Relation between Access Blood Flow and Mortality in Chronic Hemodialysis Patients

Abstract

Background: Access blood flow (Qa) measurement is a potentially important determinant of systemic hemodynamics in hemodialysis patients. High Qa may contribute to left ventricular dilation and high output heart failure. On the other hand, low Qa might lead to underdialysis, which is associated with adverse outcomes.

Methods: In this retrospective study of incident chronic hemodialysis patients treated in three Canadian cities (Edmonton, Calgary, and Halifax), the hypothesis that extremes of Qa(low or high) would be associated with increased mortality was tested. The distribution of Qa was not Gaussian, and therefore Qa was log-transformed in analyses that treated it as a continuous variable. Qa was classified into categories defined by cutpoints of 500, 1000, 1500, and 2000 ml/min. Univariate and multivariate Cox proportional hazard models were performed to examine the relation between Qa and all-cause mortality. Patients were followed from the date of Qa measurement until death; follow-up was discontinued at loss to follow-up, kidney transplantation, or end of study.

Results: Of 820 participants, those with lower levels of Qa tended to be older and to have more comorbidities. During the median follow-up period of 28 mo, 206 (25.1%) participants died and 101 (12.3%) patients received a kidney transplant. When only baseline measures of Qa were considered, there was significant association between Qa and mortality [hazard ratio (HR) per unit increase in logQa 0.81, 95% confidence interval (CI) 0.67, 0.97; adjusted HR per unit increase in logQa 0.90, 95% CI 0.72, 1.11]. The adjusted risk of mortality was similar between the different categories of baseline Qa before and after adjustment for demographic characteristics, comorbidity, and access type. In analyses that included all Qa measurements per patient as a time-varying covariate, the adjusted association between Qa and death remained nonsignificant, with no evidence of increased mortality at higher Qa (HR per unit increase in logQa 0.82, 95% CI 0.67, 1.01, P = 0.066).

Conclusion: The findings of this study do not suggest an increased risk of death at higher levels of Qa, Further studies would be needed to confirm an increased risk of death at lower Qa.

Measurement of access blood flow (Qa) as a means to detect access dysfunction has become common over the past decade. Many dialysis programs now screen arteriovenous fistula and synthetic grafts on a repetitive basis, because observational studies have shown that low or declining Qa predict subsequent access failure (1–4). Qa is potentially affected by many factors, including systemic hemodynamics (i.e. BP and cardiac output), the size and endothelial function of the vessels supplying and draining the access, and the presence of significant vascular stenosis. Therefore Qa might theoretically serve as a marker of cardiovascular health in dialysis patients (5–8) If true, this suggests that Qa could be used to predict clinical outcomes in dialysis patients.

However, any association between Qa and survival might not be straightforward. For instance, although low Qa might be linked to adverse outcomes through poor cardiac status or (in extreme cases) underdialysis (9), high Qa could also be problematic because it may contribute to left ventricular dilation and high output heart failure (10–17) This is supported by the observation that banding of high-flow arteriovenous fistulae apparently reverses heart failure and its associated hemodynamic changes (18–22). Because extremely high levels of Qa might therefore contribute to cardiovascular decompensation, regular echocardiography has been recommended for such patients (15).

Thus, although both low and high Qa might lead to adverse outcomes, no data exist to confirm this. We hypothesized that extremes of Qa (low or high) would be associated with increased mortality in hemodialysis patients.

Concise Methods

Participants

This was a retrospective cohort study of incident adult patients who underwent chronic hemodialysis treatment in three Canadian cities (Edmonton, Calgary, and Halifax) between June of 1999 and November 2005. All patients initiating dialysis during this period were eligible for inclusion, provided that they had a functioning arteriovenous access (fistula or graft) and had at least one access flow measurement performed. Patients less than 18 yr of age were excluded.

Measurement of Access Blood Flow

Access screening was performed as recommended by the Canadian Society of Nephrology (CSN) in all three study sites (23) using Transonic HD01/HD02 devices (Transonic Corporation, Ithaca, New York). The electronic data files from all of the devices at each site were combined to obtain a record of all Qa measurements performed during the study period. Using this file, we obtained the first instance of Qa measurement for each patient. Because Qa was measured at least twice in each session (23), the mean of all values on that date was used. In the primary analysis we classified Qa into categories based on arbitrarily selected cutpoints of 500, 1000, 1500, and 2000 ml/min (<500; 500 to 999.9; 1000 to 1499.9; 1500 to 1999.9; ≥2000 ml/min). Most participants had a baseline Qa of 500 to 999 ml/min, which was therefore selected as the reference category. However, we thoroughly tested the effect of alternative classification schemes in sensitivity analyses, using both data-driven classifications (such as quartiles and quintiles) as well as focusing on very high or very low levels of Qa. We also considered using Qa as a continuous variable after log transformation to account for its skewed distribution. Finally, we considered the possibility that the risk might be exponential at higher levels of flow by inserting a quadratic term in the model.

Clinical Characteristics

For all included patients, data on age, gender, and baseline comorbidity (including diabetes, hypertension, angina, previous myocardial infarction, pulmonary edema, peripheral vascular disease, cerebrovascular disease, lung disease, and malignancy) were extracted from the Canadian Organ Replacement Register (CORR), as in our previous work (24) Data were forwarded to CORR by all Canadian dialysis centers, transplant centers, and organ procurement organizations. Patients with missing data on comorbidity were excluded. Data on access type and location were obtained from the Transonic electronic files and local electronic (Edmonton, Calgary) and paper (Halifax) records collected by individuals who were unaware of the study hypothesis.

Ascertainment of All-Cause Death

Data on all-cause death were obtained from the CORR. Because (as with any registry) there is a delay between the occurrence of a clinical event and its availability in CORR, local electronic (Edmonton, Calgary) and paper (Halifax) records were used to supplement the CORR data on date of death by individuals who were unaware of Qa measurements. These data on all-cause death were then linked to the Qa data and the clinical data from CORR to create a single database on which all analyses were performed.

Statistical Analyses

Participants were followed from the date of Qa measurement until death; follow-up was discontinued at loss to follow-up, kidney transplantation (as ascertained by CORR), or end of study. Cox proportional hazard models were used to determine the independent association between Qa and mortality among study participants, with the forward selection technique (P for inclusion <0.20 and P for exclusion ≥0.05) used for model building. Factors considered in the Cox models included patient demographics, smoking status, time since dialysis initiation (days), vascular access type (graft versus fistula), access location (lower versus upper arm), and the systolic and diastolic BP at time of Qa measurement. Comorbid conditions including diabetes mellitus, coronary disease (angina, myocardial infarction, previous coronary artery bypass grafting), history of hypertension, chronic heart failure, stroke, chronic lung disease, and peripheral vascular disease (limb loss, gangrene, peripheral revascularization), and known malignancy were also included in the Cox regression analyses. We used crossproduct terms in the Cox model (25) to test for two-way interactions between Qa and each of the following: access type, access location, diabetic status, and peripheral vascular disease.

We tested the proportional hazard assumption in all Cox models by examining plots of the log-negative-log of the within group survivorship functions versus log-time as well as comparing Kaplan–Meier (observed) with Cox (expected) survival curves. Statistical analyses were performed with STATA software version 9.1 (College Station, Texas). The ethics review board at the University of Alberta approved the study.

Results

There were 842 potentially eligible patients who initiated dialysis at the three centers during the study period. Of these, 22 had missing data on one or more covariates, leaving 820 patients available for analysis (Table 1).

Table 1.

Baseline characteristics of patients

n	820
Age, yearsa	66 (23)
Gender, male	514 (62.7)
Race
Caucasian	505 (61.6)
Asian	64 (7.8)
Black	14 (1.7)
Indian subcontinent	20 (2.4)
Aboriginal	46 (5.6)
other	171 (20.9)
Cause of ESRD
diabetes	267 (32.6)
GN	114 (13.9)
hypertensive/ischemic	125 (15.2)
polycystic kidney disease	46 (5.6)
other	179 (21.8)
unknown	89 (10.9)
Blood pressure, mmHga
systolic	133 (30.5)
diastolic	70 (19)
Access type, arteriovenous fistula	658 (80.2)
Access location, lower arm	369 (45.0)
History of comorbidity
diabetes mellitus	250 (30.5)
hypertension	655 (79.9)
peripheral vascular disease	129 (15.7)
malignancy	66 (8.1)
chronic lung disease	66 (8.1)
myocardial infarction	146 (17.8)
angina	172 (21.0)
pulmonary edema	170 (20.7)
cerebrovascular disease	92 (11.2)

^aMedian interquartile range.

Patients with lower levels of Qa tended to be older and to have more comorbidities such as diabetes mellitus, prior pulmonary edema, and previous myocardial infarction (Table 2). Patients with fistulae had higher median Qa [interquartile range] than those with grafts (935 ([829] versus 887.5 [510] ml/min respectively), and Qa was higher in upper-arm accesses compared with lower-arm accesses (1065 ([890] versus 780 ([590] ml/min respectively). During the median follow-up period of 28 mo (range 2 to 69 mo), 206 (25.1%) participants died and 101 (12.3%) patients received a kidney transplant.

Table 2.

Baseline characteristics of patients by access blood flow category

Variable	Access Blood Flow in ml/min
	<500	500 to 999.9	1000 to 1499.9	1500 to 1999.9	≥2000	P value
n	156	303	204	84	73
Age, yearsa	70.5 (18.5)	67 (22)	64 (22)	57 (26)	58 (24)	<0.001
Male gender	95 (60.9)	173 (57.1)	134 (65.7)	56 (66.7)	56 (76.7)	0.02
Race	0.57
Caucasian	99 (63.5)	182 (60.1)	123 (60.3)	49 (58.3)	52 (71.2)
Asian	8 (5.1)	28 (9.2)	18 (8.8)	6 (7.1)	4 (5.5)
Black	2 (1.3)	4 (1.3)	6 (2.9)	1 (1.2)	1 (1.4)
Indian subcontinent	3 (1.9)	10 (3.3)	4 (2.0)	3 (3.6)	0 (0.0)
Aboriginal	9 (5.8)	11 (3.6)	13 (6.4)	9 (10.7)	4 (5.5)
Other	35 (22.4)	68 (22.4)	40 (19.6)	16 (19.1)	12 (16.4)
Cause of ESRD
diabetes	66 (42.3)	98 (32.3)	58 (28.4)	24 (28.6)	21 (28.8)	0.05
GN	11 (7.1)	43 (14.2)	29 (14.2)	18 (21.4)	13 (17.8)	0.03
hypertensive/ischemic	24 (15.4)	50 (16.5)	28 (13.7)	14 (16.7)	9 (12.3)	0.85
polycystic kidney disease	4 (2.6)	15 (5.0)	21 (10.3)	3 (3.6)	3 (4.1)	0.02
other	28 (18.0)	69 (22.8)	46 (22.6)	16 (19.1)	20 (27.4)	0.51
unknown	23 (14.7)	28 (9.2)	22 (10.8)	9 (10.7)	7 (9.6)	0.50
Blood pressure, mmHgb
systolic	130 (36.5)	133 (32)	135 (33)	136.5 (29.5)	134 (24)	0.35
diastolic	67.5 (20.5)	70 (18)	71 (19)	71.5 (18.5)	73 (17)	0.18
Access type, arteriovenous fistula	133 (85.3)	225 (74.3)	157 (77.0)	73 (86.9)	70 (95.9)	<0.001
Access location, lower arm	86 (55.1)	162 (53.5)	84 (41.2)	23 (27.4)	14 (19.2)	<0.001
History of comorbidity
diabetes mellitus	68 (43.6)	89 (29.4)	50 (24.5)	25 (29.8)	18 (24.7)	0.002
hypertension	128 (82.1)	249 (82.2)	164 (80.4)	64 (76.2)	50 (68.5)	0.09
peripheral vascular disease	36 (23.1)	46 (15.2)	24 (11.8)	12 (14.3)	11 (15.1)	0.06
malignancy	12 (7.7)	26 (8.6)	16 (7.8)	10 (11.9)	2 (2.7)	0.33
chronic lung disease	13 (8.3)	23 (7.6)	18 (8.8)	4 (4.8)	8 (11.0)	0.68
myocardial infarction	42 (26.9)	46 (15.2)	32 (15.7)	12 (14.3)	14 (19.2)	0.02
angina	34 (21.8)	64 (21.1)	44 (21.6)	20 (23.8)	10 (13.7)	0.58
pulmonary edema	46 (29.5)	58 (19.1)	46 (22.6)	10 (11.9)	10 (13.7)	0.006
cerebrovascular disease	24 (15.4)	39 (12.9)	18 (8.8)	6 (7.1)	5 (6.9)	0.11

^aMean ± SD and t test P value otherwise n (%) and χ² test P value.

^bMedian ± interquartile range, and P value from nonparametric test of medians in two groups.

Qa as Continuous Independent Variable

Lower levels of log-transformed Qa were significantly associated with increased mortality in univariate analyses (HR per unit increase in logQa 0.81, 95% CI 0.67, 0.97). After adjustment, no statistically significant association was seen (HR per unit increase in logQa 0.90, 95% CI 0.72, 1.11).

Qa as Categorical Independent Variable

Analysis based on our a priori Qa categories did not suggest a linear relation between Qa and the risk of death (Table 3). The risk of mortality was similar in all Qa categories after adjustment for demographic characteristics, comorbidity, and access type (Table 3).

Table 3.

Unadjusted and adjusted relation between access blood flow and death

Qa (ml/min)	Unadjusted HR (95% CI)	Adjusteda HR (95% CI)
<500	1.10 (0.78 to 1.57)	1.01 (0.71 to 1.46)
500 to 999	1	1
1000 to 1499	0.74 (0.51 to 1.07)	0.77 (0.53 to 1.13)
1500 to 1999	0.65 (0.38 to 1.13)	0.83 (0.47 to 1.46)
≥2000	0.72 (0.42 to 1.24)	0.92 (0.51 to 1.64)

Qa, access blood flow; HR, hazard ratio; CI, confidence interval

^aAdjusted for age, gender, race, center, days on dialysis at time of first Qa measurement, angina, myocardial infarction, pulmonary edema, diabetes, peripheral vascular disease, malignancy, chronic lung disease, access type, access location, and blood pressure.

Sensitivity Analyses

We tested for two-way interactions between Qa and the following patient characteristics: access type, access location, and diabetic status. All were nonsignificant, suggesting that they did not significantly modify the relationship between Qa and mortality.

In additional analyses, neither patients with very high (>2500 ml/min, n = 38) or very low (<250 ml/min, n = 41) Qa had independently increased mortality compared with those with Qa of 500 to 999 ml/min. Results were unchanged when Qa was categorized in quartiles, quintiles, or deciles, or when a quadratic term for Qa was added to the model (to address the probability that risk increased exponentially at higher levels of Qa). Results were similar when analyses were restricted to the 658 (80.2%) patients with fistulae rather than grafts (data not shown).

Finally, we conducted analyses that included all Qa measurements per patient as a time-varying covariate. A total of 6419 Qa measurements were available for the 816 eligible patients. In these analyses, higher logQa was significantly associated with a decreased risk of death in univariate analyses (HR per unit increase in logQa 0.73, 95% CI 0.61, 0.88). After multivariate adjustment, this association was of borderline significance (HR per unit increase in logQa 0.82, 95% CI 0.67, 1.01, P = 0.066).

Discussion

We found no increased risk of death associated with high levels of Qa in this cohort of incident hemodialysis patients. Although there were trends toward lower mortality at higher levels of Qa, this association was no longer present after adjustment, suggesting that patient characteristics rather than access flow per se were responsible for the higher incidence of death. For example, diabetes mellitus, prior pulmonary edema, and previous stroke were all more common in patients with lower Qa. In addition, patients with lower Qa were more likely to have a graft rather than a fistula. Because grafts have been associated with excess mortality compared with fistulas in hemodialysis patients (25–27), this may also have contributed to the apparent increased risk in unadjusted analyses.

Our findings argue against the routine use of more intensive monitoring of cardiac function in patients with high Qa who have no clinical evidence of high-output cardiac failure. However, because our study did not include data on other clinically relevant adverse outcomes such as hospitalization and cardiovascular events, further studies will be required. The clinical significance of the borderline association between lower levels of Qa and the risk of death is unclear. Given the extent to which adjustment for measured comorbidity attenuated the strong association seen in univariate analyses, we speculate that this apparent association may be due to residual confounding by severity of vascular disease rather than any effect of low Qa per se.

Because of the arbitrary nature of the categories used in our primary analysis, we fully tested the impact of selecting different categories through extensive sensitivity analyses. We used data-driven classifications such as quartiles and quintiles as well as focusing on very high or very low levels of Qa. Results were similar in all analyses, suggesting that the lack of association between Qa and the risk of death is not dependent on the categories chosen in the primary analysis.

This analysis has several limitations that should be considered when interpreting the results. First, the cause of death was unknown and so we cannot determine whether death was due to cardiovascular disease, infection, or other causes. It is possible that death due to cardiovascular disease might have been higher in patients with high or low Qa. However, the clinical significance of this association (if present) would be unclear given that Qa was not associated with all-cause death. Although it would have been preferable to also include outcomes such as hospitalization, incident coronary events, or new heart failure in our analysis, we did not have access to these data. Second, in the primary analysis Qa was measured on only one occasion, meaning that some participants might have been misclassified with respect to Qa. Although absolute changes in Qa over time are relatively small (28), the error introduced by our relatively imprecise measurement of Qa may have biased the primary analysis toward the null. However, results were similar when all Qa measurements were included, suggesting that this is unlikely to have affected our conclusions. Third, this was a retrospective analysis that relied in part on registry data, which have well-known limitations. One such limitation is that the data set we used did not have information on several potentially relevant characteristics, including prior use of central venous catheters and dialysis dose, and thus we cannot exclude the possibility of residual confounding. Fourth, we cannot exclude the possibility that in some patients, surgeons may have deliberately constructed accesses that deliver lower Qa to patients known to have poor cardiac function. Finally, because patients with Qa <500 ml/min were routinely referred for fistulography (usually leading to angioplasty if stenosis was identified), it is possible that radiologic interventions could have altered the relation between low Qa and death. However, given that results were consistent whether follow-up Qa values were or were not included, we believe that the effects of percutaneous interventions are unlikely to have affected our conclusions.

In conclusion, our data do not suggest an independently increased risk of death in patients with high levels of Qa. Therefore the utility of routine Qa measurements may be limited to the known association between low levels of Qa and the risk of access loss (29,30)

Disclosures

None.