
==== Front
Ren Fail
Ren Fail
Renal Failure
0886-022X
1525-6049
Taylor & Francis

39311647
10.1080/0886022X.2024.2404237
2404237
Version of Record
Research Article
Clinical Study
SACrA score to predict the initiation of renal replacement therapy in critically ill patients: a single-center retrospective study
G. Suzuki et al.
Suzuki Ginga
Nishioka Saria
Kobori Toshimitsu
Masuyama Yuka
Yamamoto Saki
Serizawa Hibiki
Nakamichi Yoshimi
Honda Mitsuru
Critical Care Center, Toho University Omori Medical Center, Tokyo, Japan
Supplemental data for this article can be accessed online at https://doi.org/10.1080/0886022X.2024.2404237.

CONTACT Ginga Suzuki ginga.suzuki@med.toho-u.ac.jp Critical Care Center, Toho University Omori Medical Center, 6-11-1, Omori Nishi, Ota-ku, Tokyo, Japan
23 9 2024
2024
23 9 2024
46 2 240423713 7 2024
8 9 2024
9 9 2024
KnowledgeWorks Global Ltd.23 9 2024
published online in a building issue23 9 2024
© 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
2024
The Author(s)
https://creativecommons.org/licenses/by-nc/4.0/ This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which this article has been published allow the posting of the Accepted Manuscript in a repository by the author(s) or with their consent.

Abstract

Background

Acute kidney injury (AKI) is a prevalent complication in critically ill patients that affects the timing of renal replacement therapy (RRT) initiation. This study aimed to develop and validate the SACrA score for predicting non-emergent initiations (BUN ≥112 mg/dL or oliguria for >72 h) of RRT in critically ill patients.

Methods

We conducted a retrospective cohort study using data from two cohorts. The derivation cohort included patients admitted to the ICU between November 2021 and December 2023, whereas the validation cohort included patients admitted between September 2019 and October 2021. The primary outcome was non-emergent RRT initiation. The multivariate logistic regression with stepwise selection based on the Akaike information criterion finalized the model, including the variables, such as sex, albumin (Alb), creatinine (Cr), and APACHE II score (SACrA).

Results

The derivation and validation cohorts comprised 470 and 476 patients, respectively. The SACrA score showed a strong predictive performance for non-emergent RRT initiation in both the cohorts. Cohort 1 had an ROC–AUC of 0.971, with a calibration slope of 0.982 and an intercept of 0.009, whereas cohort 2 had an ROC–AUC of 0.918, with a calibration slope of 0.988 and an intercept of 0.004.

Conclusions

The SACrA score is a robust tool for predicting non-emergent RRT initiation in critically ill patients using readily available clinical variables. Though additional data are needed to validate the SACrA score, our analysis suggests the tool may help clinicians make informed decisions, reduce unnecessary RRT, and thereby improve patient outcomes.

Keywords

Renal replacement therapy
AKI
acute kidney injury
critical care
intensive care
This study received no specific grants from any funding agency in the public, commercial, or not-for-profit sectors.
==== Body
pmcIntroduction

Acute kidney injury (AKI) is a common complication in patients admitted to the intensive care unit (ICU) and is associated with poor outcomes [1–3]. Renal replacement therapy (RRT) is the primary supportive therapy for severe AKI; however, the optimal timing for initiating RRT remains unclear. Recent studies have shown no clear benefits of early initiation of RRT in AKI [4–7]. A recent randomized controlled trial (RCT) compared a non-emergent strategy (initiating RRT at blood urea nitrogen (BUN) ≥112 mg/dL or oliguria lasting more than 72 h) to a more delayed strategy [8]. A delayed strategy is associated with the potential risk of increased 60-day mortality. Therefore, initiating RRT at non-emergent situations while avoiding RRT in patients who spontaneously recover from AKI is currently a reasonable strategy. Despite this, the decision to avoid RRT still relies on clinical judgment, and there is a lack of useful tools [9–10]. To date, models have only been developed to predict actual RRT initiation based on clinical judgment [11–13]; no models exist to predict non-emergent initiation.

This study aimed to develop a predictive model for non-emergent initiation of RRT using a cohort in which non-emergent initiation was adopted as the clinical protocol. The results of this study could provide clinicians with evidence to avoid unnecessary RRT and to propose new RRT initiation criteria.

Methods

Design and setting

This single-center retrospective study was conducted at the Toho University Omori Medical Center, a university-affiliated tertiary emergency medical care facility in Tokyo, Japan. This study was conducted in accordance with the principles of the 1975 Declaration of Helsinki. The study protocol was approved by the Ethics Committee of the Toho University Omori Medical Center (approval number: M23256). The requirement for obtaining written informed consent from patients was waived because of the retrospective nature of the study.

Objective

This study aimed to develop a predictive model for non-emergent initiation of RRT using a cohort in which non-emergent initiation was adopted as a clinical protocol. Additionally, the model was validated using a different cohort.

Participants

This study included two cohorts of patients admitted to the ICU.

Cohort 1 (derivation cohort)

Patients admitted to the ICU from the emergency department (ED) with a total hospital stay >7 days between November 2021 and December 2023. During this period, a non-emergent RRT initiation protocol was adopted as part of the standard ICU care. Although this protocol was implemented, not all patients followed it strictly. However, most patients in cohort 1 underwent delayed initiation, making this cohort ideal for developing a predictive model for non-emergent RRT initiation. To develop the model, the cohort was divided into the following two subgroups.

Indication group: Patients who met the non-emergent RRT initiation criteria.

No-indication group: Patients who did not meet the non-emergent RRT initiation criteria.

Cohort 2 (validation cohort)

Patients admitted to the ICU from the ED with a total hospital stay >7 days between September 2019 and October 2021. During this period, the decision to initiate RRT was at the discretion of the attending physician, considering creatinine (Cr) levels, urine volume, and other blood data. This cohort was used to validate the predictive model developed for cohort 1. For validation, cohort 2 was divided into the following groups:

Indication group: Patients who met the non-emergent RRT initiation criteria.

No-indication group: Patients who did not meet the non-emergent RRT initiation criteria.

Non-emergent initiation of RRT protocol

A non-emergent strategy was adopted based on a recent large-scale RCT [8]. From November 2021 (cohort 1), RRT was initiated when BUN ≥112 mg/dL or oliguria <500 mL/day persisted for more than 72 h (defined as non-emergent initiation). The initiation of RRT was at the discretion of the clinician. The RRT mode (continuous renal replacement therapy (CRRT) or intermittent renal replacement therapy (IRRT)) and membrane selection were determined by the attending physician.

Exclusion criteria

For both the cohorts, the exclusion criteria were as follows: patients under 18 years of age, early discharge (within seven days), early death (within seven days), maintenance dialysis, post-kidney transplant, postrenal cause, urgent indication for RRT (serum potassium ≥5.5 mEq/L despite medical therapy, pH ≤7.15 at any time, pulmonary edema unmanageable by diuretics requiring FiO2 ≥ 50% to maintain SpO2 ≥ 95%) [4–8], patients who already met the non-emergent initiation (described below) at admission, gastrointestinal bleeding (as the BUN levels would be elevated), RRT required for treatment of drug toxicity or liver failure, patients requiring venoarterial extracorporeal membrane oxygenation (VA-ECMO), patients with do not attempt resuscitation (DNAR) orders at admission, and missing data.

Data extraction

The following data were extracted from medical records: age, sex, body mass index (BMI), cause of admission, Charlson Comorbidity Index (CCI) [14], medication use (angiotensin-converting enzyme inhibitors (ACE-I) or angiotensin receptor blockers (ARBs), diuretics, non-steroidal anti-inflammatory drugs (NSAIDs)) [15–17], use of contrast agents at admission [15], creatine kinase (CK), albumin (Alb), BUN, Cr, lactate (Lac), and Acute Physiology and Chronic Health Evaluation (APACHE) II score. We extracted the initial vital signs and laboratory data obtained from the ED. For patients whose conditions deteriorated during their hospital stay, we utilized the vital signs and laboratory data obtained immediately before their transfer to the ICU.

Outcome

The primary outcome was the non-emergent initiation (BUN ≥112 mg/dL or oliguria <500 mL/day for more than 72 h). The secondary outcomes were RRT initiation, ICU length of stay (ICU LOS), 60-day mortality, and 90-day mortality.

Statistical analysis

Cohorts 1 and 2 were divided into two groups based on whether they met the non-emergent initiation criteria (indication group) or not (no-indication group) during their hospital stay. The factors associated with non-emergent initiation were compared between the two groups. Cohort 1 was used to develop the predictive model and cohort 2 was used for external validation.

Continuous variables were tested for normality using the Kolmogorov–Smirnov test. Non-normally distributed variables were presented as medians and interquartile ranges (IQRs) and compared using the Mann–Whitney U-test. Normally distributed variables were presented as means and standard deviations (SDs) and compared using t-tests. Categorical variables are presented as percentages and were compared using the Chi-square test or Fisher’s exact test.

The candidate factors for the non-emergent initiation predictive model included age, sex, CCI, APACHE II score, and variables from previous studies (CK, Alb, BUN, and Cr) [3,18–20]. Multivariate logistic regression analysis was performed and the stepwise selection method, based on the Akaike information criterion (AIC), was used to select the final model [21]. The variables in the model were evaluated for multicollinearity using the variance inflation factor (VIF), and variables with VIF >10 were excluded [21].

For clinical use, each variable in the predictive model was scored based on its beta coefficient, which was divided by the smallest coefficient and converted to the nearest integer. The predictive score for each patient was calculated and the discriminative ability of the model was assessed using receiver operating characteristic (ROC) curves and the area under the curve (AUC). The optimism-adjusted AUC was evaluated using bootstrap resampling with 1000 iterations, and calibration was performed using the bootstrap method. The cutoff value was determined from the ROC curve using the Youden index.

The scores of the prediction model developed in cohort 1 were calculated for each patient in cohort 2, and the performance of the model was verified using AUROC and calibration plot. Both the tests were performed using the bootstrap method. The number of RRT initiations and 60-day mortality in the no-indication group were compared between cohorts 1 and 2 to evaluate the validity of the clinical protocol. We have also provided a breakdown of the RRT initiation criteria for patients in each cohort who underwent RRT.

No prior sample size calculations were performed. However, the statistical power was examined in this study. Referring to a report on the calculation of the sample size for clinical prediction models [22], the necessary sample size was calculated using the pmsampsize package in R (R Foundation for Statistical Computing, Vienna, Austria). The statistical analyses were performed using R version 4.2.0; statistical significance was set at p < .05.

Language editing

The manuscript has been proofread and edited for English by ChatGPT, an AI language model developed by OpenAI, to ensure clarity and coherence.

Results

A total of 1148 and 1198 patients were included in cohorts 1 and 2, respectively. After applying the exclusion criteria, 678 and 722 patients were excluded from the two cohorts, respectively. The major reasons for exclusion were as follows: early discharge: 337 and 349, early death: 96 and 143, urgent initiation of RRT: 68 and 46, and DNAR: 48 and 65 from cohorts 1 and 2, respectively. Finally, 470 (indication group, 30; no-indication group, 440) and 476 (indication group, 29; no-indication group, 447) patients in cohorts 1 and 2, respectively, were included in the analysis. The patient flow diagram (Figure 1) provides a detailed overview of the inclusion and exclusion processes.

Figure 1. Patient flow diagram. This diagram illustrates the inclusion and exclusion processes for patients in cohort 1 (derivation cohort) and cohort 2 (validation cohort). A total of 1143 and 1195 patients were initially screened in cohorts 1 and 2, respectively. After applying the exclusion criteria, 673 and 719 patients were excluded. The final analysis included 470 patients in cohort 1 (30 in the indication group and 440 in the no-indication group) and 476 patients in cohort 2 (29 in the indication group and 447 in the no-indication group). ICU, intensive care unit; ED, emergency department; RRT, renal replacement therapy; VA-ECMO, venoarterial extracorporeal membrane oxygenation.

The baseline characteristics and outcomes of the enrolled patients are summarized in Table 1. The key characteristics included were as follows. Age: in cohort 1, the median age was 68.0 (IQR: 58.0–76.0) years in the indication group and 68.0 (IQR: 52.5–77.0) years in the no-indication group (p = .60). In cohort 2, the median age was 70.0 (IQR: 58.3–79.0) years in the indication group and 65.0 (IQR: 51.0–77.0) years in the no-indication group (p = .07). Sex: in cohort 1, 63.3% and 65.9% of the patients in the indication and no-indication groups were males, respectively (p = .77). In cohort 2, 69.0% and 62.4% of the patients in the indication and no-indication groups were males, respectively (p = .48). In cohort 1, the median CCI was 2.5 (IQR: 2.0–3.0) in the indication group and 1.0 (IQR: 0.0–2.0) in the no-indication group (p < .01). In cohort 2, the median CCI was 3.0 (IQR: 1.8–4.0) in the indication group and 1.0 (IQR: 0.0–2.0) in the no-indication group (p < .01). Other significant differences were observed in CK, Alb, BUN, Cr, and APACHE II scores. These differences were consistent across the cohorts.

Table 1. Baseline characteristics and outcomes.

Parameters	Cohort 1	p Value	Cohort 2	p Value	
Indication group (N = 30)	No-indication group (N = 440)	Indication group (N = 29)	No-indication group (N = 447)	
Age, years	68.0 (58.0–76.0)	68.0 (52.5–77.0)	.60	70.0 (58.3–79.0)	65.0 (51.0–77.0)	.07	
Sex (male), n (%)	19 (63.3%)	290 (65.9%)	.77	20 (69.0%)	279 (62.4%)	.48	
Body mass index, kg/m2	21.2 (19.3–24.8)	21.2 (18.8–24.4)	.79	23.6 (20.8–25.4)	21.7 (18.8–24.7)	.06	
Primary disease	 	 	.61	 	 	.11	
 Cardiovascular, n (%)	4 (13.3%)	55 (12.5%)	 	11 (37.9%)	65 (14.5%)	 	
 Respiratory, n (%)	5 (16.7%)	75 (17.0%)	 	2 (6.9%)	80 (17.9%)	 	
 Digestive, n (%)	1 (3.3%)	24 (5.5%)	 	0 (0%)	3 (0.7%)	 	
 Neurological, n (%)	0 (0%)	64 (14.5%)	 	0 (0%)	78 (17.4%)	 	
 Metabolic, n (%)	0 (0%)	17 (3.9%)	 	0 (0%)	14 (3.1%)	 	
 Sepsis, n (%)	15 (50.0%)	40 (9.1%)	 	11 (37.9%)	35 (7.8%)	 	
 Abnormal body temperature, n (%)	1 (3.3%)	8 (1.8%)	 	1 (3.4%)	14 (3.1%)	 	
 Cardiopulmonary arrest, n (%)	2 (6.7%)	20 (4.5%)	 	1 (3.4%)	14 (3.1%)	 	
 Trauma, n (%)	1 (3.3%)	83 (18.9%)	 	0 (0%)	86 (19.2%)	 	
 Other, n (%)	1 (3.3%)	54 (12.3%)	 	3 (10.3%)	58 (13.0%)	 	
Charlson Comorbidity Index	2.5 (2.0–3.0)	1.0 (0.0–2.0)	<.01	3.0 (1.8–4.0)	1.0 (0.0–2.0)	<.01	
ACE-I or ARB, n (%)	3 (10.0%)	54 (12.3%)	1.00	5 (17.2%)	46 (10.3%)	.22	
Diuretics, n (%)	1 (3.3%)	13 (3.0%)	.61	3 (10.3%)	15 (3.4%)	.09	
NSAIDs, n (%)	2 (6.7%)	46 (10.5%)	.76	2 (6.9%)	29 (6.5%)	1.00	
Contrast agent use, n (%)	20 (66.7%)	238 (54.1%)	.18	13 (46.4%)	217 (48.5%)	.83	
CK, mg/dL	350.5 (160.0–851.0)	144.5 (74.0–340.0)	<.01	304.5 (149.5–995.5)	138.0 (73.3–339.5)	.02	
Alb, mg/dL	2.4 (1.8–3.0)	3.7 (3.0–4.1)	<.01	2.9 (2.0–3.5)	3.6 (3.1–4.1)	<.01	
BUN, mg/dL	52.5 (34.0–72.0)	17.0 (13.0–26.0)	<.01	39.5 (26.5–72.0)	18.0 (13.0–26.0)	<.01	
Cr, mg/dL	4.0 (2.1–5.4)	0.9 (0.7–1.2)	<.01	2.9 (1.8–4.9)	0.9 (0.7–1.2)	<.01	
Lactate, mmol/L	3.8 (1.9–8.0)	3.2 (1.7–5.5)	.15	3.6 (2.0–6.6)	2.8 (1.7–5.0)	.25	
APACHE II score	26.0 (22.0–34.0)	16.0 (11.0–21.0)	<.01	24.0 (20.0–32.0)	15.0 (11.0–21.0)	<.01	
RRT initiation, n (%)	26 (86.7%)	1 (0.2%)	<.01	22 (78.6%)	7 (1.6%)	<.01	
ICU length of stay, day	14.0 (11.0–24.0)	5.0 (2.0–8.5)	<.01	12.0 (6.0–16.5)	4.0 (2.0–9.0)	<.01	
60-day mortality, n (%)	9 (30.0%)	20 (4.5%)	<.01	11 (39.3%)	19 (4.3%)	<.01	
90-day mortality, n (%)	9 (30.0%)	20 (4.5%)	<.01	12 (42.9%)	22 (4.9%)	<.01	
ACE-I: angiotensin-converting enzyme inhibitor; ARB: angiotensin receptor blocker; NSAIDs: non-steroidal antiinflammatory drugs; CK: creatine kinase; Alb: albumin; BUN: blood urea nitrogen; Cr: creatinine; APACHE: Acute Physiology and Chronic Health Evaluation; RRT: renal replacement therapy; ICU: intensive care unit.

Continuous variables were tested for normality using the Kolmogorov–Smirnov test. Non-normally distributed variables are presented as medians and interquartile ranges (IQRs) and compared using the Mann–Whitney U-test. Normally distributed variables are presented as means and standard deviations (SDs) and compared using t-tests. Categorical variables are presented as percentages and compared using Chi-square tests or Fisher’s exact tests. All confidence intervals are calculated at 95%, and p < .05 is considered statistically significant.

The outcomes of RRT initiation, ICU LOS, and 60-day and 90-day mortalities are summarized in Table 1. In cohort 1, 86.7% of the patients in the indication group required RRT initiation compared to 0.2% in the no-indication group (p < .01). In cohort 2, 78.6% of the patients in the indication group required RRT initiation compared to 1.6% in the no-indication group (p < .01). In cohort 1, the 60-day mortality rates were 30.0% and 4.5% in the indication and no-indication groups, respectively (p < .01). In cohort 2, the 60-day mortality rates were 39.3% and 4.3% in the indication and no-indication groups, respectively (p < .01). Other outcomes (ICU LOS and 90-day mortality) also showed significant differences between the groups and were consistent across both the cohorts.

The predictive model was developed using a stepwise selection method based on AIC. eTable 1 (Supplemental File) lists the candidate models and their corresponding AIC values. The final model included the following variables: sex, Alb and Cr levels, and APACHE II scores (SACrA). The final model, which was selected based on the lowest AIC, included these variables. Based on the beta coefficient, each factor was scored as follows: sex (male) = 4, Alb = −7, Cr = 4, and APACHE II score, 1 (Table 2).

Table 2. Selected logistic regression model.

Parameters	Beta-coefficient	Score	Odds ratio	95% CI	VIF	p Value	
Sex (male)	−0.904	−4	0.405	0.117–1.404	1.020	.15	
Alb	−1.565	−7	0.209	0.092–0.475	1.118	<.01	
Cr	0.873	4	2.394	2.394–3.449	1.251	<.01	
APACHE II score	0.228	1	1.256	1.256–1.133	1.176	<.01	
CI: confidence interval; VIF: variance inflation factor; Alb: albumin; Cr: creatinine; APACHE: Acute Physiology and Chronic Health Evaluation.

A multivariate logistic regression analysis was performed. All confidence intervals are calculated at 95%, and p < .05 is considered statistically significant.

The ROC curve for the predictive score in cohort 1 is shown in Figure 2(a), with an AUC of 0.971. The calibration was performed using a calibration plot (Figure 3(a)). After 1000 bootstrapping iterations, the calibration slope was 0.982 and intercept was 0.009. The cutoff value was determined to be 14.1 using the Youden index, with a sensitivity of 0.97 and specificity of 0.90.

Figure 2. Receiver operating characteristic (ROC) curves for the predictive score in two cohorts. (a) The ROC curve for cohort 1, with an area under the curve (AUC) of 0.971. (b) The ROC curve for cohort 2, with an AUC of 0.918.

Figure 3. Calibration curves for the predictive score in the two cohorts. (a) The calibration curve for cohort 1, with an intercept of 0.009 and a slope of 0.982. (b) The calibration curve for cohort 2, with an intercept of 0.004 and a slope of 0.988.

For external validation, the ROC curve for the predictive score in cohort 2 is shown in Figure 2(b), with an AUC of 0.918. The calibration was performed using the calibration plot. After 1000 bootstrapping iterations, the calibration slope was 0.988 and intercept was 0.004. The detailed results are provided in eTable 2 (Supplemental File).

To evaluate the validity of the clinical protocol, we compared the RRT initiation and 60-day mortality between the two cohorts. In cohort 1, there was a significant reduction in RRT initiation after the introduction of the protocol (p = .03). However, there was no significant difference in the 60-day mortality (p = .83) between the cohorts (Table 3).

Table 3. Comparison of outcomes between non-indication groups in each cohort.

Parameters	Cohort 1
No-indication group (N = 440)	Cohort 2
No-indication group (N = 447)	p Value	
RRT initiation, n (%)	1 (0.2%)	7 (1.6%)	.03	
60-day mortality, n (%)	20 (4.5%)	19 (4.3%)	.83	
RRT: renal replacement therapy.

Chi-square tests were performed. All confidence intervals are calculated at 95%, and p < .05 is considered statistically significant.

We have provided a breakdown of the criteria for RRT initiation for patients in each cohort in eTable 3 (Supplemental File).

The required sample size was calculated post hoc. Eight factors were included in the model (age, sex, CCI, BUN, Cr, CK, and Alb levels, and APACHE II scores). In addition, the required sample size was calculated using the pmsampsize package in R with the C-statistic (AUC, 0.971) and outcome (0.06 (30/470). The required sample size comprised 378 participants.

Discussion

We developed and validated a predictive score for the non-emergent initiation of RRT in ICU patients. The SACrA score demonstrated a good performance, suggesting its potential to identify patients who require RRT. Additionally, our study supports a non-emergent strategy as a reasonable approach for RRT initiation.

Previous research

Several studies have predicted the initiation of RRT for AKI. Wilson et al. reported a model for predicting RRT initiation in post-noncardiac surgery patients; however, it had limited sensitivity and positive predictive value [23,24]. Meersch et al. developed a model for predicting RRT indications [25]; however, it included urgent RRT, which is typically not a wait-and-see scenario. Their model also used specific biomarkers, which limited its generalizability and lacked external validation.

Strengths and clinical impact

To the best of our knowledge, this is the first study to predict non-emergent RRT initiation. We explicitly defined the criteria for urgent RRT and non-emergent initiation, achieving high sensitivity and specificity. The internal validation was adjusted for overfitting using bootstrapping and external validation was performed using a separate cohort. Our model was based on commonly collected admissions data, making it applicable to various settings. Cohort 1 followed a non-emergent RRT initiation protocol, allowing accurate data collection without unnecessary RRT. Cohort 2 did not have a protocol similar to that of cohort 1; therefore, it was not a group for accurately evaluating the reproducibility of the model. However, it was possible to evaluate its generalizability.

This prediction model provided a strong basis for avoiding unnecessary RRT. By predicting the patients who would meet the non-emergent initiation criteria, it can provide a tool to verify the effectiveness of early intervention in these patients. The reduction in the number of RRT cases without an increase in mortality suggests the appropriateness of this protocol.

Notably, the use of specific BUN levels as a criterion for initiating RRT is controversial. Although it is unlikely to be used as a stand-alone indicator, it seems reasonable to use it in conjunction with the oliguric criteria, as shown in Table 3.

The current evidence indicates no benefit of early RRT initiation in patients with AKI, except in urgent cases [4–8]. Predicting non-emergent indications helps avoid unnecessary use of RRT and supports a wait-and-see strategy. Although the primary outcome (RRT-free days) of the AKIKI2 trial was neutral, in a pre-specified adjusted analysis, the authors found that a more delayed strategy was associated with an increased 60-day mortality compared to a delayed indication [8]. This implied that some patients may be harmed by excessive delays in RRT initiation and highlighted the importance of accurately identifying patients who truly needed RRT. This predictive model could potentially improve outcomes through the timely initiation of RRT in at-risk patients.

Interpretation

The biochemical variables included in the model (Alb, Cr) are associated with AKI progression [3,18,20], indicating inflammation, nutritional status, and renal reserve. The APACHE II score reflects the overall severity of the illness. Although sex was included in the model despite not being significant variable in the univariate analysis, sex differences have been reported in AKI risk [26–27], and the model’s overall fit improved with its inclusion.

Limitations

This study had several limitations. First, this was a single-center study with a small sample size, which limits the generalizability of the findings. Second, the validation cohort was from the same institution and requires further validation in different settings. Third, the calibration was suboptimal for patients with a predicted probability of approximately 20%, possibly due to overfitting. Finally, a RCT is needed to determine whether the SACrA score improves the patient outcomes by identifying patients who truly require RRT.

Conclusions

The SACrA score demonstrated the ability to predict non-emergent initiation of RRT. The validation in an external cohort is essential before considering further RCTs to determine whether this score can improve the patient outcomes by identifying patients who truly require RRT. However, our analysis suggests the SACrA tool may help clinicians make informed decisions, reduce unnecessary RRT, and thereby improve patient outcomes.

Supplementary Material

fig2.jpg

fig1.jpg

Suppl new.docx

fig3.jpg

Ethical approval

This single-center retrospective study was conducted at the Toho University Omori Medical Center, a university-affiliated tertiary emergency medical care facility in Tokyo, Japan. This study was conducted in accordance with the principles embodied in the 1975 Declaration of Helsinki. The study protocol was approved by the Ethics Committee of Toho University Omori Medical Center (approval number: M23256).

Author contributions

G.S. drafted the manuscript. G.S., S.N., T.K., Y.M., S.Y., H.S., Y.N., and M.H. contributed to data acquisition. G.S. designed and coordinated the study. M.H. conceived the study, participated in study design and coordination, and contributed to manuscript writing. All the authors have read and approved the final version of this manuscript.

Consent form

The requirement for the acquisition of written informed consent from the patients was waived owing to the retrospective nature of the study.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

Not applicable.
==== Refs
References

1 Uchino S, Kellum JA, Bellomo R, et al. Acute renal failure in critically ill patients: a multinational, multicenter study. JAMA. 2005;294 (7 ):813–818. doi:10.1001/jama.294.7.813.16106006
2 Bellomo R, Kellum JA, Ronco C. Acute kidney injury. Lancet. 2012;380 (9843 ):756–766. doi:10.1016/S0140-6736(11)61454-2.22617274
3 Hoste EA, Bagshaw SM, Bellomo R, et al. Epidemiology of acute kidney injury in critically ill patients: the multinational AKI-EPI study. Intensive Care Med. 2015;41 (8 ):1411–1423. doi:10.1007/s00134-015-3934-7.26162677
4 Gaudry S, Hajage D, Schortgen F, et al. Initiation strategies for renal-replacement therapy in the intensive care unit. N Engl J Med. 2016;375 (2 ):122–133. doi:10.1056/NEJMoa1603017.27181456
5 Zarbock A, Kellum JA, Schmidt C, et al. Effect of early vs delayed initiation of renal replacement therapy on mortality in critically ill patients with acute kidney injury: the ELAIN randomized clinical trial. JAMA. 2016;315 (20 ):2190–2199. doi:10.1001/jama.2016.5828.27209269
6 Barbar SD, Clere-Jehl R, Bourredjem A, et al. Timing of renal-replacement therapy in patients with acute kidney injury and sepsis. N Engl J Med. 2018;379 (15 ):1431–1442. doi:10.1056/NEJMoa1803213.30304656
7 STARRT-AKI Investigators; Canadian Critical Care Trials Group; Australian and New Zealand Intensive Care Society Clinical Trials Group, et al. Timing of initiation of renal-replacement therapy in acute kidney injury. N Engl J Med. 2020;383 :240–251.32668114
8 Gaudry S, Hajage D, Martin-Lefevre L, et al. Comparison of two delayed strategies for renal replacement therapy initiation for severe acute kidney injury (AKIKI 2): a multicentre, open-label, randomised, controlled trial. Lancet. 2021;397 (10281 ):1293–1300. doi:10.1016/S0140-6736(21)00350-0.33812488
9 Pickkers P, Ostermann M, Joannidis M, et al. The intensive care medicine agenda on acute kidney injury. Intensive Care Med. 2017;43 (9 ):1198–1209. doi:10.1007/s00134-017-4687-2.28138736
10 Meersch M, Schmidt C, Hoffmeier A, et al. 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 Med. 2017;43 (11 ):1551–1561. doi:10.1007/s00134-016-4670-3.28110412
11 Meersch M, Schmidt C, Zarbock A. Predicting acute kidney injury and renal replacement therapy in the ICU. Curr Opin Crit Care. 2016;22 :548–555.
12 Chawla LS, Davison DL, Brasha-Mitchell E, et al. Development and standardization of a furosemide stress test to predict the severity of acute kidney injury. Crit Care. 2013;17 (5 ):R207. doi:10.1186/cc13015.24053972
13 Chen JJ, Chang CH, Huang YT, et al. Furosemide stress test as a predictive marker of acute kidney injury progression or renal replacement therapy: a systemic review and meta-analysis. Crit Care. 2020;24 (1 ):202. doi:10.1186/s13054-020-02912-8.32381019
14 Charlson ME, Pompei P, Ales KL, et al. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40 (5 ):373–383. doi:10.1016/0021-9681(87)90171-8.3558716
15 KDIGO. KDIGO clinical practice guideline for acute kidney injury. Kidney Int Suppl. 2012;2 :1–138.
16 Heidland A, Gerabek W, Sebekova K. Renal disease and hypertension. Nephrol Dial Transplant. 2006;21 :1066–1073.16476721
17 Palmer BF. Renal dysfunction complicating the treatment of hypertension. N Engl J Med. 2002;347 (16 ):1256–1261. doi:10.1056/NEJMra020676.12393824
18 Zhang Z, Hong Y, Liu N. A nomogram to predict acute kidney injury in patients with acute myocardial infarction. Kidney Blood Press Res. 2018;43 :569–578.
19 Bouchard J, Mehta RL. Predictive methods for acute kidney injury in critical care. Curr Opin Crit Care. 2016;22 :519–526.
20 De Geus HR, Betjes MG, Bakker J. Biomarkers for the prediction of acute kidney injury: a narrative review on current status and future challenges. Crit Care. 2013;17 :211.23680259
21 Peng JC, Wu Y, Xing SP, et al. Development and validation of a nomogram to predict the risk of renal replacement therapy among acute kidney injury patients in intensive care unit. Clin Exp Nephrol. 2023;27 (11 ):951–960. doi:10.1007/s10157-023-02383-5.37498349
22 Riley RD, Ensor J, Snell KIE, et al. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020;368 :m441. doi:10.1136/bmj.m441.32188600
23 Wilson TA, de Koning L, Quinn RR, et al. Derivation and external validation of a risk index for predicting acute kidney injury requiring kidney replacement therapy after noncardiac surgery. JAMA Netw Open. 2021;4 (8 ):e2121901. doi:10.1001/jamanetworkopen.2021.21901.34424303
24 Nishimoto M, Murashima M, Kokubu M, et al. External validation of a prediction model for acute kidney injury following noncardiac surgery. JAMA Netw Open. 2021;4 (10 ):e2127362. doi:10.1001/jamanetworkopen.2021.27362.34661665
25 Meersch M, Weiss R, Gerss J, et al. Predicting the development of renal replacement therapy indications by combining the furosemide stress test and chemokine (C-C motif) ligand 14 in a cohort of postsurgical patients. Crit Care Med. 2023;51 (8 ):1033–1042. doi:10.1097/CCM.0000000000005849.36988335
26 Neugarten J, Golestaneh L, Kolhe NV. Sex differences in acute kidney injury requiring dialysis. BMC Nephrol. 2018;19 (1 ):131. doi:10.1186/s12882-018-0937-y.29884141
27 Neugarten J, Golestaneh L. Sex differences in acute kidney injury: epidemiology, underlying mechanisms and clinical implications. Adv Chronic Kidney Dis. 2018;25 :83–92.
