Modifiable risk factors for perioperative AKI: A post-hoc analysis of the RELIEF randomised trial

Statistical Analysis Plan

Background:

Perioperative acute kidney injury (AKI) has been variably reported to occur in approximately 6-39% of patients undergoing non-cardiac surgery1-4 and is associated with a significant burden of morbidity and mortality.

  • No specific preventive or therapeutic strategies yet clearly identified and confirmed
  • Recent efforts have focused on broad strategies of “risk-minimisation” with limited evidence to support individual components of such bundles.5, 6
  • RELIEF study database provides a unique opportunity to explore a high-fidelity, prospectively collected dataset for important and potentially modifiable variables that may be associated with perioperative AKI and thus identify strategic targets for subsequent testing to reduce the incidence of perioperative AKI

Hypothesis:     Our primary hypothesis is that modifiable perioperative risk factors can be identified that are associated with postoperative AKI in patients undergoing major abdominal surgery.

Aim: To identify potential kidney-protective targets and strategies for testing in a subsequent prospective clinical trial.

Methods:         This is a post-hoc observational analysis of the RELIEF trial cohort.7 The RELIEF trial was an international, randomised, controlled trial of a liberal versus restrictive perioperative intravenous (IV) fluid regimen during and up to 24 hours after surgery, enrolling 3000 patients at increased risk of complications undergoing major abdominal surgery between May, 2013 and September, 2016.

Results of the primary analysis of the RELIEF study were published in 2018, finding no difference in primary outcome of disability-free survival at 1 year postoperatively. However, a 71% increase in relative risk of moderate to severe (KDIGO stage 2 or 3) AKI was identified in the group assigned to a restrictive IV fluid regimen (adjusted risk ratio 1.71, 95% CI 1.29-2.27).

Study cohort:  The study cohort will comprise the entire RELIEF cohort analysed for the primary outcome (n=2983) with additional exclusion of any patients meeting the following criteria:

  • Identified as undergoing renal/urologic surgery
  • Missing preoperative serum creatinine (SCr) [Anticipated n=41 based on primary manuscript]
  • No postoperative SCr available prior to discharge [Anticipated n=60 based on primary manuscript]  

Outcomes:      The primary outcome for the current analysis is the maximal stage of AKI identified prior to hospital discharge, defined and staged according to current KDIGO criteria, and calculated as the difference between peak postoperative SCr prior to discharge (36.9 or 36.10) and preoperative SCr (5.5 or 5.6).

Any use of renal replacement therapy (RRT) within 30 days (39.7) will be considered to have occurred in hospital will thus also define patients as having reached stage 3 AKI prior to hospital discharge. Consistent with the analysis of RELIEF data in the original manuscript urine output criteria will not be included in the definition of AKI.8 In contrast to the principle analysis reported in the original manuscript we will NOT adjust SCr measured on POD1 and POD3 for perioperative fluid balance for our primary analysis, instead doing this only as part of a pre-specified sensitivity analysis.

Secondary outcomes will include RRT within 90 days of surgery (41.9), mortality within 90 days of surgery (41.1) and a composite of RRT or mortality within 90 days of surgery (to explore for consistency of results across more robust [but less frequent] outcomes that may reflect perioperative renal injury).9

Analysis:         The incidence of any AKI in the RELIEF cohort was approximately 23% and we therefore anticipate more than 600 events for analysis in the current study proposal. The primary outcome will be analysed using multivariable ordinal logistic regression to explore the association between pre-specified potentially modifiable perioperative risk factors for AKI and ordered stages of postoperative AKI. Secondary outcomes will be analysed using multivariable logistic regression.

With the primary aim of identifying potentially important and modifiable risk factors for postoperative AKI our analytic approach allows for development and testing through a large model rather than seeking to develop a parsimonious model for the most efficient prediction of AKI. Potentially modifiable perioperative variables were selected from the existing RELIEF dataset by investigators based on a combination of existing literature,2, 4, 10-22 expert opinion and perceived feasibility of testing the effect of modifying each variable in a subsequent clinical trial. The following variables were considered by the investigators to be biologically plausible and potentially modifiable contributors to postoperative AKI:

PREOPERATIVE

  • Preoperative ACE/ARB administration, including administration on day of surgery. This will be evaluated as a 3-level ordinal variable as follows: Not on a regular ACE/ARB, usually on an ACE/ARB but DID NOT TAKE day of surgery; usually on an ACE/ARB and DID receive on day of surgery (4.4 and 4.4a).
  • Preoperative PPI or H2-blocker (4.6)
  • Preoperative Hb (5.2 or 5.3)
  • Bowel prep used (4.1)
  • Time since most recent oral fluids (4.11)

INTRA- and POST-OPERATIVE

  • Use of a Goal-Directed Haemodynamic Management (GDHM) device, evaluated as a binary Y/N variable (9.0)
  • PEEP. This will be evaluated as a continuous variable (cm H2O). (13.1)
  • Neuraxial blockade with LA. This will be evaluated as a binary variable. (11.5)
  • Surgical Approach. This will be evaluated as a 3-level categorical variable: Open, Laparoscopic. Lap-assisted/converted (10.4)
  • Vasoactive drug administration. This will be evaluated as a categorical variable (0-5) based on number of different vasoactive agents used. (11.3 and 11.4)
  • Diuretic administration (furosemide or other). This will be evaluated as an ordinal variable (0-3) where 0=no diuretic administered intraoperatively or in PACU or POD-1; 1=diuretic administered intraoperatively only; 2=diuretic administered postoperatively only [PACU or POD-1]; 3=diuretic administered both intraoperatively and postoperatively [in either PACU or POD-1]. (11.2, 18.8, 23.1 and 23.2)
  • NSAID or Cox-II inhibitor administration. This will be evaluated as an ordinal variable (0-3) where 0=no NSAID or Cox-II administered intraoperatively or on POD-1; 1= NSAID or Cox-II administered intraoperatively only; 2= NSAID or Cox-II administered on POD-1 only; 3= NSAID or Cox-II administered both intraoperatively and on POD-1. (11.6 and 23.3)
  • Lowest intraoperative SBP (continuous variable) (12.6)
  • Highest intraoperative SBP (continuous variable) (12.7)
  • Difference between lowest intraoperative SBP and pre-specified lowest acceptable SBP (continuous variable) (12.6 – 8.0)
  • Lowest intraoperative MAP (continuous variable) (12.9)
  • Highest intraoperative MAP (continuous variable) (12.10)
  • Temperature at end of surgery (continuous variable) (17.11)
  • Lowest SBP in PACU (continuous variable) (18.3)
  • Highest SBP in PACU (continuous variable) (18.4)
  • Lowest DBP in PACU (continuous variable) (18.5)
  • Highest DBP in PACU (continuous variable) (18.6)
  • Highest temperature within 24 hours of surgery (continuous variable) (23.10)
  • Lowest Hb concentration on POD-1 (24.1 or 24.2)
  • Transfusion of pRBCs (total volume intraoperative / PACU / POD1) (16.8 + 20.8 + 26.8)
  • Total 0.9% saline administered intraoperatively, PACU and POD-1 (16.2 + 20.2 + 26.2)
  • Total non-saline crystalloid administered intraoperatively, PACU and POD (16.0 + 16.1 + 16.3 + 20.0 + 20.1 + 20.3 + 26.0 + 26.1 + 26.3)
  • Total intraop, PACU and POD-1 colloid volume (individually evaluated as starch-based (16.5 + 20.5 + 26.5), gelatin-based (16.4 + 20.4 + 26.4), albumin-based (16.6 + 20.6 + 26.6) or other (16.7 + 20.7 + 26.7))

We do not intend to perform a detailed analysis on fluid volume strategy as a potentially modifiable risk factor for AKI because the primary trial analysis has already addressed this question through randomization of patients to a liberal vs. restrictive strategy. However, we have included actual volume and type of specific fluid administered as a potentially modifiable variable due to the potential influence of TYPE and COMPOSITION of administered fluid as a plausible contributor to postoperative AKI.

Additional potentially confounding variables for which to further adjust models will be selected a priori by investigators based on a combination of expert opinion, previously described risk factors, 2, 4, 10, 12, 13, 16-18, 21, 22 a previously published predictive score for AKI after non-cardiac surgery, 23 and available data within the RELIEF study database (including the assigned intervention within the study). The following variables will be considered as potential confounders for inclusion in regression models:

  • RELIEF group assignment
  • Age (1.1)
  • Body Mass Index (BMI) (1.2 and 1.4)
  • Sex (1.0)
  • ASA physical status (1.3)
  • PHx of hypertension (2.0); heart failure (2.2); PVD (2.4); COPD (2.9); solid malignant disease (Yes/No) (2.17); liver disease (binary variable: 0=no history of liver disease or history of mild liver disease; 1= history of moderate or severe liver disease) (2.20, 2.21 and 2.22); diabetes (binary variable: 0=no diabetes or diabetes without pharmacologic treatment; 1=diabetes on oral pharmacotherapy or insulin) (2.23 and 4.9 and 4.10); current infection/sepsis (2.26)
  • Preoperative diuretic therapy (4.3)
  • Preoperative albumin (5.7 or 5.8), SCr concentration (5.5 or 5.6) (with calculated eGFR using Cockcroft-Gault – indexed per 1.73 m2 of BSA [calculated by Du Bois formula), HbA1c% (5.0 or 5.1)
  • Type of surgery (5-level categorical variable [Renal/urol excluded) (10.1)
  • Surgical field (binary variable (clean/clean contaminated or contaminated/dirty) (10.4)
  • Surgical urgency (binary variable: expedited or elective) (10.3)
  • Surgical duration (17.12 – 10.2)
  • PLANNED postoperative location. This will be evaluated as a 4-level ordinal variable (0=ward; 1=HDU; 2=ICU - NOT ventilated; 3=ICU - ventilated). This variable may function as an integrated marker of baseline perioperative risk using otherwise ‘unmeasured’ patient and surgical factors) (10.5 and 10.6)
  • Cancer surgery. This will be evaluated as a binary Y/N variable (17.5)
  • Country in which each patient underwent surgery

Given existing uncertainty regarding the optimal method of model construction we will develop two separate and distinct models, with the specific purpose of evaluating the robustness of our results through consistency across these two techniques for variable selection. Model 1 will be based on variable selection guided by expert opinion and existing literature regarding risk factors for postoperative AKI. An initial assessment for evidence of important collinearity between potentially modifiable variables, and separately for potentially confounding variables, will be conducted through correlation analysis and paired scatterplots, boxplots, and mosaic plots, and assessments of multicollinearity (e.g., using variance inflation factors). A multiple R2 or Spearman rank-order correlation coefficient >0.90 will be considered to represent substantial collinearity.24, 25 Where substantial collinearity is identified, selection of variables for exclusion from the multivariable model will be determined by the expert opinion of investigators. All remaining potentially modifiable risk factors will be included in the multivariable model with the additional inclusion of covariates hierarchically selected by investigators, to maintain a ratio of outcome events for each degree of freedom included in the model. Optimism in the Brier score will be assessed using 10-fold cross-validation to evaluate for evidence of model over-fitting. If there is optimism greater than 10% in the Brier score, covariates will be excluded in an order selected a priori by investigators.

Model 2 will use Lasso regression techniques to construct a multivariable model from all identified potentially modifiable risk factors for AKI and potential covariates. The lasso penalty will be selected by minimizing the 10-fold cross-validated Brier score. Regression analyses will be implemented using the ‘ordinal’ and ‘ordinalNet’ packages for R. Secondary endpoints will be evaluated using Lasso regression techniques only.

Pre-planned sensitivity analyses:

  1. We will evaluate the robustness of our results to variations in the definition of (staged) AKI, including:
  • AKI re-defined with adjustment for fluid balance within the first 3 postoperative days (as done in the primary RELIEF manuscript)
  • AKI defined using only SCr values from the first 3 postoperative days (to exclude AKI first developing beyond this time-period that may reflect aetiologies unrelated to the immediate perioperative period) (24.4 or 24.5 [POD-1] and 32.4 or 32.5 [POD-3])
  • AKI defined only as stage 2 or 3 AKI (prior to discharge)
  • Restricting the analysis to include only those patients with SCr measured on both POD1 and POD3 (to minimise any effect of ascertainment bias) (24.4 or 24.5 [POD-1] and 32.4 or 32.5 [POD-3])
  • In the subgroup of patients undergoing surgery outside of North America we will repeat the primary analysis after redefining preoperative eGFR according to the CKD-EPI equation under the assumption that all of these patients are non-black, to explore for sensitivity of results to the baseline eGFR estimating equation used.
  1. Subgroup-analyses:          Based on an existing understanding of AKI pathophysiology several subgroup effects or interactions are biologically plausible and will be evaluated for their association with the primary outcome of AKI. These include:
  • Preoperative ACE I /ARBs (including day of surgery) and metrics of perioperative hypotension (as defined earlier)
  • Perioperative NSAID and RELIEF group assignment (ie Liberal or Restrictive)
  • Perioperative NSAID and metrics of hypotension (as defined earlier)
  • Use of GDHM (9.0) and RELIEF group assignment (ie Liberal or Restrictive)
  • Neuraxial blockade with LA and RELIEF group assignment (ie Liberal or Restrictive)
  • Neuraxial blockade with LA (11.5) and metrics of hypotension (as defined earlier)

Based on our findings additional exploratory analyses may be undertaken to better understand these results and inform subsequent trial design.

Missing data: 

Uncertainty due to missing data for covariates thought to be missing at random in the multivariable model will be quantified using the chained equations method of multiple imputation. Five completed datasets will be generated using chained equations and predictive mean matching. These completed datasets will then be analysed using the described regression methods. Estimates and standard errors for regression coefficients across the five completed datasets will be combined into single estimates and standard errors using Rubin’s rules. The ‘mice’ package for R will be used to implement multiple imputation and other regression methods.26 Missing outcome data will not be imputed and these cases will be excluded from regression analyses. Variables thought to be missing not at random will be treated using the ‘missingness indicator’ approach, which encodes an interaction between missing status (present/absent) and the observed value. Sensitivity analyses for the effect of imputation of missing data will be conducted by repeating the analysis for our primary outcome after case-wise deletion.

Model assumptions and checking:

The effects of quantitative variables on the ordinal odds of AKI stage (and the odds of secondary outcomes) will be modelled using a four-knot natural spline with knots at the 20th, 40th, 60th, and 80th percentiles of the independent variable. No interaction between covariates will be considered. For ordinal logistic regression, the proportional odds assumption will be assessed using the repeated dichotomization method. For logistic regression, the Hosmer-Lemeshow goodness-of-fit test will be considered.

The robustness of newly identified associations (i.e., between risk factors and AKI) to unmeasured confounding will be assessed using the E-value concept.27, 28 The E-value is defined as the minimum strength of association on the odds ratio scale that an unmeasured confounder would need to have with both the risk factor (assuming a binary risk factor) and AKI, conditional on the measured covariates, to fully explain away a specific risk factor–AKI association.

Based on the observed results an attributable risk fraction may be calculated for identified modifiable risk factors to further inform the direction and development of future interventional studies.

Mediation Analysis

The primary analysis of the RELIEF trial data indicated that when compared to a liberal fluid management strategy a perioperative restrictive fluid management strategy resulted in an increased relative risk of postoperative AKI. Exploratory mediation analyses will be undertaken to identify potentially modifiable variables that may mediate this effect. The input variable will be RELIEF trial group assignment (restrictive or liberal perioperative fluid strategy) and the output outcome will be maximum stage of AKI identified prior to hospital discharge. Possible mediators of the relationship between perioperative fluid strategy and AKI for analysis will include preoperative ACE I/ARB use, intraoperative hypotension (assessed using both lowest intraoperative SBP and lowest intraoperative MAP), intraoperative vasopressor therapy, and perioperative NSAID/COX-II and diuretic administration. The Preacher-Hayes method of testing for indirect effects29 will be used to assess for evidence of mediation by these variables on the association between perioperative fluid strategy group assignment and maximum stage of postoperative AKI.

Reporting:       Results will be reported as odds ratios with 99% confidence intervals. P-values less than 0.01 will be considered evidence of statistical significance. In view of the hypothesis generating nature of the analysis, and the absence of a formal familywise hypothesis, no formal statistical adjustment will be made for multiple testing. However,we have selected a type-I error rate of 1% to mitigate the risk of type-I errors. In addition, we will consider the consistency of our findings across the two modeling strategies where selected covariates were based on existing published literature and expert opinion (MODEL 1) AND in a model where explanatory variables were included based on data driven statistical techniques alone (MODEL 2).


References:

1.         Bihorac A, Brennan M, Ozrazgat-Baslanti T, Bozorgmehri S, Efron PA, Moore FA, Segal MS and Hobson CE. National surgical quality improvement program underestimates the risk associated with mild and moderate postoperative acute kidney injury. Critical care medicine. 2013;41:2570-83.

2.         Garg AX, Kurz A, Sessler DI, Cuerden M, Robinson A, Mrkobrada M, Parikh CR, Mizera R, Jones PM, Tiboni M, Font A, Cegarra V, Gomez MF, Meyhoff CS, VanHelder T, Chan MT, Torres D, Parlow J, Clanchet Mde N, Amir M, Bidgoli SJ, Pasin L, Martinsen K, Malaga G, Myles P, Acedillo R, Roshanov PS, Walsh M, Dresser G, Kumar P, Fleischmann E, Villar JC, Painter T, Biccard B, Bergese S, Srinathan S, Cata JP, Chan V, Mehra B, Wijeysundera DN, Leslie K, Forget P, Whitlock R, Yusuf S, Devereaux PJ and Investigators P-. Perioperative aspirin and clonidine and risk of acute kidney injury: a randomized clinical trial. JAMA : the journal of the American Medical Association. 2014;312:2254-64.

3.         Sun LY, Wijeysundera DN, Tait GA and Beattie WS. Association of intraoperative hypotension with acute kidney injury after elective noncardiac surgery. Anesthesiology. 2015;123:515-23.

4.         Walsh M, Devereaux PJ, Garg AX, Kurz A, Turan A, Rodseth RN, Cywinski J, Thabane L and Sessler DI. Relationship between intraoperative mean arterial pressure and clinical outcomes after noncardiac surgery: toward an empirical definition of hypotension. Anesthesiology. 2013;119:507-15.

5.         Gocze I, Jauch D, Gotz M, Kennedy P, Jung B, Zeman F, Gnewuch C, Graf BM, Gnann W, Banas B, Bein T, Schlitt HJ and Bergler T. Biomarker-guided Intervention to Prevent Acute Kidney Injury After Major Surgery: The Prospective Randomized BigpAK Study. Annals of surgery. 2018;267:1013-1020.

6.         Meersch M, Schmidt C, Hoffmeier A, Van Aken H, Wempe C, Gerss J and Zarbock A. Prevention of cardiac surgery-associated AKI by implementing the KDIGO guidelines in high risk patients identified by biomarkers: the PrevAKI randomized controlled trial. Intensive care medicine. 2017;43:1551-1561.

7.         Myles PS, Bellomo R, Corcoran T, Forbes A, Peyton P, Story D, Christophi C, Leslie K, McGuinness S, Parke R, Serpell J, Chan MTV, Painter T, McCluskey S, Minto G, Wallace S, Australian, New Zealand College of Anaesthetists Clinical Trials N, the A and New Zealand Intensive Care Society Clinical Trials G. Restrictive versus Liberal Fluid Therapy for Major Abdominal Surgery. The New England journal of medicine. 2018;378:2263-2274.

8.         Kidney disease: Improving global outcomes (KDIGO) Acute Kidney Injury Work Group. KDIGO Clinical Practice Guideline for Acute Kidney Injury

. Kidney international. 2012;Suppl 2:1-138.

9.         McIlroy DR, Bellomo R, Billings FTt, Karkouti K, Prowle JR, Shaw AD and Myles PS. Systematic review and consensus definitions for the Standardised Endpoints in Perioperative Medicine (StEP) initiative: renal endpoints. British journal of anaesthesia. 2018;121:1013-1024.

10.       Abelha FJ, Botelho M, Fernandes V and Barros H. Determinants of postoperative acute kidney injury. Crit Care. 2009;13:R79.

11.       Abrahamsson A, Oras J, Snygg J and Block L. Perioperative COX-2 inhibitors may increase the risk of post-operative acute kidney injury. Acta anaesthesiologica Scandinavica. 2017;61:714-721.

12.       Bang JY, Lee JB, Yoon Y, Seo HS, Song JG and Hwang GS. Acute kidney injury after infrarenal abdominal aortic aneurysm surgery: a comparison of AKIN and RIFLE criteria for risk prediction. British journal of anaesthesia. 2014;113:993-1000.

13.       Bell S, Dekker FW, Vadiveloo T, Marwick C, Deshmukh H, Donnan PT and Van Diepen M. Risk of postoperative acute kidney injury in patients undergoing orthopaedic surgery--development and validation of a risk score and effect of acute kidney injury on survival: observational cohort study. BMJ. 2015;351:h5639.

14.       Courtney PM, Melnic CM, Zimmer Z, Anari J and Lee GC. Addition of Vancomycin to Cefazolin Prophylaxis Is Associated With Acute Kidney Injury After Primary Joint Arthroplasty. Clin Orthop Relat Res. 2015;473:2197-203.

15.       Giglio M, Dalfino L, Puntillo F and Brienza N. Hemodynamic goal-directed therapy and postoperative kidney injury: an updated meta-analysis with trial sequential analysis. Crit Care. 2019;23:232.

16.       Hallqvist L, Granath F, Huldt E and Bell M. Intraoperative hypotension is associated with acute kidney injury in noncardiac surgery: An observational study. Eur J Anaesthesiol. 2018;35:273-279.

17.       Meersch M, Schmidt C and Zarbock A. Perioperative Acute Kidney Injury: An Under-Recognized Problem. Anesthesia and analgesia. 2017;125:1223-1232.

18.       Romagnoli S, Zagli G, Tuccinardi G, Tofani L, Chelazzi C, Villa G, Cianchi F, Coratti A, De Gaudio AR and Ricci Z. Postoperative acute kidney injury in high-risk patients undergoing major abdominal surgery. Journal of critical care. 2016;35:120-5.

19.       Salmasi V, Maheshwari K, Yang D, Mascha EJ, Singh A, Sessler DI and Kurz A. Relationship between Intraoperative Hypotension, Defined by Either Reduction from Baseline or Absolute Thresholds, and Acute Kidney and Myocardial Injury after Noncardiac Surgery: A Retrospective Cohort Analysis. Anesthesiology. 2017;126:47-65.

20.       Vaara ST and Bellomo R. Postoperative renal dysfunction after noncardiac surgery. Current opinion in critical care. 2017;23:440-446.

21.       Vaught AJ, Ozrazgat-Baslanti T, Javed A, Morgan L, Hobson CE and Bihorac A. Acute kidney injury in major gynaecological surgery: an observational study. BJOG. 2015;122:1340-8.

22.       Walsh M, Garg AX, Devereaux PJ, Argalious M, Honar H and Sessler DI. The association between perioperative hemoglobin and acute kidney injury in patients having noncardiac surgery. Anesthesia and analgesia. 2013;117:924-31.

23.       Kheterpal S, Tremper KK, Heung M, Rosenberg AL, Englesbe M, Shanks AM and Campbell DA, Jr. Development and validation of an acute kidney injury risk index for patients undergoing general surgery: results from a national data set.[see comment]. Anesthesiology. 2009;110:505-15.

24.       Kutner MH, Nachtsteim CJ and Neter J. Applied linear regression models 4th ed: McGraw-Hill Irwin; 2004.

25.       Hair Jr JF, Anderson RE, Tatham RL and Black WC. Multivariate data analysis. 3rd ed. New York: Macmillan; 1995.

26.       Van Buuren S G-OK. mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software. 2011;45:1-67.

27.       Mathur MB, Ding P, Riddell CA and VanderWeele TJ. Web Site and R Package for Computing E-values. Epidemiology. 2018;29:e45-e47.

28.       Haneuse S, VanderWeele TJ and Arterburn D. Using the E-Value to Assess the Potential Effect of Unmeasured Confounding in Observational Studies. JAMA : the journal of the American Medical Association. 2019;321:602-603.

29.       Preacher KJ and Hayes AF. SPSS and SAS procedures for estimating indirect effects in simple mediation models. Behav Res Methods Instrum Comput. 2004;36:717-31.

Copyright © 2020 REstrictive versus LIbEral Fluid Therapy in Major Abdominal Surgery (RELIEF). All Rights Reserved.
Joomla! is Free Software released under the GNU General Public License.