Gfaa 017
Gfaa 017
Gfaa 017
doi: 10.1093/ndt/gfaa017
REVIEW
Jule Pinter1, Charles Chazot2, Stefano Stuard3, Ulrich Moissl3 and Bernard Canaud4
Naþ content might improve cardiovascular outcomes in these EuCliDV (European Clinical Data System) is an interna-
patients. tional electronic health record repository that allows continu-
ous point-of-care data collection of routine clinical practice and
lab test information in HD patients from Fresenius Medical
THE BURDEN OF FLUID OVERLOAD AND
Care (FMC) clinics across 20 neighbouring countries across
CARDIAC DISEASE IN HAEMODIALYSIS
Europe, the Middle East and Asia. Among 31 349 incident
ECF overload is a major factor in morbidity of the HD popula- (2010–14) chronic HD patients in the EuClid database, Cox
tion. In a prospective study among 176 790 prevalent HD models were used to prospectively study the association be-
patients, Arneson et al. [43] have reported that, during a 2.5- tween FO, pre-BP and all-cause mortality risk, controlling for
year follow-up, 14% of the patients required hospital admission differences in demographics (age and gender), diabetes, conges-
for one or more episodes of FO, heart failure or pulmonary oe- tive heart failure and body mass index.
dema necessitating urgent fluid removal. Also, Plantinga et al. We report an inverse relationship between FO, pre-BP and
[44] have reported from a US cohort including 215 251 preva- all-cause mortality risk (FMC, unpublished data on file,
lent HD patients a 23% rate of readmission during a 30-day pe- Figure 1).
riod after discharge, 44% of them being related to pulmonary As Figure 1 indicates, the highest all-cause mortality risk was
oedema. When pulmonary oedema was the cause of the first ad- found in patients presenting with high FO but low pre-BP. The
mission, then pulmonary oedema represented 70% of the cause second highest risk was found with both high pre-BP and FO,
of readmission. This suggests that fluid excess may have been while high pre-BP but normal FO related to only moderate risk.
inadequately handled when transitioning from the hospital to This finding, supported by Zoccali et al. [6], described above,
the dialysis unit. In a report by the Dialysis Outcomes and points out that neither pre-BP nor FO should be treated as
Practice Patterns Study (DOPPS), Goodkin et al. [45] identified
the prevalence of congestive heart failure in, respectively, 46%
and 25% of the patients in the USA and in Europe. In the same
study, hypertension was present in 83 and 73% of the US and
European patients. It may be hypothesized that fluid excess was
the major underlying determining factor. Long-term exposure 12
8
in dialysis patients [46]. ECF excess is present early during the
6
CKD progression as shown by Essig et al. [47], and acts as a
continuum all along the CKD progression. These data partly ex- 4
Pre OH
Normohydration (HS = 0 L)
TAFO
TAFO
Post OH
Post OH
Standard HD TAFO = 0 HD
FIGURE 2: During the standard dialysis session, the prescription of post-dialysis body weight targets normal extracellular fluid balance at the
end of the dialysis session. Yet, this exposes the patient to fluid accumulation all along the interdialytic interval, realizing a ‘time-averaged fluid
overload’ (TAFO ¼ half of the interdialytic weight gain). To avoid the exposure to TAFO during the interdialytic period (TAFO ¼ 0), the post-
overloaded. A moderately negative target weight would be de- A clear modifiable source of Naþ exposure in HD patients
sirable in both of these groups of patients. and a promising intervention to improve cardiovascular mortal-
We are aware that this concept may appear highly provoca- ity is to reduce DNaþ during HD treatment [13].
tive bearing in mind how deleterious the effects of overesti- The recent Cochrane review evaluated 12 randomized trials
mated fluid removal rate can be. Yet we believe that the concept of low (<138 mmol/L) versus neutral (138–140 mmol/L) or
expands our understanding of how to reduce the large cardio- high (>140 mmol/L) DNaþ for HD patients, including 310
vascular morbidity and mortality in HD patients. patients, did not examine hard clinical endpoints such as car-
diovascular or all-cause mortality [66]. The authors rated the
quality of evidence as low and concluded that the effect of the
DIALYSATE SODIUM PRESCRIPTION, intervention on overall patient health and well-being is cur-
SODIUM BALANCE AND SODIUM REMOVAL rently unknown [66].
A recent cross-sectional snapshot of all EuCliD patients (data The questions at heart remain whether lowering DNaþ pre-
from 52 000 HD patients, as of July 2019, FMC unpublished vents cardiac remodelling and sudden cardiac death by improv-
data on file) demonstrated that dialysate Naþ (DNaþ) pre- ing Naþ regulation and FO or whether the benefit of lowering
scriptions range from 134 to 143 mmol/L. A DNaþ concentra- DNaþ in reducing left ventricular mass is offset by increased
tion of 138 mmol/L is utilized in roughly 50% of all patients and myocardial stunning and micro-injury [10, 13]. The ongoing
140 mmol/L is the second most common, with 15%. global Randomised Evaluation of SOdium dialysate Levels on
Whereas Eastern European countries tend to utilize lower Vascular Events (RESOLVE) Trial investigates whether lower
DNaþ concentrations, and Russia seems to prescribe the lowest DNaþ may improve cardiovascular outcomes and will deter-
concentrations between 135 and 137 mmol/L in two-thirds of mine comparative effectiveness of two default DNaþ concen-
patients, Italy prefers higher DNaþ concentrations, trations (ClinicalTrials.gov Identifier: NCT02823821).
140 mmol/L. On the contrary, a proof of principle study of the ‘0 Na diffu-
Future studies are necessary to understand cultural behav- sion’ concept—an option developed for newer HD machines —
iour, ethnic factors, common beliefs or eminence to explain the was recently undertaken and proved that automated DNaþ in-
unit-, region- and provider-specific DNaþ prescription pattern dividualization by ‘Naþ control’ approaches isonatraemic dial-
practices in Europe and bordering the Mediterranean Sea. ysis in the clinical setting without the need to determine
In HD patients, Naþ balance depends on dietary intake and the plasma Naþ concentration [67]. Automated Naþ control
Naþ removal during HD [64]. An excessive Naþ load is associ- holds the promise to avoid diffusive Naþ load or removal
ated with high mortality [25]. A post hoc analysis from The during HD and future studies are needed to determine whether
Hemodialysis Study considering 1800 chronic HD patients isonatraemic dialysis could have any effect on hard clinical end-
showed a significantly increased risk of death with a dietary points [67].
Naþ load >2.5 g/die [25]. Patient education for low salt diet
may reduce IDWG by 30% [65]. However, recent data suggest HOW TO BEST APPROACH EUVOLAEMIA—
that educating patients with kidney disease to reduce dietary STATE OF THE ART 2020
Naþ does not lead to the desired outcome [14]. The heterogeneity in the ESRD population challenges the ne-
The dialyser must serve as the salt-excretory function and phrologist in HD practice to combine disease management with
should precisely remove the amount of Naþ that has accumu- the patient’s attainable treatment goal and prognosis [68].
lated during the interdialytic period. Naþ removal during HD The context is complex and a multidimensional approach to
occurs via convective (78%) and diffusive losses (22%) be- Naþ, fluid and pressure management in HD patients may be
tween dialysate and plasma Naþ concentration [64]. The dif- required to improve cardiovascular outcomes in HD patients
fuse Naþ gradient during HD ‘fine tunes’ Naþ balance. (Figure 3). The pillars of clinical, instrumental and patient
FIGURE 3: Multidimensional approach to sodium, fluid and pressure management in haemodialysis patients. US, ultrasound; 23Na MRI,
magnetic resonance imaging of tissue sodium stores.
management need to complement each other to correctly assess Yet, before we know for certain, we need rigorous cardiovas-
the patient’s dry weight, and achieve BP control and haemody- cular outcome studies and thorough study interventions such
namic tolerance of HD therapy, while reassuring patient accep- as the above-mentioned (e.g. effect of UF rate thresholds) on
tance of HD duration and sequence. volume status, fluid-related hospitalizations and cardiovascular
Some dialysis facilities have implemented incremental dialy- outcomes while ensuring the patient’s acceptance to therapy.
sis as a preventive measure at the start of renal replacement
therapy. Other dialysis facilities permit a degree of initial over- ADVANCED ANALYTICS MANAGEMENT
hydration to preserve residual renal function, which may im-
Artificial intelligence (AI) is already successfully applied to sup-
prove outcomes [69]. A recent DOPPS analysis [70] confirmed
port physicians in the decision-making process of patients’ care
the previous finding [71] for incident HD patients that urine presenting either with chronic (e.g. anaemia management) or
output of at least 1 cup daily was associated with better survival acute kidney failure (e.g. fluid resuscitation in the intensive care
and it may be of value to routinely ask patients a simple unit) [76–79]. AI relies today on two main paths, known as
question about urine output. Diuretics may lessen chronic FO symbolic reasoning (expert systems, symbolic AI) and machine
in patients with preserved residual kidney function, but this learning (deep learning, neural networks, connectionist AI),
approach has not been evaluated [72]. each with their own advantages and limitations. Expert systems
Some experts have suggested imposing UF rate thresholds require perfect knowledge of physiological processes and inter-
(such as 13 mL/h/kg) to reduce UF-related risk [73]. UF profil- actions to provide reliable prediction based on rational algo-
ing (decelerating UF rate to match declining plasma refill rate) rithms. Machine learning (connectionist AI) requires a large
and sequential dialysis (isolated UF followed by combined HD amount of data (big data) to learn and define its own rules
and UF) are other potential HD prescription changes to reduce (open and not guided), in order to provide reliable prediction
the harm of rapid UF rate [72]. but lack of model interpretability. A third pathway is emerging,
Finally, as once demonstrated by the Tassin group, increased which combines the two approaches of symbolic reasoning
frequency or duration of HD sessions may bring high volume (symbolic AI) and machine learning (connectionist AI) (such
status, left ventricular mass and BP under control by more ef- as deep symbolic learning and/or enabling neural networks).
fectively reducing ECF load than conventional HD [9]. Yet, it is too early today to define what will be the best approach
Unfortunately, classic randomized controlled trials in HD for future predictive and supportive medicine [80].
thus far have failed to yield any meaningful information in the Expert systems based on AI outperform experienced neph-
area of dose and frequency of HD, mainly due to methodologi- rologists in assessing dry weight in HD patients [81].
cal issues, such as statistical errors, unfeasible trial efficiency, re- Preliminary and proof-of-concept studies based on machine
cruitment challenges and applicability of results to the research learning AI have been performed in the field of fluid and haemo-
question [74]. Cluster randomization may be a novel and dynamic management of HD patients to train and validate mod-
advantageous trial method and a potential mean to overcome els on a large patient dataset [82]. As shown, AI is able to predict
these barriers [75]. reliably individual session-specific patient haemodynamic