Estimated Glomerular Filtration Rate and Postoperative Mortality in Patients Undergoing Non-Cardiac Surgery: A Single-Center Retrospective Study.

Introduction: There is limited evidence to clarify the specific relationship between preoperative estimated glomerular filtration rate (preop-eGFR) and postoperative thirty-day mortality in patients undergoing non-cardiac surgery. We aimed to investigate details of this relationship. Methods: We reanalyzed a retrospective analysis of the clinical records of 90,785 surgical patients at the Singapore General Hospital from January 1, 2012 to October 31, 2016. The main outcome was postoperative thirty-day mortality. Results: The average age of these recruited patients was 53.96 ± 16.88 years, of which approximately 51.64% were female. The mean of preop-eGFR distribution was 84.45 ± 38.56 mL/min/1.73 m 2 . Multivariate logistic regression analysis indicated that preop-eGFR was independently associated with thirty-day mortality (adjusted odds ratio: 0.992; 95% confidence interval [CI]: 0.990–0.995; P < 0.001). A U-shaped relationship was detected between preop-eGFR and thirty-day mortality with an inflection point of 98.688(P for log likelihood ratio test < 0.001). The effect sizes and confidence intervals on the right and left sides of the inflection point were 1.013 (1.007 to 1.019) [P < 0.0001] and 0.984 (0.981 to 0.987) [P < 0.0001], respectively. Preoperative comorbidities such as congestive heart failure (CHF), type 1 diabetes, ischemic heart disease (IHD),and anemia were associated with the odds ratio of preop-eGFR to thirty-day mortality (interaction P < 0.05). Discussion: The relationship between preop-eGFR and thirty-day mortality is U-shaped. The recommended preop-eGFR at which the rate of the thirty-day mortality was lowest was 98.688 mL/min/1.73 m 2 .


Introduction
The unmet global burden of surgical disease is enormous [1] With the popularity of surgery, it is especially important to optimize the safety of the operation [2]. Thirty-day mortality is one of the most important indicators of perioperative mortality (POMR) and can be used to effectively indicate the safety of surgery and the risk of postoperative complications [3]. Preoperative renal dysfunction is a ac-knowledgeable risk factor for postoperative mortality, and the risk of patients with moderate to severe kidney insufficiency increases dramatically [2,4]. Estimated glomerular filtration rate (eGFR), describing filtrate flow through the kidneys, is a universal surrogate indicator for assessing renal function. It has been widely used in the clinical diagnosis of chronic kidney disease (CKD) [5].
Studies have shown that preop-eGFR is a moderately effective predictor of thirty-day mortality in hospitalized surgical patients [4]. However, the current research population is mainly concentrated on patients who have undergone critical surgery such as cardiac surgery [6] [7][8] [9][10] [11][12][13][14], lack of other surgeries. The ethnicity of these studies is also rarely related to Asians. Moreover, the current studies mainly focus on the relationship between preop-eGFR and perioperative mortality in patients with renal insufficiency [7] [15], but several reports address high preop-eGFR levels or any other asymptomatic patient. Only Takashi Ui at.al made it clear among patients undergoing gastrointestinal malignancies that high preop-eGFR is associated with poor surgical outcomes, and also indicated a U-shaped relationship of preop-eGFR and thirty-day mortality [16] .However, data related to non-cardiac and non-neuron surgery are scarce.
Our study was designed to explore the details of relationship between preop-eGFR and thirty-day mortality in Asian patients undergoing non-cardiac surgery. Not only limited to patients with renal insufficiency, but also in patients with high preop-eGFR.

Data source
We downloaded the raw data from the DATADRYAD database (www.datadryad.org). As Diana Xin Hui Chan, et al. [20] have uploaded the original data and authorized the ownership to the website, we can perform secondary data analysis on this data to verify different scientific assumptions.(Dryad data package: Chan, Diana Xin Hui et al. (2018), Data from: Development of the Combined Assessment of Risk Encountered in Surgery (CARES) surgical risk calculator for prediction of post-surgical mortality and need for intensive care unit admission risk -a single-center retrospective study, Dryad, Dataset, https://doi.org/10.5061/dryad.v142481).

Study population
It is important to note that Chan, Diana Xin Hui et al. [20]completed data collection. They conducted a singlecenter retrospective study at Singapore General Hospital, a 1,700-bed tertiary academic hospital [20]. These clinical data of all surgical patients from January 1, 2012 to October 31, 2016 were obtained from the clinical information system (Sunrise Clinical Manager, Allscripts, Illinois, USA) and stored in SingHealth-IHiS Electronic Health Information System (eHINTS), a data storage and analysis system. Based on exclusion criteria, the study recruited a total of 79,609 surgical cases [20]. Exclusion standards: (1) patients with no surgery performed; (2)patients undergoing cardiac surgery, neurosurgery, transplantation and burn surgery; (3) pediatric cases;(4) cases under local anesthesia;(5) cases under cadaveric harvesting; (6) cases with missing important variables. As the personal information of patients was anonymous, no informed consent was required. Details of ethical license can be found in the data source article [20].

Measurement of preop-eGFR, thirty-day mortality and other covariates
Variables of the database file included: demographic information; preoperative comorbidities;preoperative laboratory perioperative blood transfusion data; anesthesia type ;priority of surgery ; surgical risk classification( based on the 2014 ESC/ESA guidelines on non-cardiac surgery [18] [19]);and postoperative thirty-day prognosis. Preoperative comorbidities were consist of anemia,CKD, and medical history [cerebrovascular accidents (CVA), ischemic heart disease (IHD), congestive heart failure (CHF),type1 diabetes].Those medical histories mentioned above were also associated with Revised Cardiac Risk Index (RCRI) score [17].The latest results of preoperative laboratory test were mainly about eGFR, red blood cell distribution width (RDW)(Levels above 15.7% were defined as high RDW ,as the normal reference range was 10.9-15.7% [22]), hemoglobin.We listed in detail the covariates used in this study. In short, their reasons for inclusion included as follows: (1) demographic data; (2) variables that can affect preop-eGFR or thirty-day mortality reported by previous literature [20]; (4) based on our clinical experiences. The most important outcomes was postoperative thirty-day mortality, including deaths from the date of surgery to one month later [3].As this is a retrospective study, reducing the possibility of selection bias and observation bias.

Statistical analysis
In data analysis, we represented continuous variables as median (quartile) (skewed distribution) or mean ± standard deviation (normal distribution), and categorical variables as a percentage or frequency. In the process of multivariate regression analysis, there are some confounders with partial missing data. If it is a categorical variable, the missing data would be treated directly as a new independent group; if it is a continues variable, the missing data would be replaced with an average or median value. We use Kruskal-Wallis H test (skewed distribution), one-way ANOVA (normal distribution) or χ2 (categorical variables)to calculate differences between different preop-eGFR group.
Our study supposed to figure out the specific relationship between preop-eGFR and postoperative thirty-day mortality (linear or non-linear), and then find out variables interfering with or modifying the relationship between them. After excluding the effects of these potential modifiers and confounders, the independent effect of preop-eGFR on postoperative thirty-day mortality can be determined.
Based on analytical principles mentioned above, univariate and multiple linear regression models were used to assess relationships with preop-eGFR and postoperative thirty-day mortality. Three models(an unadjusted model, a demographically adjusted model, and a fully-adjusted model) were constructed according to the STROBE statement [24].As for the fully-adjusted model, those adjusted variables, reported in previous studies, are related covariates that may affect preop-eGFR and/or thirty-day mortality [22] [20].Moreover, subgroup analyses were performed by stratified linear regression models. Subgroup interaction test was performed to verity the effect modification by subgroup and then a likelihood ratio test was conducted.
The following sensitivity analysis was performed in our research to ensure that the results of the data analysis were reliable. Firstly, the continuous variable preop-eGFR was converted into a categorical one by quartile to observe the possibility of nonlinearity. Secondly, if a nonlinear relationship exited between preop-eGFR and postoperative thirty-day mortality, it would be processed using a generalized additive model. Thirdly, we measured the threshold effect of preop-eGFR on the thirty-day mortality rate via a two-segment linear regression model based on a smoothing graph, determined the saturation of preop-eGFR by a recursive algorithm, and then detected the inflection point to obtain the maximum model likelihood. Based on the p-value of the log-likelihood ratio test, the best-fit model can be determined.

The selection of participants
The original data(N = 90,785) for this study was recruited according to exclusion standards by Diana Xin Hui Chan, at al [20]. The missing data on preop-eGFR and thirty mortality was about 10830 cases. And after excluding patients with preop-eGFR outliers[25](N = 346), 79 609 cases were included to our study(shown in Schedule1).

Univariate and multivariate analysis
Univariate analysis results were seen in Table 2.These results indicates that age, transfusion rate frequency, anemia, priority of surgery, surgical risk ,RDW and comorbidities(CVA, IHD, CHF, DM on insulin) was correlated with higher thirty-day mortality. We also find that there was no significant difference of thirty-day mortality with different race, whereas preop-eGFR and female were significantly associated with thirty-day mortality in a negative way. The results of univariate and multivariate linear regression models are shown in  In the sensitivity analysis, we converted preop-eGFR to a categorical variable classified by quartile to observe P of trend (Table 3).Compared to the reference group(Q1),the effect size of preop-eGFR on thirty-day mortality in group Q3 (preop-eGFR:96.68 ± 4.67)was the smallest among different adjusted model. For instance, in the fullyadjusted model, the effect size of group Q2,Q3 and Q4 were 0.520, 0.356, 0.616 respectively, compared with group(Q1).This kind of non-equidistant changes in effect size indicated a non-linear relationship between preop-eGFR with thirty-day mortality. Figure 1 shows the U-shaped non-linear correlation between preop-eGFR and postoperative thirty-day mortality. This non-liner relationship was verified by smooth curve of the generalized additive model. And the P-value of the log likelihood ratio test is less than 0.05 in Table 4,which further indicates the two-part linear regression model should be used to fit the relationship, rather than the linear regression model (through linearly fitting) .The inflection point was calculated to be 98.688 by a two-part linear regression model and a recursive algorithm. On the left side of the inflection point, preop-eGFR is one of the independently protective factors of thirty-day mortality (OR = 0.984, 95% CI: 0.981 to 0.987, P < 0.0001). While on the right side, it acts as a risk factor (OR = 1.013, 95% CI: 1.007 to 1.019, P < 0.0001).

The results of subgroup analyses and interaction analysis
As is shown in Table 5, the interaction test was significant for patients comorbidities, as DM, CHF, IHD or anemia (P = 0.0397, 0.0357, 0.0168, 0.0001 for interaction, respectively), while the interaction of other covariates was not statistically significant (interaction P value is greater than 0.05). Under the influence of preoperative comorbidities, such as CHF, DM and IHD, the effect of preop-eGFR on the thirty-day postoperative mortality was gradually increased. For patients with CHF, an increase of the preop-eGFR unit related to a 1.2% reduction in thirty-day mortality (0.988 (0.984, 0.992)). Without CHF, there's decreased by 0.5% for each additional unit of Preop-eGFR (0.995 (0.992, 0.998)). The same trend has also been seen in patients with DM (1.3% reduction of thirty-day mortality with DM vs. 0.5% reduction without DM) and IHD (0.9% with IHD vs. 0.4% without IHD). However, the opposite trend was seen in people with anemia. For patients without anemia, an increase unit of preop-eGFR would cost a 2.3% reduction in thirty-day mortality (0.977, 95%CI: 0.971, 0.984). For patients with anemia, the thirty-day mortality was decresaed by 0.5% with each additional unit of preop-eGFR.

Discussion
In this study, we confirmed an independent nonlinear relationship between preop-eGFR and postoperative thirtyday mortality. A stable U-shaped trend can be seen in this relationship. When preop-eGFR ≤ 98.688 mL/min/1.73 m 2 , the thirty-day mortality would be decreased by 1.6% for each additional unit of Preop-eGFR. While preop-eGFR > 98.688 mL/min/1.73 m 2 ,there would be a 1.3% increase of thirty-day mortality rate for each additional unit of preop-eGFR. CHF, DM, IHD and anemia complications (CHF, DM, IHD acting as a promoting factor, while anemia as an inhibitory factor) interfere with the effect of preop-eGFR on postoperative thirty-day mortality.
Studies has already been proved that preop-eGFR is a powerful and independent predictor of thirty-day morbidity risk after surgery [4,10,13,[26][27][15] [16,26]. Even in some literature reports, preop-eGFR is the strongest predictors of posttransplant survival [26]. Preop-eGFRs is a important indicator of many adverse surgical outcomes [16], as acute kidney injury, significantly related to higher mortality. At the same time, DM, IHD, CVF, and blood transfusion also are risk factors for poor postoperative prognosis [6]. The current research population is mainly concentrated in transplant [26], cardiac [10,13] and neuro [14] surgery. There is still a lack of research on other surgery. There are two articles that define the study population as non-cardiac surgery patients. Jacek B. Cywinski, at al evaluated 92,888 patients undergoing non-cardiac surgery, and confirmed preop-eGFR is a scientifically feasible predictor of postoperative thirty-day mortality [4].J. R. Prowle et al. reported that significantly increases the risk of death after non-cardiac surgery, according to the data of 36 779 cases [2].
Previous studies mostly focused on the patients of renal insufficiency to verify the important regulatory role of preop-eGFR [12][13]15]. High preop-eGFR levels have also been connected with greater mortality among nonsurgical patients indicating a potential U-shaped association of preop-eGFR with poor prognosis [28][29][30]. A recent study revealed the association of the specific trend between preop-eGFR and thirty-day mortality in patients undergoing surgery for gastrointestinal malignancies, without clarifying the inflection point [16]. In addition, the current research population is mainly Europeans and Americans, and rarely Asians.
To our best knowledge, it is the first time that the specific U-shaped relationship between preop-eGFR and postoperative thirty-day mortality has been clearly identified in Asian patients undergoing non-cardiac and nonneuro surgery, ranging from minor day cases to major surgeries.
Strengths of our study are mentioned as follows: firstly, the generalized additive model was used to evaluate non-linear relations, instead of using the generalized linear model to illustrate the linear relationship only. Secondly, as an observational study, there were some unavoidable potential confounders included in this study.
In order to minimize residual confounding, strict statistical adjustment was performed. What's more, effect modifier factor analysis makes the use of data better. Sensitivity analysis was performed of these data to ensure reliability.
The findings of this study should be helpful for reducing the risk of postoperative death. The preop-eGFR at which the rate of the perioperative Mortality was lowest was 98.688. It suggests that regulation of preop-eGFR can effectively reduce perioperative mortality, especially with CHF, DM, IHD comorbidities. While comorbid with anemia, it becomes the same important to control anemia for reducing mortality.
This study has several acknowledged limitations. First, as for our study is a secondary analysis based on the published data, we cannot exclude some residual and/or unmeasured confounders (such as socioeconomic factors and inflammatory markers), that may bias the estimated relationship. Secondly, the study population, which only included Asian patients, can be further expanded to conduct multi-center research to increase the reliability of the data. Our choice of outcomes and variables is also limited. We could not investigate the relationship between preop-eGFR with long-term outcomes. What's more, when it comes to high preop-eGFR, the results would be much more accurate formula based on cystatin C, instead of basing on creatinine. However, cystatin C haven't be widely used in clinical practice right now [31].

Conclusion
In patients undergoing non-cardiac and non-neurological surgery, the level of preop-eGFR is associated with operative adverse events in a U-shape trend. The preop-eGFR with the lowest perioperative mortality was 98.688. CHF, DM, IHD and anemia comorbidities (CHF, DM, IHD as a promoting factor, anemia as an inhibitor) interfere with the effect of preop-eGFR on postoperative thirty-day mortality.

Ethics approval and consent to participate
In the previously published article, Diana Xin Hui Chan, et al. has clearly stated that: the study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all Participants. Since the study was based on a secondary analysis of past data and the patient's personal information in the original data was anonymous, there was no need for informed consent from the participants. The ethical license has been elaborated in the published paper. Figure 1 Relationship of e-GFR and thirty-day mortality