PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Kidney Int. Author manuscript; available in PMC 2010 October 1.
Published in final edited form as:
Published online 2010 January 20. doi:  10.1038/ki.2009.523
PMCID: PMC2864594
NIHMSID: NIHMS196230

Shorter dialysis times are associated with higher mortality among incident hemodialysis patients

Abstract

There is an association between hemodialysis session length and mortality independent of the effects of session duration on urea clearance. However, previous studies did not consider changes in session length over time nor did they control for the influence of time-dependent confounding. Using data from a national cohort of 8552 incident patients on thrice-weekly, in-center hemodialysis, we applied marginal structural analysis to determine the association between session length and mortality. Exposure was based on prescribed session length with the outcome being death from any cause. On the 31st day after initiating dialysis, the patients were considered at-risk and remained so until death, censoring, or completion of 1 year on dialysis. On primary marginal structural analysis, session lengths <4 h were associated with a 42% increase in mortality. Sensitivity analyses showed a dose–response relationship between session duration and mortality, and a consistency of findings across prespecified subgroups. Our study suggests that shorter hemodialysis sessions are associated with higher mortality when marginal structural analysis was used to adjust for time-dependent confounding. Further studies are needed to confirm these findings and determine causality.

Keywords: adequacy, dialysis, epidemiology, mortality, outcome, session length

Over the past several decades, dialysis dose has been measured in terms of, and titrated with respect to, urea kinetics. In 1985, Gotch and Sargent1 introduced the concept of single pool Kt/V, a measure of dialysis adequacy that attempts to match the degree of urea clearance to metabolic needs. In the intervening years, single pool Kt/V, and the related equilibrated Kt/V (eKt/V), have been largely adopted as the primary indices of dialysis adequacy.2

Over the past decade, the validity of Kt/V (single pool and/or equilibrated) as an outcome measure has been debated. Lowrie et al. have argued that since both greater urea clearance (K*t) and greater volume of distribution (V) are each associated with improved survival,3 dividing the two (that is, placing V in the denominator) is counter-intuitive, and guiding therapy by the Kt/V metric may therefore lead to disadvantageously low levels of urea removal in patients with already poor projected survival on the basis of small body size.4 In addition, in the present era of high-efficiency dialyzers (which enable rapid urea removal), the singular focus on urea kinetics may result in session lengths that are too short to enable adequate clearance of larger, charged, or protein-bound molecules, and moderate, well-tolerated rates of fluid removal for many patients.5,6

Several studies have examined the association between dialysis session length and mortality independent of (that is, adjusted for) Kt/V.5,7,8 Although many results lent credence to an association between shorter session length and mortality, these studies were beset by methodological shortcomings. Specifically, all studies considered session length as of a baseline value, which does not reflect the commonplace clinical practice of titration in response to clinical and biochemical circumstances. Moreover, none of the studies accounted for the potential of time-dependent confounding of the session length–mortality association. Time-dependent confounders are variables that are influenced by previous exposure, that then influence subsequent exposure, and are also associated with the outcome.912 It has been shown that in the presence of time-dependent confounding, standard methods of statistical adjustment (as used by the previous studies cited above) yield biased (that is, inaccurate) estimates of causal association, even when covariates are time-updated in the statistical model. Fortunately, novel statistical methods have been developed that enable more accurate estimation of causal associations in the presence of time-dependent confounding, among them marginal structural analysis.11,12

To determine whether shorter session length is associated with higher mortality while considering time-dependent confounding, we undertook a series of marginal structural analyses in which we examined the associations between session length and mortality, considering the former as a time-varying exposure. For comparative purposes, we also present the results of standard analyses—those in which session length was considered as of baseline levels, and those in which it was considered as a time-varying exposure using standard methods of statistical adjustment—to emphasize the potential advantages of marginal structural analysis as a means of more accurately quantifying associations when exposures are titrated over time.

RESULTS

The source cohort contained data on 10,044 patients who initiated incenter maintenance hemodialysis at any Fresenius Medical Care North America facility between June 2004 and August 2005. After excluding minors (n = 14), patients who did not survive until the start of at-risk time (dialysis day 31; n = 603) and those with insufficient covariate data for analysis (n = 875), 8552 subjects qualified for inclusion in the study. The mean age was 62.3±15.4 years, 54.8% were male and 40.5% were of nonwhite race (Table 1). Excluded patients were significantly more likely than study participants to be older, nonwhite, dialyze through a catheter, have lower blood pressure, serum albumin, and creatinine, be prescribed shorter session lengths, and less likely to have congestive heart failure (CHF) (Table 1).

Table 1
Baseline description of study participants and excluded patients

The distribution of baseline session length (that is, the last observed over dialysis days 0–30) is shown in Figure 1a. Given the discrete pattern observed, session length was categorized as shorter (<240 min) or longer (≥240 min) in primary analyses; secondary analyses considered session length categorized as ≤180, 181–239, or ≥240 min and as a continuous variable. Representative changes in session length during at-risk time are depicted in Figure 1b. For clarity, the figure shows transitions occurring between months 0, 3, and 6; analytically, session length was updated with each new order (time-updated analyses) or at monthly intervals (marginal structural analyses). Overall, 1698 subjects (19.9%) transitioned at least once between shorter (<240 min) and longer (≥240 min) session length; 2478 (29.0%) transitioned at least once among ≤180, 181–239, and ≥240 min session length strata.

Figure 1
Observed hemodialysis session lengths among study cohort

At-risk time began on dialysis day 31 and continued until each subject died, was censored, or completed 1 year on dialysis (the last day of complete follow-up). Censoring criteria included transfer of care (n = 851), recovery of kidney function (n = 236), withdrawal from dialysis (n = 336), and receipt of a kidney transplant (n = 213). Overall 8552 subjects contributed 78,798 patient-months of at-risk time, during which 1047 died; 5869 subjects remained alive and on dialysis as of the end of study.

Baseline and Time-updated Analyses

In baseline analyses, exposure and covariates were defined as the last value observed during the baseline exposure period (dialysis days 0–30), and used to predict survival in the subsequent at-risk period. Over the baseline exposure period, last observed session length was highly correlated with mean session length (r = 0.94), suggesting that classification of exposure in terms of the former was reflective of the cumulative exposure to dialysis during this interval. Shorter session length (<240 min) was observed for 54.7% of subjects.

On unadjusted analysis, shorter session length was significantly associated with higher mortality: HR (95% CI) 1.24 (1.10–1.41; P = 0.001) (Figure 2). On multivariable adjustment, the association between shorter session length and mortality was attenuated and no longer statistically significant: HR (95% CI) 0.89 (0.78–1.02; P = 0.09) (Figure 2).

Figure 2
Hazard ratios (95% CIs) for all-cause mortality for shorter session length (<240 min) in baseline, time-updated, and marginal structural analyses

As session length is not static, but rather, typically titrated in response to clinical circumstances, we fit time-updated proportional hazards models to estimate the association between session length and mortality. In these models, we updated session length with each new order, along with blood pressure, eKt/V, presence or absence of CHF, serum albumin, and creatinine (as described in Materials and Methods). Shorter session length (<240 min) was observed in 52.9% of all patient-time.

On unadjusted analysis, shorter session length was significantly associated with higher mortality: HR (95% CI) 1.33 (1.18–1.51; P<0.001) (Figure 2). On multivariable adjustment, the association between shorter session length and mortality was attenuated and no longer statistically significant: HR (95% CI) 1.00 (0.87–1.14; P>0.9) (Figure 2).

Marginal Structural Analysis

To better account for the influence of time-dependent confounding on the session length–mortality association, we applied marginal structural analyses. In these analyses, session length and time-varying covariates were updated at each month of study. Overall, shorter session length (<240 min) was observed in 52.4% of all patient-months.

In the unweighted (that is, unadjusted) model, there was a significant association between shorter session length and death similar to that seen in the unadjusted time-updated model: HR (95% CI) 1.38 (1.21–1.57; P<0.001) (Figure 2). However, unlike with baseline and time-updated models, on multivariable adjustment (through application of stabilized weights), the association between shorter session length and mortality was not attenuated, and remained highly statistically significant: HR (95% CI) 1.42 (1.24–1.62; P<0.001) (Figure 2).

To examine the robustness of this finding, we examined the association between shorter session length and mortality in prespecified subgroups of study patients using marginal structural models analogous to those above. In each model, weights were reestimated in the restricted population using models otherwise analogous to those from the full population. Adjusted estimates of association between shorter session length and mortality were qualitatively similar in all 12 subgroups, and reached conventional levels of statistical significance in 10 of 12 (Figure 3).

Figure 3
Adjusted hazard ratios (95% CIs) for all-cause mortality for shorter session length (<240 min) in prespecified subgroups

To assess for a potential dose response, we fit models in which session length was categorized as ≤180, 181–239, or ≥240 min (observed in 12.2, 40.2, and 47.6% of patient-months, respectively). In both the unadjusted (unweighted) and multivariable adjusted (weighted) models, incremental decrements in session length corresponded to incrementally higher risks of mortality; all values were statistically significant relative to the referent group receiving session length ≥240 min (Figure 4). In addition, we fit a model in which session length was considered as a continuous variable. On marginal structural adjustment, each 15 min decrement in session length was associated with a hazard ratio (95% CI) for all-cause mortality of 1.12 (1.08–1.16; P<0.001).

Figure 4
Association between incremental decrements in session length and all-cause mortality

DISCUSSION

In this study, we sought to determine whether hemodialysis session length was associated with mortality under a thrice-weekly paradigm. The primary finding was that shorter session length was associated with higher mortality. Our findings differed depending on the analytic strategy adopted; using marginal structural modeling, we were better able to adjust for time-dependent confounding on the basis of eKt/V and markers of nutritional (serum albumin and creatinine), hemodynamic status (blood pressure and CHF), and case-mix. Findings seemed robust on sensitivity analyses in which prespecified subgroups of patients were considered.

Numerous observational studies have examined the relation between hemodialysis session length and mortality, with conflicting results. In a sample of prevalent hemodialysis patients from Australia and New Zealand, Marshall et al.8 showed that shorter session length (therein defined as <3.5 h) was associated with significantly higher mortality (HR 1.75; 95% CI 1.27–2.40) compared with a referent group receiving 4–4.4 h treatments. Similarly, in a multinational sample of prevalent hemodialysis patients from the Dialysis Outcomes and Practice Patterns Study (DOPPS), Saran et al.5 showed that session lengths <211 and 211–240 min were associated with significantly higher mortality (HR 1.34, P<0.001 and 1.19, P = 0.01, respectively) compared with a referent group receiving >240-min treatments. In a national sample of prevalent Fresenius Medical Care patients from 2002, Lowrie et al.7 showed a near linear association between shorter session length and a greater log-hazard for all-cause mortality. Conversely, several studies failed to show a session length and mortality,13,14 however these studies were based on data from the late 1980's and early 1990's—an era when shorter treatment times, non-volumetric ultrafiltration and acetate-based buffers were the norm—and their applicability to contemporary hemodialysis patients remains uncertain.

Our study differed from previous studies in several important ways. First, unlike previous studies, which considered session length at baseline only, our study examined session length as a time-varying exposure, which minimizes the potential for exposure misclassification bias stemming from manipulation of session length, as is commonplace in clinical practice. Second, our study considered a cohort of strictly incident hemodialysis patients, thereby minimizing the potential for survivor bias (that is, depletion of susceptibles), and informing decisions regarding session length at a time when dialysis-associated mortality is highest and consideration of session length most acute. Most importantly, we used marginal structural analysis to appropriately adjust for the influence of time-dependent confounding. Time-dependent confounders are variables that (like typical confounders) are associated with future exposure and outcome, but (unlike typical confounders) are influenced by previous exposure. It has been proven that in the presence of time-dependent confounding, conventional statistical methods yield biased (that is, incorrect) estimates of causal association.11,12 To understand this, consider the example of eKt/V. Measured eKt/V in month 2 will influence session length in month 3, and will also be associated with future survival. In this sense, eKt/V functions as a typical confounder, and necessitates statistical adjustment lest estimates be biased. However eKt/V in month 2 will be influenced by session length in month 1, and this pathway (session length→eKt/V→mortality) is one means by which session length might influence mortality. In this sense, eKt/V functions as a causal pathway intermediate, and any statistical adjustment for it will lead to biased (toward the null) estimates of the session length–mortality association. Conventional statistical methods cannot selectively adjust for the confounding effects of eKt/V without simultaneously (and inappropriately) adjusting for its pathway-intermediate effects, thus ensuring that results will be biased. Though presented in the context of eKt/V, these considerations are true for time-dependent confounders in general.

Marginal structural modeling is one statistical technique capable of adjusting for confounding by time-dependent confounders without inappropriately adjusting for their pathway-intermediate effects;11,12 its use in this regard has become increasingly common in the nephrology literature.1518 Under the untestable assumption that all time-dependent confounders are included and appropriately specified in the statistical model (an assumption no different from that implicit in standard multivariable adjustment), marginal structural models render estimates that approximate randomized controlled trials.11 In these analyses, use of standard proportional hazards analyses suggested no association between shorter session length and mortality, which likely resulted from the inappropriate ‘adjusting away’ of the effects of session length on mortality mediated through improvements in nutritional status, small molecule clearance and hemodynamics. Conversely, on marginal structural analysis, it seemed that shorter session length was potently associated with higher mortality.

Ultimately, the best strategy to eliminate the influence of time-dependent confounding is through randomization of subjects to fixed treatment protocols. To date, only two large randomized trials of the effect of dialysis dose on patient outcomes have been conducted. The National Cooperative Dialysis Study randomized patients in a 2 × 2 factorial design to high or low blood urea nitrogen levels and shorter or longer session lengths.19 Findings suggested a near-significant association between shorter session lengths and increased morbidity (P = 0.056), but no effect on mortality. However, the National Cooperative Dialysis Study was underpowered to examine mortality as an endpoint, in part due to early termination at the time of interim analysis. The Mortality and Morbidity in Hemodialysis (HEMO) study randomized subjects to high or standard equilibrated Kt/Vurea (and high versus low flux dialysis membranes), and found no association between either intervention and mortality.20 It is important to note that subjects in the HEMO study were not randomized to various levels of session length per se (rather to levels of Kt/Vurea, which can be manipulated through multiple means); thus the study does not constitute a true randomized trial of session length.21 As such, conclusions with respect to the influence specifically of session length are subject to some of the same limitations (for example, bias, confounding) as would accompany an observational study.

The precise mechanism through which session length might influence mortality is not clear. It has been shown that longer session lengths result in improved removal of urea, creatinine, phosphorous, and beta-2 microglobulin even when Kt/Vurea is held constant.6 Therefore, it is possible that some of the true effects of longer session length stem from enhanced solute clearance beyond what is reflected by Kt/Vurea. In addition, longer session length enables ultrafiltration to proceed at a slower rate, which may improve blood pressure control and facilitate more consistent attainment of dry weight.22,23 These effects may, in turn, attenuate left ventricular hypertrophy, fibrosis, and diastolic dysfunction, all of which are associated with mortality in patients on maintenance dialysis.2427 Some have argued that, given the inherent inter-relatedness of session length and Kt/V, that the association between session length and mortality may stem from inappropriately low Kt/V among patients receiving shorter session lengths.28 Nonetheless, that the association between session length and mortality remained potent and statistically significant among the subgroup of patients with eKt/V>1.2 (Figure 3) is evidence that session length may influence mortality even when eKt/V conforms to the current standard of care.2

Several limitations of this study bear mention. Most notably, our study was designed to measure the association between session length and mortality within the context of conventional thrice-weekly, incenter hemodialysis. Specifically, we did not include patients receiving short-daily or extended dialysis (incenter/at home, diurnal/nocturnal), and therefore, we cannot determine the associations of session length or changes in session length among patients on more frequent or more intensive hemodialysis schedules. Although our findings support the hypothesis that more dialysis might reduce mortality and/or morbidity, definitive conclusions must await the results of dedicated clinical trials, such as the ongoing Frequent Hemodialysis Network trials.29 Studies such as ours cannot replace clinical trials (which may fall short in power and generalizability) but can and should contribute to the evidence based on which clinical decisions are made in practice.

As with any observational study, our study is subject to the potential influence of residual confounding. To minimize this concern, we adjusted estimates for many factors that may confound the association between session length and mortality such as demographics, anthropometrics, and indices of small molecule clearance, hemodynamics and nutrition. Of note, we lacked data on interdialytic weight gain, which may influence decisions on session length and may also be directly related to mortality and morbidity. However, inasmuch as patients with larger interdialytic weight gains are more likely to have longer session lengths and are more likely to die,3033 any bias would be toward the null and would not explain our findings. In addition, we tracked prescribed (not delivered) session length. This approach, though necessitated by the available data, is consistent with intention-to-treat principles, and residual confounding on this basis would be expected to bias estimates toward the null assuming either that early termination of dialysis sessions was unrelated to prescribed duration, or (as might be more likely) that patients prescribed longer session lengths were more likely to terminate prematurely. Again, this would not explain the observed association between shorter session length and mortality, and may have rendered our findings conservative. Finally, exclusion of some patients may limit generalizability of findings to certain patient groups, most notably those who have extremely limited survival on dialysis. Nonetheless, given the modest (30 day) survival requirement, and our large and diverse study population, we believe that our results do generalize to the vast majority of chronic dialysis patients for whom consideration of session length is of issue.

In conclusion, when applied to data from this large cohort of incident hemodialysis patients, marginal structural analysis suggests that shorter prescribed hemodialysis session length is potently and independently associated with higher all-cause mortality at 1 year. Differences in findings between marginal structural and proportional hazards analyses may point to the ability of the former to appropriately adjust for time-dependent confounding. Further studies are needed to confirm and extend these findings.

MATERIALS AND METHODS

Study Population

This protocol was deemed exempt by the Partners Health Care Institutional Review Board. We conducted a nonconcurrent cohort study of patients in the ArMORR cohort. Details of the source cohort have been previously published.15,34 Briefly, the Accelerated Mortality on Renal Replacement (ArMORR) cohort contains data on 10,044 patients who initiated incenter maintenance hemodialysis at any Fresenius Medical Care Unit in North America (FMC-NA) between June 2004 and August 2005. Of these, we excluded minors (age <18 years), patients who did not survive until the start of at-risk time (dialysis day 31), and those with insufficient covariate data for analysis.

Study Data

All study data were collected prospectively. Demographic and comorbid disease characteristics were recorded as of dialysis initiation. Time-updated data were available for session length (updated with each new order), seated pre-dialysis blood pressure (updated at each treatment), and laboratory data (updated with each measurement, typically at intervals of 2–4 weeks). All serum and urine tests were performed in a centralized laboratory (Spectra Laboratories, Rockleigh, NJ, USA). Equilibrated Kt/V was calculated by formal urea kinetic modeling. Presence of pre-existing CHF was assessed at dialysis initiation, and patients’ status were updated (where necessary) at the time of attributed hospitalization (International Classification of Disease-9 codes 276.6, 402.x1, 404.01, 404.11, 404.91, 404.03, 404.13, 404.93, 425.xx, 428.xx); all hospitalization data were rigorously collected and reconciled by dialysis unit personnel as previously described.15

Statistical Analysis

All analyses were conducted using STATA 10.0MP (College Station, TX, USA). Continuous variables were examined graphically, described in terms of their means and standard deviations or medians and inter-quartile ranges, and compared across groups using Student's t-test or the Wilcoxon rank–sum test as appropriate. Categorical variables were described by frequency distribution and compared across groups using contingency tables and χ2 testing.

Unadjusted associations between predictors and outcome were explored using Kaplan Meier methods, log rank testing, and unadjusted proportional hazards models. Adjusted associations between predictors and outcome were estimated by fitting multivariable proportional hazards models. Covariates were selected on the basis of biological plausibility and clinical precedent; all were forced into multivariable models. For continuous covariates the linearity assumption was tested graphically (by Martingale residuals) and by comparison of model fit (by Akaike's Information Criterion) between non-nested models in which the covariate was specified as a linear and categorical term. Those variables demonstrating gross violation of the linearity assumption were entered as categorical terms in the final multivariable models. The overall covariate vector in the primary analysis consisted of the following: age, sex, race (white, nonwhite), vascular access (fistula, graft, catheter), body mass index (≤20, 20–25, 25–30, >30 kg/m2), CHF, eKt/V (≤1.0, 1.0–1.2, 1.2–1.4, 1.4–1.6, >1.6), systolic blood pressure, serum albumin, and creatinine. Results of all analyses were nearly identical when body mass index and eKt/V were specified as linear terms, when diastolic blood pressure was substituted for systolic (strong colinearity precluded inclusion of both simultaneously; r = 0.72), and when height and weight were substituted for body mass index (data not shown).

We estimated the parameters of a marginal structural proportional hazards model using a pooled logistic regression model as previously described.11,12 In these analyses, each patient's experience was decomposed into individual patient-months. For each patient-month, the probability of survival was estimated as a function of the patient's session length measured at month's start, and the month of dialysis (to allow for differential baseline probability of survival according to time on dialysis); robust variance estimates were used to account for nonindependence of observations within patient.

Marginal structural models were adjusted by application of stabilized weights based on the probability that a patient had his/her observed session length and on the probability of censoring.11,12 Stabilized probability of exposure weights were defined as the ratio of two terms: the numerator the probability that each subject had his/her observed session length based on previous session length and baseline covariates; the denominator the probability that each subject has his/her observed session length based on previous session length, and covariates considered both at baseline and as of the previous month. In the primary marginal structural analyses, each was estimated using a separate pooled logistic regression model in which session length was categorized as <240 or ≥240 min. Secondary analyses considered session length as ≤180, 181–239 or ≥240 min; multinomial logistic regression models were used to estimate the probability of exposure. Stabilized probability of censoring weights was estimated in a completely analogous manner except that no term was included for the previous month's session length. Fuller detail regarding specification of marginal structural models is given in online supplemental materials.

Supplementary Material

Suppl Methods

ACKNOWLEDGMENTS

This work was presented in abstract form at the American Society of Nephrology Annual Meeting 2009, San Diego, CA, USA.

Footnotes

DISCLOSURE

This work was supported by awards from the NIH/NIDDK (DK079056 to SMB, DK84974 to RT). SMB has received consulting fees from C.B. Fleet Company; his spouse is employed by Genzyme. GMC serves on the Scientific Advisory Board for DaVita Clinical Research, receives research support from Amgen and has received research support and honoraria from Genzyme.

SUPPLEMENTARY MATERIAL

Supplementary material is linked to the online version of the paper at http://www.nature.com/ki

REFERENCES

1. Gotch FA, Sargent JA. A mechanistic analysis of the National Cooperative Dialysis Study (NCDS). Kidney Int. 1985;28:526–534. [PubMed]
2. Hemodialysis Adequacy 2006 Work Group Clinical practice guidelines for hemodialysis adequacy, update 2006. Am J Kidney Dis. 2006;48(Suppl 1):S2–S90. [PubMed]
3. Lowrie EG, Chertow GM, Lew NL, et al. The urea [clearance x dialysis time] product (Kt) as an outcome-based measure of hemodialysis dose. Kidney Int. 1999;56:729–737. [PubMed]
4. Lowrie EG. Prescribing and monitoring hemodialysis dose. Kidney Int. 2008;74:262–264. [PubMed]
5. Saran R, Bragg-Gresham JL, Levin NW, et al. Longer treatment time and slower ultrafiltration in hemodialysis: associations with reduced mortality in the DOPPS. Kidney Int. 2006;69:1222–1228. [PubMed]
6. Eloot S, Van Biesen W, Dhondt A, et al. Impact of hemodialysis duration on the removal of uremic retention solutes. Kidney Int. 2008;73:765–770. [PubMed]
7. Lowrie EG, Li Z, Ofsthun N, et al. Measurement of dialyzer clearance, dialysis time, and body size: death risk relationships among patients. Kidney Int. 2004;66:2077–2084. [PubMed]
8. Marshall MR, Byrne BG, Kerr PG, et al. Associations of hemodialysis dose and session length with mortality risk in Australian and New Zealand patients. Kidney Int. 2006;69:1229–1236. [PubMed]
9. Robins JM. Correction for non-compliance in equivalence trials. Stat Med. 1998;17:269–302. discussion 387–269. [PubMed]
10. Robins JM, Blevins D, Ritter G, et al. G-estimation of the effect of prophylaxis therapy for Pneumocystis carinii pneumonia on the survival of AIDS patients. Epidemiology. 1992;3:319–336. [PubMed]
11. Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11:550–560. [PubMed]
12. Hernan MA, Brumback B, Robins JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology. 2000;11:561–570. [PubMed]
13. Owen Jr WF, Lew NL, Liu Y, et al. The urea reduction ratio and serum albumin concentration as predictors of mortality in patients undergoing hemodialysis. N Engl J Med. 1993;329:1001–1006. [PubMed]
14. Held PJ, Port FK, Wolfe RA, et al. The dose of hemodialysis and patient mortality. Kidney Int. 1996;50:550–556. [PubMed]
15. Teng M, Wolf M, Ofsthun MN, et al. Activated injectable vitamin D and hemodialysis survival: a historical cohort study. J Am Soc Nephrol. 2005;16:1115–1125. [PubMed]
16. Feldman HI, Joffe M, Robinson B, et al. Administration of parenteral iron and mortality among hemodialysis patients. J Am Soc Nephrol. 2004;15:1623–1632. [PubMed]
17. Weinhandl ED, Rao M, Gilbertson DT, et al. Protective effect of intravenous levocarnitine on subsequent-month hospitalization among prevalent hemodialysis patients, 1998 to 2003. Am J Kidney Dis. 2007;50:803–812. [PubMed]
18. Brunelli SM, Joffe MM, Israni RK, et al. History-adjusted marginal structural analysis of the association between hemoglobin variability and mortality among chronic hemodialysis patients. Clin J Am Soc Nephrol. 2008;3:777–782. [PubMed]
19. Lowrie EG, Laird NM, Parker TF, et al. Effect of the hemodialysis prescription of patient morbidity: report from the National Cooperative Dialysis Study. N Engl J Med. 1981;305:1176–1181. [PubMed]
20. Eknoyan G, Beck GJ, Cheung AK, et al. Effect of dialysis dose and membrane flux in maintenance hemodialysis. N Engl J Med. 2002;347:2010–2019. [PubMed]
21. Depner T, Daugirdas J, Greene T, et al. Dialysis dose and the effect of gender and body size on outcome in the HEMO Study. Kidney Int. 2004;65:1386–1394. [PubMed]
22. Charra B, Calemard M, Laurent G. Importance of treatment time and blood pressure control in achieving long-term survival on dialysis. Am J Nephrol. 1996;16:35–44. [PubMed]
23. McGregor DO, Buttimore AL, Nicholls MG, et al. Ambulatory blood pressure monitoring in patients receiving long, slow home haemodialysis. Nephrol Dial Transplant. 1999;14:2676–2679. [PubMed]
24. Zoccali C, Benedetto FA, Mallamaci F, et al. Left ventricular mass monitoring in the follow-up of dialysis patients: prognostic value of left ventricular hypertrophy progression. Kidney Int. 2004;65:1492–1498. [PubMed]
25. Krane V, Heinrich F, Meesmann M, et al. Electrocardiography and outcome in patients with diabetes mellitus on maintenance hemodialysis. Clin J Am Soc Nephrol. 2009;4:394–400. [PubMed]
26. Ahmed A, Rich MW, Sanders PW, et al. Chronic kidney disease associated mortality in diastolic versus systolic heart failure: a propensity matched study. Am J Cardiol. 2007;99:393–398. [PMC free article] [PubMed]
27. Aoki J, Ikari Y, Nakajima H, et al. Clinical and pathologic characteristics of dilated cardiomyopathy in hemodialysis patients. Kidney Int. 2005;67:333–340. [PubMed]
28. Gotch FA, Yarian S, Keen M. A kinetic survey of US hemodialysis prescriptions. Am J Kidney Dis. 1990;15:511–515. [PubMed]
29. Suri RS, Garg AX, Chertow GM, et al. Frequent Hemodialysis Network (FHN) randomized trials: study design. Kidney Int. 2007;71:349–359. [PubMed]
30. Stegmayr BG, Brannstrom M, Bucht S, et al. Minimized weight gain between hemodialysis contributes to a reduced risk of death. Int J Artif Organs. 2006;29:675–680. [PubMed]
31. Kalantar-Zadeh K, Regidor DL, Kovesdy CP, et al. Fluid retention is associated with cardiovascular mortality in patients undergoing long-term hemodialysis. Circulation. 2009;119:671–679. [PMC free article] [PubMed]
32. Holmberg B, Stegmayr BG. Cardiovascular conditions in hemodialysis patients may be worsened by extensive interdialytic weight gain. Hemodial Int. 2009;13:27–31. [PubMed]
33. Movilli E, Gaggia P, Zubani R, et al. Association between high ultrafiltration rates and mortality in uraemic patients on regular haemodialysis. A 5-year prospective observational multicentre study. Nephrol Dial Transplant. 2007;22:3547–3552. [PubMed]
34. Thadhani R, Tonelli M. Cohort studies: marching forward. Clin J Am Soc Nephrol. 2006;1:1117–1123. [PubMed]