Back to table of contents
Thursday, 13. February 2003 12:20 PM GMT
Adequacy of dialysis refers to the compromise between therapeutic outcome &
cost & inconvenience.
Assessment of adequacy of dialysis
The prescription of dialysis requires knowledge of the normal function of the
kidney, of patient metabolism and physiology and of dialysis technology. Central
to this process is the need to define and quantify both the dialysis procedure
and the needs of the patient. Uncertainties about the reliability of the
dialysis, the response of the patient to the procedure and pathophysiological
changes within the patient require that objective measurements of dialysis
adequacy are made regularly. These measurements are designed to quantify the
effect of dialysis on the patient. This allows continuous adaptation of the
treatment to changing patient needs and the rapid detection of treatment failure
in each patient. The exercise also allows each dialysis unit to audit its
performance, guiding systematic improvement in therapy. Since the functions of
dialysis are diverse, measurement of dialysis adequacy is multi-dimensional.
Monitoring includes clinical assessment and objective measurement, including
weight, blood pressure, laboratory investigations and some measure of the amount
of solute cleared during the dialysis process (Tattersal et.al., 1998).
Are you a doctor or a nurse?
Do you want to join the Doctors Lounge online medical community?
Participate in editorial activities (publish, peer review, edit) and
give a helping hand to the largest online community of patients.
Click on the link below to see the requirements:
Doctors Lounge Membership
Detection of under dialysis by clinical parameters depends on awareness and
frequency of control; the risk that under dialysis is overlooked is substantial.
Inadequate therapy can remain unrecognized when therapeutic decisions are
exclusively based on clinical parameters. Nevertheless, the inverse is as true,
and follow up of dialysis adequacy should never be restricted to static or
dynamic biochemical parameters (Solute concentration, clearance, kinetic
modeling) when estimating adequacy, clinical signs of under dialysis should
never be ignored (Raja et.al., 1978).
Kjellstrand et al.(1978) have suggested that there is an increase of
morbidity and mortality when dialysis is started late. The current trend to
restrict protein in pre-end stage renal disease may reduce subjective complaints
and postpone the start of dialysis, but will not necessarily improve overall
outcome, especially as protein restriction per se may increase morbidity on
dialysis (Degoulet et al., 1988).
Clinical signs that may reflect underdialysis:
(Vanholder and Ringoir, 1992)
Static biochemical tests:
Single "Static" evaluations are influenced by other factors than
dialysis adequacy as well. The concentration of retention products is
proportional to 1/clearance rate. If the retention product is toxic (e.g. K)
then you need to be sure that the concentration of the product remains below the
toxic range. When protein intake is deficient, pre-dialysis urea will remain low
inspite of inadequate dialysis, this may incorrectly instigate a decrease in the
dialysis quantity concentration as unstable indices (Shapiro et.al., 1983). In a
study by Degoulet et.al.,(1982), low urea corresponded to a high mortality risk.
In parallel, Oksa et.al.,(1987), demonstrated that survival was higher for a low
urea/creatinine. Lowrie and Lew (1990), demonstrated that low serum albumin and
low serum creatinine were associated with a high death risk. All these findings
refute the general belief that bad outcome is only found at high pre-dialysis
serum solute concentrations.
Serum levels of albumin, creatinine and body mass index were found to have
significant survival predictive value, other variables such as total
cholesterol, serum uric acid, total protein had no significant predictive value.
Potassium, phosphate and calcium, may contribute to survival, high serum
creatinine was associated with low risk of death, but was also associated with
high dialysis dosage. While increased dialysis may lead to greater patient body
mass and generation of creatinine, it is also possible that physicians consider
serum creatinine value and provide dialysis at greater doses to better nourished
patients. If albumin and creatinine values rather than diabetes mellitus were
applied as predictors of mortality risk. It might be found that the extremely
high mortality associated with diabetes mellitus is related as much to
undernutrition as it is to diabetes mellitus per se (Iseki et.al., 1993).
One of the functions of the human kidney is to regenerate bicarbonate.
Haemodialysis achieves this by including a supra-physiological concentration of
bicarbonate in the dialysis fluid. During dialysis, bicarbonate diffuses into
the blood across the dialyzer membrane. One of the aims of dialysis is to
normalise the serum bicarbonate concentration. Optimal survival has been shown
to relate to normal pre-dialysis serum bicarbonate concentration (Lowrie and Lew,
1990). This is achievable if the dialysis is correctly prescribed and planned.
Bicarbonate has a similar molecular weight as urea so the relative mass of
bicarbonate transferred into the blood will be proportional to the Kt/V (will be
discussed in the dynamic biochemical tests) and the bicarbonate concentration
gradient across the membrane. To some extent, a natural homeostasis occurs in
that those patients who are acidotic will have lower bicarbonate and a higher
concentration gradient. A dialysate bicarbonate concentration of 40 mM will keep
the patients bicarbonate concentration at 22-26 mM in virtually all cases if the
Kt/V is 1. If the Kt/V is increased, a proportional reduction in concentration
gradient is required so that for a Kt/V of 1.5, the dialysate bicarbonate
concentration should be reduced to 35 mM. Individual variations in the dialysate
bicarbonate concentration may be required to cope with differing metabolic
rates, diets and oral base intake (e.g. calcium carbonate). Most dialysis
machines allow individualised control of the dialysis fluid bicarbonate
concentration from the front control panel. Assessment of dialysis adequacy is
not complete without measurement of the pre-dialysis plasma bicarbonate
concentration. In the adequately dialysed patient, the bicarbonate concentration
should be in the normal range (Tattersal et.al., 1995).
Static biochemical parameters used in the assessment of dialysis adequacy: (Vanholder
and Ringoir, 1992)
Red blood cell count
Dynamic biochemical tests:
Kt/V is the mainstay of dialysis adequacy (Gotch, 1990). Originally the term
was function of dialyzer clearance (K), dialysis time (t) and urea distribution
volume (V). An approximate value from K can be read of a graph on the dialyzer
data sheet or calculated from the dialyzer performance characteristics, blood
and dialysate flow rates. (V) is equal to the body water volume and can be
calculated approximately from patient measurements. Kt/V is precisely the
exponential term describing the decline in BUN during dialysis for the case in
which we assume no interdialytic weight gain. Kt/V may be interpreted as the
fractional urea clearance (Levine and Bernard, 1990). This approach has the
* The dose of dialysis predicted to be delivered by any dialysis prescription
can be calculated from (K), (t) and (V). This is termed the prescribed Kt/V.
* The dialysis time needed to achieve a target dialysis dose can be
calculated for a given patient size (from which V can be calculated) and a given
combination of dialyzer type, blood and dialysate flow rate (from which K can be
* The dose of dialysis actually delivered to the patient can be calculated
directly from measurements of blood urea made pre- and post-dialysis. This is
termed the delivered Kt/V.
* The comparison of the prescribed Kt/V and delivered Kt/V quantifies any
inefficiency in the process or any errors in K or V (which cannot easily be
measured precisely). This information can be used for diagnosis and
trouble-shooting and to fine-tune the dialysis prescription to ensure that
adequacy targets are met.
The equation can be re-arranged to return Kt/V from only measurements of
pre-and post-dialysis urea thus;
This is the delivered Kt/V. it is calculated directly from blood
concentration measurements and no hard-to-measure values such as V or K are
needed. This simple equation assumes that V remains constant and that no urea is
generated during dialysis. To take account of these factors, the equation is
expanded to take the form;
Where Vpre and Vpost are the urea distribution volumes pre- and post-
dialysis and UG is the mass of urea generated during the dialysis. In effect
Kt/V is the natural logarithm of the mass of urea in the patient at the start of
dialysis divided by (the mass of urea in the patient at the end of dialysis
minus any urea generated during the dialysis). The effect of the urea generation
is small but important. If it is ignored, long dialyses will be significantly
underestimated. In an extreme example, continuous dialysis will always result in
a Kt/V=0 if urea generation (which is opposing the fall in urea concentration
due to dialysis) is ignored. The effect of ultrafiltration is also small but
important. The contribution of ultrafiltration dose depends on V but since its
contribution is small, errors in V are relatively unimportant.
It is now known that other factors within the patient which limit the rate of
transfer of solutes from peripheral parts of the body into the fistula are
important. These factors include the rate of diffusion and blood flow between
body compartments. These factors reduce the effective K and, therefore Kt/V and
result in the post-dialysis rebound. To take account of these factors, Kt/V
should, ideally, be calculated using a post-dialysis sample taken 30-60 minutes
after dialysis when the urea concentrations have re-equilibrated. It is not
usually practical to measure this post rebound concentration directly but it can
be calculated with acceptable accuracy from the pre-and post- dialysis
concentrations and the time between these samples. The effects of
ultrafiltration and urea generation can be included by using the correct
If rebound is ignored, Kt/V will be overestimated by up to 30%, especially in
short dialyses. For this reason, recently, the term Kt/V has been expanded to
include the rebound effects. The combination of the urea generation and rebound
corrections ensure that treatments with the same Kt/V and the same frequency per
week will remove the same mass of urea (relative to the pre-dialysis mass)
regardless of the actual values of K, t and V.
Blood compartment clearance:
Clearance can be defined as the volume of blood that is purified per unit of
time, which depends both on diffusion and convection. Higher clearance equals
principally better adequacy (Vanholder and Ringoir, 1992).
Clearance is calculated based on blood flow (QB) determinations and
concentration in blood samples collected from the inlet and outlet bloodlines.
Several influencing factors should be considered. The error in blood flow
determinations might be substantial. QB Can most readily be estimated by
injecting an air bubble into a disposable additional track of known cross
sectional area and length the measured transit time over a known length yields a
reliable blood flow calculation (Flanigan et al., 1991).
A mass ratio between red blood and plasma of 0.85 has been reported for urea
which acknowledges that respective plasma and erythrocyte water contents are 93%
and 79% (Flanigan et al., 1991). The use of photoelectric cells may reduce
observer error (Peoples et al., 1973).
An alternative is the Electro-magnetic flow probe, but this approach
necessitates frequent calibration and sterilization. Ultrafiltration adds extra
solute flux to diffusion, so that elimination is increased by a quantity that
does not match ultrafiltration rate (QF) (Gupta and Jaffrin, 1984), its negative
impact on diffusive clearance was first recognized in coil dialyzers (Husted et
al., 1976) and appeared later to be present for all types of dialyzers and
should always be considered for the calculation of QB based clearances. The
additional clearance delivered by ultrafiltration equals ultrafiltration,
multiplied by 100% minus percentage removal. The greater the diffusive removal,
the less is the impact of ultrafiltration removal and vice versa. Recirculation
occurs when dialyzed blood is recycled through the dialyzer inlet. It is
considered to be a typical drawback of single needle (SN) dialysis. But is as
well present in two needle dialysis, especially high efficiency dialysis
(Collins et al., 1988 and Sherman et al., 1991) whereby blood flow and
recirculation are related. Enhancing conditions are:
* High dialyzer blood flows.
* Vascular access in flows lower than dialyzer blood flow.
* Stenosis at the access outflow.
* Common or close inlet and outlet vascular pathways.
* Increased length of bloodlines.
* Increased compliance of dialyzer outflow.
* Incorrect position of the needle in arterio-venous fistula.
* Small stroke volumes (in single needle dialysis).
* Small needle and tubing diameter.
The non-homogenesity of blood may hinder removal by imposing a slower
transport from blood cells to blood water than from blood water to dialysate.
This necessitates erythrocyte equilibration with plasma (Skalsky et a1.,1978)
and/or cell lysis (Bass et al., 1973), before determination of concentration to
calculate blood clearance, or determination of plasma clearances after immediate
processing before re-equilibration (Basile et al., 1986).In summary, blood cells
side clearance are subjected to important flows, which necessitates correct
blood flow and concentration measurements and correction for recirculation and
ultrafiltration, protein binding and red blood cells / plasma disequilibrium (Skalsky
et al., 1978). Most errors will result in an overestimation of clearance and
create a false feeling of security. Even if determined cautiously, it is
suggested to match blood side clearances to simultaneous dialysate clearances (Sargent
and Gotch, 1989).
Clearance based on distribution volume based on first order kinetic
principles, this calculation multiplies the ratio of single pool distribution
volume (V) over dialysis time with the negative logarithm of the ratio
concentration pre/post dialysis (Sargent et al., 1975 and Casino et al., 1990).
The source of error with this approach is minimal and depends on only two
factors, concentration and distribution volume. This approach used for the
determination of urea clearance. Volume of urea may be estimated from its
distribution over body water, which, however, occupies a variable fraction of
body weight (Moore et al., 1952).
In addition, urea distribution volume doesn't always concur with body water.
The current trend of considering Vurea (volume distribution of urea) as 58% of
total body weight is thus a source of error. The use of morphometeric data may
result in more reliable calculations although those data are currently based on
non-uraemic subjects. Exact estimations of Vurea can only be obtained by
injection of radioactive urea, or by the use of multiple pool model .In
conclusion, clearance values based on pre and post dialysis concentrations and
distribution volume, will give acceptable results, as far as the approximation
of V is reliable. Whatever the method, the clearance approach estimates solute
elimination in function of the dialyzer and the dialysis technique, but doesn't
take into account patient (body size), solute (generation) and/or kinetic
characteristics (protein binding, cell membrane transport) (Vanholder and
The principal is the same as for genuine "urinary" renal clearance:
collection of all dialysate and calculation of total solute waste. Calculations
of mean pre and post dialysis values may result in an underestimation of
clearance, as a linear decay of blood solute concentration is presumed. Further
error sources are low dialysate concentrations and incorrect determinations of
dialysate volume. Specific devices that collect small timed aliquots of
dialysate may be less cumbersome than large graduated tanks (Garred et. al.,
1989). Dialysate clearance, by its concept, is not subject to the vagaries of QB
determination, ultrafiltration or recirculation (Gibson and Gotch, 1976).
The dialysate collection method has the advantage of avoiding the need for a
rebound or urea generation correction, even a post-dialysis urea sample is not
needed, however, unlike the blood concentration method, the dialysate method
relies heavily on an accurate value for (V) (Argiles et.al., 1997).
In the dialysate collection method, the total mass of solute removed by
dialysis including ultrafiltration is calculated from the total dialysate output
volume (dv) multiplied by the average solute concentration in the dialysate
(dc). In this way, the dialysis is quantified in exactly the same way as in the
Urea kinetic model:
The use of mathematical models to describe the physical processes involved in
dialysis therapy is well known (Abrecht and Prodany, 1971).
During recent years urea kinetic modeling has been widely used in dialysis
therapy as an analytic tool to improve clinical understanding of the uraemic
syndrome and to prescribe and deliver reproducible and quantified doses of
Several models have been developed in the past with two major goals: first,
the understanding and quantitative analysis of the physiological processes of
patients on hemodialysis (Abrecht and Prodany, 1971) second, the estimation of
patient parameters in order to deliver adequate doses of dialysis by controlling
the removal of toxic solutes (Sargent and Gotch, 1975) thus reducing possible
complications since the rigorous description of uraemic state would require the
knowledge of the kinetics of all toxic substances . Urea is usually assumed to
be a marker solute for all toxins with low molecular weight (Gotch and Keen,
Single pool models:
Single pool models relate to small water-soluble compounds, with equal
distribution. Distribution volume (V) and generation rate (G) are interactive
and are calculated by titration from intra- and interdialytic shifts (Gotch,
1976). The single pool model assumes:
* Instantaneous mixing without shifts within the pool.
* Constant generations.
* Constant rate of interdialytic weight gain.
These assumptions are correct only by approximation. Pitfalls of the single
pool urea kinetics was incomplete delivery if dialysis prescription (Sargent,
1990) induces, when not recognized, flaws in kinetic calculation. Errors in
registration of dialysis time due to connection, disconnection, turning down of
blood flow and interruptions, manufacturer clearances that overestimate true
clearance, dialysis devices that overestimate true blood flow, and lack of
correction for recirculation all will result in a misconception of dialysis
dose, and hence of G and V (Aebischer et al., 1985),
Multiple pool models:
Many solutes and potential uraemic toxins distribute over several pools,
substances with slow transport from intracellular to extracellular, protein
binding, and/or a middle to high molecular weight such as peptides (Peeters et
al., 1974) or guanidine (Giovannetti and Barsotti, 1974). The site of generation
may affect concentration both in the cell and in the plasma, especially when the
cell membrane restricts transport, When metabolite is generated in one small
well perfused specialized organ (for example the liver for urea, generation can
be considered to be extracellular. Use of more than two pools may add little
accuracy (Popovitch et al., 1975).
Multiple pool models can predict changes in body compartments normally
inaccessible to the clinician. Since toxicity will mainly affect intracellular
systems, It may be clinically more relevant to focus on predicted intracellular
or intratissular concentrations, rather than on plasma concentrations, according
to the concept of biologically active distribution volume. Two pool models
should thus not be discarded as overly complex, as they may disclose important
new information. Even the intradialytic behavior of urea is better described by
a multiple than by a one compartment model (Keshaviah et. al., 1985) whereby
transcapillary exchange between intravascular and slow equilibrating
interstitial fluid spaces may be the rate limiting step (Bowsher et. al., 1985).
Mathews and Downey, (1984) demonstrated that both pools are nearly equal in
volume and much larger than the intravascular volume of a normal adult they
postulated that the urea of the blood rapidly mixes with intracellular water of
well-perfused organs, whereas it mixes more slowly with poorly perfused organs,
and interstitial water. If an exact scientific approach of the intradialytic
behavior of urea is to be pursued, a two-pool model may thus be more
appropriate. A single pool model is more attractive for routine clinical
conditions, because of its easy applicability. A phenomenon related to the two
compartment behavior of urea is the postdialysis rebound that is especially
observed with short high efficiency dialysis (Pedrini et al., 1988).
The urea rebound may not be caused exclusively by transtissular
disequilibrium, but may be affected by an increase in protein catabolism (Farell,
1986) induced by loss of amino acids and glucose and/or a catabolic effect
induced by blood contact with the dialysis membrane (Gutierrez et al., 1990).
Lim and Flanigan,(1989) showed patients on a constant protein diet with decrease
in protein catabolic rate when interdialytic intervals were increased from two
to three days, which suggests a catabolic effect of dialysis. The occurrence of
urea rebound, even with radiolabeled exogenous urea, however, suggests a
The Solute removal index:
In order to correct for the intermittency and rebound effects in
haemodialysis, a new method of quantifying small solute removal has been
proposed. The solute removal index (SRI) is the ratio of mass of urea removed by
dialysis to the mass present at the start of dialysis (Keshaviah and Star,
A dialysis strategy aimed at achieving the greatest SRI is required. In
continuous treatments and in normal renal function, SRI is almost the same as
Kt/V (it is reduced slightly by 2-pool effects). However, in intermittent
treatments (such as hemodialysis), SRI may be much less than Kt/V, depending on
the frequency and duration of dialysis (the degree of intermittency) and the
total weekly Kt/V delivered. For a given weekly Kt/V, daily or very long
thrice-weekly dialyses have greater SRI than the shorter treatments (Tattersal
The solute removal method (urea or other solutes) becomes the preferred and
accurate method to quintitate dialysis as it abolishes all the errors of urea
kinetic modeling, so it should be regarded as the gold standard of dialysis
quantification, but its draw-back was the difficulty in collecting the whole
spent dialysate (Shohate and Boner, 1997).
Using urea appearance as a measure of
Assessing the nutritional status of patients with chronic renal failure can
be extremely difficult (Schoenfeld et. al., 1983). Since urea is the predominant
nitrogenous product of protein catabolism, a number of investigators have
examined the question whether the easily measured urea appearance rate could be
used as an index of protein catabolism, and hence dietary protein intake? (Maroni
et. al., 1985). PCR (protein catabolic rate) diverges from protein intake
estimated by delivery anamnesis (Panzetta et. al., 1990). The assumed cause is
that outpatient food intake records are of limited value (Schoenfled et. al.,
1983). Patients may overestimate a low food intake and vice versa, Imbalance
may, however, also be due to the unsuspected presence of catabolism or
anabolism. Controls in metabolic wards of the reliability of PCR as an index of
protein intake are rant (Cogan et.al., 1981). Asymmetry in protein intake,
dialysis efficiency and/or dialysis schedule may render kinetic results
unreliable. Gross changes (infection, pericarditis, fluid overload) may require
larger increases in dialysis quantity than those proposed by the National
Cooperative Dialysis Study (NCDS) in mostly well equilibrated patients (Depner
and Chear, 1989).
For a fixed relationship to exist between the urea appearance rate (G) and (PCR)
there are two basic requirements. First, all urea degraded within the gut must
be quantitatively reconverted back to urea; this guarantees that all of the urea
appearance reflects protein catabolism. Second, non-urea nitrogen appearance
defined as the sum of the accumulation of non-urea forms of nitrogen (e.g.,
creatinine, urate, etc) and of the total-body excretion of non-urea nitrogenous
wastes (e.g. fecal nitrogen, non-urea urinary nitrogen, etc) must be fairly
constant irrespective of a patient's diet or level of renal function. The
following equation, describing the fixed relationship between the urea
appearance rate and the protein catabolic rate, was derived by (Borah et. al.,
G = 0.154 PCR
G represents the urea appearance rate in grams of urea nitrogen per day; the
symbol (G) is used because it is the net generation of urea (G), as defined by (Sargent,
1983) and PCR represents the rate of protein catabolism (from both dietary and
endogenous protein sources) in grams of protein per day. Although this equation
was derived for dialyzed patients, it has wide applicability. The slope of 0.154
implies that for each 10 gram of protein that is catabolized, 1.54 gram of urea
nitrogen will be generated. This is approximately 96% of the nitrogen content of
the protein (assuming that on average the complete degradation of 6.25 g of
protein will yield 1.0 gram of nitrogen), and is consistent with the observation
that urea represents the predominant nitrogenous waste product of protein
catabolism. When G equals zero, there is still a positive PCR of 11.04 g/day.
This positive PCR, deposit a urea appearance rate of zero, represents an
obligatory catabolism of protein leading to non-urea nitrogen. When PCR equals
zero, we calculate a value for G of 1.7 g/d, implying a net consumption of urea,
but G always has a positive value.
The final equation PCR= 9.35 * G + 0.294 * V
(V represents the kinetically determined volume of distribution of urea)
Using PCR to assess nitrogen balance:
The definition of balance can be given in terms of either tissue nitrogen or
tissue protein stores. The relevant equations are: - (Sargent et al., 1978)
(Tissue N stores) = Dietary N. intake - Non-protein N. Appearance
= Dietary N. intake (urea N. Appearance + Non-urea N. Appearance)
= Dietary N intake (G + Non-urea N. Appearance)
(Tissue protein stores) = DPI (dietary protein intake)- PCR
(N. Represent nitrogen)
In a stable hemodialysis patients, a (tissue protein stores) will be zero,
and the kinetically derived value of PCR may be used as an estimate of DPI.
Since dietary histories and records are notoriously inaccurate, this calculated
value of PCR may in fact prove a more reliable estimate of DPI (Wineman et al.,
1977). Therefore, by permitting a reliable estimate of DPI to be obtained, urea
kinetic modeling represents a powerful tool for ensuring adequate nutrition
among dialyzed chronic renal failure patients.
The threshold of adequacy:
While it is now generally accepted that Kt/V relates to outcome, the optimal
value for Kt/V is still under debate. Most of this debate followed on from Gotch
and Sargent, (1985) analysis of the NCDS study suggesting that a threshold Kt/V
value divided adequate from inadequate treatment. This threshold Kt/V was
described as being in the region between 0.8 and 1. While it was considered
necessary to ensure that a patient receives the adequate Kt/V, there was
considered to be little extra benefit to be gained by increasing Kt/V further.
The concept of the threshold of adequacy is theoretically plausible for
hemodialysis. At a Kt/V of 1.1, 65 % of the total urea in the patient has
already been removed. There are rapidly diminishing returns from increasing
Kt/V, and an infinite increase in Kt/V is required to remove the remaining 35%.
However, the concept of this "threshold" has been challenged.
Re-analyses of the NCDS (Hakim et. al., 1992) have suggested that there may be
progressive improvement in outcome as Kt/V increases above 1. A number of
dialysis units around the world routinely deliver Kt/V up to 1.6 by thrice
weekly hemodialysis (Charra et. al., 1992). These units appear to have
significantly better outcome, even when the effect of co-morbidity has been
corrected. Retrospective analyses of quality-adjusted survival (Hornberger,
1993), have suggested that there may be a continuous improvement in outcome as
Kt/V increases beyond 1. Limitations of the NCDS study have been recognized
(Lindsay and Spanner, 1989). These included relatively small numbers of
patients, a short follow-up period and a fairly atypical study population as
patients with significant co-morbidity were excluded. The study was not designed
to detect differences in long-term morbidity and mortality between patients
receiving Kt/V of 1 and higher values. Pending the conclusion of further studies
current opinion is moving towards acceptance of "adequate" Kt/V values
as being much higher than 1 in hemodialysis.