Impact of Prior to Intensive Care Unit Statin use on Patients with Acute Kidney Injury

Purpose: To evaluate if prior to intensive care unit (ICU) statin use improve the clinical outcomes, for critically ill patients with acute kidney injury (AKI). Materials and Methods: Patients with AKI were selected from the Medical Information Mart for Intensive Care IV v1.0 database for this retrospective observational study. The primary outcome was 30-day ICU mortality. 30-day in-hospital mortality and ICU length of stay (LOS) were considered as secondary outcomes. Comparison of mortality between pre-ICU statin users with non-users was conducted by multivariable cox proportional hazards model. Comparison of ICU LOS between two groups was implemented by multivariable linear model. Three propensity score methods were used to verify the results as sensitivity analyses. Stratication analyses were conducted to explore whether the association between pre-ICU statin use and mortality differed across various subgroups classied by sex and different AKI stages. Results: 3821 pre-ICU statin users and 9690 non-users were identied. In multivariable model, pre-ICU statin use was associated with reduced 30-day ICU mortality rate [Hazara ratio (HR) 0.68 (0.59,0.79); P<0.001], 30-day in-hospital mortality rate [HR 0.64 (0.57, 0.72); P<0.001] and ICU LOS [Mean Difference -0.51(-0.79, -0.24); P<0.001]. The conclusions were consistent in three propensity score methods. In Subgroup analyses, pre-ICU statin use was associated with decreased 30-day ICU mortality and 30-day in-hospital mortality in both sexes and AKI stages, only except for 30-day ICU mortality in AKI stage 1. Conclusions: Patients with AKI who were administered statins prior to ICU admission might have lower mortality rate during ICU or hospital stay and shorter ICU LOS. for IV; SQL, Structured Query Language; SD, standard deviation; IQR, interquartile range; HR, hazard ratio; PSM, propensity score matching; Logistic-OW, overlap weighting with logistic regression; GBM-OW, overlap weighting with generalized boosted models; KS, Kolmogorov-Smirnov statistics; ASMD, absolute standardized mean difference.


Introduction
Due to the abrupt decline in kidney function, including reversible or irreversible, acute kidney injury (AKI) can lead to retention of metabolic waste products within a short time [1]. Patients with AKI experience water and sodium retention, oliguria, or even anuria, hyperkalemia, metabolic acidosis, acute pulmonary edema, cerebral edema, and other complications. Owing to multiple etiologies, AKI is common in hospitalized patients and intensive care unit (ICU) patients. Patients with multiple risk factors, such as sepsis, surgery, shock, diabetes, hypertension, heart failure, advanced age, use of contrast agents, and nephrotoxic drugs, and those critically ill in the ICU often have a higher prevalence of AKI and increased mortality rates [2]. Approximately 20% of critically ill patients develop AKI in hospital, and approximately 10% of them eventually require renal replacement therapy (RRT). Mortality rates of patients with AKI range from 16-50%, depending on AKI stage [3]. AKI progresses rapidly, and given the current lack of speci c pharmacological treatments, current treatment guidelines focus on supportive care and dialysis [1].
However, statins have a protective effect against AKI according to previous animal studies [4,5]. By reducing serum cholesterol levels, statins reduce the risk of cardiovascular death [6]. Furthermore, the "pleiotropic effects" [7] of statins, including anti-in ammatory, antithrombotic, as well as immunomodulatory effects [8,9] and so on, reported to reduce the incidence of AKI caused by multiple etiologies, such as surgery [10,11], contrast agent use [12], and sepsis [13]. There is con icting evidence on the pre-admission statin use among patients with AKI, while some studies reported an association with improved clinical outcomes [14,15], others reported no effects [16][17][18]. Given this inconsistent data, further research is required to investigate this association. Herein, we sought to examine the following clinical outcomes in pre-ICU statin users and non-users in a large sample of ICU patients with AKI: (1)

Study population
Patients in the database meeting the criteria below were selected into this study: (1) rst ICU admission of rst hospitalization; (2) ICU LOS ≥ 2 days; (3) age 18 and above; (4) had AKI according to the Kidney Disease Improving Global Outcomes (KDIGO) criteria [20]. Patients with statin prescription records after admission to the ICU and none prior to admission were excluded. The process of the cohort selection was showed in Fig. 1. There are 69211 records of ICU admissions in MIMIC-IV v1.0 database, of which we selected 53150 patients with rst ICU admission on the rst hospitalization. After excluding 28301 patients with ICU LOS < 2 days, 6990 patients with a statin prescription record after admission to the ICU but none prior to admission, and 4348 patients without AKI, we obtained 3821 pre-ICU statin users and 9690 non-users, a total of 13511 patients as nal study sample. Data extraction was performed using Structured Query Language (SQL) in PostgreSQL (v13.0). The SQL script codes used to extract patients with AKI were obtained based on the SQL script retrieved from the GitHub website (https://github.com/MIT-LCP/mimic-iv).

Medication exposure
We de ned pre-ICU statin users as patients with records of statins, including the brand and generic names of medicines, otherwise as non-users. The type of statin was referred to other authors' study based on MIMIC-III database [21], it is also applicable to MIMIC-IV database through careful veri cation. The prescriptions of the medicines were recorded in the MIMIC-IV v1.0 "PRESCRIPTION" le.

Outcome measures
The primary outcome was 30-day ICU mortality, determined by using three variables in the MIMIC-IV: the "DOD" (date of death) variable in the "PATIENTS" le, and "INTIME" (ICU admission time) and "OUTTIME" (ICU discharge time) variables in the "ICUSTAYS" le. Patients with records of duration from "INTIME" to "DOD" within 30 days and "DOD" between "INTIME" and "OUTTIME" were considered as having a primary endpoint. Other patients were de ned as censors, and the time of censoring was chosen as a minimum of 30 days and ICU LOS. We also conducted an analysis of 30-day ICU mortality excluding patients whose "DOD" was later than the "OUTTIME" as sensitivity analysis, the result is shown in supplement.
The two secondary outcomes were 30-day in-hospital mortality and ICU LOS. Since the "DOD" was only the time of death in-hospital, 30-day in-hospital mortality was de ned as survival time from "INTIME" to "DOD" within 30 days. The censor time was chosen as a minimum of 30 days and hospital LOS. Hospital LOS was calculated based on the time between "INTIME" from the "ICUSTAYS" le and "DISCHTIME" (hospital discharge time) from the "ADMISSIONS" le. ICU LOS was calculated based on the time between "INTIME" and "OUTTIME."

Covariates
In this study, we extracted 28 covariates including sex; age; admission type; AKI stage; 8 comorbidities, 3 in-hospital procedures, 3 severity of illness scores, 3 vital signs and 7 laboratory indexes. All the covariates are listed in the rst column of Table 1. Abbreviations: ASMD, absolute standardized mean difference; SD, standard deviation; AKI, acute kidney injury; CHF, congestive heart failure; CBVD, cerebrovascular disease; COPD, chronic pulmonary disease; MI, myocardial infarct; RRT, renal replacement therapy; CCI, Charlson comorbidity index; SOFA, sequential organ failure assessment; SAPS, simpli ed acute physiology score; MBP, mean blood pressure; WBC, white blood cell; bpm, bytes per minute.

Statistical analysis
Since there were missing values in the extracted dataset, R package "MICE" [22] was used to process missing values by multiple imputation initially. The missing ratio of each variable is shown in Table 1 Strati cation analyses were conducted to explore whether the association between pre-ICU statin use and mortality differed across various subgroups classi ed by sex and different AKI stages.
In addition to multivariable adjustment analysis, we also used propensity score matching (PSM), overlap weighting with logistic regression (Logistic-OW), and overlap weighting with generalized boosted models (GBM-OW) methods to analyze the outcomes for sensitivity analyses. PSM was performed by nearestneighbor matching using a caliper with 0.05 SD of the logit of the estimated PS value. Patients were matched in a 1:1 ratio, such that each patient with pre-ICU statin use was matched to one patient without.
In Logistic-OW, we used logistic regression to estimate the PS value as in PSM. The weight of each patient without pre-ICU statin use was equal to its PS value, and the weight of each patient with pre-ICU statin use was equal to 1 minus its PS value, called overlap weighting, which focuses on the population where patients in two groups have the most similar characteristics [23]. In GBM-OW, we used GBM [24] to estimate PS values. PS values were estimated by GBM iteration, implemented by the R package "GBM" [25]. To assess weight quality, we compared the maximum Kolmogorov-Smirnov statistics (KS.max) of all covariates among iterations. We choose the iteration with the minimum KS.max indicating the best balance of all iterations to obtain the PS value and then calculated the overlap weights as Logistic-OW [26].Absolute standardized mean differences (ASMDs) were calculated to evaluate the e ciency of PS matching and weighting in reducing the differences between groups. We considered the covariate as a balance, as its ASMD was < 0.1 [27]. After PS adjustment, the outcomes were further analyzed using univariate analysis, causing all covariates to be balanced. All covariates with their main effects were included in multivariable analysis, PSM, and Logistic-OW. All covariates were included in the GBM-OW model without the need to set their relationship because the GBM model could automatically estimate the nonlinear relationship between covariates.
We used R software (version 4.0.3) and SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) to conduct all statistical analyses. A p value was considered statistically signi cant at P < 0.05.

Baseline characteristics
The baseline characteristics of the two groups are shown in Table 1. Most of the covariates have signi cant differences between two groups. Compared with non-users, pre-ICU statin users were more likely to be male, elderly, admitted non-emergency, and AKI stage 1 or 2. All comorbidities, except cancer, were more common in pre-ICU statin users. RRT, ventilation, and vasopressin use were more common among non-users than users.
Supplemental le Table S1-S3 summarize the characteristics of the study population after PS adjustment. As mentioned above, most covariates were imbalanced between the two groups before PS adjustment. Nevertheless, after PS matching or weighting, the ASMDs were < 0.10 for all covariates (Fig. 2), implying that after PS adjustment, the included covariates were balanced across two groups. Note that if PS values were estimated from a logistic regression model, any included covariates will achieve exact balance after overlap weighting, such that the ASMDs of all included covariates after Logistic-OW adjustment are close to 0 [23], which is an interesting and meaningful feature of overlap weighting.    (Table 2). Consistent results were obtained with sensitivity analyses (Table 3).

Subgroup analyses
As shown in Fig. 3, pre-ICU statin use was associated with improved 30-day ICU mortality in both sexes, as well as in patients with AKI stages 2 and 3. Statin use was associated with improved 30-day inhospital mortality in men and women, as well as in patients with AKI stages 1 to 3. The p value for the interaction suggested no sex differences in the association of pre-ICU stain use with mortality, but it did differ with AKI stage.  [21]. Our study was based on ICU patients with AKI, similar but different population with Chinaeke's, also showed an association between pre-ICU statin use and outcomes. The present study further illustrates the pleiotropic effect of statins in critically ill patients. Although the pathophysiological mechanisms are not exactly the same, both AKI and sepsis are common in ICU patients [28]. Many critically ill patients simultaneously have both, namely septic AKI. Previous treatment guidelines for sepsis and AKI focused more on antibiotics, aggressive uidbased therapy, vasoactive drugs [28]. However, these two studies provided a reference for the treatment of AKI and sepsis, which could be further veri ed in future randomized controlled trials to explore the optimal dose and type of statins.
Wu et al. showed that statin use reduced risks of 1-year and in-hospital mortality in 6091 hospitalized patients with dialysis-requiring AKI [14], and Li et al. showed that statin use reduced the occurrence of AKI and AKI-related mortality among patients undergone cardiac surgery [15]. These two studies were conducted on speci c patients with AKI. AKI can occur for a variety of reasons in critically ill patients, and patients with AKI have a variety of manifestations and comorbidities. Therefore, we cannot exclude the possibility that the association of outcomes with statin use in patients with AKI is due to other factors.
Our study included all patients with AKI in the MIMIC-IV database, regardless of etiology, manifestations, comorbidities, and AKI severity. Therefore, we cannot con rm whether the association is indirect, while the exact mechanism by which statins affect AKI remains obscure. However, the present study did show an association of positive outcomes with pre-ICU statin use in a complete population with AKI, not just in a speci c group.
Statins have pleiotropic effects. Although the exact mechanism behind the effect of statins in AKI patients is not clear, some animal studies may provide clues. A rat study indicated that atorvastatin use could reduce endoplasmic reticulum stress and apoptosis [29], and another demonstrated that pravastatin reduced urinary protein excretion and retained the renal function and expression of nephrin in doxorubicin-induced nephropathy rats, concluding that pravastatin protects from and treats adriamycininduced renal injury [30]. These studies suggest different mechanisms of action for statins in AKI, but this is a question which deserves further research.
Since this is an observational study, many variables were imbalance between the two groups, we utilized PS matching and weighting approaches, which ensure that patients were pseudo-randomized across two groups, as in a typical randomized controlled trial. Considering the possible interactions between variables, we also used the GBM model with PS weighting. GBM models can automatically nd the relationships between covariates, such that we did not need to set the interaction between covariates. By PS methods, the conclusions were consistent with the multivariable model, proving that the results were robust.
Our study has some limitations: (1) The study is a retrospective observational study using existing data and not randomized. Although we extracted some related covariates and conducted three sensitivity analyses with PS methods, unobserved confounders may still exist that could lead to bias in the results.
(2) Some individuals may have non-recorded pre-ICU statin use; which cannot be checked. (3) Some covariates in the data were missing, and multiple imputations were used for the missing values. This might have led to different results. For this reason, we conducted an analysis using the complete data set without missing values, obtaining consistent results (supplemental le Table S4). (4) We did not conduct a long-term effect analysis because the database did not have complete long-term follow-up, so we focused on 30-day mortality.

Conclusions
Through analysis, we found that pre-ICU statin users had reduced 30-day ICU mortality, 30-day in-hospital mortality, and ICU LOS compared to none users, suggest that pre-ICU statin use can improve the clinical outcomes of critically ill patients with AKI. The conclusion was consistent in multivariable model and PS matching and weighting model.  The ow chart of the cohort selection process. ICU, intensive care unit; MIMIC-IV, Medical Information Mart for Intensive Care IV; LOS, length of stay; AKI, acute kidney injury.