Bioimpedance-defined overhydration predicts survival in end stage kidney failure (ESKF): systematic review and subgroup meta-analysis

Both overhydration and comorbidity predict mortality in end-stage kidney failure (ESKF) but it is not clear whether these are independent of one another. We undertook a systematic review of studies reporting outcomes in adult dialysis patients in which comorbidity and overhydration, quantified by whole body bioimpedance (BI), were reported. PubMed, EMBASE, PsychInfo and the Cochrane trial database were searched (1990–2017). Independent reviewers appraised studies including methodological quality (assessed using QUIPS). Primary outcome was mortality, with secondary outcomes including hospitalisation and cardiovascular events. Of 4028 citations identified, 46 matched inclusion criteria (42 cohorts; 60790 patients; 8187 deaths; 95% haemodialysis/5% peritoneal dialysis). BI measures included phase angle/BI vector (41%), overhydration index (39%) and extra:intracellular water ratio (20%). 38 of 42 cohorts had multivariable survival analyses (MVSA) adjusting for age (92%), gender (66%), diabetes (63%), albumin (58%), inflammation (CRP/IL6–37%), non-BI nutritional markers (24%) and echocardiographic data (8%). BI-defined overhydration (BI-OH) independently predicted mortality in 32 observational cohorts. Meta-analysis revealed overhydration >15% (HR 2.28, 95% CI 1.56–3.34, P < 0.001) and a 1-degree decrease in phase angle (HR 1.74, 95% CI 1.37–2.21, P < 0.001) predicted mortality. BI-OH predicts mortality in dialysis patients independent of the influence of comorbidity.


Search Strategy and Selection Criteria. MEDLINE, EMBASE, PsychINFO and the Cochrane Register
for Controlled Trials (CENTRAL) were searched from 01/01/1990 through to 06/11/2017 to identify relevant citations. 1990 was chosen as the lower cut off as BI machines were not routinely available prior to this date. Search terms included both medical subject headings (MeSH) and agreed a priori individual search terms. Reference lists from identified citations and selected manual journal searching was used to identify any further relevant studies that matched the inclusion criteria prior to data extraction. The search strategy used for CENTRAL is available as an appendix (see supplementary material).

Study Selection and Data Extraction.
All retrieved citations were imported from the citation library into a central database (using Microsoft Excel 2011). Citations were assessed at the title and abstract level by two independent reviewers (MT and EE or SJD) using exclusion criteria: the study was in the paediatric population; the wrong BI measure was used (i.e. segmental and intrathoracic BI methods); the outcome of interest for our review was not reported (i.e. mortality and hospitalisation); there was no full paper available; or there was no English translation available. During abstract review, if the citation suggested that the study assessed prognostic outcome data, or if it was unclear from the abstract what the study outcomes were, then the citation was accepted for full paper review. Full paper review was conducted again using two independent reviewers (MT and MD or SJD), using the same exclusion criteria. Data extraction occurred both at the individual study level, using piloted study summaries (on Microsoft Word 2011), and in the form of review summary tables (on Excel 2011). No a priori assumptions were made regarding data quality.
Classification of BI method for expressing fluid status. Studies were sub-grouped according to whether they used phase angle (PA)/BI-vector analysis (BIVA), normalised ECF (ECF/TBW) or the overhydration index (OHI) as previously described 3 .

Risk of Bias (ROB) Assessment. Risk of bias (ROB) within studies was assessed by two independent
reviewers (MT and MD or SJD) using the Quality in Prognostic Studies (QUIPS) tool 10 , which grades six separate study domains (selection of participants, study attrition, prognostic factor measurement, outcome measurement, study confounding and statistical analyses) according to their risk of Bias (low, medium or high risk of bias). If disagreements occurred then this was resolved following discussion between MT and MD.
Statistical Analyses and Meta Analysis Methods. All cohorts reporting multivariable survival analyses (MVSA) for outcome data (mortality odds ratio, risk ratio or hazard ratio with 95% confidence intervals) were considered for quantitative pooling in a meta-analysis. More than one cohort had to report the same BI-OH method and cut off value defining overhydration within MVSA to be included for pooling. In all PA cohorts mortality hazard ratios were expressed for every 1-degree increase in PA in initial MVSA. Therefore, to better reflect the effect of increasing overhydration defined by PA on mortality, individual cohort mortality HR and 95% confidence intervals were mathematically reciprocated before pooled summaries were produced. Random effects pooled summaries, using the generic inverse variance method, were produced using Review Manager 5.3 (Nordic Cochrane Centre). The I 2 statistic was used to assess statistical heterogeneity, with I 2 values between 30%-60% representing moderate levels of heterogeneity 11 . Data availability. The datasets generated during the current study are available from the corresponding author on reasonable request, including the full search strategy used to identify citations and the QUIPS paper summaries.
In four observational studies using MVSA BI-OH did not independently predict adverse outcomes 22,33,57,59 . A Brazilian cohort demonstrated BI-OH was predictive of cardiovascular (CV) event rate in diabetics but not in non-diabetics 22 . This study, however, was relatively underpowered and reported cardiovascular event rate as opposed to mortality. A Polish cohort determined the effect of dialysis vintage on survival, with secondary analyses exploring the effects of BI-OH, echocardiographic data and troponin levels on survival 33 . In this analysis, which included the cardiac biomarker troponin, BI-OH did independently predict mortality, but when adjusted for albumin, cholesterol and intraventricular septum thickness on echocardiography, this relationship was not seen. A Romanian cohort, which explored the additive value of BNP and relative overhydration (ROH) in predicting mortality in HD patients 57 found that while these had an additive effect in predictiting survival, ROH alone was not an independent predictor, possibly due to relatively small numbers in this sub-group analysis. In a Chinese cohort of PD patients, increased extra-intra cellular water ratio was predictice of worse survival in MVSA, except of r the final models which incorporated C-reactive protein; again this was likely underpowered given the number of covariates used and low number of deaths 59 .   Table 1. Summary of studies fulfilling the search criteria. Individual patient cohorts listed according to author(s), year of publication (where multiple studies from the same cohort are identified, lead authors of each study and year of most recent study cited) and geographical location of cohort. For each cohort, the BI-OH markers, dialysis modalities, follow up period, number of patients within the cohort, primary/secondary outcomes, number of endpoints and summary of findings are provided. Summaries for each cohort are given, along with the appropriate BI-OH marker and its utility within the cohort to predict survival (denoted by the numbers of * by the BI-OH marker): * shows that the BI-OH marker is an independent predictor of the primary outcome (but does not report a hazard ratio/risk ratio/odds ratio and confidence interval), ** shows the BI-OH marker is an independent predictor of the primary outcome (and reports hazard ratio/risk ratio/ odds ratio and confidence interval), *** shows the BI-OH is an independent predictor of secondary but not primary outcome and **** shows that BI-OH is a univariate predictor of primary outcome. QUIPS risk of Bias summaries are provided for each cohort, with QUIPS domains coded as "L" for low risk of bias, "M" as medium risk of bias and "H" as high risk of bias. QUIPS domains include SP = Study participation, SA = Study attrition, PFM = Prognostic factor measurement, OM = Outcome measure, SC = Study confounding and SAR = Statistical analysis reporting. Randomised controlled trials (RCT) were not quality appraised using QUIPS as this is not a valid method of appraising methodological quality in this study design. In one study (highlighted ^), the follow up period was reported ambiguously and may have reflected cumulative follow up. Association of BI-defined overhydration with morbidity and mortality in Heart Failure. Five heart failure cohorts demonstrated an association between BI-OH and adverse patient outcomes ( 13,18,21,51,60 , Table  1). One cohort undertook univariate analyses and demonstrated BI-OH values predicted the severity of HF symptoms (NHYA classification) in heart failure 18 . Four cohorts included MVSA, adjusting for covariates such as age (100% of MVSA), diabetic status (25%), renal dysfunction (50%), LVEF (50%), BNP (25%) and haemoglobin/ haematinics (50%). In all four cohorts with a MVSA BI-OH was independently predictive of all cause mortality 13,21,60 and adverse events 51 .
Methodological quality of studies. Methodological quality varied widely between studies. Where more than one study was present within a cohort, the study where data was extracted for the review was appraised for methodological quality. Within all studies (Fig. 4)

Discussion
This systematic review provides strong narrative evidence, supported by quantitative evidence from a subgroup meta-analysis, that bio-impedance defined overhydration (BI-OH) is an independent predictor of mortality in ESKF patients. It is the first systematic review exploring this question and the first to demonstrate that different BI-OH metrics, such as phase angle or overhydration index (OHI), act as similar predictors of outcome, with overhydration defined by PA or OHI in the subgroup meta-analysis being associated with approximately double  Table 2. Summary of cohorts reporting multivariate analyses with a stated hazard ratio, (HR) risk ratio (RR) or odds ratio (OR), 95% confidence intervals, (CI) lower limit (LL) and upper limit (UL). Authors highlighted with * or ** had their studies included within the final subgroup meta-analysis. Censoring, where used within MVSA, are stated, with reasons including: Transfer to another RRT modality (1), transplantation (2), loss to follow up (3), transfer to another dialysis facility (4), withdrawal from RRT (5) or, in the case of one paper death due to non-cardiovascular cause (6).
SCIeNtIfIC RepoRts | (2018) 8:4441 | DOI:10.1038/s41598-018-21226-y the risk of mortality compared with normohydrated patients. Furthermore, this is the first review to demonstrate that whole body BI-OH is an independent predictor of mortality in HF, suggesting a role for overhydration as a useful prognostic marker across different chronic disease groups. BI-OH remained independently predictive of mortality or hospitalisation in all ESKF cohorts following adjustment for body mass index (BMI; 1,17,23,24,40,43,52,53,61 ), subjective global assessment (SGA 19,30,39,52,53 ), normalised protein nitrogen appearance (nPNA 39,52,53 ) and malnutrition inflammation score (MIS 16,30 ). This suggests that the additional predictive value of BI-OH is not confined to its ability to identify lean body mass cachexia but that it is also identifying absolute or relative expansion of the extracellular fluid volume as an independent risk. The association of ECFv expansion with malnutrition is not new, having been observed previously, using gold standard techniques of volume measurement in populations with cachexia due to poverty-related starvation 62 . However there are a number of additional explanations for this in the ESKD population, some of which were adjusted for in studies included in this review. Chronic inflammation (c-reactive protein or interleukin-6) was adjusted for in multiple cohorts ( Table 2) and was itself an independent predictor of mortality in 43% of cohorts, without nullifying BI-OH as a predictor of mortality. This association 63,64 is likely explained by the observation that inflammation drives lean body mass cachexia, with such changes being potentially driven by translocation of bacterial endotoxins across an oedematous bowel wall in overhydrated ESKF patients 65,66 . Chronic inflammation also contributes towards hypoalbuminaemia, which in our cohorts was demonstrated to a predictor of mortality in ESKF in half of all cohorts adjusting for it in MVSA; a finding consistent with previous studies that suggest hypoalbuminaemia may contribute towards intradialytic hypotension in haemodialysis and extravascular tissue oedema in peritoneal dialysis 64,67,68 . And yet, as demonstrated with chronic inflammation, hypoalbuminaemia did not nullify the ability of BI-OH to predict mortality in most cohorts, again suggesting the influence of overhydration on mortality in ESKF is synergistic. Cachexia, inflammation and hypoalbuminaemia is a common triad in many chronic diseases 69 , including in chronic kidney disease 69,70 , supporting the argument that overhydration in chronic diseases, as opposed to a catabolic metabolism, is contributing to poor outcomes.
Echocardiographic abnormalities are common in ESKF; one previous study estimating the prevalence of left ventricular hypertrophy and systolic dysfunction in dialysis patients to be 74% and 15% respectively 71 . Our review demonstrates BI-OH remains an independent predictor of mortality even in the presence of abnormal left ventricular ejection fraction (LVEF 44 ) and mass index (LVMI 56 ). Although cautious interpretation is warranted given the small number of cohorts including echocardiogram data, our findings add weight to the developing narrative that cardiac structural disease, particularly left ventricular systolic dysfunction (LVSD), may not be the sine qua non underlying excess mortality in overhydrated ESKF patients. The link between BI-OH and adverse cardiovascular events has been previously noted; BI-OH previously being correlated with endothelial dysfunction 72 , arterial stiffness 43 and the development of left ventricular hypertrophy (LVH 73,74 ). In studies exploring sudden cardiac death in ESKF, LVH was predictive of mortality even when adjusting for blood pressure, whereas LVSD played no such role in predicting mortality. What cannot be answered by current evidence is whether LVH precipitates sudden cardiac death or whether LVH merely acts as a surrogate for overhydration, since the Onofriescu et al. study demonstrated improvements in LVH correlated with improvements in BI-OH measurements 43 . Furthermore, a recent systematic review and meta-analysis by Badve et al. suggests that in CKD, interventions to reduce LVH through altering volume status are not consistently effective, and even where they do reduce LVH (for example through improving haemoglobin or renin-angiotensin blockade), no survival benefit has been seen 75 . The role of BNP as a predictor of overhydration and mortality was explored in four cohorts and data from one suggested a role for both BNP and BI-OH as independent predictors of mortality, albeit echocardiography was not included in this analysis. One hypothesis is that mortality in overhydrated ESKF patients may be driven by the dialysis prescription 76 , with greater ultrafiltration rates during dialysis having been previously demonstrated to induce HD-induced cardiac injury in the form of regional wall motion abnormalities subclinical myocardial ischaemia 77,78 . However, given that two cohorts demonstrated a role for BI-OH and not BNP as independent predictors of mortality, there is still much to be learned about the interaction of cardiac biomarkers and overhydration in predicting outcomes in dialysis patients.
Two recently published large international studies are included in our review 23,61 . Dekker et al. demonstrated in a European multinational cohort using data from 5450 selected HD patients that baseline pre-dialysis BI-OH (where the definition of severe fluid overload was >2.5 L absolute overhydration) predicted increased mortality when adjusted for multiple demographic and co-morbidity covariates. Furthermore this study demonstrated an additive risk of mortality in overhydrated patients with chronic inflammation 23 . The second study, by Zoccali et al., demonstrated in an multinational cohort using data from 39,566 ESKF patients that when adjusted for multiple demographic and co-morbidity covariates that overhydration at baseline, defined as an OHI > 15% for men and >13% for women, is an independent predictor for mortality 61 . They also explored the well established J-shaped relationship between pre-dialysis blood pressure and mortality, finding that higher mortality in overhydrated compared with normohydrated patients is observed across all blood pressure strata, and demonstrated that overhydration remained an independent predictor of mortality with cumulative BI-OH measurements over a one year period 61 . The inclusion of these studies adds significant value to our narrative and pooled summaries, establishing that in approximately 50,000 dialysis patients baseline OHI > 15% is predictive of mortality despite adjustment for multiple comorbidities and inflammation.
Our review has several strengths, including the use of systematic methods to identify studies, independent reviewers throughout the study selection, review and quality appraisal process and the inclusion of heart failure studies as a comparator group, to explore the role of BI-OH in different chronic disease states. It is the first attempt to our knowledge, to summarise and compare the utility of different BI-OH measures in predicting mortality. The review does however have several limitations. Methodological heterogeneity within the studies was considerable, with common sources of bias including unclear study design 14 2,17,18,20,22,25,28,29,31,32,37,46,47,[49][50][51]55,56 , lack of clarity regarding the protocol for the measurement of BI-OH 16,17,19,[31][32][33]45,46,55,56 , exclusion of clinically relevant covariates from MVA 26,[30][31][32][33]37,40 and a lack of clarity regarding the statistical methods used during survival analysis 2,14,16,[18][19][20][21]26,29,39,46,47 . In some studies there was a failure to adjust for HIV status in cohorts where prevalence of HIV is high or where the large proportions of the population is African-Americans; 20,22,29,38,47,49 importantly BI-defined cachexia is associated with HIV infection and therefore potentially confounds the association of BI-OH and mortality. Finally considerable heterogeneity within BI-OH method reporting, and particularly the use of different BI devices which use different "normal populations" to define their BI-OH cut offs, limited the scope for performing a comprehensive pooled survival analysis. This particularly explains why all cohorts expressing BI-OH using the ECWR method could not be pooled, as they all depend on the algorithms used for total body water estimation, which differs between devices 3 and is potentially confounded by ethnicity. Given the anticipated heterogeneity within our pooled analysis we followed the recommendation of Higgins et al. when planning our meta-analysis 79 , including the use of a random effects method, assessing for a consistent pattern in the directionality of the results in included studies and the use of studies which adjust for the effects of covariates on the outcome variable.
This review clearly establishes BI-OH as a predictor of survival in ESKF patients, independent of the effect of malnutrition, inflammation, multimorbidity and within a few cohorts, cardiac structural disease. Similar conclusions are noted in HF patients, suggesting a role for overhydration in predicting poor outcomes in other chronic diseases -a hypothesis which should be tested in other disease groups. The evidence presented necessitates  further investigation into the pathogenic role of overhydration, for example through real-time cardiac imaging and ultrafiltration rate during dialysis or the prognostic value of BI-OH in preventing volume related deaths contributing to the increased mortality observed during the 3-day break. Likewise, it does not establish the value of BI-OH as a tool for goal directed fluid management. Although recent trials suggest that use of BI can improve fluid status and blood pressure, as summarised by Covic et al. in a systematic review and in the recent UK NICE guidelines [80][81][82] , with further studies are on-going 83 , there is no clear benefit on all-cause mortality.