Modeled Urea Distribution Volume and Mortality in the HEMO Study

Abstract

Summary

Background and objectives

In the Hemodialysis (HEMO) Study, observed small decreases in achieved equilibrated Kt/V_urea were noncausally associated with markedly increased mortality. Here we examine the association of mortality with modeled volume (V_m), the denominator of equilibrated Kt/V_urea.

Design, setting, participants, & measurements

Parameters derived from modeled urea kinetics (including V_m) and blood pressure (BP) were obtained monthly in 1846 patients. Case mix–adjusted time-dependent Cox regressions were used to relate the relative mortality hazard at each time point to V_m and to the change in V_m over the preceding 6 months. Mixed effects models were used to relate V_m to changes in intradialytic systolic BP and to other factors at each follow-up visit.

Results

Mortality was associated with V_m and change in V_m over the preceding 6 months. The association between change in V_m and mortality was independent of vascular access complications. In contrast, mortality was inversely associated with V calculated from anthropometric measurements (V_ant). In case mix–adjusted analysis using V_m as a time-dependent covariate, the association of mortality with V_m strengthened after statistical adjustment for V_ant. After adjustment for V_ant, higher V_m was associated with slightly smaller reductions in intradialytic systolic BP and with risk factors for mortality including recent hospitalization and reductions in serum albumin concentration and body weight.

Conclusions

An increase in V_m is a marker for illness and mortality risk in hemodialysis patients.

Introduction

In the Hemodialysis (HEMO) Study, patients assigned to receive an equilibrated Kt/V (eKt/V) of either 1.05 or 1.45 three times weekly showed no significant differences in survival, hospitalization, nutritional status, or health-related quality of life (1,2). These results contrast markedly with observational studies that have consistently shown associations between higher dialysis dose and survival (3–5). In a previous analysis of HEMO Study data, we found strong associations between the achieved dose and survival within each of the two randomized dose groups (6). The magnitude of the association greatly exceeded the 95% confidence limits of the dose–survival relationship indicated by the intention-to-treat analysis. Within each of the two randomized dose groups, patients achieving a higher dialysis dose apparently shared clinical characteristics that were favorably associated with survival independent of dose itself. We referred to this bias as dose-targeting bias. Such a bias might be related to the numerator of dose (Kt) influenced directly by the dialysis prescription or to the denominator, the modeled urea volume, V_m. This study explores the association between mortality and modeled urea volume (V_m) as another potential source of confounding of the dose–mortality association.

Patients and Methods

Study Design

The HEMO Study was a multicenter randomized clinical trial designed to compare survival, hospitalization, nutritional status, and health-related quality of life in subjects randomized to different levels of dialysis dose (i.e., eKt/V_urea 1.05 or 1.45) and membrane flux (1). A total of 1846 subjects from 72 dialysis units affiliated with 15 dialysis centers in the United States were randomized between 1995 and 2001. The dose intervention was administered with centrally generated dialysis prescriptions containing combinations of blood flow, dialysate flow, and treatment time that resulted in an expected eKt/V of 1.05 or 1.45 (single-pool Kt/V levels of approximately 1.3 and 1.7). Prescribed eKt/V_urea was computed based on the subject's running mean V_m over the prior four kinetic modeling (KM) sessions in the standard dose group and six KM sessions in the high dose group. Prescriptions were updated after changes in the running mean V_m of ≥5% in the standard dose group and after increases of ≥5% or decreases of ≥7.5% in the high dose group. Urea KM in the HEMO trial (7–10) is described in more detail in the Appendix.

Measurements

Baseline evaluation included five case mix factors: age, gender, race (black versus nonblack), diabetes, and years of dialysis before the trial. Anthropometric total body water volume (V_ant) was estimated using the Watson formula (11). Comorbidity was summarized using the Karnofsky index (12), the overall score from the Index of Coexisting Disease (ICED) (13,14), and eight subindices from the ICED evaluating specific domains. Vascular access type (fistula versus graft versus catheter), occurrence of access revisions during the previous month, number of hypotensive episodes, pre- and postdialysis weight, pre-, intra-, and postdialysis systolic and diastolic BP, interruptions, and other deviations from the dialysis prescription were extracted from the dialysis facility's flow sheets. Hospitalizations were identified by chart review and patient interview and cross-checked with Medicare records (15).

Data Analysis

We stratified all survival analyses by clinical center and censored follow-up time at transplantation or 4 months after transfer to nonparticipating dialysis facilities. We restricted analyses of V_m to modeled sessions with <15 minutes of interruption time and no obvious blood urea nitrogen (BUN) measurement errors (7,10).

Correlates of V_m, V_ant, and V_m/V_ant

We used mixed effects models (17,18) to relate modeled and anthropometric volume measured monthly to individual clinical factors after controlling for the randomized dose and flux groups and five case mix factors: age, gender, race, diabetes, and vintage (years on dialysis). Vascular access type was also included as a covariate in analyses relating volume measurements to BP. Clinical factors considered and the technical details of the mixed models are described in the Appendix.

We applied a 1-month lag to the 4-month moving averages to avoid coupling with the volume-related measurements. We examined three different models for V_m and V_ant to distinguish among relations involving changes in V_ant and relations involving changes in V_m as distinct from V_ant. The first two models related V_m and V_ant as outcome variables to the predictor variables listed above while controlling for baseline V_ant. The third model related the follow-up V_m/V_ant ratio to the predictor variables while controlling for follow-up V_ant.

Association of V_m with Mortality

We used Cox regression with time-dependent covariates (18) to relate mortality at each time point to the latest 4-month running mean values of V_m, V_ant, or V_m/V_ant while controlling for the five baseline case mix variables, randomized treatment, and either baseline or follow-up V_ant (see above). Our primary assessments of the association of V_m with mortality were based on analyses that both factored V_m for V_ant and incorporated V_ant as a covariate. The ratio of V_m/V_ant has a weaker correlation with V_ant than does V_m; hence, this model provides greater robustness for assessing relations between V_m and mortality than analyses relating mortality to V_m while controlling for V_ant. These analyses were performed in all 1846 randomized subjects.

Association of Change in V_m with Mortality

We also used Cox regression with time-dependent covariates to relate mortality to the change in the 4-month moving average of V_m over the preceding 6 months while controlling for the five baseline case mix variables, randomized treatment, and the 6-month lagged running mean V_ant. We used a two-slope linear spline to estimate the mortality hazard ratios (HRs) separately for increases and decreases in V_m. The analyses of change in V_m were restricted to 1704 subjects with at least one KM session after 6 months of follow-up. The basic Cox regression for change in V_m was extended by adding interaction terms to evaluate effect modification by (1) gender, (2) lower or higher V_ant (V_ant ≤ 35 L versus V_ant > 35 L), (3) diabetes, (4) subjects randomized to the standard or high dose groups, and (5) subjects with and without occurrence of a vascular access issue during the preceding 10 months (including the 6 months over which the change in V_m was evaluated plus the prior 4 months). We defined a vascular access issue as one or more of the following: (1) access procedure, (2) access hospitalization, (3) change of access type, and (4) use of venous catheter. An extension of this analysis evaluated the relation between change in V_m and mortality separately for subjects with and without occurrence of either an access issue or a reported shortfall in treatment time. A similar model with interaction terms was used to evaluate the HR for the 6-month change in the 4-month average V_m separately for subjects with contemporaneous increases (≥2 kg) in postdialysis weight, stable weight (absolute value of weight change <2 kg), or weight decreases (≥2 kg), where weights were averaged over 4-month windows and changes evaluated over 6 months, similarly to V_m.

Role of V_m in Accounting for Dose Targeting Bias

To align each measure of V_m and eKt/V_urea, we used a Cox regression with time-dependent covariates to relate mortality to quintiles of prescribed eKt/V_urea within each dose group while controlling for the running mean V_m that was used for the current prescription, along with case mix, baseline V_ant, and the randomized dose and flux groups. For details, please see the Appendix.

Results

Patient Characteristics

Baseline characteristics of the HEMO subjects have been previously described (2). Treatment characteristics and volume-related parameters at 4 months are shown in Table 1. In 1716 subjects who had at least one KM session with valid estimates of V_m at month 4 or later, the 4-month average values of V_m, V_ant, and V_m/V_ant at the first KM session at or after month 4 were 31.6 ± 6.3, 35.0 ± 6.1, and 0.90 ± 0.12 (SD) L, respectively. At 4 months, the Pearson correlation coefficients of mean V_ant with V_m and V_m/V_ant were +0.76 and −0.15, respectively.

Table 1.

Patient characteristics at 4-month follow-up (n = 1716)

Factor	Mean (SD) or Percentage
	Standard Dose Group	High Dose Group
Age (years)	57.6 (14.0)	57.5 (14.1)
Percent diabetic	44%	44%
Percent black	64%	62%
Percent female	55%	57%
Percent with history of CHF	39%	39%
4-month predialysis SBP (mmHg)	153 (21)	151 (21)
4-month predialysis DBP (mmHg)	82 (12)	81 (12)
URR (%)	66.7 (7.2)	75.6 (5.7)
spKt/V	1.33 (0.17)	1.72 (0.22)
eKt/V	1.16 (0.15)	1.54 (0.19)
Dialysis time (minutes)	190 (24)	219 (25)
Blood flow at 30 minutes (ml/min)	325 (75)	410 (61)
Dialyzer clearance (ml/min)	218 (28)	250 (22)
4-month modeled Vm (L)	31.6 (6.5)	31.5 (6.0)
4-month Watson Vant (L)	35.3 (6.2)	34.8 (5.9)
4-month Vm/Vant ratio (%)	89.9 (11.3)	91.1 (12.5)
4-month body weight (kg)	70.2 (14.8)	69.1 (14.7)
Percent venous catheter	6%	6%
Percent AV graft	57%	60%
Percent AV fistula	35%	33%

Means (SD) are presented for continuous variables and percents for dichotomous factors. Access type was classified as other or unknown for approximately 2% of patients. Averages of measurements over months 1 through 4 of follow-up are shown for V_m, V_ant, V_m/V_ant, and body weight.

Predictors of V_m, V_ant, and V_m/V_ant

Baseline factors.

In multivariable analyses relating V_ant to randomization group and case mix, follow-up V_ant was significantly associated with gender (8.3 ± 0.2 L higher for men, mean ± SEM), diabetes (1.2 ± 0.2 L higher for subjects with diabetes), vintage (0.13 ± 0.02 L lower per year of dialysis), and age (0.4 ± 0.08 L lower per decade). After controlling for follow-up V_ant, the follow-up V_m/V_ant ratio was associated with race (4.2 ± 0.6% higher for blacks) and diabetes (2.7 ± 0.5% higher for subjects with diabetes).

Follow-up factors.

Table 2 shows the associations of V_m and V_ant with selected follow-up factors after controlling for case mix and dose and flux randomized groups. Compared with subjects with fistulas, graft use was associated with a lower V_m (−1650 ml) but a higher V_ant (+460 ml), whereas use of catheters was associated with a higher V_m (+2550 ml) but lower V_ant (−570 ml).. Vascular access– and non–vascular access–related hospitalizations were associated with a slightly higher V_m and lower V_ant; lower serum albumin concentrations were also associated with higher V_m and lower V_ant. After adjustment for V_ant, graft use was associated with a 5.5% lower V_m/V_ant, catheter use with a 8.6% higher V_m/V_ant, vascular access– and non–vascular access–related hospitalizations with a 3.4 and 1.6% higher V_m/V_ant, respectively, and reductions in serum albumin and weight were respectively associated with increases in V_m/V_ant of 12.8% per 0.5 mg/L decrease and 2.0% per 5 kg decrease.

Table 2.

Relationship of follow-up volume parameters with selected follow-up factors

Model	Predictor Variable	Current Vant (L)		Vm (L)		Vm/Vant (%) Controlling for Follow-up Vant
		Regression Coefficient	95% CI	Regression Coefficient	95% CI	Regression Coefficient	95% CI
1	Graft versus fistula	0.46a	(0.29 to 0.62)	−1.65a	(−2.04 to −1.27)	−5.48a	(−6.61 to −4.36)
2	Catheters versus fistula	−0.57a	(−0.71 to −0.43)	2.55a	(2.11 to 3.00)	8.56a	(7.20 to 9.91)
3	Vascular access-related hospitalization in last 4 months	−0.22a	(−0.30 to −0.13)	1.06a	(0.79 to 1.32)	3.42a	(2.60 to 4.24)
4	Nonvascular access-related hospitalization in last 4 months	−0.51a	(−0.56 to −0.45)	0.24a	(0.07 to 0.41)	1.64a	(1.15 to 2.13)
5	Change in serum albumin from baseline (per 0.5-g/dl decrease)	−3.60a	(−4.07 to −3.14)	1.84a	(0.73 to 2.95)	12.76a	(9.64 to 15.9)
	Change in serum albumin from baseline (per 0.5-g/dl increase)	1.40a	(0.87 to 1.93)	−0.10	(−1.52 to 1.31)	−2.99	(−6.92 to 0.94)
6	Change in postdialysis weight from baseline (per 5-kg decrease)	−1.36a	(−1.40 to −1.32)	−0.39a	(−0.62 to −0.17)	2.04a	(1.35 to 2.74)
	Change in postdialysis weight from baseline (per 5-kg increase)	1.30a	(1.25 to 1.34)	0.43a	(0.23 to 0.63)	−1.34a	(−1.90 to −0.79)

Shown are results of separate time-dependent mixed effects regression analyses relating each of the three follow-up volume parameter to six individual follow-up predictor variables after controlling for baseline case mix factors, access type, and randomized treatment group. Linear spline models were used to fit separate relationships for increases and decreases in serum albumin and weight. The analyses of V_ant and V_m also controlled for baseline V_ant. The changes in serum albumin and postdialysis weight were computed based on 4-month averages and lagged by lagged by 1 month.

^aP < 0.01.

Table 3 shows the associations of V_m and V_ant with direct measures of BP and ultrafiltration. After adjustment for V_ant, there was no association between V_m/V_ant and either the predialysis systolic BP (SBP) or diastolic BP (DBP). However, postdialysis SBP and DBP exhibited weak but statistically significant direct correlations with the V_m/V_ant ratio. There was also a statistically significant direct correlation between change in BP during dialysis and V_m/V_ant, whether this was expressed as the decline from predialysis SBP to the lowest intradialytic SBP or as the decline from predialysis to postdialysis SBP. The magnitude of the relationship was relatively small, however; e.g., a 10% greater decline from the pre- to postdialysis SBP was associated with a 0.8% lower V_m/V_ant. Intradialytic hypotension and higher ultrafiltration volume relative to postdialysis body weight were also associated with lower V_m/V_ant ratios. Figure 1A shows the more pronounced intradialytic SBP decline associated with lower V_m/V_ant; Figure 1B shows the relationship relative to the total variation in intradialytic decline in SBP.

Table 3.

Relations of follow-up volume parameters with BP-related variables averaged over the preceding 4 months

BP-Related Predictor Variable	Outcome
	Vant (L)		Vm (L)		Vm/Vant (%) Controlling for Follow-Up Vant
	Regression Coefficient	95% CI	Regression Coefficient	95% CI	Regression Coefficient	95% CI
Predialysis systolic BP (per 10 mmHg increase)	0.02	(−0.01 to 0.04)	−0.00	(−0.05 to 0.05)	−0.06	(−0.22 to 0.09)
Predialysis diastolic BP (per 10 mmHg increase)	0.05a	(0.01 to 0.09)	0.08	(−0.01 to 0.16)	0.10	(−0.15 to 0.35)
Postdialysis systolic BP (per 10 mmHg increase)	−0.08a	(−0.10 to −0.06)	0.09a	(0.04 to 0.14)	0.36a	(0.22 to 0.51)
Postdialysis diastolic BP (per 10 mmHg increase)	−0.04	(−0.07 to −0.00)	0.15a	(0.06 to 0.23)	0.44a	(0.19 to 0.70)
Drop from predialysis to postdialysis systolic BP (per 10% greater drop)	0.17a	(0.14 to 0.20)	−0.18a	(−0.26 to −0.09)	−0.80a	(−1.05 to −0.56)
Drop from predialysis to min. intradialytic systolic BP (per 10% greater drop)	0.14a	(0.11 to 0.18)	−0.27a	(−0.37 to −0.18)	−1.06a	(−1.34 to −0.78)
Hypotensive episodes (≥1 episode versus 0 episodes)	0.35a	(0.22 to 0.48)	−0.92a	(−1.26 to −0.58)	−0.43a	(−0.63 to −0.24)
Uft/Wt (per 10% increase)	−0.49a	(−0.80 to −0.19)	−1.50a	(−2.25 to −0.76)	−3.30a	(−4.29 to −2.31)

Shown are results of separate time dependent mixed effects regression analyses relating follow-up volume parameters to individual follow-up BP-related predictor variables averaged over the preceding 4 months (not including the modeling session used to compute V_m or V_ant) after controlling for baseline case mix factors, access type, and randomized treatment group. Analyses of V_ant and V_m also controlled for baseline V_ant

^aP < 0.01.

Figure 1.

(A) Unadjusted mean ± SEM levels of the intradialytic decrease in systolic BP by V_m/V_ant quintile. (B) Full distribution of the intradialytic drop in systolic BP (10th, 25th, 50th, 75th, and 90th percentiles) by V_m/V_ant quintile.

Associations among V_ant, V_m, and V_m/V_ant and Mortality

Table 4 presents the relationship of mortality with the three volume indices. An inverse association was present between V_ant and mortality. The relative hazard of a subject with a V_ant >40 L was only 0.3 compared with a patient with V_ant <30 L. For modeled volumes (V_m), the association was reversed, such that a patient with V_m >40 L had a 50% higher mortality hazard compared with a subject with V_m <30 L. After controlling for V_ant, the association between V_m/V_ant and mortality was magnified, showing an approximate threefold increase in risk for subjects in whom the V_m/V_ant ratio was >1.1 (110%) compared with those with V_m/V_ant < 0.80. Throughout follow-up, 16.6, 34.1, 28.8, 12.8, and 7.8% of V_m/V_ant measurements were <0.80, 0.80 to 0.90, 0.90 to 1.00, 1.00 to 1.10, and >1.10, respectively. A total of 34.3% of subjects had a 4-month mean V_m/V_ant >1.10 at least once during follow-up.

Table 4.

Association of all-cause mortality with time-dependent volume indices in the HEMO Study

Effect of Vant (L)				Effect of Vm (L)				Effect of Vm/Vant Controlling for Vant (L)
Vant (L)	HR	95% CI	P	Vm (L)	HR	95%CI	P	Vm/Vant (as %)	HR	95% CI	P
<30 La	1.00	—	—	<30 La	1.00	—	—	<80%a	1.00	—	—
30 to 35 L	0.60	(0.50 to 0.72)	<0.001	30 to 35 L	1.11	(0.93 to 1.33)	0.26	80 to 90%	1.07	(0.84 to 1.38)	0.57
35 to 40 L	0.43	(0.34 to 0.55)	<0.001	35 to 40 L	1.06	(0.85 to 1.33)	0.61	90 to 100%	1.27	(0.99 to 1.64)	0.06
>40 L	0.28	(0.21 to 0.38)	<0.001	>40 L	1.51	(1.17 to 1.94)	0.001	100 to 110%	2.23	(1.70 to 2.92)	<0.001
—								>110%	3.41	(2.60 to 4.49)	<0.001
HR per 5 L (continuous)	0.64	(0.58 to 0.70)	<0.001	HR per 5 L (continuous)	1.10	(1.04 to 1.17)	0.002	HR per 10% (continuous)	1.27	(1.22 to 1.33)	<0.001

All HRs are based on time-dependent Cox regressions relating mortality to 4-month averaged in the indicated volume parameters and adjusted for baseline case mix variables and randomized treatment group. The three top panels provides HRs from separate Cox regressions relating mortality to V_ant, V_m, and V_m/V_ant categories relative to the respective reference categories. The three bottom panels provide overall summary HRs under separate Cox regressions relating mortality to the respective volume parameters under linear models. The dose–response curves did not differ significantly from linearity for either V_ant (P for test of linearity = 0.48) or V_m (P = 0.39), but indicated a stronger relationship at higher V_m/V_ant ratios than at smaller V_m/V_ant ratios (P < 0.001). Each analysis was stratified by clinical center.

^aReference category.

A further graphical analysis of the joint association of mortality with V_ant and V_m/V_ant is presented in Figure 2. Higher levels of V_m/V_ant were associated with marked increases in mortality risk among patients at any given level of V_ant.

Figure 2.

Shown are HRs resulting from a time-dependent Cox regression jointly relating mortality to subgroups defined by the indicated levels of V_m and V_m/V_ant after controlling for the five case mix variables and the randomized dose and flux groups, with stratification by clinical center. The Cox model included separate indicator variables for each of the 19 nonreference subgroups, thus accounting for any differences in the relationship of mortality with V_m/V_ant across different V_ant levels.

In contrast to the strong cross-sectional association of V_m with V_ant, across all follow-up visits, the association of change in 4-month mean V_m with contemporaneous change in the 4-month V_ant was negligible (Pearson r = 0.07). Hence, we evaluated the association of change in V_m with mortality without statistical adjustment for change in V_ant. Overall results and results in specific clinical circumstances are presented in Table 5. In the overall analysis (top panel), mortality was not associated with decreases in V_m but was directly associated with increasing V_m. The association of increasing V_m with mortality was evident regardless of whether the subject's weight was increasing, decreasing, or relatively constant, but was strongest when the subject's weight was increasing. The hazard associated with increasing V_m was evident in men and women and was more pronounced in women (interaction P = 0.008). The mortality hazard associated with increasing V_m was evident in “smaller” and “larger” patients. Diabetes did not modify the association of increasing V_m with mortality, nor did age, vintage, or randomized dose group. The hazard associated with increasing V_m was evident during time periods with and without reported vascular access issues and with and without either a reported vascular access issue or time shortfall of reported versus prescribed treatment time.

Table 5.

Association of mortality with 1-L increase in 4-month running mean of V_m in different subgroups or situations

Factor	Subgroup or Situation	HR	95% CI	P in Subgroup or Condition	Interaction P
Direction of change in Vm	Decrease in Vm	1.00	(0.97 to 1.04)	0.90	—
	Increase in Vm	1.08	(1.06 to 1.11)	<0.001
Weight change	>2 kg	1.13	(1.06 to 1.21)	<0.001	0.04
	−2 to 2 kg	1.10	(1.06 to 1.14)	<0.001
	< −2 kg	1.05	(1.02 to 1.08)	0.004
Gender	Male	1.05	(1.02 to 1.08)	0.004	0.008
	Female	1.11	(1.08 to 1.15)	<0.001
Baseline Vant	Vant < 35 L	1.08	(1.05 to 1.11)	<0.001	0.95
	Vant > 35 L	1.08	(1.04 to 1.12)	<0.001
Diabetic status	Nondiabetic	1.09	(1.05 to 1.12)	<0.001	0.63
	Diabetic	1.07	(1.04 to 1.11)	<0.001
Dose group	Standard dose	1.10	(1.06 to 1.13)	<0.001	0.24
	High dose	1.07	(1.04 to 1.10)	<0.001
Vascular access issue	No access issues	1.07	(1.04 to 1.11)	<0.001	0.92
	≥1 access issue	1.07	(1.04 to 1.11)	<0.001
Vascular access issue or time or shortened time	No access or time issues	1.06	(1.03 to 1.10)	<0.001	0.40
	≥1 access or time issue	1.08	(1.05 to 1.11)	<0.001

All HRs are based on time-dependent Cox regressions relating mortality to changes in the 4-month mean V_m levels over the preceding 6 months, adjusting for the case mix variables, baseline anthropometric volume, the assigned dose, and flux groups, and the 4-month running mean volume lagged by 6 months. The analyses were stratified by clinical center. Increases and decreases in the running mean V_m were modeled separately using a linear spline model; only HRs for increases in V_m are shown for the asterisked rows. The HRs evaluate the change in risk associated with a 1-L increase in V_m, which represents 25% of 1 SD (4.01 L) of the changes in the 4-month running mean V_m.

Effect of Controlling for V_m on Dose-Targeting Bias

The previously reported bias in the relation between achieved eKt/V and mortality (6) was eliminated in the standard dose group and attenuated in the high dose group when the analysis was performed on prescribed eKt/V and adjusted for the running mean modeled volume (V_m) at the time of the preceding prescription change (Figure 3).

Figure 3.

Shown are the HRs from a time-dependent Cox regression analysis relating mortality to quintiles of prescribed eKt/V within each dose group while controlling for the running mean V_m that was used for the patient's latest dialysis prescriptions, while controlling for the five baseline case mix factors, baseline V_ant, and the randomized dose and flux groups. The analysis was stratified by clinical center.

Sensitivity Analyses

The results shown in Tables 4 and 5 were essentially unchanged after additionally controlling for a more complete set of baseline covariates, including the Karnofsky score, serum albumin, the equilibrated normalized protein catabolic rate, predialysis serum creatinine, predialysis SBP, predialysis DBP, and eight subindices of the ICED (but not the overall ICED itself, to avoid collinearity). Similar results for the association of mortality hazard with V_m to those shown in Table 4 and Figure 2 were obtained when the mortality hazard was related to log-transformed V_m while controlling for log transformed V_ant as a separate covariate.

Discussion

Our results suggest that subjects in the HEMO Study, whether they were assigned to receive the conventional or higher dose of dialysis, were at increased risk of death if they had a high value for modeled urea volume (V_m) or if their value for V_m was increasing with time. The association between increasing V_m and mortality was magnified after adjusting for patient size (as V_ant) because the relation between V_m and mortality and V_ant and mortality was in opposite directions, as others have reported (19,20). The association of mortality with changes in V_m was evident across a wide age range, in both genders, in blacks and non-blacks, in subjects with or without diabetes, and across other comorbidities.

To explain the association of V_m with mortality, we considered whether increases in V_m reflected some technical problem with dialysis delivery that might also be linked to patient risk. An increased in modeled urea volume may not always represent a true increase in total body water, but may be caused by overestimation of the amount of urea removed during dialysis. This can be caused by overestimation of dialysis session length (because of unaccounted-for interruptions) or to overestimation of dialyzer clearance (because of overestimation of blood or dialysate flow rate, dialyzer membrane mass transfer area coefficient [K₀A]), or time-averaged intradialytic BUN [e.g., because of access recirculation (20)]. Several scenarios associated with overestimation of dialytic urea removal might conceivably be linked to mortality. For example, subjects with dialysis-associated hypotension resulting in interruptions or reductions in blood flow might have both overestimation of urea removal and a poorer prognosis. We believe that overestimation of urea removal was not the major cause of the observed high values for V_m for the following reasons. During each monthly modeled dialysis, clinical symptoms, lowest intradialytic BP, and any interruptions, including those requiring a reduction in blood flow rate, were carefully recorded. We limited our V_m mortality analyses to modeled sessions during which treatment interruptions totaled no more than 15 minutes. Higher V_m was associated with higher, rather than lower, intradialytic BP, suggesting that hypotension-related interruptions were not the principal cause of increasing V_m. For each modeled session, the K₀A was adjusted for the estimated effects of reprocessing on clearance. Finally, the association between V_m and mortality persisted in the conventional dose group, where relatively low blood flow values were used and where there dialysis delivery parameters were not being stressed to deliver the maximum possible dose of dialysis.

The use of intravenous catheters for vascular access has been consistently associated with a lower Kt/V_urea relative to fistulas and grafts, because of the inability to deliver higher blood flow rates and possibly because of increased degrees of access recirculation. An increase in V_m/V_ant was associated with use of a venous catheter and also with use of a fistula versus a graft (perhaps explained by more recirculation at high blood flows in some fistulas). However, the relation of change in V_m with mortality was similar in patients using each of the three vascular access types (data not shown) and was also similar during periods with and without vascular access issues (defined as reported access procedures, access-related hospitalizations, or use of an intravenous catheter). Thus, vascular access issues could not adequately explain the association of V_m with mortality. Nevertheless, it is possible that some portion of the increases in V_m observed were caused by overestimation of urea removal in the kinetic modeling process.

A second hypothesis was that an increased V_m is a reflection of fluid overload. We studied the V_m–mortality relation under conditions when body weight was increasing, staying more or less constant, or decreasing (Table 5), and in all three scenarios, increases in V_m were associated with mortality. In the situation where both V_m and body weight were increasing, the most plausible explanation is that the increase in V_m was primarily caused by an increase in extracellular fluid volume. In fact, under these conditions, the relation between increasing V_m and mortality risk was the strongest (Table 5). In the situation where V_m increases but weight is not changed, there are two possible explanations: (1) overestimation of urea removal or (2) patient's “flesh weight” (both muscle and fat) has been lost and has been replaced with fluid. When body weight decreases but V_m increases, the increase in fluid more than offsets the loss of flesh weight. In the latter situation, the association between V_m and mortality was attenuated, possibly because the V_m-associated risk was overshadowed by the high mortality risk associated with a falling body weight.

Although not associated with predialysis SBP or DBP per se, higher V_m/V_ant ratios were associated with higher postdialysis BPs (both systolic and diastolic). Also, higher V_m/V_ant ratios were associated with smaller falls in SBP during dialysis, whether calculated using the postdialysis SBP or the lowest measured intradialytic BP. These results are consistent with data from Inrig et al. (21,22), who found that the highest survival in patients with the greatest intradialytic fall in BP, whereas patients in whom BP did not fall or in whom BP increased, mortality risk was increased. It is tempting to speculate that the increase in V_m/V_ant and the lower decrement in intradialytic BP in patients in whom V_m/V_ant was increased were both related to fluid overload. However, our support for this hypothesis must be tempered by the relatively weak strength of the relationships between various BP metrics and V_m/V_ant.

As shown in Figure 3, the association between achieved eKt/V and mortality was eliminated in the conventional dose group after adjusting for V_m, but only moderately attenuated in the high dose group, which may be more similar to the currently prescribed dialysis dose. Thus, another, as yet unexplored, factor that may further confound the Kt/V–mortality relation may be the numerator of the Kt/V equation; namely, dialysis clearance (K) and session length (t). The relations between changes in K or t and mortality remain a subject for current (23–25) and future study and analysis.

In summary, we identified a strong and consistent association between modeled urea distribution volume (V_m) and mortality. Based on a series of analyses presented herein, we believe that the association is most likely related to an increase in extracellular water. This interpretation remains speculative, but the association of increased V_m with mortality does agree with the relation between bioimpedance-derived evidence of fluid overload and mortality recently published by others (26).

Disclosures

None.

Appendix

Details of KM

KM sessions included central measurements of pre- and postdialysis BUN and predialysis serum albumin determined by nephelometry (Spectra East, Rockleigh, NJ). Expected dialyzer clearance was computed at the monthly KM sessions from blood flow, dialysate flow, and centrally determined in vitro K₀A (dialyzer mass transfer area coefficient) values using standard formulae with appropriate adjustments for blood water content and ultrafiltration, as well as an adjustment for measured (in vivo)/(in vitro) clearance derived from a subsample of modeled dialyzes where in vivo cross-dialyzer clearance was measured (6–10,27). Single pool urea kinetic volume (V_m) and single spool Kt/V (spKt/V) were computed by applying the Depner and Cheer 2-BUN algorithm (8) to the expected dialyzer clearance. A previously described adjustment was then applied to V_m to better approximate two-pool volume (7). During the trial, eKt/V was computed using the Daugirdas-Schneditz rate equation eKt/V = spKt/V − 0.6 K/V + 0.03 (6–8,27). Based on substudies performed during the trial, a modified rate equation, eKt/V = spKt/V − 0.40 K/V {+ 0.01 for patients in the high dose group}, was developed for reporting the trial results (7,10).

Details of Statistical Models for Predictors of V_m, V_ant, and V_m/V_ant

We estimated coefficients of each mixed model using normality restricted maximum likelihood with a compound symmetry covariance assumption to account for correlations of repeated measurements. The compound symmetry assumption is equivalent to fitting a model with a random effect for each patient. We used robust sandwich estimates for SEs to account for possible deviations of the true correlation structure from the assumed compound symmetry model (28). Linear splines were used to estimate the regression coefficients for the changes in serum albumin and weight separately for increases and for decreases in these parameters. Each mixed effects analysis was performed for 49,080 modeled dialyzes, with valid estimates of V_m obtained in 1716 patients with at least one KM session performed after 4 months of follow-up.

Details in the Analysis of the Role of V_m in Accounting for Dose Targeting Bias

In this analysis, prescribed eKt/V was defined as the predicted eKt/V based on the patients running mean V_m at the time of the most recent centrally administered prescription change in conjunction with the dialysis prescription parameters, blood flow, dialysate flow, dialyzer K₀A, and treatment time at the current KM session. We used prescribed eKt/V and the running mean V_m at the most recent prescription to minimize confounding effects resulting from recent fluctuations in V_m.