Predictors and prognostic significance of persistent fluid overload: A longitudinal study in Chinese peritoneal dialysis patients

Background: Cross-sectional studies showed that fluid overload (FO) measured by bioimpedance spectroscopy (BIS) predicted adverse outcomes in patients on peritoneal dialysis (PD). We aimed to describe the longitudinal change in volume status in Chinese PD patients and determine its relation with clinical outcomes. Methods: We performed a single-centre, retrospective analysis of all PD patients who underwent repeated BIS from 2010 to 2015. FO was defined by relative hydration index (RHI; volume of overhydration adjusted by extracellular water >7%). Variability of volume status (VVS) was denoted by the standard deviation of all RHI. The association of time-averaged RHI and VVS on patient and technique survival was explored by a competing risk model. Results: A total of 269 patients were followed for a median of 47.1 months. Mean time-averaged RHI was 17.6 ± 10.2%. Multivariable mixed linear regression revealed that RHI was significantly associated with diabetes, time-varying systolic blood pressure, and inversely with time-varying albumin level, lean tissue index and fat tissue index (p <0.0001 for all). Time-averaged RHI independently predicted patient survival (subdistribution hazard ratio (SHR) 1.05, 95% CI 1.03–1.07, p <0.0001) and technique survival (SHR 1.04, 95% CI 1.02–1.06, p <0.0001), whereas VVS did not. The mortality risk for patients with persistent FO was consistently higher than the corresponding risk estimated from baseline FO of the same extent. Conclusions: Persistent FO was a strong predictor of patient and technique failure. Repeated bioimpedance measurements to monitor volume status may provide additional prognostic information in PD patients. Graphical Abstract This is a visual representation of the abstract.


Introduction
Maintaining euvolemia is one of the most important treatment goals for patients on peritoneal dialysis (PD), who have exceedingly high burden of cardiovascular diseases (CVD) [1,2]. However, traditional clinical assessments are often unreliable and subjective, especially in detecting occult uid overload (FO).
In contrast, bioimpedance study enables quantitative measurement of body composition which are noninvasive and highly reproducible [3,4]. Published evidences suggested that FO de ned by bioimpedance methods predicted patient survival, technique survival and CVD in PD patients [5][6][7][8][9]. However, some of these studies only assessed hydration status at baseline [5][6][7], while others with repeated bioimpedance measurement were limited by small sample size or relatively short duration of follow up [8,9].
On the other hand, the relationship between variation of volume status and clinical outcomes was seldom studied. A retrospective study showed that higher standard deviation (SD) of extracellular water/intracellular water (E/I ratio), which indicated more uctuation in hydration status, was associated with mortality and technique failure [10]. However, the association became insigni cant after adjustment of nutrition and in ammation. Recently, the Initiative for Patient Outcomes in Dialysis-Peritoneal Dialysis (IPOD-PD) study reported that the volume status of incident PD patients tended to stabilize over time; and baseline clinical parameters and PD prescription did not predict change in volume status over rst 6 months [11]. Repeated measurements of volume status over a longer period of follow up may provide additional insight on the variability of volume status and potential modi able factors on the course of hydration status.
In the present study, we aimed at describing the longitudinal changes in volume status of Chinese PD patients, and to explore predictive factors of volume status. Second, we evaluated the prognostic value of repeated bioimpedance measurements compared with single baseline measurement. Third, we examined the association between variability in volume status and clinical outcomes.

Study design
This study was approved by the Joint Chinese University Hong Kong-New Territories East Cluster Clinical Research Ethics Committee. All studies procedures are in compliance with the Declaration of Helsinki. We retrospectively reviewed all PD patients who were followed up in a tertiary hospital from January 2010 to December 2015. Patients with a minimum of two body composition measurements were included into the study. Both incident (≤ 4 weeks after initiation of PD) and prevalent (> 4 weeks after PD) patients were eligible. Patients who had history of metallic prosthesis or pacemaker implantation were contraindicated for bioimpedance study and thus were excluded.

Data collection
Baseline demographics and PD prescriptions were retrieved by reviewing medical records of patients.
Charlson's Comorbidity Index (CCI) was calculated to re ect the burden of comorbidities [12]. Systolic blood pressure (SBP), diastolic blood pressure (DBP) and body weight (BW) were measured during baseline and at subsequent bioimpedance studies, which were performed approximately every 12 months during clinic follow up. Laboratory data within two months of bioimpedance study were recorded for analysis. Patients with baseline urine output <200ml were considered as anuric.
We determined the volume status of PD patients by using a validated multi-frequency bioimpedance spectroscopy device (Body Composition Monitor [BCM], Fresenius Medical Care, Germany) [13]. In essence, BCM measures impedance by passing electric current with 50 different frequencies via two electrodes attached on the patients in a spine position. Based on a three-compartment model which assumes normally hydrated adipose and lean tissues, excessive uid can be estimated and expressed in absolute liters of volume of overhydration (OH) [14]. We also calculated the relative percentage of excessive uid by dividing OH by extracellular water (ECW), which was known as relative hydration index (RHI) [15].
Patients were considered to have no, mild, moderate and severe FO if their RHI was £7%, >7.0 to 15.0%, >15.0 to 25.0% and >25.0%, respectively [15,16]. Volume status of individual patient was represented by baseline RHI and time-averaged RHI (de ned as the weighted mean of all RHI of each patient, with weights representing the time elapsed since the previous measurement). On the other hand, variability of volume status (VVS) was denoted by standard deviation (SD) and of all RHI.
In addition to volume status, nutritional parameters including fat tissue index (FTI) and lean tissue index (LTI) were automatically computed by BCM. Clinicians were allowed to adjust PD prescriptions at their discretion based on BCM data.

Follow up and outcomes
All patients were followed until 31 st July 2017. Outcomes included patient survival and technique survival, which was de ned as a composite of death or transfer to long term hemodialysis (HD).

Statistical analyses
Continuous data are presented as mean ± SD or median (inter-quartile range [IQR]). Baseline characteristics of patients with different degree of FO were compared by one way analysis of variance (ANOVA) or Chi-square test where appropriate. Then we constructed a linear mixed model to present the change of RHI from baseline to the fourth year of follow up (BCM measurements in Year 5 were excluded due to limited number of data). Subjects were included as random intercept, while time of BCM measurements (number of years after baseline assessment) were included as xed effect. Baseline FO (RHI ≤7% vs. >7%), residual urine output (anuric vs. non-anuric), baseline diabetes, baseline albumin level (albumin level ≤30 vs. >30g/L), gender, and the interaction of these factors with time were included as xed effects. Parameter(s) which achieved P ≤ 0.1 in univariate analysis were included as time-varying co-variate(s). An autoregressive covariance was assumed.
Kaplan-Meier method with log-rank test was used to compare survival curves between patients in different categories of FO de ned by time-averaged RHI. The association between RHI and patient and technique survival was then examined by Fine and Gray competing risk model [17], with adjustment for age, gender, CCI, body mass index, serum albumin, weekly Kt/V and residual urine output. Change to HD and transplantation were considered as competing risks of patient survival; and transplantation was considered as competing risk of technique survival. SD of OH/ECW, the surrogate of VVS, were natural log-transformed due to skewed distribution. Baseline RHI together with potential confounders measured at baseline (as stated above), and time-averaged RHI with the same time-averaged parameters were analyzed in two separate multivariate competing risk model. VVS was subsequently forced into the timeaveraged model. The difference in predictive power of each model was compared by Wald's test. In a sensitivity analysis, we repeated our survival analyses with patients with ≥3 RHI measurements.
All statistical analyses were performed by SPSS for Windows software (version 24.0. IBM Corporation, Armonk, NY) and Stata version 15 (StataCorp LP, College Station, Tx). A P value of less than 0.05 was considered signi cant. All probabilities were two-tailed.

Results
Bioimpedance studies were performed in 337 patients during the study period, in which 272 patients had at least two repeated measurements. 269 patients with complete body composition data were included in the nal cohort. At baseline, 158 (58.6%) was male and the mean age was 59.7 ± 11.5 years; 13.4% of them were on automated peritoneal dialysis (APD) ( Table 1). The overall prevalence of diabetes was 54.6%. First BCM study showed that 58 (21.6%) patients had mild FO, while 164 (61.0%) had moderate to severe FO . Patients with FO were older yet with a shorter dialysis vintage. They had higher SBP, higher prevalence of diabetes and heavier burden of comorbidities. Besides, FO was accompanied by a signi cantly lower albumin level, and a trend towards anemia. Nevertheless, there was no signi cant difference between weekly Kt/V, residual glomerular ltration rate (GFR), or urine output (Table 1 and S1).
Change in volume status over time A total of 699 bioimpedance studies were performed during the study period. The number of bioimpedance studies that a patient underwent ranged from 2 to 5. The median number of measurements were 2 (IQR 2-3); 40.5% of patients were studied 3 times or more. The median interval between consecutive BCM measurements was 12.1 (IQR 10.5-13.3) months.
In subsequent multivariate mixed linear regression analysis, we found that RHI was signi cantly associated with baseline diabetes, time-varying SBP, and inversely with time-varying albumin level, weekly Kt/V, LTI and FTI (Table 2). In contrast, baseline peritoneal transport status (P=0.89) and residual urine output (P=0.45) were not associated with change in volume status.
Among the pre-speci ed subgroups, we found that baseline hydration status (RHI >7% vs. ≤7%, P =0.001) and residual urine output (<200ml/day vs. ≥ 200ml/day, P =0.04) signi cantly modi ed the relationship between volume status and year of assessment (Table 3). Patients who had FO (RHI >7%) at the beginning of study had signi cantly higher RHI compared with the others, but the adjusted difference between two groups diminished with time ( Figure 2a). The adjusted mean difference in change in RHI between euvolemic versus FO patients was 3.2% per year (95% CI 1.5 to 4.9%, P =0.0002) ( Table 3). In addition, the adjusted difference in RHI between anuric versus non-anruic group was 1.4% per year (95% CI -0.9 to 3.7%) ( Figure 2B), which may suggest progressive uid accumulation in anuric patients, although the difference did not reach statistical signi cance. On the other hand, there was no interaction between time and other covariates (age, gender, diabetes and baseline albumin level  (Table S3).
To evaluate the prognostic value of VVS and repeated BCM measurement, we created three competing risk models which were adjusted for the same confounders as  [11]. Therefore it may not be surprising that initial improvement of volume status was largely observed in the latter two studies, in which the BCM ndings may prompt the clinicians for optimization of uid status. However, while volume status tended to stabilize with roughly 50% of patients rendered euvolemic in IPOD-PD study, our results indicated that over 70% of PD patients remained hypervolemic in subsequent follow up (Table S2). Instead of a 'regression-to-mean' phenomenon (patients with hypo-and hypervolemia both progressed to euvolemia) [11], our study may suggest a different trajectory of volume status which was signi cantly modi ed by baseline volume status (Figure 2A , was important and this may also need to be considered in patients who were initially euvolemic. Similar to the ndings of previous cross-sectional studies [7,15,19], our study ( Table 2) showed that diabetes (P <0.0001), hypoalbuminemia (P <0.0001), lower weekly Kt/V (P =0.006) were independent predictors of hypervolemia. Of note was that our longitudinal study with repeated measurements over a long duration of follow up provided more robust evidence to a rm their associations. On the other hand, residual urine volume and baseline peritoneal transporter status were not associated with volume status. This suggested that dietary compliance and appropriate adaption of dwell length to transporter status may play a greater role in maintaining normohydration. Interestingly, we found that time-varying FTI (P <0.0001) and LTI (P <0.0001) were inversely associated with volume status. This combination of low fat and lean body mass, and hypoalbuminemia constituted the phenotype of protein-energy wasting, which was previously shown to correlate with overhydration [19]. Systemic in ammation, which is prevalent among dialysis patients, may result in unnoticed reduction in FTI or LTI and inaccurate estimation of dry weight [20], culminating in uid retention. Reciprocally, FO may indirectly cause muscle wasting by aggravating in ammation via increased bacterial translocation through bowel wall [21,22].
Previous studies suggested that visit-to-visit blood pressure variability was associated with increase in mortality and cardiovascular events in HD patients [23,24]. This variability was, at least partially, attributed by the drastic and non-physiological change in extracellular volume during HD, which was followed by interdialytic uid accumulation. Likewise, aggressive uid control in overhydrated PD patients might lead to rapid uctuation of uid status, which causes depletion in intravascular volume and organ hypoperfusion. It had been reported that PD patients with greater uid variation had faster decline in GFR and urine output [25]. Another single-center Chinese study examined the association of SD of E/I ratio (as the proxy of magnitude of changes in volume status over time) and clinical outcomes [10].
However, the prognostic value of E/I ratio was confounded by nutritional state and C-reactive protein [10]. The predictive power of RHI (in the present study) was known to be independent of nutrition and in ammation [5]. Nevertheless, we were still unable to demonstrate any signi cant association between VVS and patient or technique survival. One of the possible reasons may be that relative long interval between BCM masked the underlying variability. The fact that approximately half of patients only underwent BCM twice may also fail to unravel VVS. However, sensitivity analysis which included ≥3 BCM measurements produced similar results.
In a multivariate competing risk model adjusted for age, gender, comorbidities, dialysis adequacy and nutrition (Table 4), every 1% increase in time-averaged RHI was signi cantly associated with 5% increase in mortality (SHR 1.05, 95% CI 1.03 to 1.08) and 4% increase in technique failure (SHR 1.04, 95% CI 1.01 to 1.06), respectively. When volume status was analyzed as a categorical variable, persistent moderate and persistent severe FO independently predicted a signi cant increase in mortality by 2.7 and 4.2 times, respectively. This risk was considerably higher compared to that associated with baseline FO within the same category (Table S3). This suggested that persistent FO was a stronger risk factor than FO based on cross-sectional measurement, which was a consistent nding from previous observational studies in both PD and HD patients [8,26]. This was particularly relevant because initial improvement in volume status was not uncommon (as in our cohort) when clinicians attempted to correct FO after knowing the rst BCM data, and that may attenuate the predictive power of baseline FO. While cumulative exposure to FO was proved to increase the risk of HD conversion in short to medium term [8,27], our study expanded the existing evidence that persistent FO remained a signi cant, and more importantly, modi able predictor for technique failure after a median follow up of 47 months. Although analysis using time-averaged RHI as categorical variable seemed to fall short of signi cance (Table S3), the strong association when timeaveraged RHI was analyzed as continuous variable may suggest there was no reliable or universal cut-off to de ne FO in predicting technique failure.
Our study had a number of limitations. First, the inherent limitation of a retrospective observational study did not allow us to establish causality. Nevertheless, the longitudinal design with repeated, objective assessment by BCM provided more robust estimate between volume status and associated factors compared with cross-sectional studies. Second, survival bias may be present because only survivors would undergo repeated BCM measurements. However, given that CVD was one of the major causes of death and was closed related with FO, such bias was more likely to underestimate the risk. Third, data on ultra ltration volume and salt intake were missing in many patients and thus not included for analyses. Nevertheless, ultra ltration volume had not been shown to predict hypervolemia in both incident and prevalent patients [7,15]; and there was no simple surrogate for dietary sodium intake in clinical practice. Besides, we did not evaluate the impact of change in PD modality nor prescription because the use of APD and icodextrin were, to some extent, governed by availability. This was further complicated by the fact that interventions to ameliorate FO were often complex and multiple, such that the effect of single intervention was di cult to analyze even in the setting of randomized controlled trial [28]. Finally, we could not exclude type II error concerning the effect of VVS on clinical outcomes, given the limited number of BCM measurements.
In conclusion, persistent FO was associated with increase in mortality and technique failure. Despite the additional prognostic value brought about by repeated bioimpedance measurements, there is unfortunately insu cient evidence that supports routine bioimpedance-guided uid management would improve clinical outcomes [29,30]. Future studies are warranted to identify the subgroup that will bene t most for bioimpedance-guided uid management.

Funding
This study was supported by the Chinese University of Hong Kong (CUHK) research accounts 6901031 and 7101215. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Con icts of interest
Prof. Philip Li received honorarium from Fibrogen. Other authors declared no con ict of interests.
Ethics approval

Data availability statement
The dataset generated and/or analyzed in this article will be shared upon reasonable request to the corresponding author.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. STROBEchecklistcohort.docx