Perioperative Normal Saline Administration and Delayed Graft Function in Patients Undergoing Kidney Transplantation: A Retrospective Cohort Study

This is a retrospective cohort study to investigate the association of perioperative normal saline versus balanced crystalloid administration on delayed graft function in patients undergoing kidney transplantation between 2005 and 2015 at a single academic medical center. Patients undergoing combined transplant surgeries were not eligible for inclusion. After Institutional Review Board approval and a waiver of informed consent was received (University of California San Fransisco, San Francisco, California), we submitted a finder file to the Scientific Registry of Transplant Recipients. The Scientific Registry of Transplant Recipients is a national database under custody of the U.S. Department of Health and Human Services (Washington, D.C.). All U.S. transplant centers are required to report their data. The Scientific Registry of Transplant Recipients manages these data, and supplies data, summary reports, and analyses to the transplant community to support ongoing evaluation of the national organ transplant system and to facilitate research. The Scientific Registry of Transplant Recipients data system includes data on all donors, wait-list candidates, and transplant recipients in the United States, submitted by the members of the Organ Procurement and Transplantation Network. The Health Resources and Services Administration, U.S. Department of Health and Human Services, provides oversight to the activities of the Organ Procurement and Transplantation Network and Scientific Registry of Transplant Recipients contractors.

The finder file submitted to the Scientific Registry of Transplant Recipients contained name, date of surgery, and transplant number to identify institutional cases within the national database. The file was a broad and untested extraction from our institutional transplant database that was readily available but not linked to the anesthesia databases. It contained 3,955 identifiers of kidney transplant recipients from 2004 to 2015. Of those, 3,891 records could be linked to Scientific Registry of Transplant Recipients data.

Simultaneously, we extracted kidney transplant records from two consecutive institutional electronic anesthesia record databases (Picis Clinical Solutions, Inc., USA, from 2005 to 2012; and Epic, USA, from 2012 to 2015) and merged both datasets to create the file of available institutional anesthesia records from 2005 to 2015. The merged file of anesthesia records contained 3,183 records.

Subsequently, we linked our final anesthesia record dataset (n = 3,183) with the identified Scientific Registry of Transplant Recipients data and obtained 2,951 records. After exclusion of combined transplants (n = 46) and pediatric patients (n = 166), our eligible cohort consisted of 2,739 records. Of those, 2,515 had any crystalloids charted perioperatively and were included in the data analysis (fig. 1, flowchart).

Our original protocol included pediatric patients (age less than 18 yr). However, we later realized that body mass index, number of transfusions, and crystalloid volume would all need to be normalized to age or weight for pediatric patients to have a physiologic interpretation similar to adults. Therefore, for simplicity, we restricted our main analyses to adults and report these results in this manuscript.

At our institution, the choice of crystalloid is solely at the discretion of the anesthesia provider. The recommendations for intraoperative fluid administration during kidney transplantation are accessible on the departmental intranet since 2009 and suggest 40 ml/kg of crystalloids but do not recommend a specific type. The routinely used hemodynamic endpoint is the noninvasive mean arterial blood pressure to be within 20% of the baseline value. Synthetic colloids are discouraged. Albumin 5% can be used if indicated. The preferred vasopressor, if indicated, is dopamine. Since 2009, routine placement of central venous lines or arterial lines is not recommended. Adjustment of these general recommendations based on patients’ comorbidities is at the discretion of the anesthesia provider (Supplemental Digital Content 1, https://links.lww.com/ALN/C654).

The primary exposure of interest was the perioperative administration of normal saline. The perioperative time period was defined as the time from anesthesia start to discharge from the postanesthesia care unit. Volume and type of crystalloids were extracted from the anesthesia records. In both anesthesia databases, crystalloids are usually charted in increments from 100 to 1,000 ml throughout the perioperative period. The sum of each type of crystalloid administered during the perioperative period was extracted from the anesthesia databases based on an algorithm, and 101 charts were then manually reviewed to validate the extraction process.

Because we expected that a significant proportion of patients received a combination of different crystalloid solutions, we defined the percent normal saline of the total amount of crystalloids as our primary exposure of interest. This was an a priori definition based on the rationale that the percent normal saline administered perioperatively is solely the choice of the anesthesia provider. In contrast, the total volume of crystalloids administered is often dictated by perioperative conditions and patients’ characteristics such as body weight.

After accessing the data, we could not confirm a linear association between the percent normal saline and the log odds of delayed graft function. Therefore, we categorized the variable into a low normal saline group (30% or less), an intermediate normal saline group (greater than 30% but less than 80%), and a high normal saline group (80% or more). We chose these cutoffs based on a bimodal distribution of the percent normal saline variable and clinical considerations: The less than 30% category roughly represents all patients receiving less than 1 l of normal saline.

Patients’ demographic data such as age, sex, and body mass index, and perioperative data such as surgery duration, medications, type and volume of crystalloids, and all other fluids administered were extracted from the electronic anesthesia records. Transplant and donor data were derived from the Scientific Registry of Transplant Recipients Standard Analysis File for kidney transplant recipients.

The a priori defined primary outcome was the occurrence of delayed graft function, defined as the need for dialysis within 1 week after transplant surgery. Delayed graft function is a well-established endpoint for allograft dysfunction in kidney transplantation and often used as a primary endpoint in research studies.^18,19  All outcome data were derived from the Scientific Registry of Transplant Recipients Standard Analysis File for kidney transplant recipients.

To explore possible heterogeneity of our mixed cohort of deceased and living donor transplants, we re-ran our primary logistic regression model for both subgroups separately and tested for interactions between the primary exposure and the type of donor.

All analyses were performed based on an a priori study protocol unless indicated as post hoc analysis. A data analysis and statistical plan was written after the data were accessed.

Bivariate associations were assessed with the chi-square test, Fisher exact test, Mann–Whitney U test, or independent samples t test. Data were mostly complete except that cold ischemia time was missing in a relevant number of records. Assuming missingness at random, where the propensity for a data point to be missing is related to some of the observed data, we used multiple imputation with ordinal logit regression and imputed missing cold ischemia time categories (less than or equal to 2 h, greater than 2 to 10 h, greater than 10 to 20 h, greater than 20 h).²⁰  The imputation model included all independent variables plus the outcome variable of the analysis models. Twenty imputed datasets were generated.

To compare delayed graft function among the low, intermediate, and high normal saline groups, we determined the unadjusted risks by study group and for relevant subgroups. To adjust for potential confounding from nonrandom treatment assignment, we then fit a multivariable logistic regression model. Model development was based on four consecutive steps to investigate for potential confounders and covariates for inclusion into the model in a forward selection process.

1. Creation of a directed acyclic graph and identification of the minimal sufficient adjustment set before starting data analysis.

2. Consideration of variables that showed unadjusted statistical differences (P ≤ 0.2) between the normal saline groups (primary exposure) as well as between the outcome categories (delayed graft function or not).

3. Identification and inclusion of variables extremely strongly associated with the outcome to reduce variability of the model.

4. Consideration of variables of general interest that commonly are included into models of delayed graft function in the literature.

If nonlinearity of the association between continuous covariates and the log odds of the outcome delayed graft function was suspected, we categorized the covariates, either based on clinically established cutoff values or based on sample quartiles. We verified that no significant interactions were present. We assessed the overall model fit with the Hosmer–Lemeshow test using deciles of fitted risks, and we report the C-statistic as a marker of the discriminative ability of our model. We also assessed the contribution of the normal saline variable to the overall model fit with the likelihood ratio test.

No statistical power calculation was conducted before the study. The sample size was solely based on the available data. A minimum clinically meaningful effect size was not defined a priori.

To further investigate the robustness to misspecification of our primary parsimonious logistic regression model, we developed an alternative regression model with adjustment for treatment propensity scores. Using a logistic regression model, we calculated the propensity of treatment with 50% or more versus less than 50% normal saline. We used all covariates of the primary model to fit the treatment model. We created restricted cubic splines of the propensity score with five knots. We then regressed delayed graft function on the two normal saline groups adjusting for cubic splines of the propensity score.

To address concerns that our primary exposure variable, percent normal saline, does not reflect the amount of normal saline administration, we created three strata of total normal saline administration: no normal saline, 2 l or less, and more than 2 l. We then used this alternative exposure variable with our primary logistic regression model.

To address concerns that our decision to split the percent normal saline variable at 30% and 80% cutoffs might introduce bias, we have created an alternative variable of percent normal saline groups with cutoffs at 20%, 40%, 60%, and 80% normal saline. We then used this alternative exposure variable with our primary logistic regression model.

To explore a possible effect of kidney transplant recipients who have not been on dialysis at the time of transplant, we reran our primary model excluding this subgroup.

Data are presented as mean ± SD. All P values are drawn for two-sided hypothesis testing, and statistical significance was evaluated at the significance level of 0.05. We used Stata 14.2 (Stata Corp., USA) for all statistical analyses.

The dataset included 2,515 kidney transplant cases. Twenty-two patients underwent two kidney transplant procedures between 2005 and 2015. Mean ± SD age of the cohort was 52 ± 14 yr, 58.1% (1,461) were male, and 58.5% (1,472) received a deceased donor graft. Sixteen (0.6%) donor kidneys were machine-perfused before transplant. Three hundred eighty-five cases (15%) received 30% normal saline or less (low normal saline group), 646 cases (26%) received greater than 30% but less than 80% normal saline (intermediate normal saline group), and 1,484 cases (59%) received 80% or more normal saline (high normal saline group). Of the 2,515 cases, 216 (9%) did not receive any normal saline, and 1,402 (56%) received normal saline exclusively. Over the years, the proportion of patients receiving 80% or more normal saline decreased from 75% (348/465) in the period from 2005 to 2007, to 27% (106/397) from 2014 to 2015.

Baseline characteristics in the low, intermediate, and high normal saline groups are displayed in table 1. The mean ± SD volume of normal saline administration was 264 ± 384 ml in the low percent normal saline group and 2,861 ± 1,344 ml in the high normal saline group. The mean normal saline volume, mean balanced crystalloid volume, and mean volume of crystalloids per percent normal saline group are displayed in table 2. We also assessed the bivariate association of individual covariates with the outcome delayed graft function. As expected, several patient- and donor-related characteristics were associated with delayed graft function. Of the anesthesia- and surgery-related characteristics, surgery duration and the normal saline group were associated with delayed graft function (Supplemental Digital Content 2, https://links.lww.com/ALN/C655).

The overall incidence of delayed graft function was 19.3% (485/2,515). In the subgroups of deceased donor transplants, the incidence of delayed graft function was 29.2% (430/1,472), and in living donor transplants 5.3% (55/1,043). For the full cohort, the unadjusted incidence of delayed graft function in the low, intermediate, and high normal saline group was 15.8% (61/385), 17.5% (113/646), and 21% (311/1,484), respectively. Stratified by donor type, delayed graft function in the low, intermediate, and high normal saline group for deceased donor transplants was 24% (54/225), 28.6% (99/346), and 30.8% (277/901), respectively; and for living donor transplants, 4.4% (7/160), 4.7% (14/300), and 5.8% (34/583), respectively. The incidence of delayed graft function over time is displayed in Supplemental Digital Content 3 (https://links.lww.com/ALN/C656).

The primary logistic model regressed delayed graft function on the three normal saline groups. Based on the directed acyclic graph, we included year of transplant surgery, total crystalloid volume, duration of surgery, the perioperative use of a vasopressor infusion, and erythrocyte transfusions as covariates. In addition, we included recipient sex, age, body mass index, race, number of human leukocyte antigen mismatches, and dialysis vintage; and donor type, age in quartiles, and sex. Age thresholds for quartiles of donor age are shown in Supplemental Digital Content 4 (https://links.lww.com/ALN/C657). The adjusted odds ratio of delayed graft function was 1.24 (95% CI, 0.85 to 1.81; P = 0.258) for the intermediate normal saline group and 1.55 (95% CI, 1.09 to 2.19; P = 0.014) for the high normal saline group compared with the low normal saline group (table 3).

The model fit was satisfactory (Hosmer–Lemeshow P = 0.576). The discriminative ability of our model was 0.791 (C-statistic). The likelihood ratio test revealed a significantly better fit of the model that included the normal saline variable as compared to the same model without the normal saline variable (P = 0.049).

Other covariates significantly associated with delayed graft function in our model were male recipients, higher recipients’ body mass index, black race compared with white, dialysis vintage, deceased donor compared with living donor, donor age, and male donor (table 3). We did not find significant interactions between the primary exposure variable and any covariate in the model. Specifically, there was no evidence for interaction between the normal saline groups and donor type, and between normal saline groups and total crystalloid volume.

In bivariate analysis, cold ischemia time was significantly associated with the primary exposure normal saline group and with the outcome delayed graft function and therefore a potential confounder (table 1 and Supplemental Digital Content 2, https://links.lww.com/ALN/C655). However, values for cold ischemia time were missing in 423 cases (16.8%). All missing values were from living donor transplantations between 2005 and 2012. After multiple imputation of cold ischemia time as described above, we fit a second logistic regression model that included imputed data for cold ischemia time. In the imputed model, the odds ratio of delayed graft function was higher with longer cold ischemia times, but this association was not statistically significant (Supplemental Digital Content 5, https://links.lww.com/ALN/C658). We compared the two logistic regression models with and without imputed cold ischemia time. This comparison revealed no relevant difference in point estimates and variances for the primary exposure and most covariates (Supplemental Digital Content 5, https://links.lww.com/ALN/C658). For simplicity, we therefore used the primary best-fit model without imputed cold ischemia time for all further analyses.

As defined in our analysis plan, we repeated our primary logistic regression model for the subgroups of deceased (n = 1,472) and living (n = 1,043) donor transplants separately. The odds ratio of delayed graft function was 1.52 (95% CI, 1.05 to 2.21; P = 0.028) for the high normal saline group compared with the low normal saline group for deceased donor transplants; and 1.66 (95% CI, 0.65 to 4.25; P = 0.287) for living donor transplants (Supplemental Digital Content 6, https://links.lww.com/ALN/C659). Within the full model, the interaction terms of the normal saline groups with donor type were nonsignificant (P = 0.733 and P = 0.808).

Adjusting for cubic splines of the propensity of treatment with high versus low normal saline, the odds ratio of delayed graft function was 1.33 (95% CI, 1.02 to 1.74; P = 0.036) for the group receiving 50% or more normal saline as compared with less than 50% normal saline.

This alternative logistic regression model regressed delayed graft function on the total amount of normal saline (no normal saline, 2 l or less normal saline, or more than 2 l normal saline). The odds ratio of delayed graft function was 1.48 (95% CI, 0.95 to 2.29; P = 0.080) for the administration of more than 2 l of normal saline as compared to the no normal saline group (Supplemental Digital Content 7, https://links.lww.com/ALN/C660).

Splitting the percent normal saline variable into five groups in intervals of 20%, 12.2% (307) transplant cases received less than 20% normal saline, 7.6% (190) received between 20% and less than 40%, 8.2% (206) received between 40% and less than 60%, 12.3% (310) received between 60% and less than 80%, and 59.7% (1,502) received 80% or more normal saline. The odds ratio of delayed graft function was 1.13 (95% CI, 0.65 to 1.97; P = 0.671) for the 20% to less than 40% normal saline group, 1.06 (95% CI, 0.63 to 1.80; P = 0.828) for the 40% to less than 60% group, 1.54 (95% CI, 0.96 to 2.46; P = 0.71) for the 60% to less than 80% group, and 1.61 (95% CI, 1.10 to 2.35; P = 0.014) for the group that received 80% or more normal saline as compared with the group that received less than 20%. Estimated point estimates of all other covariates were comparable to the primary model. The full model is shown in table 4.

In our cohort, 400 patients received a preemptive kidney transplant, i.e., did not require dialysis before transplant. Of those, 13 (3.3%) developed delayed graft function (four living donor recipients and nine deceased donor recipients). Using our primary logistic regression model excluding patients who were not on dialysis at the time of transplant (n = 2,113), the odds of delayed graft function were 56% (95% CI, 1.1 to 2.22; P = 0.014) higher for the high as compared with the low normal saline group. All covariates in the model showed point estimates and variances comparable to our primary model (Supplemental Digital Content 8, https://links.lww.com/ALN/C661).

In patients undergoing kidney transplantation, a high percentage of 80% or more normal saline administered perioperatively demonstrated a higher crude risk and a higher adjusted odds of delayed graft function. In our dataset, the low, intermediate, and high percent normal saline groups overlapped with the mean total volume of normal saline administered while the total volume of crystalloids was comparable between groups (table 2).

Delayed graft function occurs in about 20 to 40% of deceased donor kidney transplants and in about 5% of living donor kidney transplants in the United States.^18,21,22  It is highly relevant as an outcome in kidney transplant because it is associated with a higher risk of acute rejection, poorer long-term graft survival, and higher healthcare costs.^18,19,23 

The robustness of our results is supported by an a priori subgroup analysis and post hoc analyses. We did not find evidence for a heterogenous effect of the percent normal saline group on delayed graft function caused by the inclusion of deceased as well as living donor transplants or caused by the inclusion of patients not on dialysis before transplant. Additionally, the post hoc analyses based on propensity score regression and on alternative primary exposure variables showed comparable odds of delayed graft function with high percent or high absolute volume of normal saline as we have demonstrated with our primary model. Our post hoc regression model with an alternative primary exposure variable of percent normal saline categories with cutoffs at regular 20% intervals specifically addressed concerns of bias in our primary model because the choice of cutoffs for the percent normal saline categories at 30% and 80% was made after accessing the data (table 4). The very comparable results do not indicate bias based on the choice of the exposure categories. Overall, the similar results of our subgroup and post hoc analyses support the meaningfulness of our choice of primary exposure variable and statistical analysis.

Our findings are in line with the growing evidence of an unfavorable effect of normal saline on kidney function in specific populations. We confirmed the results of one previous study in kidney transplant recipients comparing normal saline with balanced crystalloids on graft function.¹⁵  Previous animal research and one volunteer study have shown that administration of chloride rich fluids such as normal saline reduces kidney perfusion and glomerular filtration rate, and prolongs time to micturition.^24,25  A large propensity-matched cohort study in patients undergoing major abdominal surgery found an almost fivefold higher need for postoperative dialysis in patients receiving normal saline compared with balanced crystalloids on the day of surgery.¹³  In 2018, the results from the Saline Against Lactated Ringer’s or Plasma-Lyte in the Emergency Department (SALT-ED) trial, a large pragmatic crossover trial, were reported.¹⁵  The authors investigated the effects of normal saline as compared with balanced crystalloids in noncritically ill adult surgical and medical patients admitted to the hospital via the emergency department. They found a statistically significant 0.9% higher absolute risk of their composite endpoint of major adverse kidney events in their normal saline group.

We report a markedly larger unadjusted absolute effect size of more than 7% for deceased donor transplants, which we attribute to the higher vulnerability of the transplanted kidney as compared with the native organs studied in the SALT-ED trial. In addition, the median crystalloid volume administered in the SALT-ED trial (1.1 l; interquartile range, 1 to 2) was roughly one third of what we found in our cohort (3 l; interquartile range, 2.2 to 3.6). Even though the percentages of normal saline administered during the study period are comparable between our high normal saline group (80% or more) and the normal saline group in the SALT-ED trial (93% adherence), the higher total crystalloid volume in our cohort translates into a higher absolute normal saline volume. That might also contribute to the larger effect size found in our study.

The recently published Saline or Lactated Ringer’s (SOLAR) trial assigned alternating cohorts of patients undergoing elective orthopedic or colorectal surgery to receive either lactated Ringer’s solution or normal saline.¹⁶  This trial did not find a meaningful difference in postoperative acute kidney injury between study groups. Even though these results seem to contradict an effect of normal saline on kidney function, the authors note that key differences of their trial compared to previous publications are the relatively healthy elective surgery patient population and the low amount of intraoperative fluid administration (median, 1.9 l, interquartile range, 1.3 to 2.6; compared with median, 3 l, interquartile range, 2.2 to 3.6 in our study). Considering these distinctly different cohorts, the SOLAR trial does not contradict our study results.

Our study also demonstrates that fluid management practice during kidney transplantation already changed over the years at our institution. In the time period from 2014 to 2015, only 27% of all kidney transplant patients received 80% or more normal saline, down from 75% from 2005 to 2007. This is in accordance with a national trend.¹² 

The marked increase in delayed graft function between 2005 and 2013 (Supplemental Digital Content 3, https://links.lww.com/ALN/C656) has previously been reported.²⁶  This is attributed to the increasing utilization of expanded criteria donors in an effort to meet the high demand of kidney donations.

In our primary model, several covariates show a significant association with the outcome delayed graft function. These findings are consistent with the transplant literature.^27,28  However, we cannot rule out additional confounding or interactions between these covariates and delayed graft function. Interpretation of these findings should therefore be cautious.

This observational study has limitations. It relies on three different databases. The purpose of the Scientific Registry of Transplant Recipients dataset is primarily research and quality assurance and includes mandatory data validation. On the other hand, both anesthesia databases used routinely collected clinical data extracted from the electronic anesthesia record. Data entry is human operator–dependent, and especially fluid administration data might be imprecise. There is no objective data check, validation, or control mechanism in place to validate the original data entries. However, errors in the underlying perioperative intravenous fluid documentation are likely to be similar across exposure groups and therefore unlikely to lead to a false association between normal saline load and delayed graft function.

The outcome variable delayed graft function, defined as the need for dialysis within 1 week after transplant, has limitations. Criteria and thresholds for initiation of dialysis can vary between centers and also between providers. Based on our clinical routine, we believe that the criteria for initiation of dialysis are relatively consistent within our center, and our outcome measurement is valid. However, varying clinical practice and criteria used to ascertain the outcome could impact the results. Our dataset did not include complete data on alternative outcome measures such as a decline in creatinine after transplant.

In addition, given the nearly sixfold variation in the incidence of delayed graft function across living versus deceased donor renal transplantation, the observations should be interpreted with caution in the living donor population. Although similar adjusted effect sizes were observed, the subgroup analysis of living donors did not demonstrate statistical significance.

The retrospective nature of this study limits the availability of data that might yet be of relevance. In our dataset, the variables “recipient diabetes,” “donor hypertension,” and “donor terminal creatinine” were highly missing but are described in the literature as strong predictors of the outcome.²⁷  As such, these variables would have qualified for inclusion in our regression model, even though we did not classify them as confounders. Inclusion of these variables might have modified the effect size but are unlikely to reverse the effect. Another potential predictor of the outcome is type and volume of fluid administration after discharge from the recovery room. However, our datasets only included fluid data for the perioperative period, defined from anesthesia start to discharge from the recovery unit.

In addition, these are single center data that might not reflect practices at other centers, reducing the generalizability of our observations. The type and volume of crystalloids administered perioperatively at our center might differ from fluid management practice elsewhere. Our results are not applicable to centers that use no normal saline or less total crystalloids during kidney transplantation. On the other hand, our dataset decreases variability in surgical technique since a consistent group of only eight kidney transplant surgeons performed all procedures from 2005 to 2015 with an equal distribution within every year. Variability in surgical technique, often a point of criticism in research investigating delayed graft function as the primary outcome, is therefore unlikely to affect the results.

In conclusion, our study demonstrated an association of high percentages of normal saline with delayed graft function in patients undergoing kidney transplantation.

The data reported here have been supplied by the Hennepin Healthcare Research Institute (Minneapolis, Minnesota) as the contractor for the Scientific Registry of Transplant Recipients (U.S. Department of Health and Human Services, Washington, D.C.). The interpretation and reporting of these data are the responsibility of the authors and in no way should be seen as an official policy of or interpretation by the Scientific Registry of Transplant Recipients or the U.S. Government.

This study was supported by an internal grant from the Department of Anesthesia and Perioperative Care, University of California, San Francisco (San Francisco, California).

The authors declare no competing interests.