Usefulness of serum creatinine-to-albumin ratio for 28-day all-cause mortality in critically ill patients with atrial fibrillation: A retrospective study based on the MIMIC-IV database

Abstract

Atrial fibrillation (AF) is the most common type of cardiac arrhythmia and is associated with increased short-term mortality in critically ill patients. The creatinine-to-albumin ratio (CAR) is associated with adverse outcomes in various clinical settings. This study investigated the association between CAR and 28-day all-cause mortality and evaluated its contribution to a clinical prediction model. A total of 4501 critically ill patients with AF were identified from the Medical Information Mart for Intensive Care IV database. The optimal cutoff value for CAR was determined using X-tile software, and the patients were categorized into high (>0.5) and low (≤0.5) CAR groups. The Kaplan–Meier analysis demonstrated a significantly lower 28-day survival in patients with elevated CAR. The CAR showed good discriminative ability for 28-day mortality. In multivariate Cox regression analyses, CAR remained independently associated with 28-day mortality, both as a continuous variable (hazard ratio = 1.211, 95% confidence interval = 1.061–1.381, P = .004) and as a categorical variables (>0.5 vs ≤0.5; hazard ratio = 1.377, 95% confidence interval = 1.175–1.615, P < .001). Restricted cubic spline analysis revealed a significant nonlinear dose–response relationship between CAR and mortality risk. Subgroup analyses showed consistent associations across most prespecified subgroups. A multivariate prediction model incorporating CAR was used to construct a nomogram that demonstrated good discrimination, calibration, and clinical utility. Overall, elevated CAR was independently and consistently associated with an increased risk of short-term mortality in critically ill patients with AF. In addition, the nomogram incorporating CAR demonstrated strong predictive performance. Given its routine availability and biological plausibility, CAR may serve as a simple and practical biomarker for early risk stratification and individualized prognostic assessment in this high-risk population.

1. Introduction

Atrial fibrillation (AF) is the most common cardiac arrhythmia in clinical practice, with a global prevalence of approximately 1% to 2%, which is increasing significantly with age.^[1,2] AF is closely associated with various cardiovascular diseases and their complications, such as heart failure and stroke, and has approximately doubled the risk of all-cause mortality (ACM) and quadrupled the risk of stroke.^[3,4] In an intensive care unit (ICU), the AF incidence can increase to approximately 15%, often secondary to conditions such as sepsis, electrolyte disturbances, or multiple organ dysfunction, leading to prolonged hospitalization and poor outcomes.^[5,6] Previous studies have reported that AF in ICU patients increases the risk of 90-day mortality by approximately 1.38-fold compared with patients without AF.^[7] Moreover, new-onset AF is associated with a 1.6-fold higher risk of in-hospital and 1-year post-discharge mortality, and serves as an independent predictor of both short- and long-term mortalities.^[8] These findings highlight the urgent need for reliable prognostic tools to enable risk stratification and optimize the management of critically ill patients with AF.

Renal dysfunction, systemic inflammation, and malnutrition are independent predictors of mortality in critically ill patients and are strongly associated with increased cardiovascular morbidity and mortality.^[9,10] Serum creatinine (Cr), a classic marker of glomerular filtration rate, is widely used to assess renal function,^[11] whereas serum albumin (Alb) reflects a range of physiological states, including acute inflammatory responses and nutritional deficiencies.^[12] Both biomarkers are closely associated with adverse cardiovascular events and other critical illnesses.^[13,14] The creatinine-to-albumin ratio (CAR), a novel composite biomarker, integrates information on inflammation, oxidative stress, nutritional status, and hepatic-renal function. Compared to single indicators, CAR offers superior diagnostic efficiency and operational simplicity. Recent studies have identified CAR as an independent predictor of poor prognosis in several diseases, including heart failure and stroke.^[15,16]

However, the prognostic value of CAR in critically ill patients with AF remains unclear. Therefore, this study aimed to investigate the association between baseline CAR levels at ICU admission and 28-day ACM in critically ill patients with AF and to develop a clinical prediction model to facilitate early individualized risk assessment and improve prognostic management.

2. Methods

2.1. Data source

This retrospective cohort study was conducted using the Medical Information Mart for Intensive Care IV (MIMIC-IV; version 3.1) database, a publicly accessible critical care database developed by the Laboratory for Computational Physiology, Massachusetts Institute of Technology. The MIMIC-IV contains comprehensive, de-identified clinical data from patients admitted to ICUs of the Beth Israel Deaconess Medical Center between 2008 and 2019. All protected health information was removed to ensure complete de-identification. Therefore, this study was exempt from Institutional Review Board approval, and the requirement for informed consent was waived. Access to the database was granted after the completion of a Collaborative Institutional Training Initiative training program (certification number: 64227218).

2.2. Cohort selection

Adult patients (aged ≥ 18 years) diagnosed with AF were identified using the International Classification of Diseases, Ninth and Tenth Revision (ICD-9 and ICD-10) diagnostic codes. For patients with multiple ICU admissions, only their first ICU stay was included to avoid repetition. Patients were excluded if they died within the first 24 hours after ICU admission or if Cr or Alb measurements within the first 24 hours of ICU admission were unavailable, as these were required to compute the CAR. After applying inclusion and exclusion criteria, 4501 eligible patients were included in the final analysis. The patient selection process is illustrated in Figure 1.

Figure 1.

The flowchart of patient selection in this study. CAR = creatinine-to-albumin ratio, ICU = intensive care unit, MIMIC-IV = Medical Information Mart for Intensive Care IV.

2.3. Data collection

Data were extracted using Structured Query Language via Navicat Premium (version 17.1.12; PremiumSoft CyberTech Ltd.). Demographic variables included age, sex, race, and body weight. Vital signs collected during the first 24 hours after ICU admission included heart rate, systolic and diastolic blood pressure, mean blood pressure, and saturation of peripheral oxygen. Laboratory variables obtained from the first available measurements within 24 hours of ICU admission included glucose levels, complete blood count parameters, Alb, and renal and liver function markers, such as blood urea nitrogen, Cr, alanine aminotransferase, and aspartate aminotransferase. Comorbidities, including myocardial infarction, congestive heart failure, cerebrovascular and peripheral vascular diseases, chronic pulmonary disease, sepsis, diabetes, renal dysfunction, and malignancy, were identified using ICD codes. Illness severity was assessed using recognized scoring systems available in the database, including the Charlson Comorbidity Index (CCI), Simplified Acute Physiology Score II (SAPS II), Oxford Acute Severity of Illness Score, and Sequential Organ Failure Assessment (SOFA) scores. Data on therapeutic interventions during ICU stay, including the use of invasive mechanical ventilation, dialysis, and vasopressor agents, were also collected.

Primary outcome was 28-day ACM, defined as death occurring within 28 days of ICU admission, as ascertained from the MIMIC-IV mortality records. CAR was calculated as Cr divided by Alb. To ensure temporal consistency across patients, Cr and Alb values used to calculate the CAR were obtained from the first available laboratory measurements within the first 24 hours after ICU admission. If multiple measurements were available during this period, the earliest value was selected.

Variables with more than 20% missingness were excluded from the analysis. For variables with ≤20% missing values, the missing data were imputed using a random forest algorithm to minimize potential bias. Outcome variables were not imputed.

2.4. Statistical analysis

X-tile software (version 3.6.1, Yale University) was used to identify the optimal CAR cutoff value for predicting 28-day mortality, after which patients were categorized into low and high CAR groups (Fig. 2). Continuous variables were tested for normality using the Shapiro–Wilk test. Normally distributed variables are presented as means with standard deviations and were compared using the Student t test. Non-normally distributed variables are expressed as medians with interquartile ranges and were compared using the Wilcoxon rank-sum test. Categorical variables are presented as counts and percentages and were compared using the Pearson chi-square or Fisher exact test.

Figure 2.

Determination of the optimal critical threshold for the CAR via X-tile analysis. (A) The left panel shows the X-tile plot specifically for critically ill patients with atrial fibrillation, with the optimal cutoff points marked by black circles. (B) These cutoff values are further illustrated in the histogram located in the middle panel. (C) In addition, the Kaplan–Meier survival curve is displayed in the right panel to provide a visual representation of the survival analysis. CAR = creatinine-to-albumin ratio.

Kaplan–Meier survival curves were generated to compare the 28-day survival between the groups, and differences were assessed using the log-rank test. Receiver operating characteristic (ROC) curve analysis was used to examine the discriminative ability of CAR in predicting 28-day mortality. To identify risk factors associated with mortality, least absolute shrinkage and selection operator regression was used for variable selection, followed by multivariable logistic regression to determine independent associations and construct a predictive model for 28-day mortality. Model discrimination was evaluated using the area under the ROC curve (area under the curve [AUC]), calibration was assessed using calibration plots, and clinical utility was examined using decision curve analysis. A nomogram was generated to visually represent the prediction model. Internal validation was performed using bootstrap resampling to assess model stability.

Time-to-event analyses were conducted using Cox proportional hazards regression models to evaluate the association between CAR and 28-day mortality, with CAR analyzed as continuous and categorical variables. The proportional hazards assumption was assessed using the Schoenfeld residuals, and no significant violations were observed. Restricted cubic spline (RCS) functions with 3 knots placed at predefined percentiles were applied to explore potential nonlinear dose–response relationships, using a median CAR value as the reference.

Subgroup analyses were prespecified based on clinical relevance and prior literature, including age, sex, diabetes, myocardial infarction, heart failure, cerebrovascular disease, vasopressor use, and mechanical ventilation. Interaction effects were evaluated using multiplicative interaction terms. To assess the robustness of subgroup findings, additional analyses were performed using CAR quartiles instead of dichotomized CAR. All statistical analyses were conducted using R software (version 4.5.0; R Foundation for Statistical Computing), and a two-sided P value of <.05 was considered statistically significant.

3. Results

3.1. Baseline characteristics

In total, 15,358 adult patients with AF were initially screened from the MIMIC-IV database. After applying the inclusion and exclusion criteria, 4501 patients were finally enrolled in the study. Table 1 presents the baseline characteristics of patients with AF. Based on an optimal CAR cutoff value, the patients were categorized into low (≤0.5, n = 2956) and high (>0.5, n = 1545) CAR groups. The median age of the participants was 75.9 years, and 57.6% of them were male. Compared with the low CAR group, patients in the high CAR group were slightly younger, had a higher proportion of males, and showed significantly higher body weight (all P < .01). Moreover, the prevalence rates of myocardial infarction, heart failure, liver disease, renal disease, sepsis, and diabetes were significantly higher in the high CAR group than those in the low CAR group (all P < .001). Severity scores, including the SOFA, Oxford Acute Severity of Illness Score, SAPS II, and CCI, were also markedly elevated (all P < .001). Regarding laboratory findings, compared with patients in the low CAR group, those in the high CAR group had significantly higher levels of white blood cells, blood glucose, liver function markers (alanine aminotransferase and aspartate aminotransferase), and renal function indices (blood urea nitrogen and Cr; all P < .001), while hemoglobin, platelet count, and Alb levels were markedly lower (all P < .001), resulting in a substantially higher CAR value (0.82 vs 0.29, P < .001). Regarding clinical outcomes, patients in the high CAR group had longer ICU and hospital stays (both P < .001) and significantly higher 28-day ACM compared with those in the low CAR group (33.1% vs 18.0%, P < .001), indicating that an elevated CAR was associated with increased mortality risk.

Table 1.

The baseline characteristics of patients with atrial fibrillation stratified by CAR category.

Variables	Total	Low CAR	High CAR	P values
	N = 4501	N = 2956	N = 1545
Age, years old	75.88 (66.09, 84.13)	76.21 (66.44, 84.56)	75.27 (65.69, 82.78)	.001
Gender, male, n (%)	2594 (57.6)	1566 (53.0)	1028 (66.5)	<.001
Weight, kg	79.70 (66.70, 96.50)	77.00 (64.50, 93.03)	85.00 (70.60, 100.90)	<.001
Comorbidities, n (%)
Myocardial infarct	1061 (23.6)	644 (21.8)	417 (27.0)	<.001
Congestive heart failure	2170 (48.2)	1272 (43.0)	898 (58.1)	<.001
Diabetes	1530 (34.0)	848 (28.7)	682 (44.1)	<.001
Renal disease	1385 (30.8)	448 (15.2)	937 (60.6)	<.001
Liver disease	663 (14.7)	332 (11.2)	331 (21.4)	<.001
Sepsis	2920 (64.9)	1698 (57.4)	1222 (79.1)	<.001
Cancer	711 (15.8)	441 (14.9)	270 (17.5)	.029
Cerebrovascular disease	937 (20.8)	734 (24.8)	203 (13.1)	<.001
Interventions, n (%)
Mechanical ventilation	730 (16.2)	400 (13.5)	330 (21.4)	<.001
Vasopressors	564 (12.5)	214 (7.2)	350 (22.7)	<.001
Dialysis	286 (6.4)	12 (0.4)	274 (17.7)	<.001
Scores, points
SOFA	6.00 (3.00, 9.00)	4.00 (2.00, 7.00)	9.00 (6.00, 12.00)	<.001
OASIS	35.00 (29.00, 42.00)	34.00 (28.00, 40.00)	38.00 (31.00, 46.00)	<.001
SAPS II	40.00 (33.00, 50.00)	37.00 (31.00, 44.00)	49.00 (40.00, 59.00)	<.001
CCI	7.00 (5.00, 9.00)	6.00 (5.00, 8.00)	8.00 (6.00, 10.00)	<.001
Vital signs
MBP, mm Hg	75.95 (69.78, 84.08)	77.87 (71.34, 85.72)	72.77 (67.31, 79.61)	<.001
Heart rate, bpm	85.33 (73.58, 98.42)	84.83 (73.28, 97.80)	86.24 (73.91, 99.61)	.008
SpO2, %	96.86 (95.42, 98.20)	96.77 (95.41, 98.12)	97.00 (95.49, 98.39)	.013
SBP, mm Hg	114.00 (104.81, 127.44)	117.00 (106.91, 129.96)	109.84 (101.38, 120.29)	<.001
DBP, mm Hg	61.68 (54.94, 69.08)	63.14 (56.05, 70.96)	59.10 (52.92, 65.37)	<.001
Laboratory values
WBC, 109/L	10.90 (7.90, 15.30)	10.55 (7.73, 14.45)	11.82 (8.25, 17.33)	<.001
Hemoglobin, g/dL	10.65 (9.20, 12.20)	11.10 (9.62, 12.60)	9.85 (8.55, 11.30)	<.001
Platelet, 109/L	187.75 (136.00, 252.00)	193.50 (143.73, 256.06)	174.00 (121.50, 245.00)	<.001
Glucose, mmol/L	7.36 (6.05, 9.36)	7.27 (6.03, 8.96)	7.63 (6.16, 10.13)	<.001
Albumin, g/dL	3.30 (2.80, 3.70)	3.40 (3.00, 3.80)	3.00 (2.50, 3.40)	<.001
Creatinine, mg/dL	1.20 (0.85, 1.95)	0.95 (0.75, 1.20)	2.45 (1.90, 3.72)	<.001
Potassium, mmol/L	4.20 (3.85, 4.63)	4.10 (3.80, 4.45)	4.48 (4.05, 4.97)	<.001
ALT, IU/L	27.00 (16.00, 60.50)	25.92 (16.00, 50.00)	32.00 (17.00, 100.33)	<.001
AST, IU/L	41.00 (25.00, 96.00)	38.00 (24.00, 74.00)	56.00 (27.00, 171.00)	<.001
BUN, mg/dL	26.67 (17.50, 44.33)	20.50 (15.00, 28.40)	50.00 (36.50, 71.33)	<.001
CAR	0.37 (0.26, 0.65)	0.29 (0.22, 0.37)	0.82 (0.63, 1.28)	<.001
Outcomes
ICU LOS, day	3.12 (1.89, 6.00)	2.93 (1.81, 5.37)	3.69 (2.12, 7.15)	<.001
Hospital LOS, day	8.83 (5.36, 15.07)	8.10 (5.08, 13.69)	10.66 (5.92, 17.98)	<.001
28-day mortality, n (%)	1043 (23.2)	532 (18.0)	511 (33.1)	<.001

Medians and interquartile ranges (25th and 75th percentiles) were calculated for continuous variables and frequencies and percentages for categorical variables. The Wilcoxon rank-sum test was used to compare group differences for continuous variables and chi-squared tests for categorical variables.

ALT = alanine aminotransferase, AST = aspartate aminotransferase, BUN = blood urea nitrogen, CAR = creatinine-to-albumin ratio, CCI = Charlson Comorbidity Index, DBP = diastolic blood pressure, ICU = intensive care unit, LOS = length of stay, MBP = mean blood pressure, OASIS = Oxford Acute Severity of Illness Score, SAPS II = Simplified Acute Physiology Score II, SBP = systolic blood pressure, SOFA = Sequential Organ Failure Assessment, SpO₂ = saturation of peripheral oxygen, WBC = white blood cell.

3.2. Survival analysis and predictive performance

The Kaplan–Meier survival analysis demonstrated that patients in the high CAR group had a significantly greater 28-day mortality risk than those in the low CAR group (log-rank P < .001; Fig. 3A). ROC analysis also indicated that CAR had a moderate differentiation ability for 28-day ACM prediction (AUC = 0.62, 95% confidence interval [CI] = 0.601–0.639; Fig. 3B). Notably, CAR outperformed individual predictors such as Cr (AUC = 0.597) and Alb (AUC = 0.398), underscoring its superior prognostic capability.

Figure 3.

(A) Kaplan–Meier survival curves for the high- and low-CAR groups. (B) CAR predicts 28-day mortality in atrial fibrillation. AUC = area under the curve, CAR = creatinine-to-albumin ratio, ROC = receiver operating characteristic curve.

3.3. Logistic regression analysis of CAR and mortality

The least absolute shrinkage and selection operator regression model initially identified 11 potential risk predictors (Fig. 4). All these variables were significantly associated with 28-day ACM in univariate logistic regression analyses. In the multivariate logistic regression analysis (Table 2), the following variables remained independent predictors of 28-day mortality: age (odds ratio [OR] = 1.032, 95% CI = 1.023–1.041, P < .001), body weight (OR = 0.992, 95% CI = 0.988–0.996, P < .001), heart rate (OR = 1.014, 95% CI = 1.009–1.019, P < .001), white blood cell (OR = 1.015, 95% CI = 1.008–1.023, P < .001), cerebrovascular disease (OR = 1.583, 95% CI = 1.291–1.939, P < .001), SOFA score (OR = 1.065, 95% CI = 1.032–1.098, P < .001), SAPS II score (OR = 1.025, 95% CI = 1.015–1.035, P < .001), CCI (OR = 1.139, 95% CI = 1.099–1.180, P < .001), need for mechanical ventilation (OR = 8.912, 95% CI = 7.221–11.035, P < .001), and CAR (OR = 1.267, 95% CI = 1.040–1.543, P = .018).

Figure 4.

Identification of vital factors correlated with 28-day mortality among critically ill patients with atrial fibrillation utilizing LASSO logistic regression analysis. (A) Examination of the variable coefficient’s variability patterns. (B) Optimization of the LASSO regression model’s λ parameter through a rigorous cross-validation approach. LASSO = least absolute shrinkage and selection operator.

Table 2.

Logistic regression analysis for the risk factors of 28-day all-cause mortality selected by LASSO regression.

	Univariate		Multivariate
	OR (95% CI)	P	OR (95% CI)	P
Age	1.029 (1.023–1.035)	<.001	1.032 (1.023–1.041)	<.001
Weight	0.993 (0.990–0.996)	<.001	0.992 (0.988–0.996)	<.001
Heart rate	1.012 (1.008–1.016)	<.001	1.014 (1.009–1.019)	<.001
WBC	1.029 (1.021–1.037)	<.001	1.015 (1.008–1.023)	<.001
Cerebrovascular disease	1.523 (1.294–1.788)	<.001	1.583 (1.291–1.939)	<.001
CCI	1.182 (1.150–1.215)	<.001	1.139 (1.099–1.180)	<.001
SOFA	1.174 (1.155–1.195)	<.001	1.065 (1.032–1.098)	<.001
OASIS	1.085 (1.077–1.094)	<.001	1.004 (0.991–1.017)	.572
SAPS II	1.061 (1.056–1.067)	<.001	1.025 (1.015–1.035)	<.001
Mechanical ventilation	9.617 (8.082–11.466)	<.001	8.912 (7.221–11.035)	<.001
CAR	2.252 (1.955–2.594)	<.001	1.267 (1.040–1.543)	.018

CAR = creatinine-to-albumin ratio, CCI = Charlson Comorbidity Index, CI = confidence interval, LASSO = least absolute shrinkage and selection operator, OASIS = Oxford Acute Severity of Illness Score, OR = odds ratio, SAPS II = Simplified Acute Physiology Score II, SOFA = Sequential Organ Failure Assessment, WBC = white blood cell.

A nomogram incorporating these independent predictors was developed to illustrate the cumulative risk profile, in which higher scores indicated a greater 28-day mortality risk (Fig. 5). The model showed robust discrimination, achieving an AUC of 0.833 (95% CI = 0.819–0.847) for mortality prediction (Fig. 6A). The calibration assessment demonstrated close agreement between model estimates and actual outcomes, with a mean absolute error of 0.015 (Fig. 6B). In addition, a decision curve analysis confirmed a meaningful net clinical benefit across a wide range of decision thresholds (Fig. 6C).

Figure 5.

The nomogram displays the scoring system of the predictive model for 28-day mortality among critically ill patients with atrial fibrillation. CAR = creatinine-to-albumin ratio, CCI = Charlson Comorbidity Index, SAPS II = Simplified Acute Physiology Score II, SOFA = Sequential Organ Failure Assessment, WBC = white blood cell.

Figure 6.

(A) ROC curve analysis of the nomogram for 28-day mortality prediction. (B) The calibration curve for predicting 28-day mortality. (C) DCA of the nomogram to predict 28-day mortality. AUC = area under the curve, CAR = creatinine-to-albumin ratio, DCA = decision curve analysis, ROC = receiver operating characteristic curve.

3.4. Cox proportional hazards analysis

To assess robustness, 4 Cox proportional hazard models were constructed. After adjusting for multiple confounders, CAR remained an independent predictor of 28-day ACM (Table 3). When modeled as a continuous variable, CAR consistently showed a significant association with mortality across all models: Model 1 (hazard ratio [HR] = 1.350, 95% CI = 1.249–1.458, P < .001), Model 2 (HR = 1.526, 95% CI = 1.406–1.655, P < .001), Model 3 (HR = 1.499, 95% CI = 1.338–1.681, P < .001), and Model 4 (HR = 1.211, 95% CI = 1.061–1.381, P = .004). When CAR was treated as a categorical variable, the high CAR group showed a significantly increased mortality risk compared to the low CAR group: Model 1 (HR = 2.045, 95% CI = 1.811–2.309, P < .001), Model 2 (HR = 2.238, 95% CI = 1.976–2.535, P < .001), Model 3 (HR = 1.968, 95% CI = 1.702–2.275, P < .001), and Model 4 (HR = 1.377, 95% CI = 1.175–1.615, P < .001). These findings indicate that elevated CAR levels are strongly associated with increased 28-day ACM, with mortality risk increasing progressively with higher CAR values.

Table 3.

Multivariate Cox regression analysis for 28-day mortality.

Methods	HR (95% CI)	P value
For continuous variable, CAR
Model 1	1.350 (1.249–1.458)	<.001
Model 2	1.526 (1.406–1.655)	<.001
Model 3	1.499 (1.338–1.681)	<.001
Model 4	1.211 (1.061–1.381)	.004
For categorical variable, CAR
Model 1	2.045 (1.811–2.309)	<.001
Model 2	2.238 (1.976–2.535)	<.001
Model 3	1.968 (1.702–2.275)	<.001
Model 4	1.377 (1.175–1.615)	<.001

Model 1: Unadjusted model. Model 2: Adjusts for baseline demographic characteristics, including age, gender, race, and weight. Model 3: Further adjusts for comorbidities and treatment, including myocardial infarction, congestive heart failure, diabetes, renal disease, cerebrovascular disease, CCI, vasopressors, dialysis, and mechanical ventilation. Model 4: Further adjusts laboratory results, and clinical scores, including MBP, heart rate, SpO₂, hemoglobin, platelet, glucose, WBC, ALT, AST, SOFA, OASIS, and SAPS II score.

ALT = alanine aminotransferase, AST = aspartate aminotransferase, CAR = creatinine-to-albumin ratio, CCI = Charlson Comorbidity Index, CI = confidence interval, HR = hazard ratio, MBP = mean blood pressure, OASIS = Oxford Acute Severity of Illness Score, SAPS II = Simplified Acute Physiology Score II, SOFA = Sequential Organ Failure Assessment, SpO₂ = saturation of peripheral oxygen, WBC = white blood cell.

Furthermore, the RCS analysis demonstrated a significant nonlinear relationship between CAR and 28-day ACM in critically ill patients with AF (both overall and nonlinearity P < .001; Fig. 7). The estimated HR increased with rising CAR levels in a nonlinear pattern. Across the lower range of CAR values, the HR remained near unity, whereas higher CAR levels were associated with progressively greater mortality risk, consistent with a nonlinear dose–response relationship.

Figure 7.

RCS curve analysis for the CAR and 28-day mortality. CAR = creatinine-to-albumin ratio, CI = confidence interval, RCS = restricted cubic spline.

3.5. Subgroup analysis

Subgroup analyses were conducted to examine the consistency of the association between CAR and 28-day ACM across clinically relevant strata (Fig. 8). The patients were stratified according to age (<65 vs ≥65 years), sex, diabetes, myocardial infarction, heart failure, cerebrovascular disease, vasopressor use, and mechanical ventilation status. In all the prespecified subgroups, higher CAR levels were consistently associated with an increased risk of 28-day mortality, and no statistically significant interactions were observed (all P for interaction > .05).

Figure 8.

The forest plot of the subgroup analysis for the CAR and 28-day all-cause mortality among critically ill patients with atrial fibrillation. CAR = creatinine-to-albumin ratio, CI = confidence interval, F = female, HR = hazard ratio, M = male.

To further explore potential dose–response patterns and assess the consistency of the association across increasing CAR levels, additional exploratory subgroup analyses were performed using CAR quartiles (Table S1, Supplemental Digital Content). Across most prespecified subgroups, higher CAR quartiles were directionally associated with an increased risk of 28-day mortality, with the highest quartile generally exhibiting the greatest hazard. A statistically significant interaction was observed between CAR quartiles and mechanical ventilation status (P for interaction = .018). Among patients not receiving mechanical ventilation, increasing CAR quartiles were associated with a progressive increase in mortality risk, with the highest quartile showing a substantially elevated hazard compared with the lowest quartile (HR for Q4 vs Q1 = 1.93, 95% CI = 1.39–2.66, P < .001). In contrast, no consistent dose-related pattern was observed among mechanically ventilated patients, in whom CAR quartiles were not significantly associated with 28-day mortality.

4. Discussion

In this retrospective cohort of critically ill patients with AF, we demonstrated that elevated CAR at ICU admission was independently associated with increased 28-day ACM. This association remained robust when the CAR was modeled as either a continuous or categorical variable after extensive adjustment for demographic characteristics, comorbidities, laboratory parameters, illness severity scores, and major ICU interventions. RCS analysis further identified a nonlinear dose-response relationship between CAR and mortality, with progressively increasing risk at higher CAR levels. Importantly, incorporation of CAR into a clinically applicable nomogram prediction model showed excellent discrimination, calibration, and clinical utility.

Previous studies have demonstrated that CAR has independent prognostic value across a range of clinical settings, including heart failure, stroke, and acute pancreatitis.^[14,17,18] In both ischemic and hemorrhagic stroke populations, CAR has consistently outperformed Cr and Alb alone in predicting short-term mortality, underscoring the advantage of integrating renal dysfunction with nutritional and inflammatory status into a single composite index. Moreover, the results demonstrated a nonlinear association between CAR and short-term mortality, with clear threshold effects, indicating a disproportionate increase in mortality risk beyond specific CAR levels.^[17,19] Evidence from sepsis and cardiac surgery cohorts further supports the robustness of CAR under conditions of severe inflammation and multi-organ dysfunction.^[20,21] Our findings are concordant with these observations and extend the existing literature by focusing on critically ill patients with AF, a population in whom acute physiological derangements and multi-organ dysfunction are particularly prevalent. By embedding CAR within a nomogram tailored to ICU patients with AF, the present study enhances the clinical translatability of this biomarker.

Subgroup analyses based on categorized variables showed generally consistent associations between elevated CAR and 28-day ACM, with no significant interactions across most clinical strata, supporting the overall robustness of CAR as a prognostic marker. Exploratory analyses using CAR quartiles yielded results that were directionally consistent with the primary binary analysis. A significant interaction was observed in the mechanical ventilation subgroup, where a graded increase in mortality risk across CAR quartiles was more evident in patients without mechanical ventilation. This finding may reflect the profound illness severity and high baseline mortality risk among mechanically ventilated patients, in whom the prognostic contribution of individual biomarkers may be diminished by competing risks.^[22] In addition, interventions such as mechanical ventilation may influence Cr and Alb levels, potentially modifying the discriminatory capacity of CAR.^[22] Importantly, this interaction does not contradict the independent association between CAR and mortality observed in the overall cohort, but rather suggests that the relative prognostic contribution of CAR may vary according to illness severity and treatment intensity. Consistent with prior CAR studies in stroke and acute pancreatitis populations, these exploratory findings underscore the potential value of CAR in risk stratification while highlighting the need for cautious interpretation of subgroup interactions.^[17–19]

The observed association between elevated CAR and increased 28-day ACM in critically ill patients with AF likely reflects its role as an integrative marker of systemic disease severity. By simultaneously capturing cardiorenal dysfunction and nutritional-inflammatory imbalance, CAR summarizes the convergence of multiple adverse pathophysiological processes in critical illness.^[23] Elevated Cr levels are indicative of impaired renal function and are commonly accompanied by fluid overload, electrolyte disturbances, and neurohormonal activation, whereas hypoalbuminemia contributes to malnutrition, impaired intravascular oncotic pressure, reduced antioxidant capacity, and heightened inflammatory burden.^[24–26] The coexistence of these abnormalities may exacerbate microcirculatory dysfunction and tissue hypoxia, creating a systemic pathological milieu, which is a key driver of adverse outcomes in critically ill patients.^[27,28] In contrast, increased CAR also represents increased inflammatory and oxidative stress activity, which may directly influence atrial structure and electrophysiological stability.^[29] Inflammatory signaling and oxidative stress can facilitate atrial fibrosis through epigenetic modulation and activation of profibrotic pathways while simultaneously impairing ion channel function and disrupting intracellular calcium homeostasis.^[30,31] The coupling of structural and electrical remodeling, together with a concomitant autonomic nervous system imbalance characterized by sympathetic–parasympathetic dysregulation, may further reinforce a proarrhythmic substrate.^[32] This process promotes AF persistence and progression, ultimately contributing to an increased risk of short-term mortality in critically ill patients.^[33]

From a clinical perspective, an elevated CAR may identify a subgroup of critically ill patients with AF characterized by an unfavorable inflammatory-metabolic and cardiorenal profile. Such patients may benefit from intensified rhythm surveillance and more comprehensive supportive management early in the disease course.^[34] Interventions targeting systemic inflammation, metabolic dysfunction, and cardiorenal interactions, including therapies with established cardiovascular protective effects, may represent potential avenues for improving outcomes in this high-risk population.^[35] In addition, dynamic changes in CAR may offer additional prognostic insight. Declining CAR levels could reflect attenuation of systemic inflammation or recovery of renal function, whereas persistently elevated CAR may indicate ongoing pathophysiological deterioration requiring intensified clinical attention.^[36] Importantly, CAR is readily available, inexpensive, and easily calculated from routinely obtained laboratory parameters, making it suitable for early risk stratification in an ICU setting.^[37]

This study has several limitations. First, this was a single-center retrospective study based on the MIMIC-IV database, which precludes causal inference despite comprehensive adjustment for confounders. Second, only baseline CAR values at ICU admission were analyzed; dynamic changes in CAR over time and their relationship with outcomes could not be assessed. Third, patients with missing Cr or Alb data were excluded, which may have introduced selection bias and affected the generalizability of the conclusions. Fourth, owing to the lack of data on therapeutic interventions, such as Alb replacement therapy, which may directly affect Alb levels and CAR values, these confounding factors were not fully adjusted for in the analysis, thereby affecting the reliability of the study conclusions. Finally, cause-specific mortality data were unavailable, which limited the exploration of AF-related versus non-AF-related mortalities. Future multicenter, prospective studies with longitudinal measurements are warranted to validate these findings and further elucidate the clinical utility of CAR in critically ill patients with AF.

5. Conclusions

This study demonstrated that an elevated CAR was independently associated with increased 28-day ACM in critically ill patients with AF. A CAR-based nomogram incorporating other key variables showed strong predictive performance. Owing to its simplicity, availability, and low cost, CAR may serve as a practical tool for early risk stratification in ICU settings. Nevertheless, given the inherent limitations of this study, prospective multicenter studies are warranted to further validate these findings.

Acknowledgments

The authors would like to express sincere gratitude to the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center for maintaining and providing access to the MIMIC-IV database.

Author contributions

Data curation: Wenbo Li, Liang Guo.

Formal analysis: Wenbo Li.

Investigation: Wenbo Li.

Methodology: Wenbo Li.

Resources: Wenbo Li, Liang Guo.

Software: Wenbo Li.

Validation: Wenbo Li.

Visualization: Wenbo Li.

Writing – original draft: Wenbo Li.

Writing – review & editing: Wenbo Li, Liang Guo.

Conceptualization: Liang Guo, He Huang.

Funding acquisition: He Huang.

Project administration: He Huang.

Supervision: He Huang.