Greater fluid overload and lower interdialytic weight gain are independently associated with mortality in a large international hemodialysis population

Abstract

Background: Fluid overload and interdialytic weight gain (IDWG) are discrete components of the dynamic fluid balance in haemodialysis patients. We aimed to disentangle their relationship, and the prognostic importance of two clinically distinct, bioimpedance spectroscopy (BIS)-derived measures, pre-dialysis and post-dialysis fluid overload (FOpre and FOpost) versus IDWG.

Methods: We conducted a retrospective cohort study on 38 614 incident patients with one or more BIS measurement within 90 days of haemodialysis initiation (1 October 2010 through 28 February 2015). We used fractional polynomial regression to determine the association pattern between FOpre, FOpost and IDWG, and multivariate adjusted Cox models with FO and/or IDWG as longitudinal and time-varying predictors to determine all-cause mortality risk.

Results: In analyses using 1-month averages, patients in quartiles 3 and 4 (Q3 and Q4) of FO had an incrementally higher adjusted mortality risk compared with reference Q2, and patients in Q1 of IDWG had higher adjusted mortality compared with Q2. The highest adjusted mortality risk was observed for patients in Q4 of FOpre combined with Q1 of IDWG [hazard ratio (HR) = 2.66 (95% confidence interval 2.21-3.20), compared with FOpre-Q2/IDWG-Q2 (reference)]. Using longitudinal means of FO and IDWG only slightly altered all HRs. IDWG associated positively with FOpre, but negatively with FOpost, suggesting a link with post-dialysis extracellular volume depletion.

Conclusions: FOpre and FOpost were consistently positive risk factors for mortality. Low IDWG was associated with short-term mortality, suggesting perhaps an effect of protein-energy wasting. FOpost reflected the volume status without IDWG, which implies that this fluid marker is clinically most intuitive and may be best suited to guide volume management in haemodialysis patients.

FIGURE 1

Measurements of FO and IDWG.

FIGURE 2

Derivation of the study cohort.

FIGURE 3

Distribution of exposure measures and their mortality effect estimates, by quartile. (A–C) Density estimates of the distribution of time-varying measurements of FO and IDWG calculated by Epanechnikov Kernel functions (719 497 repeated measurements in N = 38 614 patients). Black lines refer to the quartile cut-points. Note that IDWG is expressed in percentage of body weight, whereas FO is expressed in percentage of ECV. For absolute and relative values by quartile (means over all measurements), refer to Table 2. (D–F) Forest plots depicting estimates for the effect of IDWG, FOpre and FOpost on all-cause mortality (see also Supplementary data, Table S4). Estimates were obtained using different prognostic models that included the FO markers and IDWG as time-varying predictor variables, aggregated as 1-month or 12-month moving averages or as fixed, patient-averaged predictor variables, the latter being entitled average long-term exposure (Cox regression analyses 1A and 1B). HRs are relative hazards to the second quartile. The effect estimates of time-varying exposure was adjusted for age, sex, BMI, albumin, phosphate, haemoglobin, HDL cholesterol, triglycerides, protein catabolic rate, Kt/V, haemodialysis treatment modality, vascular access, diabetes, congestive heart failure, coronary artery disease, tumour, dementia, diuretics and treatment of hypertension. The effect estimates were additionally adjusted for IDWG (or FO, respectively); these data are shown in the lower half of each panel.

FIGURE 4

Association pattern of exposure measures. Predicted fitted marginal means of time-varying IDWG as a function of time-varying FOpre or FOpost obtained from a fractional polynomial regression model. Note that IDWG is expressed in percentage of body weight, whereas FO is expressed in percentage of ECV.

FIGURE 5

Association between IDWG and mortality, by quartile of FO. (A, B) Cubic spline models estimating multivariate adjusted HRs relative to the IDWG concentration in the second FO quartile (IDWG = 0.06). The knots were placed using Harrell’s recommended percentiles [35]. FO quartiles were calculated based on longitudinal means over all measures of a patient during the follow-up. HRs were adjusted for age, sex, BMI, albumin, phosphate, haemoglobin, HDL cholesterol, triglycerides, protein catabolic rate, Kt/V, haemodialysis treatment modality, vascular access, diabetes, congestive heart failure, coronary artery disease, tumour, dementia, diuretics and treatment of hypertension. HRs for IDWG in FO Q1, Q3 and Q4 were multiplied by the marginal HRs of the FO quartile relative to the second FO quartile (Cox regression analysis 4). (C, D) Estimates for the combined effect of IDWG and FOpre/FOpost, obtained from Cox regression analysis 2B, which included a combination dummy variable, considering the joint bivariate exposure of time-varying IDWG and FOpre (A) and FOpost (B), aggregated as 1-month or 12-month moving averages (see also Supplementary data, Table S5). HRs were calculated relative to the subgroup with measurements in the second quartile for both FOpost and IDWG. The effect estimates were adjusted for age, sex, BMI, albumin, phosphate, haemoglobin, HDL cholesterol, triglycerides, protein catabolic rate, Kt/V, haemodialysis treatment modality, vascular access, diabetes, congestive heart failure, coronary artery disease, tumour, dementia, diuretics and treatment of hypertension.