Jurnal
Jurnal
CLINICAL PRACTICE
1
Department of Anesthesiology & Perioperative Care, UCI Medical Center, 101 the City Drive South, Orange, CA, USA
2
Department of Anesthesiology and Intensive Care, Louis Pradel Hospital, Hospices Civils de Lyon, 60 avenue du Doyen Lepine, Bron 69300, France
3
Departments of Anesthesiology, Surgery, and Biomedical Informatics, Vanderbilt University Medical Center, 1211 Medical Center Drive,
Nashville, TN, USA
* Corresponding author. E-mail: jrinehar@uci.edu
Background. Variation in clinical practice in the perioperative environment and intensive care
Editor’s key points unit is a major challenge facing modern medicine. The objective of the present study was to
† Fluid management in analyse intraoperative crystalloid administration practices at two academic medical centres
abdominal surgery can in the USA.
affect perioperative Methods. We extracted clinical data from patients undergoing intra-abdominal procedures
outcome. performed at UC Irvine (UCI) and Vanderbilt University (VU) Medical Centres. Limiting data
† A retrospective to uncomplicated elective surgery with minimal blood loss, we quantified variability in fluid
two-centre analysis administration within individual providers, between providers, and between types of
evaluated fluid procedures using a corrected coefficient of variation (cCOV). Regression was performed
administration in relation using a general linear model to determine factors most predictive of fluid administration.
to provider and procedure Results. For provider analysis and model building, 1327 UCI and 4585 VU patients were used.
type. The average corrected crystalloid infusion rate across all providers at both institutions was
† Large variability in 7.1 (SD 4.9) ml kg21 h21, an overall cCOV of 70%. Individual providers ranged from 2.3 (SD 3.7)
crystalloid administration to 14 (SD 10) ml kg21 h21. The final regression model strongly favoured personnel as
was observed both within predictors over other patient predictors.
and between anaesthesia Conclusions. Wide variability in crystalloid administration was observed both within and
providers. between individual anaesthesia providers, which might contribute to variability in surgical
† Further studies are outcomes.
necessary to determine
Keywords: fluid therapy; resuscitation; safety
effects on clinical
outcome. Accepted for publication: 15 October 2014
Variation in clinical practice in the perioperative environment crystalloid strategy improves outcome, reducing both morbidity
and intensive care unit has been reported.1 2 This is commonly and hospital length of stay.5 9 – 13 Moreover, there is increasing
defined today as ‘variation in the utilization of health care ser- evidence that intraoperative fluid administration has a definite
vices that cannot be explained by variation in patient illness or effect on surgical patient outcomes (at least in moderate- to
patient preferences’.3 In 2001, the National Institutes of Health high-risk patients).14 However, it is unclear what volume of crys-
identified variance in health-care practices across different talloid anaesthetists are administering during routine proce-
institutions and different providers as one of the challenges dures and whether significant practice variability exists.
of modern medicine.4 5 In 2010, the British National Health We analysed intraoperative crystalloid administration prac-
Service published the Atlas of Variation in Healthcare, which tices at two academic medical centres in the USA without
was discussed in a series of articles.5 – 7 departmental guidelines. We sought to determine whether
This is considerable literature on perioperative fluid manage- patient, provider, surgeon, and operative factors were asso-
ment in patients undergoing abdominal surgery,8 and multiple ciated with crystalloid infusions rates or total volumes
studies have suggested that a restrictive perioperative during the most common types of abdominal surgery.
†
This article is accompanied by Editorial Aev067.
& The Author 2015. Published by Oxford University Press on behalf of the British Journal of Anaesthesia. All rights reserved.
For Permissions, please email: journals.permissions@oup.com
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
BJA Lilot et al.
Additionally, we sought to quantify variability both within and Vanderbilt University Medical Center, Nashville, TN USA). Insi-
between specific anaesthesia providers. We hypothesized that tutional review board approval was obtained at VU, and the
total crystalloid administered during abdominal surgery is study was considered insitutional review board exempt at
consistent in the absence of explanatory patient or surgical UCI. Neither institution had general departmental guidelines
factors. on fluid administration during these periods (i.e. goal-directed
fluid therapy was not practiced, nor were there any depart-
mental policies towards fluid administration in moderate- or
Methods high-risk surgery).
In a two-centre retrospective analysis, we extracted clinical
data from patients undergoing abdominal procedures per-
formed at UC Irvine (UCI) and Vanderbilt University (VU), Patient selection
from January 2009 to December 2011 at UCI and from Patient inclusion and exclusion criteria are shown in Figure 1.
January 2009 to December 2012 at VU using Perioperative In- After sorting by procedure description, all procedures per-
formation Management Systems (at UCI, SISTM , Surgical Infor- formed in the specified time frames at both institutions were
mation Systems, Alpharetta, GA, USA; and at VU, GasChart, hand reviewed to exclude procedures that were substantially
All surgeries
(UCI 2009–2011,
VU 2009–2012)
Limit to:
Appendectomy
Cholecystectomy
Colectomy Remove cases with:
Hysterctomy >500 EBL
Nephrectomy 12 982 cases Emergencies
Pancreatectomy <40 kg
Prostatectomy SRNA cases
Patients receiving transfusions
Patients receiving colloids
Patients on TPN
Providers with <6 patients
Cases <1 hour duration
ASA PS ≥4
Use of PAC
Analyses by
procedure
8294 cases
Modelling
Remove prostatectomies
Institutional
analyses
5912 cases
Analyses by
provider
Fig 1 Flow chart of patient selection. Study protocol showing how patients were selected and limited to uncomplicated, low-blood-loss operations.
Blue boxes are filtering steps and green boxes are analysis steps. ASA PS, American Society of Anesthesiologists Physical Status Classification; EBL,
estimated blood loss; PAC, pulmonary artery catheter; SRNA, Student Registered Nurse Anesthetist (excluded because only one institution has
SRNA’s); UCI, University of California Irvine; VU, Vanderbilt University.
768
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
Variability in crystalloid administration BJA
different from the simple type. For example, ‘cholecystectomy
with intraoperative cholangiogram’ would be included in Table 1 Variables assessed in the regression model
‘cholecystectomy’, but ‘hepatic segmentectomy, cholecystec- Patient factors
tomy’ would not. Prostatectomies were specifically excluded Age
from the provider analysis at both institutions because of the ASA Class
specific resuscitation guidelines for those procedures at UCI, Sex
but were kept for the procedural analysis and modelling as a Height
comparison group (Fig. 1). Median central venous pressure (if present)
In order to reduce baseline variance in procedures, criteria Median haemoglobin
were developed to narrow the patient pool to uncomplicated Median heart rate
surgical courses. Emergency procedures were excluded, and Median mean arterial pressure
nephrectomy for living-donor and transplant procedures Median temparature
were excluded. Rectosigmoid procedures were included in Minimal central venous pressure (if present)
colectomy procedures, but procedures using a transanal ap- Minimal haemoglobin
proach were excluded. Other exclusion factors are shown in Minimal mean arterial pressure
Figure 1. Given that analysis of crystalloids was done on a Minimal temperature
millilitre per kilogram per hour basis, procedures of ,60 min Nothing per os time
duration were excluded to reduce artificial bias from ‘frontload- Preoperative haemoglobin
ing’ of crystalloid during induction. Procedure duration was Urine output
defined as anaesthesia start to anaesthesia end times. Proce- Weight
dures where a Student Registered Nurse Anesthetist provided Personnel factors
care were excluded because only VU had Student Registered Anaesthesia attending
Nurse Anesthetists providing care during the study time Anaesthesia in-room provider
frame. Procedures done under only spinal or epidural anaes- Anaesthesia in-room provider type (resident vs certified
thesia were excluded, though procedures where an epidural registered nurse anaesthetist)
was placed for postoperative pain management only were Surgeon
retained. After the previously listed exclusion criteria were Procedural factors
applied, we then excluded all procedures by providers or sur- Arterial line used (yes/no)
geons with fewer than six patients in the data set (Fig. 1). Central venous line used (yes/no)
Duration (operating room-in to operating room-out)
Data collection Estimated blood loss
For each patient, we extracted the data shown in Table 1. In Epidural used (yes/no)
order to make comparisons across providers and procedure Laparoscopic (yes/no)
types, crystalloid administration was corrected using a Surgery type
neutral fluid balance approach.9 Urine output and blood loss University
were accounted for using the following formula: Interactions
Attending by in-room provider
Corrected crystalloid = crystalloid − estimated blood loss Attending by procedure
Attending by surgeon
− urine output
In-room provider by procedure
In-room provider by surgeon
In order to adjust for procedure time and patient body weight, University by procedure
results are expressed as millilitres per kilogram per hour except
for the modelling (see next subsection). For analysis, proce-
dures were grouped both by in-room anaesthesia provider (only positive values are possible, and the distribution is posi-
(in order to examine variability within and between specific tively skewed), so a generalized linear model using a g distribu-
providers) and by procedure type (to examine variability tion and log-link for response was used for modelling, which
within and between procedures). also allowed inclusion of both categorical and scalar independ-
ent variables. We first performed a whole model analysis for
Modelling of fluid administration the entire data set that included only main effects of the
Finally, after performing the analyses above, we sought to patient features listed in Table 1. After running the initial
build a model using the variables in Table 1 to examine model, plausible variable interactions were tested to account
factors associated with crystalloid administration at each insti- for multilevel effects (e.g. attending by in-room provider, pro-
tution. As the model would include duration and patient weight cedure by university) using the corrected Akaike’s information
as independent variables, the uncorrected (total) crystalloid criterion15 to guide model selection because this measure in-
administration was used as the dependent model variable. trinsically penalizes more complicated models over more
Total crystalloid administration is not normally distributed simple models, reducing over fitting. The final model was
769
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
BJA Lilot et al.
considered the ‘best’ set of predictors of fluid administration the entire study sample was used as the denominator. The cor-
from the given data, and independent variables in the rected measure preserves ease of interpretation while making
models were then characterized in terms of significance, expo- comparisons of variability between providers, procedures, and
nentiated coefficients, and 95% confidence intervals (CIs). institutions more meaningful.
For clarity, ‘attending’ anaesthetist refers to the licensed
Statistics anaesthesia provider supervising a patient, ‘resident’ refers
to a physician anaesthetist trainee, ‘CRNA’ refers to a Certified
Statistical analysis was performed with SPSS 19 (IBM, Armonk,
Registered Nurse Anesthetist, ‘in-room provider’ refers to
NY, USA) and Excel 2010 (Microsoft, Redmond, WA, USA). Data
either a resident or a CRNA being supervised, and ‘provider’ in
are presented as mean (SD) unless otherwise specified. Com-
general refers to any of the above.
parisons between groups or providers were made using
Student’s unpaired t-test (for two-way comparisons) and ana-
lysis of variance (ANOVA; for three or more comparisons). Confi- Results
dence intervals are reported at the 95% level. Patient selection
Variability in administration is expressed as a corrected
coefficient of variation (cCOV). Coefficient of variation (COV) is After all inclusion and exclusion criteria were applied and after
defined as sample standard deviation divided by sample exclusion of prostatectomies, 1327 UCI and 4585 VU patients
mean. The advantage of COV is that it is easily interpreted were obtained for provider analysis and model building. They
and puts the range of sample distribution in context of the were performed by 70 UCI (22 CRNAs and 48 residents) and
mean.16 Unfortunately, when making comparisons across 164 VU in-room providers (89 CRNAs and 75 residents). Each
samples with large relative differences in the mean, COV will provider had an average of 17 (11) patients at UCI and 26
by definition be larger for samples with smaller means even (20) patients at VU. The demographic data by institution is pre-
if the absolute range of the sample distribution is smaller. sented in Table 2.
Thus, for the purposes of describing variability in administra-
tion within and between providers and procedure types, COV Overall findings
was used for its ease of interpretation, but instead of dividing Average corrected crystalloid infusion rate across all providers
by the mean of the specific sample in question, the mean of at both institutions was 7.1 (4.9) ml kg21 h21, an overall cCOV of
Table 2 Patient characteristics and procedural data. Data are presented as count (percentage of total) or median [25th,75th]
770
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
Variability in crystalloid administration BJA
70%. At UCI, the average rate was 6.7 (5.0) ml kg21 h21 and at These volumes are not corrected for procedure mix or other
VU the average rate was 8.2 (5.3) ml kg21 h21, representing a procedural factors.
cCOV of 70 and 74%, respectively, for the two institutions.
The difference in administration rates between institutions Procedural variation
was significant at P,0.0005 (95% CI 1.2–1.8). A scatter plot When grouped by surgical procedure and including prostatec-
of fluid administration rates showing the duration of proce- tomies, 1590 UCI and 6704 VU patients were identified. The
dures is shown in Figure 2. lowest mean corrected crystalloid volume among procedure
The overall administration rate in ASA Physical Status types was for prostatectomy at both UCI and VU, which was
Classification I patients was 9.9 (6.2) ml kg21 h21, with a 3.6 and 5.3 ml kg21 h21, respectively. The highest average for
cCOV of 87%, while ASA Physical Status Classification III each institution was for appendectomies: 10.4 ml kg21 h21 at
patients received an average of 6.9 (4.7) ml kg21 h21, with a UCI and 9.6 ml kg21 h21 at VU. The lowest cCOV for both insti-
cCOV of 66%. These means were significantly different with tutions was seen in prostatectomies; at UCI this was 34% and
P,0.0005 (95% CI 2.6– 3.5). at VU this was 40%. The highest cCOV at both institutions was
seen in appendectomies; at UCI this was 97% and at VU this
Provider variation was 87%.The full data are shown graphically in Figure 3.
Individual providers at each institution exhibited a wide
range of fluid administration patterns, both within indi- Modelling
vidual providers across procedures and between providers. The significant independent variables, exponentiated coeffi-
At UCI, the lowest provider mean corrected crystalloid cients, and 95% confidence intervals resulting from the model-
volume was 2.3 (3.7) ml kg21 h21 (seven patients) and ling process are shown in Table 3. There were no significant
the highest was 11.8 (5.0) ml kg21 h21 (15 patients). At interactions (e.g. attending by in-room provider or similar)
VU, the lowest provider mean corrected crystalloid volume found in the modelling process that improved overall model
was 2.9 (1.8) ml kg21 h21 (nine patients) and the highest fit according to corrected Akaike’s information criterion com-
was 14 (10) ml kg21 h21 (11 patients). The lowest cCOV pared with direct effects only. The exponentiated coefficient
within individual providers was 27% at UCI and 26% at for each variable can be interpreted as the percentage
VU, and the highest was 128% at UCI and 141% at VU. change expected in total crystalloid administration for each
40.00
Corrected crystalloid (ml kg–1 h–1)
30.00
20.00
10.00
0.00
0 200 400 600 800
Duration (min)
Fig 2 Scatterplot of surgery duration vs corrected crystalloid volume. The graph shows the wide distribution of corrected crystalloid administration
rates across all uncomplicated procedures at both UCI and VU. Rates range from 25 to 40 ml kg21 h21, with most between 5 and 15 –20 ml kg21 h21.
Of note, this graph suggests that some degree of ‘frontloading’ is occurring (patients receive large volumes early in procedures, with tapering as
procedures go on), but also demonstrates that this phenomenon is highly inconsistent, with many patients receiving no frontloading at all. Thus,
frontloading is not a sufficient explanation of the variability obvious in the figure.
771
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
BJA Lilot et al.
34
*
32 University *
*
UCI * *
30 *
VU *
28 *
*
26 **
Corrected crystalloid infusion (ml kg–1 h–1)
24 *
*
22
*
20
18 *
**
16
14
12
10
8
6
4
2
0
–2
–4
–6
Prostatectomy Pancreatectomy Hysterectomy Collectomy Nephrectomy Cholecystectomy Appendectomy
Procedure
Fig 3 Fluid administration by surgical procedures. Corrected crystalloid infusion rates for procedures at both UCI and VU. Each boxplot is the median
and range. For most procedures, about 50% of patients received between 4 and 10 ml kg21 h21 crystalloid; the other 50% obviously fell outside this
wide range. Of note, UCI has a specific protocol for crystalloid administration during prostatectomies, and this group had the smallest range of any
of the analysed procedures, suggesting that directed protocols can be effective in reducing variability.
unit change in the predictor variable, with the prediction cor- patient, and some amount of variation in administration will
rected for all of the other elements. For example, from be explained by factors not captured in the present study, it
Table 3 we see that heart rate has an exponentiated coefficient is unlikely that both the ranges of 0– 3 and 15–20 ml kg21
of 1.002 per beat min21, meaning that for each increase in h21 of corrected crystalloid volume are ideal, but based on
heart rate of 10 beats min21 a patient would be expected to these data standard practice encompasses that range of
receive 1.00210 ¼1.02 times more fluid during a procedure. administration and wider.
Administration between providers at both institutions has
providers whose interquartile ranges for corrected crystalloid
Discussion administration do not substantially overlap. It is certainly pos-
This two-centre retrospective observational study assessed sible that these providers have different procedure distribu-
crystalloid administration in routine abdominal procedures in tions, even given the limited procedure types used in this
the absence of departmental fluid administration guidelines, analysis. However, in looking at the model for total crystalloid
and found wide inter- and intraprovider variability in infusion administration (Table 3), it is notable that the odds ratios for
volumes. This variability does not appear to be explained by providers are very widely distributed, especially compared
patient or individual surgical factors, and could result from with patient and procedural factors such as duration and
random fluctuations in practice from provider to provider and blood loss, which while significant have exponentiated coeffi-
within providers from patient to patient. There has been a rela- cients very close to one. Based on these results the, majority
tive resurgence of research interest in fluid administration of total crystalloid administration volume is predicted by per-
strategies in recent years, especially with outcome improve- sonnel factors as opposed to patient or surgical factors.
ments shown by goal-directed protocols and more restrictive Additionally, and independently of the absolute volumes,
approaches to fluid administration.9 17 18 While there are no the range of corrected crystalloid infusions within individual
definitive ‘best practices’ for any given surgical procedure or providers is also fairly wide. While some providers were
772
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
Variability in crystalloid administration BJA
Table 3 Crystalloid model parameters and significance. CRNA, certified registered nurse anesthetist; CVP, central venous pressure; Hb,
haemoglobin; HR, heart rate; MAP, mean arterial pressure; VU, Vanderbilt University. *The exponentiated coefficient for each variable can be
interpreted as the percentage change expected in total crystalloid administration for each unit change in the predictor variable, with the prediction
having been corrected for all of the other elements in the model. Bold text P-values are those that are statistically significant
773
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
BJA Lilot et al.
consistent in their administration, with cCOV’s ,40%, probably virtually all durations, and in particular, procedures ,300
a reasonable level given all the variances in procedures and min in duration (the majority) show the greatest variability.
patients, both institutions have providers with cCOV’s .100% Moreover, while ‘frontloading’ seems readily accepted by
(interquartile ranges as wide as 15 ml kg21 h21). For this many providers a priori, it is obvious from the data in Figure 2
range (assuming 0–15 ml kg21 h21 absolute volume), a that there is no consistency in the application of this practice.
patient weighing 75 kg who has a 4 h procedure with 400 ml Unwarranted variation in practice has been identified as a
blood loss and 1 ml kg21 h21 urine output will receive anything major driver of both waste and suboptimal outcomes in health-
between 700 and 5400 ml of crystalloid. Some patients care systems.2 3 19 – 23 Part of the recent National Health Service
received as much as 20– 30 ml kg21 h21, representing 6.7 – reform focus has been reduction of unwarranted variation24 in
9.7 litres for the example patient. Ignoring absolute values of order to ensure equivalent care across hospitals and providers.
the volumes, if a provider is applying a consistent methodology The results of this retrospective analysis suggest that there is a
for fluid administration, we should expect a narrower range of high degree of unwarranted variation in the volume of intrao-
corrected volumes administered, especially as this data set perative crystalloids. This might be especially relevant given
was restricted to low-blood-loss, uncomplicated procedures the evolving literature on perioperative fluid administration.
that did not require colloids or blood products. The wide Moreover, fluid administration studies are made all the more
ranges seen indicate that most providers are in fact fairly in- complicated by the fact that administration (both volume
consistent in their individual approaches. Moreover, as most and type) remains more or less ad-lib.
intraoperative haemodynamic and patient variables were Even the most recent ‘state-of the-art’ literature on the topic
either non-significant in the model or had very weak effects, shows significant variability in the way that baseline crystalloid
it is difficult to argue that the fluid differences result from infusion is administered. For example, the study by Challand
haemodynamic differences during surgery. and colleagues25 focusing on fluid management in major
In the total crystalloid model, the strongest effect types are surgery used a baseline crystalloid infusion of 10 ml kg21 h21,
those related to personnel (attending, in-room provider, and while the OPTIMIZE study by Pearse and colleagues26 used a
surgeon), while patient-specific factors or surgery-specific fac- crystalloid administration of 1 ml kg21 h21. Both studies were
tors, such as blood loss, duration, ASA status, and haemo- conducted in the UK and in similar surgeries but used dramat-
dynamics, had very low strengths, even when significant. For ically different fluid regimens. Thus, the variability in practice
example, the most liberal to most restrictive anesthesia observed in our study from practitioner to practitioner is also
attendings have effect sizes from 2.42 to 0.70, respectively observed from study to study and probably from institution to
(Table 3), indicating that a patient would receive on average institution.
four times more fluid from the liberal attending compared Interestingly, prostatectomy patients at UCI do have a spe-
with the restrictive attending. The effect of blood loss, how- cific fluid administration protocol dictated by the surgeons
ever, is ,1.0005 per millilitre lost, meaning that even a 500 ml that restricts fluids intraoperatively and then requires more
difference in blood loss between two patients would result in liberal fluid administration after re-anastamosis of the ure-
no more than a 1.2–1.3 factor difference in fluid administra- thra through the postoperative period. It might not be coinci-
tion. This suggests that crystalloid administration volume is a dence that this procedural subgroup had the lowest cCOV
result of the persons giving the fluid as opposed to patient and of any procedure (34%). This was reflected in the surgical com-
procedural factors. Alternatively, if practice is indeed based on parisons (Fig. 3), where the UCI prostatectomies had the lowest
specific patient factors, there is so little consistency between pro- corrected crystalloid infusion rate and the smallest range of
viders in how haemodynamics and other patient and procedural variation among procedures at either institution. This suggests
variables are used to guide fluid management that these effects that specific protocols can be effective in clinical practice in
appear to be random elements in the model. Either way, it reducing variation between providers. It is likely that imple-
appears that there is a great deal of inconsistency in how these mentation of guidelines such as crystalloid restriction and
variables are managed in terms of fluid resuscitation. goal-directed therapy would help to standardize the way in
Despite having eliminated procedures ,60 min in duration, which fluid is administered.
the argument could be made that there is a risk of bias from
‘frontloading’, in which shorter procedures appear to receive
more volume because of the fluids typically given during the Limitations
induction period and immediately thereafter. Moreover, there First, we attempted to control for procedural and patient
is a trend in the data towards higher infusion volumes (in milli- factors as best as possible, but it is still possible that one or
litres per kilogram per hour) in short vs long procedures (Fig. 2), more significant factors that might explain the observed vari-
which supports this argument. This is an unsatisfactory explan- ability was not taken into accounted. Second, some sources
ation of the total variability, however. First, the existence of recommend replacement of blood lost at a ratio of 3:1 with
frontloading alone does not negate the existence of variability. crystalloid, whereas in the present study we used a 1:1 ratio
In looking at Figure 2, while there is a decrease in total millilitres for the corrected crystalloid calculation. The low coefficient
per kilogram per hour and in variability as procedures grow for blood loss from the modelling, however (,1.0005), sug-
longer, it is also immediately apparent that there is a huge gests that blood loss is weakly linked to crystalloid administra-
range of fluid volumes given to patients in procedures of tion and that using a 3:1 model would not have changed
774
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
Variability in crystalloid administration BJA
findings significantly. It is also possible that one or more com- has received separate research funding from Edwards Life-
binations of variables were missed during model building that sciences and Masimo Corp.
might have had more explanatory power than the resultant
models in this study. Vasopressor use, for example, was not
included but could have played a role in some anaesthetists’
Funding
approaches to resuscitation. The types of crystalloid adminis- No funding outside of University support.
tered were not taken into account in the present study. In
general, the vast majority of the fluid administration in the op-
References
erating rooms is Ringer’s lactate, normal saline, or Plasmalyte
1 Wijeysundera DN, Austin PC, Beattie WS, Hux JE, Laupacis A.
(Baxter healthcare, Australia). As specific solutions and re-
Variation in the practice of preoperative medical consultation
spective volumes were not analysed, however, it is possible for major elective noncardiac surgery: a population-based study.
that there is some unrecognized contribution of these factors Anesthesiology 2012; 116: 25– 34
to our analysis. Finally, we did not link the variability observed 2 Pravinkumar SE. Physician practice variation in critical care: a stum-
in the present study to any postoperative outcomes. While bling block. Crit Care Med 2009; 37: 1178–9
this is of interest for future studies, we believe that the fact 3 Wennberg JE. Time to tackle unwarranted variations in practice. Br
that such variability exists is in itself an important issue to Med J 2011; 342: d1513
tackle in order to improve quality of care. 4 Medicine IO. Crossing the Quality Chasm: A New Health System for
the 21st Century. Washington, DC: National Academy of Sciences,
ed., 2001
Conclusion 5 Nisanevich V, Felsenstein I, Almogy G, Weissman C, Einav S, Matot I.
Despite recommendations in the literature for fluid manage- Effect of intraoperative fluid management on outcome after
ment in abdominal operations, we find a large patient-to- intraabdominal surgery. Anesthesiology 2005; 103: 25–32
patient variability in fluid administration practices between 6 Westert GP, Faber M. Commentary: the Dutch approach to
unwarranted medical practice variation. Br Med J 2011; 342:
and within anaesthesia providers and surgical procedures at
d1429
both the University of California Irvine and Vanderbilt Univer-
7 QIPP/Right Care. The NHS Atlas of Variation in Healthcare, 2010.
sity Medical Center. The regression model analysis showed Available from http://www.rightcare.nhs.uk/index.php/atlas/atlas-
that the strongest predictor for corrected crystalloid infusion of-variation-2010 (accessed 9 December 2014)
rate for a specific patient was the anaesthesia provider and 8 Corcoran T, Rhodes JE, Clarke S, Myles PS, Ho KM. Perioperative fluid
surgeon. Further studies are needed to assess change in management strategies in major surgery: a stratified meta-
patient outcomes with implementation of specific fluid man- analysis. Anesth Analg 2012; 114: 640–51
agement protocols on a large scale. 9 Brandstrup B, Svendsen PE, Rasmussen M, et al. Which goal for fluid
therapy during colorectal surgery is followed by the best outcome:
near-maximal stroke volume or zero fluid balance? Br J Anaesth
Authors’ contributions 2012; 109: 191– 9
M.L.: study design, data collection and analysis, and manu- 10 Joshi GP. Intraoperative fluid restriction improves outcome after
script preparation. J.M.E.: data collection and manuscript prep- major elective gastrointestinal surgery. Anesth Analg 2005; 101:
601– 5
aration. C.L.: data collection and analysis, and manuscript
11 Forget P, Lois F, de Kock M. Goal-directed fluid management based
preparation. B.H.: data collection and analysis. M.C.: study
on the pulse oximeter-derived pleth variability index reduces
design, data analysis, and manuscript preparation. J.R.: study lactate levels and improves fluid management. Anesth Analg
design, data analysis, and manuscript preparation. 2010; 111: 910–4
12 Chappell D, Jacob M, Hofmann-Kiefer K, Conzen P, Rehm M. A ration-
al approach to perioperative fluid management. Anesthesiology
Acknowledgements 2008; 109: 723–40
The authors wish to thank Zhaoxia Yu for help with the 13 Rahbari NN, Zimmermann JB, Schmidt T, Koch M, Weigand MA,
statistics. Weitz J. Meta-analysis of standard, restrictive and supplemental
fluid administration in colorectal surgery. Br J Surg 2009; 96:
331– 41
Declaration of interest 14 Cecconi M, Corredor C, Arulkumaran N, et al. Clinical review: Goal-
M.L., J.M.E., and B.H.: none declared. C.L.: Department has directed therapy - what is the evidence in surgical patients? The
received separate research funding from Edwards Lifesciences effect on different risk groups. Crit Care 2013; 17: 209
and Masimo Corp. M.C.: Sironis, Inc., equity interest; consulted 15 Claeskens G, Hjort NL. Model Selection and Model Averaging. Cam-
bridge Series in Statistical and Probabilistic Mathematics. Cambridge
and/or has prepared continuing medical education (CME)
and New York: Cambridge University Press, 2008. xvii, 312 p
materials for Covidien, Draeger, Philips Medical System, Gauss
16 Tian L. Inferences on the common coefficient of variation. Stat Med
Surgical, Edwards Lifesciences, Fresenius Kabi, Masimo Corp.,
2005; 24: 2213– 20
and ConMed; Department has received separate research 17 Scheeren TW, Wiesenack C, Gerlach H, Marx G. Goal-directed intrao-
funding from Edwards Lifesciences and Masimo Corp. J.R.: perative fluid therapy guided by stroke volume and its variation in
Sironis, Inc., equity interest; consulted and/or has prepared high-risk surgical patients: a prospective randomized multicentre
CME materials for Surgiquest and Masimo Corp.; Department study. J Clin Monit Comput 2013; 27: 225–33
775
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018
BJA Lilot et al.
18 Futier E, Constantin JM, Petit A, et al. Conservative vs restrictive indi- 23 Stiefel M, Feigenbaum P, Fisher ES. The Dartmouth Atlas applied to
vidualized goal-directed fluid replacement strategy in major ab- Kaiser Permanente: analysis of variation in care at the end of life.
dominal surgery: a prospective randomized trial. Arch Surg 2010; Perm J 2008; 12: 4 –9
145: 1193–200 24 Richards RG. Tackling practice variation. How to do it. Br Med J 2011;
19 Byrnes J. 7 tactics to reduce variation in clinical practice. Healthc 342: d2285
Financ Manage 2012; 66: 146, 148 25 Challand C, Struthers R, Sneyd JR, et al. Randomized controlled trial
20 Dudley N. Tackling practice variation. Threats and opportunities of of intraoperative goal-directed fluid therapy in aerobically fit and
the NHS reforms. Br Med J 2011; 342: d2269 unfit patients having major colorectal surgery. Br J Anaesth 2012;
21 Mays N. Reducing unwarranted variations in healthcare in the 108: 53– 62
English NHS. Br Med J 2011; 342: d1849 26 Pearse RM, Harrison DA, MacDonald N, et al. Effect of a periopera-
22 Miller MG, Miller LS, Fireman B, Black SB. Variation in practice for dis- tive, cardiac output-guided haemodynamic therapy algorithm on
cretionary admissions. Impact on estimates of quality of hospital outcomes following major gastrointestinal surgery: a randomized
care. JAMA 1994; 271: 1493– 8 clinical trial and systematic review. JAMA 2014; 311: 2181–90
776
Downloaded from https://academic.oup.com/bja/article-abstract/114/5/767/234948
by guest
on 14 March 2018