HDCN Article Review/Hyperlink

Harty J, Boulton H, Faragher B, Venning M, Gokal R

The influence of small solute clearance on dietary protein intake in CAPD patients: a methodologic analysis based on cross- sectional and prospective studies

Am J Kidney Dis (Oct) 28:553-560 1996

Because one of the most prominent effects of uremia, and its modern counterpart, inadequate dialysis, is suppression of the appetite, the dose of dialysis should logically be advanced until appetite is maximized. The success of increasing the dose should be demonstrable within a population of dialysis-dependent patients when the dialysis dose ceases to correlate with protein intake, i.e., the appetite is no longer dialysis-dependent. Since the dose is easily measured as Kt/V and appetite is mirrored by protein intake which is also easily measured as the protein catabolic rate, this hypothesis is testable; a study of this type could shed light on the question of how much dialysis is adequate.

Many investigators have addressed this issue and several widely divergent interpretations have been offered for the observed correlations. Nearly all have found some degree of correlation between Kt/V and nPCR but the meaning of this remains controversial. The article by Harty et al reports yet another study of this type but in addition to simply correlating Kt/V with nPCR in a single cross-sectional analysis of their patients as most others had done, these investigators prospectively increased the dose of peritoneal dialysis by adding an extra 0.5 liter of dialysate per exchange in half the patients and observed the effect on nPCR after one year. They also generated 20 sets of 2000 artificial pairs of Kt/V and nPCR from random values of plasma and dialysate urea concentration, dialysate volume and patient volume calculated from random values of height, weight, and age. And finally, they measured dietary protein intake from 3-day food diaries submitted by the patients.

All paired values of Kt/V and nPCR were highly correlated, whether they were obtained by cross sectional analysis of their patients' blood and dialysate, from dietary food records, or generated artificially. The highest correlation coefficients (r) were found in the artificially generated data (r = 0.64 to 0.74) and the lowest was found in the food record data (r = 0.36). The authors also observed a significant positive correlation between the change in Kt/V and the change in nPCR in the prospective part of their study.

Comment: As emphasized by investigators in the past, the prospective interventional approach taken by these investigators is the only fool-proof method for establishing a physiological relationship between Kt/V and nPCR. The positive relationship demonstrated in this study adds strength to the argument that Kt/V can affect protein intake, but the small number of patients and limited increase in Kt/V achieved did not permit examination of a possible ceiling effect as discussed above. Nonetheless the work by Harty, et al, is a significant step in that direction.

These relatively important findings are buried in a lengthy discussion of the strong correlation between Kt/V and nPCR found in their simulated database. In fact, the authors downplayed their prospective data because of what they perceive to be an inescapable high artifactual correlation caused by mathematical coupling. They argue that most of the correlation results not from physiologic events reflecting a beneficial effect of dialysis on appetite, but rather from an artifactual mathematical phenomenon caused by the fact that nPCR and Kt/V are calculated using common variables like the dialysate urea and the patient's urea distribution volume. They support their argument by pointing to the high artificial correlation that appeared in the paired Kt/V and nPCR values generated from randomly distributed measured components.

The assumption that the measured components of Kt/V and nPCR such as dialysate urea concentration, dialysate volume, plasma urea concentration, and the patient's urea distribution volume are all random values, not associated with each other, can be challenged. When normally distributed values of Kt/V and nPCR are randomly paired, the variables from which they are generated are not randomly distributed. This means that values for Kt/V and nPCR generated from randomly distributed values of dialysate urea, plasma urea, dialysate volume, and patient volume can be expected to be correlated just as the authors' data shows but the correlation is an artifact of the method used to generate the data. This explains the higher degree of correlation found in the authors' simulated data compared to their real data. The real data from which nPCR and Kt/V are generated are not randomly distributed.

Simple patient examples can be used to show that G (urea generation) and V (urea distribution volume) are not independent of one another. If G were independent of V then larger patients would eat the same amount of protein as smaller patients. In people with normal kidney function, urea clearance is linked to patient size. Clearance is the ratio of urea generation to plasma urea concentration, so if urea clearance is increased in larger patients, and urea generation is the same as in smaller patients, then the plasma urea concentration must be lower in larger patients. This is, of course, not the case. The assumption that G is independent of V creates an inappropriate correlation between Kt/V and nPCR that does not exist in nature. A similar argument can be made for the association between plasma urea concentration and Kd and between nPCR and plasma urea concentration. A strong link exists between both of these pairs of variables in nature. If one assumes that this link doesn't exist, then nPCR and Kt/V become linked inappropriately.

The authors' point that coupling exists among the calculated variables is obviously correct but their artificial simulations do not provide a meaningful estimate of what the effect of the coupling is. Their simulated data differs from real data because it appears that it is not normally (or otherwise) distributed about a mean value. Uehlinger has shown, in an elegant analysis of this question using normally distributed random variables, that errors in measured values for V can cause an artificial correlation between Kt/V and nPCR in peritoneally dialyzed patients (1). Using his approach we demonstrated a similar effect from measurement errors in the predialysis and postdialysis BUN in hemodialyzed patients (2). In the present study, the authors' use of raw random numbers is a source of additional inappropriate correlations and leads them into an unnecessarily complicated discussion of boundary conditions.

There are natural biological associations among the different input parameters in a system involving coupling such as this one. The assumption that one set of variables is mutually independent (not correlated) will generally imply that the other variables are correlated. Paradoxically, it is this weakness in the authors' simulated database that adds strength to the biological implications of their real data, that Kt/V may indeed influence nPCR.

1. Uehlinger, D.E. Another look at the relationship between protein intake and dialysis dose. J Am.Soc.Nephrol. 7(1):166-168, 1996.

2. Depner TA: Correlation between hemodialysis Kt/V and PCRn caused by BUN measurement errors. J Am Soc Nephrol 7(9):1510, 1996 (abstract).

(Thomas A. Depner, M.D., University of California at Davis, with the assistance of Tom Greene, Ph.D., Statistician, Cleveland Clinic.)