143—151
EDITORIAL REVIEW
143
144 Sprenger et a!
40
0 84.09 83.46 83.31 80.27 85.89 84.97
•0
Co
20 30 71.30 70.90 70.80 65.02 75.49 76.42
LJ
60 68.49 67.73 66.84 60.29 72.94 70.63
20 40 60 80 100 120 140 160 180'2O0'20 240
90 66.90 66.95 65.21 56.96 71.02 66.03
Time, mm
Fig. 1. Progress of average filtration rates (QF) during hemodiafiltra- 120 66.07 65.52 62.09 53.89 68.84 62.49
tion treatments lasting 3 to 4 hr in three of six patients. The measured
filtration rates of the six patients are summarized in Table 1. In all 150 65.22 64.05 59.62 51.28 68.64 58.74
patients the filtration rate changes substantially, particularly in the first
30 mm. It decreased on an average by 30.6%, from 84 mI/mm in the first 180 63.48 61.46 59.90 50.62 66.90 57.13
5 mm to 58 mI/mm after 3.5 hr. The points of measurement lie on an
exponential curve described by a function of the type y = A + i3e". 210 62.68 58.33 50.02 64.50 54.82
Differences in the rate of decrease of filtration rate between patients
(curves I to 3) may be attributed to differences in blood composition 240 61.40
(hematocrit, concentration of fat, albumen, and fibrinogen). Symbols are:
A, markers of curve 1; •, markers of curve 2; •, markers of curve 3.
y=A+ l3e'
a —5.6 —4.5 —2.8 —4.6 —4.4 —1.6
In hemodiafiltration (HDF), that is, a simultaneous hemofil- 3 20.8 22.0 30.7 30.2 19.8 41.4
A 84.09 83.46 83.31 80.27 85.89 84.97
tration/hemodialysis [14, 15], however, the clearance decreases
significantly during treatment, as our own in vivo measure- a These numbers represent the mean values SD of N = 50 measured
ments have shown. This is caused by a decrease of the filtration data; Table I supplements Figure 1.
rate and of the sieving coefficient. The observed decrease in
filtration rate during hemodiafiltration (Fig. 1, Table 1)' is
probably caused by a protein gel layer of "secondary mem- recommended that dialyzer clearance should be determined
brane" forming on the surface of the dialyzer membrane at a again for each patient, because the changes in plasma concen-
rate dependent upon the filtration rate and in turn causing an trations can be predicted only for an individual dialysis treat-
increasing resistance to filtration [16]. A steady state may occur ment if the clearance is known as a function of time.
when the rate of deposition is balanced by the rate of removal These results imply that even in hemodialysis the ultrafiltra-
by diffusion and penetration through the membrane [17]. How- tion rate may influence dialyzer clearance. As a rule, only low
ever, the sieving coefficient also decreases with protein deposi- ultrafiltration rates (5 to 15 mllmin) are necessary to attain the
tion and mass transport is further reduced. This was investigat- required weight loss and this increases the clearance of small
ed for several substances at the beginning and at the end (after 4 molecules such as urea only slightly (3 to 10%) but can increase
hr) of six hemodiafiltration treatments for the RP 6 dialyzer the clearances of middle molecules such as inulin by 30 to 40%.
under standard conditions (QB = 200 mllmin, TMP = 500 mm A small drop in the ultrafiltration rate during hemodialysis may
Hg, QD = 0 mllmin). The decrease of sieving coefficient was therefore cause a significant reduction in middle molecule
significant for creatinine (15%), glucose (9%), phosphate (17%), clearance. This effect may be diminished by devices that
and inulin (22%) but not for urea (unpublished data). The maintain a constant ultrafiltration rate rather than a constant
reduction of filtrate flow and sieving coefficient causes a transmembrane pressure.
patient-specific reduction of the total clearance of urea and Residual renal clearance (KR). The effects of even very low
creatinine by about 10 to 15% in HDF (Figs. 2 and 3, Table 2). residual renal clearances (1 to 2 mllmin) should not be underes-
For the kinetic modeling of hemodiafiltration, it is therefore timated for the calculation of substance concentrations in
dialysis patients because the natural kidney functions continu-
ously and can significantly affect blood concentrations [18],
'Six patients were hemodiafiltered in postdilution mode 50 times particularly in the middle molecular weight range (500 to 5000
each; filtrate was collected continuously in 30-mm periods. The average daltons). Care should be exercised because an error of 1/2
volume of substitution solution was 10.5 2 liters per treatment. Blood ml/min when measuring a clearance of 1 ml/min can lead to
flow (250 mI/mm), dialysate flow (500 mI/mm), and transmembrane errors of 10 to 12% in the weekly middle molecule clearance if
pressure (513 12 mm Hg) were maintained at constant levels during
all treatments. At the beginning, after 30 mm, 30 mm before the end, the clearance during dialysis is 20 mllmin [23]. The main
and at the end of six consecutive treatments heparinized blood samples sources of error are: the difficulty in passing small volumes of
were taken at the arterial and venous parts of the dialyzer (RP6). Urea urine and the change in plasma creatinine concentration be-
and creatinine concentrations were determined by routine methods. tween dialysis treatments.
Also, urea and creatinine concentrations were measured 45 and 90 mm Errors can be minimized by the collection of urine over a
after the end of three consecutive HDF and in the intertreatment
interval to cover the course of the concentration during the interval. See period substantially longer than the usual 24 hr and taking blood
the legend of Figure 1 for the progress of average filtration rates during samples at the beginning and the end of urine collection and
hemodiafiltration treatments in three of the six patients. using the average.
Kinetic modeling of artificial kidney 145
200 200
160 E 160
E
a)
C
a) 120 C 120
a)
0
C
80 80
'a
a)
a) 40 40
0
0 0 0
0 20 40 60 80 100 120 140 160 180 200 220 240 0 20 40 60 80 100 120 140 160 180 200 220 240
Time,min Time,min
Figs. 2 (left) and urea and creatinine clearance (N = 6) of three patients (Table 2) plotted against time during hemodiafiltration
3 (right). Average
treatments lasting 4 hr. Urea and creatinine clearance of the four sample points were calculated by means of equation [3]. Then clearance as a
function of time was calculated connecting these points by a spline function.
where Cb is the mean value of Cb (T) during the period from 0 to have been constructed in accordance with physiological
t, and a good estimate of Cb is given by 1/2 [Cb (0) + Cb (t)]. measurements of body compartment volumes usually in pa-
In this formula V represents the distribution volume in tients with normally functioning kidneys [21, 23—30]. Popovich
milliliters; Cb (0) and Cb (t) are plasma concentrations at the Ct al [30] estimated the volumes of the intracellular, intravascu-
beginning and end of the measurement period (t). To prevent lar, and interstitial compartments and the coefficients of mass
rebound from affecting the measurement of G the first blood transfer between these compartments by using a three-pool
sample can be taken a few hours after the end of dialysis. model to analyze the dilution curves after bolus injection of
Compartment volumes, mass transfer, and distribution coef- radioisotope labelled substances in an anephric patient between
ficients: Compartment volumes (V). Various multiple pool models dialysis treatments. The results show mass transfer coefficients
146 Sprenger et a!
and interstitial volumes which decrease when molecular weight tions of urea and creatinine, so that their distribution coeffi-
increases. Popovich et al [301 attributes the latter to the cients may be ignored in pool calculations.
omission from the model of "flow limitations." Schindhelm and
Farrell [24] repeated this method using fixed volumes from Kinetic modeling
physiological literature rather than attempting to fit both vol- Compartmental kinetic models may be used with differential
umes and mass transfer coefficients to one set of data. This equations to simulate the rates of changes in concentrations of
resulted in a much higher estimate of transcellular mass transfer solutes in the various compartments (or fluid pools) of dialysis
coefficients. It is not yet clear from available data whether the patients. Given a starting point ("infinitral condition") the
use of a multiple pool model with variable volume comes closer differential equations can be solved, such as on a computer
to reality for the prediction of mass transport than those with using numerical methods, for example, the Runge-Kutta meth-
fixed volume. od, to predict changes in the variables, for example, concentra-
Mass transfer coefficients (T). A series of measurements of tions. Reliable prediction of the probable course and results of
blood concentrations during dialysis or immediately after dialy- therapy is, however, only possible when sufficient information
sis can be used to determine the distribution of a substance in on the past and present condition of the patient is available. If
the body and the concomitant transfer resistance to transport this is the case, the model can, for example, be used to obtain
between the compartments [30]. "optimal therapy."
During treatment the high dialyzer clearance causes an im- To our present knowledge, a pool model that supplies long-
balance between concentrations in the intra- and extracellular term predictions does not exist [1, 9, 13, 19, 21, 22, 24—27, 29—
compartments. The rapid decrease in plasma concentration 31, 35—40]. It is doubtful whether a suitable model can be
leads to a considerably slower drop in intracellular concentra- developed for such a diverse group as that of dialysis patients.
tion, since only a small proportion of substance passes over Human physiology is complex and the more realistic models
from the intracellular to the extracellular space during dialysis. require too many data for general application to individual
This results in an accelerated renewed increase of plasma patients. Furthermore, small errors in the received data of the
concentration after dialysis. This increase is referred to as variables can generate accumulative errors in long-term predic-
"rebound" and can be used to calculate the cell membrane tions. Thus, models should be divided into at least two distinct
mass transfer coefficient [21, 30]. The magnitude of disequilibri- types depending on the desired application namely: (I) models
um and rebound are higher when the mass transfer coefficient is which are used primarily as research tools to check the quanti-
low (as is the case for middle molecules). Measurements of tative validity of proposed physiological mechanisms and (2)
transcellular transfer coefficients for urea, creatinine, and vita- models which can be applied in routine clinical practice or for
min B12 have been reported by several authors [19, 21, 22, 24— designing protocols to test new techniques or equipment. The
26, 28—31]. However the urea rebound may not be caused latter requires particularly simple models with a minimum of
exclusively by transcellular disequilibrium but may be affected variables to be measured and processed, but a reasonable
by an increase in protein catabolism induced by loss of glucose degree of accuracy is necessary to predict overall mass balance
and aminoacids [24]. This would make the transfer resistance (that is, pre- and postdialysis concentrations). The research
seem larger than it actually is. models, on the other hand, can be considerably more complex
For all the solutes studied the transcapillary transfer coeffi- and may be valid as broad generalizations of the system rather
cients are at least an order of magnitude higher than the than being applied quantitatively to individual patients. It is,
transcellular transfer coefficients. The latter must therefore be however, often necessary to limit the number of variables
the limiting factor in mass transport. because of the difficulty in making measurements, unless ex-
Distribution coefficients (x). The distribution coefficient (x) is trapolation from minimal data is considered acceptable. This is
the ratio of concentrations of the substance in two compart- often a necessary expedient.
ments at equilibrium. Dombeck, Klein, and Wendt [29] report- The single pooi model. In the past, the one-pool model has
ed the best fit of calculated to measured creatinine data using x been applied most frequently to dialysis-related techniques, not
= 1.63 between the intracellular and extracellular compart- only to simulate the overall mass balance of small molecules,
ment, which was the value previously measured by Giovanetti but also for middle molecules such as inulin [1, 38].
and Barsotti [32]. Although reduction of x to 1.0, which is In the single pool model, the body of the patient is conceived
usually assumed in the literature for distribution of all sub- as a homogenous compartment or "pool" of fluid in which all
stances between the intracellular volume (ICV) and extracellu- substances are uniformly distributed. The distribution volume
lar volume (ECV), had little effect on the mass transfer coeffi- should correspond to the total body water [24—28].
cient, it affected the generation rate and the dialyzer clearance The rate of change of the quantity of a substance in the pool
substantially. The latter is thereby diminished to values which
clearly lie under the measured in vivo clearance values [29].
(V), that is, !. Cb (t), during treatment can then be calculated
Distribution coefficients for urea, creatinine, and inulin be- from the dialyzer clearance (K), substance concentration (Cb),
tween red cells and plasma have been reported as 0.86, 0.73, residual renal clearance (KR), and the generation rate (G) of the
and 0.0, respectively [33]. The distribution coefficient is equal substance according to the following general model equation [41]:
to 0 for substances which cannot penetrate into the red cells,
while the distribution coefficient is equal to 1 if the concentra- [V (t) x Cb (t)] = —K(t) Cb(t) — KR Cb(t) + G. (8)
tion is equal in both compartments. However, more recent
measurements [34] on uremic erythrocytes could not detect a If the initial concentration and time intervals [and, of course,
significant difference between plasma and red cell concentra- the variables V(t), K(t), KR, G] are known, then this equation
Kinetic modeling of artificial kidney 147
C,
5 10 15 20 25 30 35 40 45 50 55
Table 5. Patient no. 2: Hemodiafiltration (HDF): 3 X 210 mm/week Table 6. Patient no. 3: Hemodiafiltration (HDF): 3 >< 180 mm/week
residual renal clearance: 1.0 ml/min; dry body weight, 62.5 kg residual renal clearance: 2.9 mI/mm; dry body weight, 86.0 kg
Filtration Filtration
Time Urea Creatinine rate Time Urea Creatinine rate
Day HDF mm mmoleslliter mmo/es//iter mi/mm Day HDF mm mmoles/iiter mmoies/iiter mi/mm
is used for urea is relatively small [13, 20, 21]; it can be as high • Mass transfer of solutes between compartments (in accord-
as 50% for middle molecules [30]. ance with coefficients T and X) is linear
The two pooi model. In the two pool model, the total body • Solute generation only takes place in the intracellular pool.
water volume (V) is divided into two homogeneous compart- Errors may result if any of these assumptions are incorrect. The
ments: the intracellular volume (ICY V1) and the extracellu- equations for the two pool model can be solved explicitly [41] if
lar volume (ECV = V2). The concentrations in these compart- the variables are constant or, as in the instance cited below on a
ments are denoted by Cb and Cb2, respectively, and the intra- to computer using numerical methods, using, for example, the
extracellular mass transfer coefficient (T) and the distribution Runge-Kutta method [41]. The interdialytic interval, including
coefficient (X) must also be known. The general differential the rebound, is of course simulated by using K = 0. The initial
equations describing the two pool model with constant volumes plasma concentration is easily measured but the intracellular
are: concentration must be estimated from the distribution coeffi-
cient Cb = XCb2. As the one pooi model simulations had only
Cb(t) = -(—TCb(t) + XTCb,(t) + G1) (9) given an imprecise description of urea and creatinine kinetics in
hemodiafiltration patients, we attempted a better evaluation of
the measurements obtained with a two pool model. In the
C2(t) = +(TCb(t) — (XI + K(t) + application of the one and two pool models the following data
for the variable factors were used (Table 3).
It can be seen in our Figures 4 and 5 that the two pool model
KR) Cb2(t) + G2) (10) simulates the observed data more accurately than the single
pool model, particularly with regard to the creatinine rebound.
where G1 and G2 are the generation rates in compartments V1 This becomes especially clear if the process is followed over
and V2. Usually the following assumptions are made: two treatment intervals such as creatinine concentration in
• V1, V2, T, X, KR, G1, and G2 are constant Figure 6 (Tables 4 to 9). The urea rebound is smaller and of
• No significant protein binding occurs shorter duration and so the difference between the two models
• Pools are homogenous — even during dialysis is less significant.
Kinetic modeling of artificial kidney 149
Table 7. Patient no. 4: Hemodiafiltration (HDF): 3 x 210 mm/week Table 8. Patient no. 5: Hemodiafiltration (HDF): 3 x 210 mm/week
residual renal clearance: 0.4 mI/mm; dry body weight, 64.5 kg residual renal clearance: 0 mI/mm; dry body weight, 75.0 kg
Filtration Filtration
Time Urea Creatinine rate Time Urea Creatinine rate
Day HDF mm mmoles/liter mmoles/liter mi/mm Day HDF mm mmoles/liter mmoles/liter mi/mm Fig.
1 1 0 31.6 1.59 87 1 1 0 30.3 1.28 80
1 30 26.1 1.34 66 1 30 28.2 1.05 75
1 170 18.5 0.99 50 1 120 21.6 0.91
1 210 17.5 0.93 48 1 180 19.1 0.82 65
255 18.0 1.00 210 15.7 0.73 62
300 18.3 1.05 390 19.4 0.88
570 20.5 0.89
2 1330 22.6 1.25 730 20.7 0.90
8. GOTCH FA: Urea guided dialysis therapy. Current clinical results. 27. WIDEROE E, GRIMSRUD T-E, BERG KJ, GODAL A, JENSEN R,
Dialysis Transplant 5:15—18, 1976 JORSTAD 5: A mathematical single-pool model for short time
9. GOTCH FA, SARGENT JA, KEEN M, LEE M: Individualized quanti- hemodialysis Proc Europ Dial Transplant Assoc. 11:136—145, 1974
fied dialysis therapy of uremia. Proc Clin Dial Transplant Forum 28. CHRISTOPHER TG, HURST P, CAMBI V, SCRIBNER BH, BABB AL:
4:27—35, 1974 Measurement of parameters for a two-compartment model of
10. GOTCH FA, SARGENT JA, KEEN M, LAM M, PROWITT M, GRADY dialysis (abstract). Proc Europ Dial Transplant Assoc 8:436, 1971
M: Clinical results of intermittent dialysis therapy (IDT) guided by 29. DOMBECK DH, KLEIN E, WENDT RP: Evaluation of two pool
ongoing kinetic analysis of urea metabolism. Trans Am Soc Artif model for predicting serum creatinine levels during intra- and inter-
intern Organs 22:175—188, 1976 dialytic periods. Trans Am Soc Artif Intern Organs 21:117—123,
11. SWENSON RS, SANFELIPPO ML, HALL DA, WALKER WE: Analysis 1975
of changing plasma concentrations of urea and creatinine in dialysis 30. POPOVICH RP, HLAVINKA DJ, BOMAR JB, MONCRIEF JW, DE-
by means of a simple mathematical model and a programmable CHERD JF: The consequences of physiological resistance on metab-
pocket calculator. Opusc Med.-Techn Ludensia XVII (Mono- olite removal from the patient-artificial kidney system. Trans Am
graph), Hechingen, West Germany, Kohlhammer and Wallishauer Soc Artiflntern Organs 21:108—I 15, 1975
GmbH, 1976 31. ABBRECHT PH, PRODANY NW: A model of patient-artificial kidney
12. GRANGER A, VANTARD G, VANTELON J, PERRONE B: A mathemat- system. IEEE Trans Biomed Eng 18:257—264, 1971
ical approach of simultaneous dialysis and filtration (SDF). Proc 32. GIOVANETTI S. BARSOTTI G: Dialysis of methylguanidine. Kidney
Eur Soc Artif Organ 5:174—177, 1978 Int 6:177—183, 1974
13. SARGENT JA, GOTCH FA: Principles and biophysics of dialysis, in 33. COLTON CK, HENDERSON LW, FORD CA, LYSAGHT MJ: Kinetics
Replacement of Renal Function by Dialysis, edited by VON of hemodiafiltration. I. In vitro transport characteristics of a
DRUKKER W, PARSONS FM, MAHER JF, Boston, Nijhoff, pp. 38—68, hollow-fiber blood ultrafilter. J Lab Clin Med 85:355—371, 1975
1978 34. NOLPH K, FELTS J, MOORE R, VAN STONE J: Differences in the
14. SPRENGER KBG, BUNDSCHU D, FIGUEROA P, FRANZ HE: Verg- distribution of urea and creatinine between red cells and plasma in
leich von Hämodiafiltration gegenuber Hamodialyse mit kontrol- normal and azotemic blood as assessed by autoanalyzer and manual
lierter Ultrafiltration in einer ABA-Langzeitstudie. Nieren-und chemical methods. mt Urol Nephrol 10:59—64, 1978
Hochdruckkrankheiter 10:226—230, 1981 35. LEWIS AED: Optimization of dialysis treatment conditions using a
15. KOHNLE W, SPRENGER K, SPOHN B, FRANZ HE: Hemodiafiltra- programmable calculator. Proc Europ Soc Artif Organs 5:182—185,
tion using readily available equipment. J Dial 3:27—39, 1979 1978
16. OKAZAKI M, YOSHIDA F: Ultrafiltration of blood. Ann Biomed Eng 36. LOTT RS, MOORHOUSE KES, WHITT CN: The application of
4:138—150, 1976 mathematical models on an interactive computer graphics display
17. HENDERSON LW, COLTON CK, FORD CA: Kinetics of hemodiafil- for renal patient management, in Technical Aspects of Renal
tration. II. Clinical characterization of a new blood cleansing Dialysis, edited by FROST TH, KERR DNS, Tunbridge Wells,
modality. fLab Clin Med 85:372—391, 1975 England, Ditman Medical Publishers, 1978, pp. 264—273
18. MILUTIN0vIc J, STRAND M, CASARETTO A, FOLLETTE W, BABS 37. SANFELIPPO ML, WALKER WE, HALL DA, SWENSON RS: Clinical
AL, SCRIBNER BH: Clinical impact of residual glomerular filtration application of a single compartment model to urea and creatinine
rate (GFR) on dialysis time. A preliminary report. Trans Am Soc kinetics in dialysis therapy. Comput Programs Biomed 8:44—50,
Artf Intern Organs 20:410—416, 1974 1978
19. FRosT TH, KERR DNS: Kinetics of hemodialysis: A theoretical 38. SAUSSE A, MAN NK, FUNCK-BRENTANO JL: A mathematical
study of the removal of solutes in chronic renal failure compared to approach to hemodialysis therapy. Proc Europ Soc Artif Intern
normal health. Kidney mt 12:41—50, 1977 1974
Organs 1:81—84,
20. GOTCH FA: A quantitative evaluation of small and middle molecule 39. SANFELIPPO ML, HALL DA, WALKER WE, SWENSON RS: Quanti-
toxicity in therapy of uremia. Dialysis Transplant 9:183—194, 1980 tative evaluation of hemodialysis therapy using a simple mathemati-
21. RASTOGI SP, FROST T, ANDERSON J, ASHCROFT R, KERR DNS: cal model and a programmable pocket calculator. Trans Am Soc
The significance of disequilibrium between body compartments in Artif Intern Organs 21:125—130, 1975
the treatment of chronic renal failure. Proc Europ Dial Transplant 40. WOLF MB, WATSON PD, BARBOUR BH: Theoretical evaluation of a
Assoc 5:102—1 15, 1968 patient-artificial kidney system using the kiil dialyzer. Math Biosci
22. BELL RL, CURTIS FK, BABB AL: Analog simulation of the patient 6:367—386, 1970
— artificial kidney system. Trans Am Soc Artif Intern Organs 41. PEYERIMHOFF A: Gewöhnliche Differentialgleichungen I, Akade-
11:183—188, 1965 mische Verlagsgesellschaft, Frankfurt, 1970, vol. 1
23. SARGENT JA, GOTCH FA: The analysis of concentration depen- 42. SARGENTJ, GOTCH F, B0RAN M, PIERCY L, SPINOZZI N, SCHOEN-
dence of uremic lesion in clinical studies. Kidney lot 7(suppl):S35— FELD P. HUMPHREYS M: Urea kinetics: A guide to nutritional
S44, 1975 management of renal failure. Am J Clin Nutr 31:1696—1702, 1978
24. SCHINDHELM K, FARELL PC: Patient-hemodialyzer interactions. 43. MAN N-K: Kurzzeitdialyse mit Membranen von hoher Permeabili-
Trans Am Soc Art ifIntern Organs 24:357—365, 1978 tat. Med KIm 71:1279—1287, 1976
25. STILLER S, MANN H, HIGA5HI F: Prediction of solute and water 44. MANN H, STILLER 5, GORICH W: Kinetik des Stoffaustausches bei
exchange between the body compartments during hemodialysis by der Hämofiltration und bei der Hamodialyse: Vergleichende Ber-
a mathematical model. Projkettragerschaften-Bericht: Materialien echnung mit einem mathematischen Model! Patient-kunstliche
und Systeme zur Organunterstutzung, Munich, Society for Re- Niere. Nieren-und Hockdruckkrankheiten 7:1—8, 1978
search on Radiation and the Environment, contract DVM 136, vol. 45. MANN H, STILLER 5, GURICH W: Kontrollierte Dialysetherapie mit
7, 1977 Hilfe mathematischer Simulation des Stoffaustausches. Nieren-
26. KING PH, BAKER WR, GINN HE, FROST AB: Computer optimiza- und Hochdruckkrankheitan 8:44—50, 1979
tion of hemodialysis. Trans Am Soc Artif intern Organs 14:389— 46. SARGENT JA, GOTCH FA: Mathematic modeling of dialysis thera-
393, 1968 py. Kidney Int lO(suppl):S2—S10, 1980