Perioperative liberal versus restrictive fluid strategies and postoperative outcomes: a systematic review and metanalysis on randomised-controlled trials in major abdominal elective surgery

Background Postoperative complications impact on early and long-term patients’ outcome. Appropriate perioperative fluid management is pivotal in this context; however, the most effective perioperative fluid management is still unclear. The enhanced recovery after surgery pathways recommend a perioperative zero-balance, whereas recent findings suggest a more liberal approach could be beneficial. We conducted this trial to address the impact of restrictive vs. liberal fluid approaches on overall postoperative complications and mortality. Methods Systematic review and meta-analysis, including randomised controlled trials (RCTs). We performed a systematic literature search using MEDLINE (via Ovid), EMBASE (via Ovid) and the Cochrane Controlled Clinical trials register databases, published from 1 January 2000 to 31 December 2019. We included RCTs enrolling adult patients undergoing elective abdominal surgery and comparing the use of restrictive/liberal approaches enrolling at least 15 patients in each subgroup. Studies involving cardiac, non-elective surgery, paediatric or obstetric surgeries were excluded. Results After full-text examination, the metanalysis finally included 18 studies and 5567 patients randomised to restrictive (2786 patients; 50.0%) or liberal approaches (2780 patients; 50.0%). We found no difference in the occurrence of severe postoperative complications between restrictive and liberal subgroups [risk difference (95% CI) = 0.009 (− 0.02; 0.04); p value = 0.62; I2 (95% CI) = 38.6% (0–66.9%)]. This result was confirmed also in the subgroup of five studies having a low overall risk of bias. The liberal approach was associated with lower overall renal major events, as compared to the restrictive [risk difference (95% CI) = 0.06 (0.02–0.09); p value  = 0.001]. We found no difference in either early (p value  = 0.33) or late (p value  = 0.22) postoperative mortality between restrictive and liberal subgroups Conclusions In major abdominal elective surgery perioperative, the choice between liberal or restrictive approach did not affect overall major postoperative complications or mortality. In a subgroup analysis, a liberal as compared to a restrictive perioperative fluid policy was associated with lower overall complication renal major events, as compared to the restrictive. Trial Registration CRD42020218059; Registration: February 2020, https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=218059. Supplementary Information The online version contains supplementary material available at 10.1186/s13054-021-03629-y.


Introduction
A worldwide effort aims to reduce postoperative complications [1], which are recognised as partially preventable events affecting long-term morbidity and impacting health and financial systems [2,3]. Several perioperative strategies have been proposed to optimise intraoperative management and postoperative care [4]. Among them, perioperative fluid therapy is a core concept. The ideal perioperative approach has been debated for decades, having the crucial role of balancing oxygen supply and demand, maintaining fluid and electrolyte homeostasis and avoiding inadequate tissue perfusion and fluid overload [5][6][7][8][9][10][11][12].
The most effective perioperative fluid management is still unclear [13][14][15]. It has been classified as restrictive (< 1.75 L per day), balanced (1.75 to 2.75 L per day) and liberal (> 2.75 L per day) [16]. However, the literature provides different, somewhat overlapping definitions (i.e. from 1.0 to 2.7 L for restrictive, compared with 2.8 to 5.4 L for liberal fluid regimens) [17] and conflicting evidence [13][14][15]. The enhanced recovery after surgery (ERAS) pathways to support early recovery among patients undergoing major surgery recommend a restrictive approach aiming at the perioperative "zero-balance" [13]. In contrast, recent findings suggest that excessively restrictive approaches could be detrimental, indicating that a moderately liberal fluid regimen (i.e. positive fluid balance of 1 to 2 L at the end of surgery) might be the best approach [14].
Interestingly, a recent large randomised controlled trial (RCT) assigning 2983 patients to either zero-balance or liberal strategy showed comparable disability-free survival outcome, although the zero-balance approach was associated with a higher rate of acute kidney injury [16]. As a matter of fact, this single study enrolled more patients than several previous RCTs regarding this topic, insofar, partially challenging previous results [18][19][20].
Therefore, we conducted an up-to-date meta-analysis of RCTs to assess the association between restrictive and liberal strategies and major adverse surgical outcomes in elective surgery.
Secondarily, we appraised the association between restrictive and liberal approaches on perioperative mortality, predefined postoperative organ-related complications and hospital length of stay. Finally, we stratified the studies according to preoperative severity score and the reported rate of complications [i.e. American Society of Anaesthesiologists (ASA) physical status score].

Materials and methods
We adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis-Protocols (PRISMA-P) guidelines [21] (Additional file 1: Table 1) and the study protocol was registered [(International Prospective Register of Systematic Reviews-PROSPERO (CRD42020218059)].

Data sources and search strategy
A systematic literature search was performed by using the following databases to identify relevant studies in indexed scientific journals: MEDLINE (via Ovid), EMBASE (via Ovid) and the Cochrane Controlled Clinical trials register, by using the terms: [("liberal" OR "restrictive" OR "zero-balance" AND ("surgery"/exp OR surgery)] with filters for humans, age (< 18 years), language (English) and time of publication (1 January 2000 to 31 December 2019).
We included RCTs 1) enrolling adult patients undergoing elective abdominal surgery; 2) comparing two different regimes of fluid administration defined as "restrictive" and "liberal" or "conventional" or "standard"; 3) starting the study protocol intraoperatively; 4) reporting postoperative complications or mortality or as primary or secondary outcomes.
The restrictive approach was defined as a modality of perioperative (i.e. intraoperative and during the first 24 h after surgery) treatment employing a specific and predefined treatment protocol to obtain in one of the enrolled populations an overall negative or zero fluid balance, as compared to the other. Accordingly, trials showing no statistically significant difference in the overall fluid intake or balance, between restrictive and liberal subgroups, were also excluded.
Studies involving cardiac, non-elective surgery, paediatric or obstetric surgeries were excluded. We also excluded editorials, commentaries, letters to the editor, opinion articles, reviews, meeting abstracts and original articles lacking abstract and/or quantitative details, or those enrolling less than 15 patients in each subgroup.
The references of all included papers, review articles, commentaries and editorials on this topic were also reviewed to identify other studies of interest missed during the primary search.

Data extraction and quality assessment
Three couples of examiners (S.S., E.M., D.B., F.I., M.B., G.D.M.) independently evaluated titles and abstracts. The articles were then subdivided into three subgroups: "included" and "excluded" (if the two examiners agreed with the selection) or "uncertain" (in case of disagreement). In the case of "uncertain" classification, discrepancies were resolved by further examination performed by two expert authors (A.M. and C.R.). We used a standardised electronic spreadsheet (Microsoft Excel, V 14.4.1; Microsoft, Redmond, WA) to extract the data from all included studies, recording trial characteristics (the complete data reporting sheet is provided in Additional file 1: Table 2). When necessary, the included studies' corresponding authors were contacted to obtain missing data related to trial demographics, methods and outcomes.

Risk of bias assessment in the included studies
Two examiners (L.C. and D.Z.) independently assessed the internal validity of the included studies and discrepancies were resolved by a third senior author (A.M. or C.R.), by using the RoB 2 (a revised Cochrane Collaboration's risk of bias tool for randomised trials) [22]. The RoB 2 considers five bias domains: 1) the randomisation process; 2) the deviations from intended interventions; 3) missing outcome data; 4) measurement of the outcome; 5) selection of the reported results. Finally, the overall risk of bias was calculated and, accordingly, studies were included in either high-risk/ some concerns /low-risk groups.

The strength of the body of evidence
The strength of the body of evidence was assessed according to the Grading of Recommendations, Assessment, Development and Evaluations (GRADE) evidence system [23].

Outcomes definitions
Our primary outcome was to appraise the effect of restrictive vs liberal approaches on the overall rate of major complications. This outcome was assessed both considering all the studies reporting it, and those in the subgroup having a low-risk of bias, according to the RoB 2 scale. Secondary outcomes were: to evaluate the association between restrictive and liberal approaches on perioperative early (≤ 30 postoperative days) and late (i.e. > 30 postoperative days) mortality and predefined postoperative major complications: renal (i.e. worsening of renal function, according to the trial definition); cardiovascular [i.e. pulmonary not infective complications, cardiac ischemic dysfunction/failure, cardiac arrhythmias; acute neurological events); infections (i.e. all infective complications reported, including the occurrence of either sepsis or septic shock). We also evaluated the length of hospital stay and stratified the studies according to the enrolled patients' ASA classification.

Statistical analysis
Statistical analysis was conducted on the summary statistics described in the selected articles (e.g. means, medians, proportions) and, therefore, the statistical unit of observation for all the selected variables was the single study and not the patient. Descriptive statistics of individual studies used different statistical indicators for central tendency and variability, such as means and standard deviations, whereas absolute and relative frequencies were adopted for qualitative variables. To show one single indicator for the quantitative variables we collected means with standard deviations (SD) or medians and inter-quartile ranges (IQR) were used, as appropriate.
The metanalysis included only those studies reporting mortality and rate of complications, according to the definition adopted in each study and expressed as rate with respect of the enrolled populations. For ASA subgroup analysis, we grouped studies with at least 50% of the included population classified as ASA I/II, compared to those having at least 50% of the included population classified as ASA III/IV. We, finally, stratified the studies according to three tertiles of the overall amount of perioperative fluid given (day 0 and day 1) and the rate of major events.
We assessed publication bias using a funnel plot for the considered outcomes. Statistical heterogeneity and inconsistency were measured using Q and I 2 tests and were deemed to be significant when p < 0.1 and I 2 > 50%. According to heterogeneity, random or fixed-effect models were used to perform metanalysis. According to Higgins et al. [24], I 2 values around 25, 50 and 75% were considered representing, respectively, low, moderate and severe statistical inconsistency.
The statistical software STATA ® 13 (StataCorp, College Station, TX, USA) was used to perform all the computations.

Perioperative fluid administration, balance and weight gain
Intraoperatively, patients in the restrictive subgroup overall received a median (IQR) of 1925 (1482-2470) of fluids, as compared to 3878 (3000-4400) of the liberal subgroup. On day 0 (i.e. considering the overall amount of fluids received on the day of the operation, including intraoperative fluid therapy), patients in the restrictive subgroup overall received a median (IQR) of 2341 ml (1635-3530) of fluids, as compared to 4,350 ml (3095-5326) in the liberal subgroup. Finally, considering day 0 and day 1, a median of 3617 (2897-5291) of fluids was administered to the restrictive subgroups, as compared to 5820 (5038-7000) to the liberal subgroups (Table 2 and Additional file 1: Table 4).

Postoperative organ-related major complications and length of stay
Pooling data from the 8 studies, the liberal approach was associated with lower overall complication renal major events, as compared to the restrictive [pooled risk difference (95% CI) = 0.06 (0.02-0.09); Chi 2 = 10.3; p value = 0.001] (Fig. 4). In this subgroup, a sub-analysis regarding the use of type of fluid used intraoperatively (colloids vs. crystalloids) showed a borderline statistical significance regarding an increased incidence of major renal events in the restrictive populations receiving colloids, as compared to the liberal ones [mean (SD) major renal events 8.0% (4.7) vs. 1.7% (1.9); p = 0.05].

Subgroup analyses
As reported in Additional file 1: Table 5, preoperative ASA risk score did not impact the postoperative complications rate, among restrictive/liberal subgroups. No difference in the overall rate of major complications was found by stratifying the in tertiles of overall perioperative fluid administered (Fig. 5).    . 2 The internal validity of the included studies was assessed by two expert authors, by using the Rob2: a revised Cochrane Collaboration's risk of bias tool for randomised trials [22]

Discussion
This metanalysis conducted on RCTs in major abdominal elective surgery regarding the effect of perioperative liberal or restrictive approach on postoperative outcomes found no difference between the two approaches in the occurrence of the overall major postoperative complications or mortality. On the contrary, the liberal fluid policy  was associated with lower overall complication renal major events, as compared to the restrictive.
Postoperative complications are common after major surgery and represent a significant financial and social burden [2,3]. Optimisation of fluid management has been extensively studied as a potentially modifiable perioperative factor by adopting specific protocols focused on modality of fluid administration and cumulative fluid balance [9][10][11]40]. The analysis of the literature in this field is rather complex due to the number of variables potentially affecting the final outcomes, which includes the overall complexity and the intrinsic risks of each specific type of the surgery.
On the one hand, a tendency towards a more restrictive approach (as supported by the ERAS pathways [13]) has been reported. In fact, a previous metanalysis [19] showed that fewer patients had a lower total complication rate and risk of infection in the restrictive group as compared to the liberal group.
On the other, the recent large RCT performed by Myles et al. challenged this concept, showing that disability-free survival at 1 year did not change in patients randomised to a zero-balance vs a liberal approach, with the first strategy being associated with higher rates of acute kidney injury, surgical site infection and need for renal-replacement therapy [16]. Interestingly, this single trial enrolled more patients than 17 RCTs combined in the previous 15 years ( Table 1). The weight of this trial specifically impacts on postoperative major renal events of the present metanalysis, which were lower in the liberal subgroup as compared to the restrictive (Fig. 4). Interestingly, renal events did not impact on overall mortality, despite the fact that acute kidney injury is recognised as an independent risk factor for mortality [41]. However, postoperative mortality is greatly affected by the results of the study of Myles et al. [16]. Moreover, the present metanalysis did not include trials on cardiac surgery, which is a clinical setting specifically associated with an increased risk of death in those patients who develop postoperative acute kidney injury [41][42][43].
Considering the impact of crystalloids/colloids use on the renal postoperative outcome, this metanalysis only suggests a possible additive effect of the colloids use in a restrictive approach. However, a recent large RCT comparing the use of low molecular weight hydroxyethylstarch vs. 0.9% saline for intravascular volume expansion in high-risk surgical patients showed no significant difference in a composite outcome of death or significant postoperative [27]. For this reason, this result should be considered with extreme caution. In the field of perioperative fluid administration, summarising the evidence available in the literature into straightforward clinical suggestions for daily clinical practice is rather complicated, and our updated results basically indicate an equivalence between the two perioperative fluid policies, suggesting a "third way". The overall amount of fluid (perioperative target) should be integrated into an individualised goal-directed fluid replacement strategy (perioperative policy), to prevent fluid overload and fluid shortage by closely monitoring the effects of each bolus administered, as long as the individual plateau of the impact on predefined flow or pressure variables is achieved [12,44]. Interestingly, this approach was incorporated into the intraoperative protocol of 5 studies [16,25,[27][28][29], including the study of Myles et al. who assessed central venous pressure and stroke volume variation in case of intraoperative hypotension to guide resuscitation (supplementary appendix of the trial [16]). Moreover, a recent metanalysis showed a trend towards the reduction of postoperative complications when a goal-directed therapy is used in patients receiving large amounts of perioperative fluids [45]. Thus, rather than choosing between a fixed-volume regimen and a goal-directed concept, an alternative approach could be to combine the two strategies. Of note, in the study of Myles et al. [16] restrictive fluid therapy had similar effects in patients treated with or without a goaldirected device.
As shown by our qualitative analysis, neither the perioperative intakes nor the fluid balance or the bodyweight gain has been consistently reported in the considered RCTs. Surprisingly, the overall fluid balance or the postoperative weight difference, which should be, in principle, the most effective perioperative variables depicting the fluid paths of restrictive and liberal subgroups, has been reported only in 5 studies (27.7%) [16,33,35,46,47]. The heterogeneity in actual overall fluid balance computation, as long the definition of postoperative complications (see further) greatly affects the comparability of the RCTs, soliciting clear standards in data reporting.
Considering the fluid intake alone, our data suggest that an overall median administration of about 4 L on days 0 and 1 may be considered restrictive, whereas about 6 L may be considered liberal. However, as pointed out by previous metanalysis [18,19], these cut-offs are associated with large inter-quartile ranges due to the lack of consistency in perioperative fluid regimen definition. Consequently, some subgroups of RCTs classified as restricted would be considered liberal in some other trials (see Fig. 5), because of overlapping perioperative volumes.
However, despite the lack of statistical significance, we found the highest rates of major complications in the highest and lowest tertiles of our sub-analysis.
Finally, the included studies all report overall low perioperative mortality rates; thus, as expected, the impact of liberal vs. restrictive strategies on perioperative mortality was not significant.

Limitations
Some limitations of this study should be acknowledged.
First, as concern the data obtained from the included studies: 1) the definition of postoperative complications and the timing in mortality assessment may vary among the included studies, implying a bias in the reported outcomes' comparability. In fact, some complications (i.e. pneumonia, wound abscess/infection/ dehiscence, pneumothorax) have been considered either as minor or major, according to study-specific criteria, which have been respected in the data analysis; 2) we computed major/minor complications according to either the criteria adopted in the study or (only for subgroup analysis) definitions clearly indicating the gravity degree (i.e. peritonitis and anastomotic leak has been always considered as severe complications); 3) the impact of restrictive/liberal policies on specific subgroups of patients is overall lacking (i.e. patients with pre-existing renal or diastolic cardiac dysfunction), limiting the generalisability of data; 4) this is not a metaanalysis based on individual data and the main author of each study (whenever needed) has been contacted only to provide more specific information regarding the results on the overall population enrolled. This is specifically important for colloids/crystalloids sub-analysis.
Third, the overall quality of the included studies reported "some concerns" in the majority of them (55.5%), mostly related to the selection of the reported results due to drawbacks in the trial registration. This is mainly due to the criteria imposed by the RoB 2 scale [22] for the trial registration. The International Committee of Medical Journal Editors policy regarding prospective trial registration was started in 2005 [48], and most of the trial published before 2010 do not fulfil the RoB 2 criteria for "low-risk" report. Moreover, a monitoring of the study protocol is reported only in three studies [16,26,33].
Finally, we adopted a database combination search strategy, including PUBMED ® , EMBASE ® and the Cochrane Controlled Clinical trials register, excluding different sources (i.e. Web of Science ® ). Although this choice should allow a reliable coverage of the published studies for the topic of interest, some RCTs could not be identified.