ml)
Dialysate sodium concentration (mmol/l)
Net sodium removal (mmol)
272±271
216±171
168±186
25±81
168±64
463±141
240±226
273±200
143±205
219±210
1.03±3.93
0.73±2.69
0.14±2.76
1.73±0.96
0.19±1.55
3.83±2.98
1.78±4.35
0.03±2.50
0.10±1.84
0.64±3.08
25.5±25.3
19.5±17.1
14.6±19.9
1.1±12.2
15.1±7.7
43.4±11.9
21.4±18.5
27.0±19.6
11.2±22.9
19.9±20.7
A detailed analysis of sodium removel by peritoneal dialysis
Table 4. Association between peritoneal ultrafiltration capacity
(derived from PET) and the selection of glucose concentration
Glucose
concentration used
Low average
High average
High
Glucose 1.36%
Glucose 2.27%
Glucose 3.86%
502±199*
451±180
471±196
437±238
450±281
388±231z
398±269
294±312
229±560y
For a given transport category, it can be seen that patients were
more likely to use a relatively hypertonic exchange if they had
worse UF capacity derived from their peritoneal equilibration test.
ANOVA (post hoc analysis): *P ¼ 0.009 compared with 2.27%,
z
P ¼ 0.07 compared with 2.27%, yP ¼ 0.03 compared with 1.36%.
469±221 ml, n ¼ 1150; 2.27%, 426±266 ml, n ¼ 431;
3.86%, 385±329 ml, n ¼ 237, ANOVA P<0.001.
These differences were in part attributable to solute
transport category, as patients with higher transport
were more likely to use hypertonic exchanges, a
difference that is taken into account in the modelling.
However, if the differences in UF capacity are broken
down by transport category (see Table 4), it can be seen
that there is a tendency for patients using hypertonic
exchanges to have worse UF capacity for a given solute
transport.
Dialysate sodium concentration. As with UF, there
was a strong and significant correlation between
observed and predicted values across the 27 categories
of exchange r ¼ 0.815, P<0.001 (Figure 2a). The
difference between observed and predicted values was
least for high transporters and long dwells, those dwells
in which closest equilibration with plasma sodium will
have occurred. The process of sodium sieving was
consistently predicted, although again, for those few
patients using 3.86% glucose in their short APD dwells,
this was less than predicted (Figure 2b).
Net sodium removal. Again, correlation between
observed and predicted sodium removal is high,
r ¼ 0.89, P<0.001. As net removal reflects the sum
effects of UF and dialysate sodium concentration, the
overestimation of UF by the model results in a proportional overestimate in sodium removal (see Figure 3).
Comparison of long overnight CAPD exchanges
with daytime APD exchanges
As both CAPD and APD patients undertook long
dwells, it was possible to make a direct comparison of
the observed and predicted fluid and sodium removal
by modality. There was no difference between the two
modalities in dialysate sodium concentration, UF or
net sodium loss (see Figure 4), in terms of both the
absolute observed amounts and the predictive power of
the model.
Discussion
This study presents a detailed analysis of UF and
sodium removal in a cross-section of CAPD and APD
1197
patients, with particular attention to the effects of
glucose strength, dwell length and membrane transport
characteristics. It is also the first attempt to conduct a
comprehensive comparison between the predicted UF
and sodium removal by the TPM using computer
simulations and that observed empirically. Although
the TPM is used in various studies to assess the
peritoneal membrane in a mathematical manner, and
is the basis for the PDC test, the model has never
been validated in such a manner for the factors that
determine sodium removal.
In discussing the observational data, it can be seen
that the values obtained for net UF, dialysate sodium
concentration and thus sodium removal are influenced
by glucose strength, dwell length and membrane
transport broadly as expected from theoretical considerations [18]. In other words, by increasing the
glucose tonicity, there is more UF and sodium sieving,
whereas the opposite tendency occurs as membrane
transport increases. The literature contains very little
comparative data of this sort, especially on a dwell-bydwell basis. The analysis by Rodriguez-Carmona and
Fontan [8] gives similar values for dialysate sodium
concentration and removal overall, but does not break
these down by glucose concentration and membrane
transport characteristics, and does not distinguish
between CAPD and APD ‘short’ dwells. It should be
emphasized that there is considerable variability,
especially in UF, within each of the 27 transport,
dwell length and glucose concentration categories, as
judged by the relatively large SDs that can be seen in
Figures 1a, 2a, 3 and 4. This variability is probably due
to several things, including the accuracy of measurement, variability in sump volume (which may be altered
by recumbency or variable programming in APD
patients), variability in plasma glucose and urea,
hydrostatic pressure and real differences in the intrinsic
properties of the membrane, such as UF coefficient. As
none of these were measured independently in this
study, they were treated as constant, fixed variables
in the modelling. The use of a single value for plasma
urea, chosen in order to simplify the modelling process,
will also have resulted in some of this variability as urea
is osmotically active, accounting for 40 ml less UF for
an increase in plasma concentration of 20–30 mmol/l
in this model. Thus, 10% of the variability in UF
between the observed and predicted values might be
accounted for by this oversimplification.
When the observational data were compared with
those predicted by the computed values generated by
the TPM, a high degree of correlation between UF,
dialysate sodium and consequently sodium removal
was observed. It is important to emphasize the nature
of the comparison between observed and predicted data
that we wished to explore in this study. This was not
an attempt to determine the reliability with which this
version of the TPM could predict UF (as for example
has been shown with the PDC) or sodium removal in
individual patients, but an attempt to see how well the
TPM described patterns of UF and sodium removal in
broad terms. Similarities would indicate that the model
1198
(a)
M. C. Aanen et al.
142
Observed (mmol/l)
137
132
127
122
117
117
122
127
132
137
142
Predicted (mmol/l)
(b)
142
Observed (mmol/l)
137
132
127
122
117
117
122
127
132
137
142
Predicted (mmol/l)
Fig. 2. Correlation between observed and predicted dialysate sodium concentration, r ¼ 0.82, P<0.001. As in Figure 1, each point
represents one of 27 different combinations of glucose concentration, dwell length and transport category. (a) Groupings according to
glucose concentration: 1.36% (open triangles), r ¼ 0.97, P<0.001; 2.27% (filled circles), r ¼ 0.90, P<0.001; and 3.86% (open circles),
r ¼ 0.85, P<0.001. (b) Data are grouped according to dwell length: short (filled diamonds), r ¼ 0.04, NS; medium (open circles) r ¼ 0.91,
P<0.001; and long (open triangles) r ¼ 0.96, P<0.001.
is generally correct in its mechanistic description of
how the membrane works, whereas differences might
point to specific parameters that are either wrong or
need to be modified for the future. The high degree of
correlation between predicted and observed data does
support the view that the model is fundamentally
correct in its mechanistic description. There were,
however, limitations to the accuracy of this description
that require further discussion.
It can be seen that there are very significant degrees
of systematic bias between the observed and predicted
data, in particular the overprediction of UF that
increases as the UF volume increases. This could be
due to one of two things: either a selection bias in the
use of increasingly hypertonic exchanges according
to patient need, or a real difference in the osmotic
conductance of the membrane associated with increasing glucose concentration. The former of these explanations is a weakness of the cross-sectional nature of
the study design in which patients were using their
typical dialysis prescription. Patients are more likely to
use hypertonic exchanges for one or more of three
reasons: to accommodate increased fluid intake, to
compensate for lost urine output or because their
A detailed analysis of sodium removel by peritoneal dialysis
1199
120
Observed (mmol/exchange)
100
80
60
40
20
0
-40
-20
0
20
40
60
80
100
120
-20
-40
-60
-80
Predicted (mmol/exchange)
Fig. 3. Correlation between observed and predicted sodium removal, r ¼ 0.89, P<0.001. As in Figure 1, each point represents one of 27
different combinations of glucose concentration, dwell length and transport category, grouped according to glucose concentration, 1.36%
(open triangles), r ¼ 0.57, P<0.005; 2.27% (filled circles), r ¼ 0.88, P<0.001; and 3.86% (open circles), r ¼ 0.89, P<0.001.
120
100
Observed (mmol)
80
60
40
20
0
-40
-20
0
20
40
60
80
100
120
-20
-40
-60
-80
Predicted (mmol)
Fig. 4. Correlation between observed and predicted sodium removal as achieved during the long exchange according to modality, CAPD
(open circles), r ¼ 0.95, P<0.001; APD (filled circles), r ¼ 0.90, P<0.001.
membranes are less efficient for a given glucose gradient
(reduced osmotic conductance). Only the last of these
reasons could explain less observed than predicted fluid
removal. Certainly these patients had been on dialysis
longer, and when we analysed the relative UF capacity
obtained from the PET, we did find that for a given
solute transport this was less for patients using
hypertonic fluids, indicating that selection for reduced
peritoneal osmotic conductance did occur in this study,
entirely in keeping with the longitudinal observations
of membrane function we have observed. This reduction in osmotic conductance is not, however, sufficient
1200
and consistent enough to explain all the discrepancy
between the observed and predicted data.
An alternative explanation would be that the
osmotic conductance, itself determined by the product
of the UF coefficient (LpS) and the reflection coefficient
to glucose (s) is influenced by the glucose concentration. It is important to understand that in this version of
the TPM, the value for LpSs is fixed for all dwells and
that the PS for glucose has been perturbed (increased)
in such a way as to reduce the driving osmotic gradient.
The reason and rationale for this are as follows: early
studies validating the TPM, in rather a small number of
patients, indicated that the amount of UF was not as
great as predicted. The rationalization is that the effects
of the interstitium dissipate the actual glucose gradient
at the peritoneal capillary wall. Data from this study
would suggest that further perturbation is required,
perhaps in a non-linear fashion, to improve the agreement between observed and predicted data.
The agreement between observed and predicted
dialysate sodium concentration was good, including
the effect of sodium sieving, but again showed bias.
This could be explained mostly by reduced sodium
sieving in a small number of patients using exclusively
3.86% glucose in the overnight short dwells of their
APD regime. These patients all had clinical and PET
evidence of UF failure resulting in reduced free water
transport as described. Again, it should be noted that in
this application of the TPM, the PS for sodium was
perturbed, on this occasion reduced, to account for the
various factors that alter free diffusion of this ion.
Finally, provided appropriate adjustments for the
overfill and flush volume for CAPD patients was taken
into account, there was no systematic difference in
the UF and sodium removal in the long exchanges
according to modality. In comparing APD and CAPD
patients in this study, care has to be taken due to the
relatively small numbers of APD patients, who were
also more likely to be anuric, longer on treatment and
thus have acquired changes in membrane function;
the smaller SDs (see Figure 4) would suggest that this
was, if anything, a more homogeneous group, however,
possibly reflecting more precise measurements of UF
that are possible with APD, and the parallel nature of
the regression lines would indicate that the fundamental relationship between UF and glucose concentration
is not different.
Acknowledgements. The authors thank Professors Bengt Rippe
and Raymond Krediet for their constructive encouragement of this
work. M.C.A. was sponsored by a grant from the Dutch National
Kidney Research Foundation. D.V. is supported by The European
Union, Contract FMRX-CT98-0219.
Conflict of interest statement. None declared.
M. C. Aanen et al.
References
1. Ates K, Nergizoglu G, Keven K et al. Effect of fluid and
sodium removal on mortality in peritoneal dialysis patients.
Kidney Int 2001; 60: 767–776
2. Brown EA, Davies SJ, Rutherford P et al. Survival of
functionally anuric patients on automated peritoneal dialysis:
the European APD Outcome Study. J Am Soc Nephrol 2003;
14: 2948–2957
3. Lamiere NH, Vanholder R, Van Loo A et al. Cardiovascular
diseases in peritoneal dialysis patients: the size of the problem.
Kidney Int 1996; 50: S28–S36
4. Rippe B. A three-pore model of peritoneal transport. Perit Dial
Int 1993; 13 [Suppl 2]: S35–S38
5. Krediet R, Lindholm B, Rippe B. Pathophysiology of peritoneal
membrane failure. Perit Dial Int 2000; 20: S22–S42
6. Yang B, Folkesson HG, Yang J, Matthay MA, Ma T,
Verkman AS. Reduced osmotic water permeability of the
peritoneal barrier in aquaporin-1 knockout mice. Am J Physiol
1999; 276: C76–C81
7. Carlsson O, Nielsen S, el-R Z, Rippe B. In vivo inhibition
of transcellular water channels (aquaporin-1) during acute
peritoneal dialysis in rats. Am J Physiol 1996; 271:
H2254–H2262
8. Rodriguez-Carmona A, Fontan MP. Sodium removal in
patients undergoing CAPD and automated peritoneal dialysis.
Perit Dial Int 2002; 22: 705–713
9. Ortega O, Gallar P, Carreno A et al. Peritoneal sodium mass
removal in continuous ambulatory peritoneal dialysis and
automated peritoneal dialysis: influence on blood pressure
control. Am J Nephrol 2001; 21: 189–193
10. Stelin G, Rippe B. A phenomenological interpretation of the
variation in dialysate volume with dwell time in CAPD. Kidney
Int 1990; 38: 465–472
11. Van Biesen W, Carlsson O, Bergia R et al. Personal dialysis
capacity (PDC(TM)) test: a multicentre clinical study. Nephrol
Dial Transplant 2003; 18: 788–796
12. Vonesh EF, Story KO, O’Neill WT. A multinational
clinical validation study of PD ADEQUEST 2.0. PD
ADEQUEST International Study Group. Perit Dial Int 1999;
19: 556–571
13. Mahon A, Fan S. Are we underestimating the problem of
ultrafiltration failure in peritoneal dialysis patients? [abstract].
Renal Association, Autumn Meeting, 2004 http://www.
triangle3.org.uk/cgi-bin/absdb/absdb_view.cgi?AbstractID=
RA4553
14. Davies SJ, Brown B, Bryan J, Russell GI. Clinical evaluation
of the peritoneal equilibration test: a population-based study.
Nephrol Dial Transplant 1993; 8: 64–70.
15. La Milia V, Di Filippo S, Crepaldi M et al. Spurious
estimations of sodium removal during CAPD when [Na](þ) is
measured by Na electrode methodology. Kidney Int 2000; 58:
2194–2199
16. Rippe B, de Arteaga J, Venturoli D. Aquaporins are
unlikely to be affected in marked ultrafiltration failure:
results from a computer simulation. Perit Dial Int 2001; 21:
S30–S34
17. Rippe B, Levin L. Computer simulations of ultrafiltration
profiles for an icodextrin-based peritoneal fluid in CAPD.
Kidney Int 2000; 57: 2546–2556
18. Wang T, Waniewski J, Heimburger O, Werynski A, Lindholm B.
A quantitative analysis of sodium transport and removal
during peritoneal dialysis. Kidney Int 1997; 52: 1609–1616
Received for publication: 13.3.04
Accepted in revised form: 8.3.05