Dialysis Time, Survival, and Dose-targeting Bias

John T Daugirdas


Kidney Int. 2013;83(1):9-13. 

Dialysis time is increasingly being appreciated as an important measure of dialysis adequacy. Increased dialysis time leads to better control of volume excess, to reduced occurrence of intradialytic hypotension, and to better control of serum phosphorus. Nevertheless, the amount of benefit obtainable by moderate increases in dialysis time in patients following a three-times-per-week schedule has not been well established, and the analysis is confounded by associations between prescribed and/or delivered dialysis time and factors related to patient mortality.

In this issue of Kidney International, analyzing a cohort of prevalent patients being dialyzed mostly three times per week with 3- to 5-hour session lengths, Flythe et al.[1] report on the association between prescribed dialysis treatment time and survival. They found a substantial mortality increase associated with slightly shorter prescribed session length.

Controlling for body size is important when analyzing effects of components of dialysis treatment on survival, because body size and dialysis prescription normally are somewhat confounded: one main target of dialysis is to achieve a minimum urea reduction ratio. In small patients a given urea reduction ratio can more easily be achieved with a relatively short dialysis session length; thus, smaller patients and patients with low total body water (such as women) typically will be dialyzed for shorter periods than larger patients and, especially, large men. For reasons not yet clear, smaller hemodialysis patients have a markedly increased mortality. If no adjustment is made for body size, when mortality is found to be increased in patients receiving shorter dialysis treatments, it is not clear whether the effect is due to treatment time alone or was partly or completely mediated by body size. On the other hand, it remains possible that the shorter treatment time usually given to smaller patients is causally related to their increased mortality risk, and in this case, adjusting the outcomes analysis for body size might result in an underestimation of the true risk of shorter treatments.

The usual method of adjusting for body size is to consider some measure of body size, be it weight, anthropometric estimates of total body water or body surface area, volume, or body mass index, as a covariate. Flythe et al.[1] used a matching strategy, in which patients of a given size being prescribed a dialysis time less than 4 h were matched with similar-sized patients being prescribed a dialysis session longer than 4 h. Secondary matching by age, sex, and vascular access type also was done. When mortality rates in the less-than-4-hour and more-than-4-hour groups were compared, there was a very substantial difference in mortality, with the group being prescribed less than 4 h (mean delivered time, 201 min) having a 26% higher mortality than the size-matched patients undergoing the longer treatments (mean delivered time, 240 min).

The concept that dialysis time per se might be an important measure of dialysis adequacy, beyond urea reduction ratio or urea Kt/V, is an old idea that has been rediscovered and is gaining increasing traction. Because urea is a small, highly diffusible molecule, urea can be rapidly removed from the body by high-efficiency dialysis. This is true especially in smaller patients, women, and children, in whom the volumes of distribution of urea are relatively small. With short, rapid dialysis, however, it is more difficult to remove partially sequestered, larger molecules such as phosphate, as well as various candidate middle molecules and protein-bound substances. At the start of a dialysis session, excess salt and water accumulated during the interdialytic interval is confined to tissue spaces and enters the circulation slowly and often incompletely during dialysis. Rapid fluid-removal attempts can result in depletion of the circulating volume and in hypotension with 'stunning' of myocardium and other organs. Both intradialytic hypotension and incomplete removal of salt and water with resulting fluid overload are associated with increased risk of death.[2,3] The potential benefits of longer dialysis times have been documented in some centers that provide dialysis schedules of three times a week for 6–8 h per session, given either during the daytime or at night. Findings include better control of blood pressure, decreased need for antihypertensive medications, fewer intradialytic hypotensive episodes, and in some studies, reduced mortality.[4] Thus, the results of Flythe et al.[1] are not inconsistent with current concepts of optimal dialysis adequacy.

There are, however, several potential concerns with the study of Flythe et al.[1] The first is the issue of biologic plausibility. It is somewhat surprising that such a modest increase in the dialysis session length on a percentage basis (about 20%) would have a 26% impact on survival. Although a high ultrafiltration rate during dialysis has been shown to have an adverse effect on patient survival,[5] in the two size-matched groups, the mean difference in ultrafiltration rate was less than 10%; moreover, there was no evidence in the more-than-4-hour group that volume control was better (in terms of lower pre- or postdialysis blood pressure), that the frequency of intradialytic hypotensive episodes was reduced, or that phosphorus control was better, in comparison with the size-matched group receiving shorter treatments. To be fair, the Dialysis Outcomes and Practice Patterns Study (DOPPS) group of investigators recently has presented such data regarding better intermediate volume and phosphorus outcomes in patients undergoing longer treatments,[6] and better volume control, especially, is almost certain to occur with longer treatments.[4]

Another potential problem with interpretation of the results of this study is lack of internal consistency. In their source population, the authors provided data that mortality in patients being dialyzed less than 4 h was only 8% higher (relative risk, 1.08) than that in patients being dialyzed more than 4 h. Because the group being dialyzed less than 4 h was enriched with smaller patients, one would expect that adjusting for body size in the usual way, as a covariate, would have further reduced this already small mortality risk of less-than-4-hour dialysis. However, when the size-matched subgroup analysis was done, this small risk of less-than-4-hour dialysis was not reduced, as expected, but paradoxically magnified. Thus, the size-matched subgroup results were not consistent with the overall dialysis time results in the population from which the size-matched subgroup was drawn. Secondly, the benefit of a longer prescribed dialysis session was monotonously similar in small versus large patients and in men versus women. One would have expected an enhanced effect in smaller patients and women. Yet another difficult-to-explain finding is the results of a secondary analysis in which patients were again matched on the basis of size and other variables, but this time into three groups, with prescribed dialysis times of less than 210 min, 210–240 min, and more than 240 min. The mortality difference in the more-than-240-min group was again lower by the same degree, but there was absolutely no difference in mortality between the groups matched to get less than 210 min and 210–240 min of dialysis. When all is said and done, when dialysis time is zero, 1-year mortality in patients with minimal residual function should approach 100%, and so the relationship between dialysis time and mortality by definition should be exponential. Thus, an exponential magnification of risk would be expected when less-than-210-minute sessions are compared with prescribed treatment times of 210–240 min; however, absolutely no increase in risk was found with the size-matching strategy used. These results suggest that there was something very specific about the 240-minute prescribed dialysis time threshold and suggest perhaps a regulatory or quality bias (see below) active at the 240-minute time point rather than a biologic adverse effect of slightly shorter treatment times.

The benefits of longer dialysis time found by Flythe et al.[1] are not consistent with randomized trials, where increases in dialysis session length of similar or greater magnitude failed to improve survival ( Table 1 ). For example, the effect of dialysis time was one of the two main intent-to-treat interventions in the randomized National Cooperative Dialysis Study (NCDS).[7] In the two longer time groups of the NCDS, mean dialysis session lengths were about 4.3 h, compared with 3.2 h in two shorter time groups. The NCDS was not powered to detect changes in mortality; a benefit of longer time on hospitalizations was found). In the HEMO trial, dialysis time was an integral part of the intent-to-treat dose intervention; the requisite differences in dialysis dose could not practically be achieved by changes in dialyzer clearance alone. In the intent-to-treat analysis of the HEMO Study results,[8] the mean session-length differences between the conventional and high-dose groups were on the order of 30 min, very similar to the time difference found between the two groups of the observational trial by Flythe et al.;[1] yet in the HEMO Study intent-to-treat analysis, no benefit of increased dose (which included time), in terms of either survival or hospitalizations, was found.

In a conventional three-times-per-week dialysis setting, data from other observational studies regarding the effect of time on survival are highly variable, to the point that results sometimes point in opposite directions. Table 1 summarizes a few of these. In one study of incident dialysis patients, Brunelli et al.[9] found opposite effects of time on survival, depending on whether time was assessed as a baseline or a time-dependent covariate. Saran et al.,[10] using the DOPPS database, found that longer dialysis sessions were indeed associated with increased survival, but the effect was enormously magnified in Japan and somewhat magnified in Europe, compared with results in US dialysis patients. Moreover, a group that included some of the same investigators[11] recently examined the effects of time on survival in a different set of US dialysis patients and found the opposite result—increased session length now was associated with decreased survival. Finally, Miller et al.,[12] looking at patients from the same large dialysis organization as the study by Flythe et al.,[1] found no benefit in terms of survival of dialysis sessions lasting longer than 3.5 h. This lack of consistency among observational studies suggests that what is being measured may be not a biologic effect of dialysis session length, but rather some confounding effect associated with both mortality and dialysis time.

After the initial intent-to-treat analysis of the HEMO Study was reported, as-treated associations between dialysis dose and survival within each assigned dose arm were analyzed.[13,14] In contrast to the negative intent-to-treat analysis, within each treatment arm, delivered dose was very powerfully associated with survival. Both the denominator (V) and the numerator (Kt) of dose contributed to this effect, which was termed 'dose-targeting bias' as it was much too large to be explainable by a biologic advantage of dose alone. Time, a component of the numerator of dialysis dose, had a marked effect on survival in the high-dose arm of the HEMO Study, and a much smaller impact in the standard-dose arm,[13,14] for reasons that remain unclear; however, in the high-dose arm the average urea reduction ratio of 75% was quite similar to the average dose of dialysis being delivered in the United States today.[13,14]

Apart from body size, what could be the explanations for how assignment of a shorter or a longer dialysis session length might be associated with changes in patient survival (Figure 1)? One prime candidate for confounding is 'regulatory,' 'quality,' or financial bias, perhaps best illustrated by the results of Saran et al.,[10] in which the benefits of a more-than-4-hour treatment time were magnified in Japan, a country where nephrologists were paid extra for each patient dialyzed for longer than 4 h. Similarly, in many European countries, regulations require that a certain percentage of patients in a given unit be dialyzed longer than 4 h. The explanation for such quality or regulatory bias is a variant of the 'good patient' and/or 'good nephrologist' hypothesis: that those patients not compliant with a quality target have something associated with them that is also linked to increased mortality risk. Flythe et al.[1] tried to get around such a 'good patient' compliance bias by evaluating prescribed, rather than delivered, dialysis time, but there are not too many nephrologists who will record a 4.5-hour dialysis prescription for patients who flatly refuse to dialyze longer than 3.5 h. In the United States, where regulatory and quality targets previously focused on the urea reduction ratio, the regulatory bias associated with longer treatment times may have been less pronounced in the past, but the recent emphasis of large dialysis organizations and end-stage renal disease networks on achieving a certain minimum dialysis time as a quality target may be increasing the importance of a time-associated regulatory bias in more recently studied US patient cohorts.

Figure 1.

Potential confounding with shorter versus longer prescribed dialysis times. Body size (lean body mass) is only one of a number of confounding factors that may be present.

In the as-treated HEMO Study results, some of which are as yet unpublished, we also found that, within the high-dose arm, slight increases in dialysis time during follow-up were associated with marked increases in survival, while slight decreases in time were associated with increased mortality risk, even during periods of stable vascular access. How can such a bias be explained? Based on clinical experience, one can concoct various scenarios, most somewhat speculative; and some explanations point in opposite directions. Several examples: Patients who are noncompliant with the dialysis prescription (who are known to have poor survival) may refuse longer dialysis times or may insist on shortening dialysis session length as their course on dialysis evolves. Dialysis session length might be increased because of difficulties with ultrafiltration, because of high interdialytic weight gain, or because of intradialytic hypotension. Time may be increased in patients who are perceived to be faring poorly. Longer dialysis time may be prescribed for patients who have lower amounts of residual function, or who have reduced blood flow due to vascular access dysfunction or who have accesses capable of only limited blood flow.

Given the above difficulties, the ability of observational studies to guide us to an optimal dialysis prescription remains limited. What, then, might be possible ways forward? One potential approach might be to work with dialysis organizations that own large numbers of dialysis units and that have substantial control of medical practice patterns within those units. Using this infrastructure, one could design and carry out randomized cluster trials, where a shorter or longer dialysis time is assigned to randomly chosen groups of dialysis units with the use of an intent-to-treat design. 'Hard' outcomes such as mortality, hospitalizations, and quality of life in each randomly assigned set of units could then be ascertained, and the two groups could be compared. Although at some level, marked increases in dialysis time may well be beneficial for most patients, I believe that we will have to await the outcome of such a randomized cluster trial before we can validly calibrate the extent of outcome benefit associated with relatively modest increases in dialysis session length in a three-times-per-week setting. Lastly, the question of whether or not higher predialysis serum bicarbonate concentrations increase mortality risk independent of nutritional adjustment remains undefined;[15] nevertheless, one needs to keep in mind that long, high-efficiency dialysis sessions against dialysates of high base content, especially when given to small patients, may increase pre- and postdialysis serum bicarbonate levels beyond an optimal range, and in such cases dialysis efficiency and/or dialysate base content may need to be adjusted appropriately.