Association between albumin-corrected anion gap and 60-day all-cause mortality in critically ill shock patients Authors: Yan-Ke Zou, Shi-Li Zhong, Li-Huan Gong, Shuai Zeng, Zhen Wang, Xiu-Qing Yao, Ya-Xi Luo Published: 2025 ABSTRACT AbstractBackgroundAlbumin-corrected anion gap ACAG can accurately reflect the patient’s acid–base status and metabolic balance, and may have the potential to predict the prognosis of shock. The purpose of this study was to evaluate the predictive value of ACAG for 60-day all-cause mortality in shock patients admitted to the intensive care unit.MethodsWe collected data from the Critical Care Medical Information Market IV (MIMIC-IV, version 3.1) database for patients who met a clear diagnosis of shock and recorded absolute values of anion gap and albumin after admission. Then, we used the ROC curve to evaluate the predictive relationship of ACAG on 60-day all-cause mortality in shock patients and derived the risk threshold, and used different methods to further evaluate the risk threshold. Kaplan–Meier survival curves were plotted to assess 60-day survival probabilities. We employed the change-in-estimate criterion to select confounding factors, ultimately incorporating SAPSII and LAC along with ACAG to construct a Cox proportional hazards model with time-dependent covariates.ResultsWe ultimately included 10,144 eligible patients from the MIMIC-IV database and described a non-linear association between ACAG and 60-day all-cause mortality. ACAG achieved an AUC of 0.68 (95% CI 0.67–0.69) for predicting 60-day mortality, with significantly superior discriminatory performance to AG (DeLong test, p < 0.001). We found that 19.25 is the risk threshold for the ACAG indicator. Kaplan–Meier survival curve showed that the 60-day survival probability in the high ACAG group was significantly lower than that in the low ACAG normal group (P < 0.001). In the Cox proportional hazards model with time-dependent covariates, the main effect of ACAG indicated that patients in the high ACAG group had a 72.5% higher risk of death compared to those in the low ACAG group (HR = 1.725, 95% CI 1.472–2.023, p < 0.001), while the time-dependent effect of ACAG was not statistically significant (p = 0.08).By combining ACAG with each scoring system (LODS, OASIS, SOFA, SAPSII) separately, it was found that the combination was better than the single ACAG value in predicting 60-day all-cause mortality in shock patients.ConclusionsACAG, derived from a rapid and low-cost formula, serves as an early predictor of mortality in shock patients. Its clinical utility is evidenced by a statistically significant association with 60-day mortality in high ACAG patients admitted to the ICU. Furthermore, integrating ACAG with established scoring systems could improve predictive accuracy.Keywords: Shock, Albumin-corrected anion gap, MIMIC-IV database, Mortality Introduction Shock is a life-threatening syndrome characterized by insufficient blood supply from the circulatory system, leading to inadequate tissue perfusion and cellular metabolic dysfunction. If left untreated, it can rapidly progress to multiple organ failure. Shock encompasses various subtypes based on etiology and physiological mechanisms, each associated with high mortality rates despite their distinctions. For example, septic shock—a form of distributive shock—has a mortality rate of 40%–50%, while cardiogenic shock can reach mortality rates as high as 75% [1, 2]. Early and timely intervention is critical, as the condition remains reversible in its initial stages [3]. However, once advanced, it imposes severe burdens, including elevated treatment costs, prolonged hospital stays, and high mortality rates, making the identification of prognostic factors essential. The anion gap (AG) is a key parameter in evaluating metabolic acidosis and mixed acid–base disorders. Defined as the difference between major cations and anions in plasma (AG [mmol/L] = [Na+ + K+] − [Cl− + HCO₃−]) [4], it provides critical insights into tissue hypoxia and impaired perfusion, especially in critically ill patients [5], Elevated AG levels often indicate poor outcomes and are frequently observed in patients with shock due to fluid loss and tissue hypoperfusion, both of which exacerbate metabolic acidosis [6–8]. Nevertheless, AG values are significantly influenced by serum albumin levels, a major unmeasured anion in plasma. Hypoalbuminemia, which occurs in over 75% of patients with shock, can lead to underestimation of AG values, resulting in false-negative assessments [9–11].Specifically, for every 1 g/dL decrease in serum albumin, the AG decreases by approximately 2.5 mmol/L [12, 13]. To address this limitation, the albumin-corrected anion gap (ACAG) has been introduced, offering a more accurate representation of acid–base disturbances by accounting for variations in albumin concentration [14]. The clinical utility of ACAG has been demonstrated in a range of critical conditions, including sepsis, cirrhosis, asthma, myocardial infarction, acute kidney injury, and cardiac arrest [5, 12, 15–18]. Recent studies have further highlighted its prognostic value in cardiogenic shock, where elevated ACAG levels correlate with higher mortality rates and poorer outcomes [19].These findings suggest that ACAG serves not only as a marker of metabolic balance, but also as a potential predictor of mortality risk in critically ill populations. Despite its established utility in specific conditions, the role of ACAG as a predictor of all-cause mortality in shock patients remains underexplored. Given the high prevalence of hypoalbuminemia and the pressing need for reliable prognostic markers, our study therefore aimed to investigate whether ACAG could predict 60-day all-cause mortality in ICU patients with shock. Materials and methods Data source This study utilized data from the Medical Information Mart for Intensive Care IV (MIMIC-IV, version 3.1) database [15], maintained by the Massachusetts Institute of Technology (MIT). The database provides comprehensive, de-identified health data from critically ill patients admitted to the ICU. One of the authors (Shili Zhong) accessed and extracted the data under approved certification (ID: 13858105). The study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for observational research [19]. Since the MIMIC-IV database is publicly available, the requirement for informed consent was waived. Patients with a confirmed diagnosis of shock, identified through International Classification of Diseases (ICD) codes from the MIMIC-IV (version 3.1) database, were included in the analysis. The inclusion criteria for this study were: a documented diagnosis of shock, and an age between 18 and 91 years. The exclusion criteria were as follows: (1) age under 18 years; (2) patients who did not receive treatment in the ICU; and (3) missing either AG or ALB results. For patients with multiple ICU admissions, only data from their first ICU admission were considered to ensure uniformity in analyses. Study variables Primary variables The primary variable of interest was ACAG. We extracted the AG and ALB of the patient’s first ICU admission and calculated the ACAG using the following formula: ACAG = AG + {4.4 − [ALB (g/dl)]} × 2.5 [20, 21]. Outcomes The primary outcome was 60-day all-cause in-hospital mortality, calculated from the date of ICU admission. Length of stay (LOS) was defined as ICU stay duration, while the outcome observation period was set to 60 days. Mortality was defined as death during the ICU stay. At the 60-day endpoint, patients were categorized into either the survivor group or the death group based on their survival status. Covariates All covariates were defined using the first measurement or assessment recorded within 24 h of hospital admission. Potential confounders were selected based on clinical judgment and existing literature. These included: gemographics: age, gender, ethnicity, body mass index (BMI). Vital signs: fastest heart rate (FHR), lowest systolic blood pressure (SBP), fastest respiratory rate (RR), and lowest peripheral oxygen saturation (SpO₂) during hospitalization. Laboratory results: AG, ALB, alanine aminotransferase (ALT), aspartate aminotransferase (AST), lactate (LAC), red blood cell count (RBC), red cell distribution width (RDW), white blood cell count (WBC), platelet count (PLT), creatinine (CREA), blood urea nitrogen (BUN), glucose (GLU), serum sodium (Na), serum calcium (Ca), total bilirubin (TBIL), prothrombin time (PT), activated partial thromboplastin time (APTT), international normalized ratio (INR). Scoring dystems: Logistic Organ Dysfunction System (LODS), Oxford Acute Severity of Illness Score (OASIS), Richmond Agitation-Sedation Scale (RASS), Glasgow Coma Scale (GCS), Simplified Acute Physiology Score II (SAPS II), and Sequential Organ Failure Assessment (SOFA) scores. Data extraction was performed using structured query language (SQL) with Postgres SQL software (v17.0.1) and Navicate Premium software (v17.0.8). Covariates with severe missingness (≥ 30% of the dataset) were also excluded from analysis. For acceptable missing data, multiple imputations were performed using the random forest algorithm implemented in the R package. Statistical analysis The Kolmogorov–Smirnov test was used to evaluate the normality of quantitative data. Normally distributed continuous data were presented as mean (SD) and compared using the Student’s t-test. Non-normally distributed data were presented as median (IQR) and compared using the Wilcoxon rank-sum test. Categorical variables were compared using the Chi-square test, with differences expressed as percentages (%). To investigate the association between ACAG and 60-day all-cause mortality in shock patients, receiver operating characteristic (ROC) curve analysis was conducted. During the determination of the optimal cutoff value for ACAG, we identified two adjacent values—19.125 and 19.375—both corresponding to an identical Youden’s index of 0.268. The mean value of 19.25 was initially selected as the ACAG threshold. A detailed justification for this threshold will be provided in the fourth paragraph of “Statistical analysis”. Patients were subsequently divided into two groups: low ACAG (ACAG ≤ 19.25) and high ACAG (ACAG > 19.25). Kaplan–Meier survival curves were plotted to assess 60-day survival probabilities. Prior to model construction, we performed the following key steps. First, multicollinearity was assessed by calculating the variance inflation factor (VIF) among the independent variables. The VIF values for all variables except prothrombin time (PT) and international normalized ratio (INR) were below 5. After excluding PT and INR, multicollinearity was within an acceptable range. Subsequently, the proportional hazards (PH) assumption was tested using univariable Cox regression for each variable and multivariable Cox regression for all variables included. While the PH assumption was satisfied in all univariable Cox regression models, the global test and the tests for the majority of covariates in the multivariable Cox model violated the assumption (p < 0.05). Due to this violation of the PH assumption, the standard Cox proportional hazards model was deemed inappropriate for our data. Consequently, we adopted a modified approach and constructed a Cox proportional hazards model with time-dependent covariates. We employed a variable selection method based on the change-in-estimate criterion, which is grounded in epidemiological principles and focuses on the impact of confounders on the effect estimate of the exposure variable. A change threshold of 15% was set—meaning that if a variable altered the hazard ratio (HR) of ACAG by more than 15%, it was considered an important confounder requiring adjustment. By integrating this change-in-estimate criterion for confounder control, we ultimately included ACAG, SAPSII, and LAC in the final Cox proportional hazards model with time-dependent covariates. To address the uncertainty surrounding the ACAG threshold, we proceeded to compare the predictive performance of different cutoff values. We observed that grouping using cutoffs of 19.25 and 19.375 produced identical ACAG classifications, suggesting no practical difference in outcomes. Therefore, we focused subsequent comparisons on the cutoffs of 19.125 and 19.25 by re-running multivariable Cox regression analyses. The C-index values were 0.75314 for a cutoff of 19.125 and 0.75325 for 19.25, demonstrating a negligible difference of 0.000113, which is considered clinically irrelevant. Both models exhibited similar predictive accuracy, with C-index values above 0.7 indicating acceptable discriminative ability. Therefore, the use of 19.25 as the final ACAG threshold is statistically and mathematically sound. Restricted cubic spline (RCS) analysis was performed to evaluate the predictive capacity and non-linear association between ACAG and 60-day mortality risk. To enhance predictive accuracy, ROC curves were plotted for combined analyses of ACAG and clinical scoring systems, and differences between combined and stand-alone ACAG predictions were assessed using the DeLong test with the “roc.test” function from the “pROC” R package. Subgroup analyses were conducted for ACAG by including variables such as age, gender, BMI, ethnicity, shock classification, and scoring systems (LODS, OASIS, RASS, SAPS II). Statistical analyses were conducted using R software (version 4.4.2; The R Foundation for Statistical Computing, Vienna, Austria; http://www.r-project.org) within the RStudio environment. A two-sided p-value of < 0.05 was considered statistically significant. Results Baseline characteristics of the study population A total of 10,144 patients were included in this study (Fig. 1), among whom 6,603 (65.1%) survived and 3541 (34.9%) died within 60 days. The key variable, ACAG, was significantly higher in the non-survivor group compared to the survivor group (20.0 vs. 17.0, p < 0.001)). Study flowchart showing the inclusion and exclusion criteria Predictive performance of ACAG and AG for 60-day mortality Receiver operating characteristic (ROC) curve analysis demonstrated that ACAG had an area under the curve (AUC) of 0.68 (95% CI 0.670.69, p < 0.01) for predicting 60-day mortality (Fig 2). The Youden index reached its maximum value (0.268) at ACAG values of 19.125 and 19.375. Given their proximity, the mean value (ACAG = 19.25) was chosen as the threshold. For AG levels, the AUC was 0.66 (95% CI 0.65–0.67, p < 0.01) with a cutoff value of 16. Comparisons using the DeLong test confirmed that ACAG had significantly better discriminatory performance than AG (p < 0.001). ROC of ACAG and AG Patients were divided into low ACAG (≤19.25) and high ACAG (>19.25) groups based on the ACAG threshold. Compared to the low ACAG group, the high ACAG group had a significantly higher mortality rate (50.83% vs. 24.99%, p < 0.001) and BMI (28.3 vs. 27.8 kg/m2, p < 0.001). Severity scores also differed significantly between groups, with higher values observed in the high ACAG group, including LODS (8 vs. 6, p < 0.001), OASIS (39 vs. 35, p < 0.001), RASS (− 1 vs. 0, p < 0.001), SAPS II (51 vs. 42, p < 0.001), and SOFA (9 vs. 7, p < 0.001). Additionally, vital signs (lower SBP and SpO₂, and higher HR and RR) and laboratory findings (liver function markers, lactate, sodium, potassium, WBC, PLT, RBC, RDW, renal function indicators, coagulation parameters) significantly differed between the high and low ACAG groups (Table 1). Table 1Clinical characteristics of different groups of patientsLevelLow ACAGHigh ACAGOverallP-value(N = 6251)(N = 3893)(N = 10,144)Age66 [55, 75]66 [55, 76]66 [55, 76]0.182Gender (n, %)Male2663 (42.60%)1708 (43.87%)4371 (43.09%)0.216Female3588 (57.40%)2185 (56.13%)5773 (56.91%)Race (n, %)White4186 (67.97%)2458 (63.14%)6644 (65.50%) < 0.001Black578 (9.25%)471 (12.10%)1049 (10.34%)Yellow179 (2.86%)105 (2.70%)284 (2.80%)Other1308 (20.92%)859 (22.06%)2167 (21.36%)Shock group (n, %)Septic shock2921 (46.73%)2030 (52.14%)4951 (48.81%) < 0.001Cardiogenic shock1291 (20.65%)947 (24.33%)2238 (22.06%)Hypovolemic shock316 (5.06%)125 (3.21%)441 (4.35%)Other shock1723 (27.56%)791 (20.32%)2514 (24.78%)Outcome (n, %)Survival4689 (75.01%)1914 (49.17%)6603 (65.09%) < 0.001Death1562 (24.99%)1979 (50.83%)3541 (34.91%)BMI(kg/m2)27.82 [23.70, 32.95]28.26 [24.10, 33.80]27.99 [23.86, 33.29] < 0.001LAC (mmol/L)1.70 [1.20, 2.50]2.50 [1.60, 4.60]1.90 [1.30, 3.10] < 0.001ALT (IU/L)27 [15, 58]40 [19, 117]31 [17, 73] < 0.001AST (IU/L)37 [22, 77]65 [30, 185]44 [24, 106] < 0.001TBIL (mg/dL)0.70 [0.40, 1.30]0.90 [0.40, 2.40]0.70 [0.40, 1.70] < 0.001RBC (m/uL)3.14 [2.73, 3.70]3.23 [2.73, 3.83]3.18 [2.73, 3.75] < 0.001RDW (%)15.70 [14.30, 17.70]16.20 [14.60, 18.40]15.90 [14.40, 18] < 0.001WBC (K/uL)10.10 [7, 14.40]13.30 [8.70, 19.40]11.10 [7.50, 16.20] < 0.001PLT (K/uL)186 [119, 273]176 [105, 264]183 [114, 270] < 0.001CREA (mg/dL)1 [0.70, 1.50]2 [1.20, 3.40]1.20 [0.80, 2.20] < 0.001BUN (mg/dL)23 [14, 37]39 [23, 63]27 [17, 47] < 0.001CA (mg/dL)8.30 [7.80, 8.80]8.20 [7.70, 8.90]8.30 [7.80, 8.80] < 0.001NA (mEq/L)139 [135, 142]137 [134, 141]138 [135, 142] < 0.001GLU (mg/dL)120 [100, 152]134 [103, 186]125 [101, 162] < 0.001INR1.30 [1.20, 1.60]1.50 [1.20, 2.00]1.40 [1.20, 1.80] < 0.001HCO3 (mEq/L)24 [21, 28]20 [16, 23]23 [19, 26] < 0.001PT (s)14.50 [12.80, 17.70]16.10 [13.50, 22.00]15 [13.00, 19.22] < 0.001APTT (s)32.70 [28.40, 43.80]35.70 [29.70, 51.30]33.80 [28.80, 47.00] < 0.001SBP (mmHg)82 [74, 90]81 [71, 89]82 [73, 90] < 0.001RR29 [25, 33]30 [26, 35]29 [25, 34] < 0.001SpO2 (%)92 [89, 94]91 [87, 94]92 [88, 94] < 0.001FHR108 [93, 124]112 [96, 128]110 [94, 126] < 0.001LODS6 [4, 8]8 [6, 10]7 [5, 9] < 0.001OASIS35 [29, 41]39 [32, 46]36 [30, 43] < 0.001RASS0 [− 3, 0]− 1 [− 4, 0]− 1 [− 4, 0] < 0.001SAPSII42 [33, 52]51 [40, 62]45 [35, 56] < 0.001SOFA7 [4, 10]9 [6, 12]8 [5, 11] < 0.001GCS15 [15]15 [15]15 [15]0.923LODS6 [4, 8]8 [6, 10]7 [5, 9] < 0.001 Clinical characteristics of different groups of patients Values are expressed as the median (IQR) or n (%).According to this criterion, patients who were alive at 60 days were classified as survivors, while those who had died were classified as non-survivors. ACAG and mortality: Kaplan–Meier and Cox proportional hazards model with time-dependent covariates Kaplan–Meier survival curves demonstrated significantly lower 60-day survival probabilities for the high ACAG group compared to the low ACAG group (Fig 3, p < 0.001). Table 2 presents hazard ratios (HRs) derived from Cox proportional hazards model with time-dependent covariates. The overall goodness of fit of the model was assessed using the likelihood ratio test (χ2 = 1765, p < 0.001), Wald test (χ2 = 2196, p < 0.001), and Score test (χ2 = 2575, p < 0.001), all of which indicated excellent model performance. In the primary effect analysis of ACAG, patients in the high ACAG group had a 72.5% higher risk of mortality compared to those in the low ACAG group (hazard ratio [HR] = 1.725, 95%CI 1.472–2.023, p < 0.001). This effect was highly statistically significant, indicating that ACAG is a strong predictor of mortality. The time-dependent effect of ACAG was not significant (p = 0.08), suggesting that its predictive effect on mortality remains relatively stable over time without substantial change. For SAPSII, the primary effect demonstrated that each 1-point increase in the score was associated with a 4.5% increase in mortality risk (HR = 1.045, 95% CI 1.041–1.050, p < 0.001). The time-dependent effect indicated that for each unit increase in log (t+1), the effect of SAPSII weakened by 1.3% (HR = 0.987, 95% CI 0.984–0.989). Regarding LAC, the primary effect showed that each 1-unit increase in lactate level was associated with a 14.7% increase in mortality risk (HR = 1.147, 95% CI 1.129–1.166, p < 0.001). The time-dependent effect revealed that for each unit increase in log (t+1), the effect of lactate weakened by 2.5% (HR = 0.975, 95% CI 0.964–0.986, p < 0.001). Kaplan–Meier survival curve of ACAG Table 2Assessment of the impact of ACAG and covariates on 60-day all-cause mortality using a Cox proportional hazards model with time-dependent covariatesHR for the main effect (95%CI)P value for the main effectHR for the time–dependent effect (95%CI)P value for the time- dependent effectACAG1.725 (1.472–2.023)p < 0.0010.929 (0.856–1.009)p = 0.080SAPSII1.045 (1.041–1.050)p < 0.0010.987 (0.984–0.989)p < 0.001LAC1.147 (1.129–1.166)p < 0.0010.975 (0.964–0.986)p < 0.001 Assessment of the impact of ACAG and covariates on 60-day all-cause mortality using a Cox proportional hazards model with time-dependent covariates We included ACAG, SAPSII, and LAC to construct the optimal model. ACAG was defined as a binary variable (low ACAG group = 1, high ACAG group = 2). The main effects and time-dependent effects of these variables were investigated. A p-value less than 0.05 was considered statistically significant. Restricted Cubic Spline of ACAG Levels versus 60-Day All-Cause Mortality Restricted cubic spline (RCS) analysis revealed a linear association between ACAG levels and 60-day mortality risk (χ2 = 2.2, p for non-linearity = 0.332). As depicted in Fig 4, mortality risk increased steadily with rising ACAG levels, highlighting its potential as a robust prognostic biomarker. RCS analysis of ACAG Subgroup analysis Subgroup analyses demonstrated that ACAG remained significantly associated with mortality across multiple strata (Fig 5). Among ethnic groups, Black patients had the lowest risk of adverse outcomes (HR 1.87, 95% CI 1.51–2.30), while other ethnic groups had the highest risk (HR 2.60, 95% CI 2.27–2.97). In shock subgroups, no significant difference was found in hypovolemic shock (p for interaction = 0.113), while septic shock showed the lowest risk of adverse outcomes (HR 1.95, 95% CI 1.77–2.41). Patients with low LODS (≤8) had a higher HR (2.12, 95% CI 1.93–2.32) compared to those with high LODS (>8) (HR 1.73, 95% CI 1.00–1.29). Similarly, patients with low RASS (≤ − 1) exhibited a higher HR (2.18, 95% CI 2.01–2.37) than those with high RASS (> − 1) (HR 1.89, 95% CI 1.69–2.11). No significant interactions were observed between ACAG and age, gender, BMI, OASIS, SAPS II, or SOFA scores (p for interaction > 0.05). Subgroup analysis of ACAG Enhanced predictive accuracy through combined models Combining ACAG with clinical scoring systems, including LODS, OASIS, SOFA, and SAPS II, significantly enhanced predictive accuracy for 60-day mortality. As shown in Fig 6, the AUCs of combined models were significantly greater than those of ACAG alone (p < 0.001), further validating its utility in clinical risk stratification. ROC curve of ACAG combined clinical score Discussion In this large-scale study of 10,144 ICU-admitted shock patients, we demonstrated a significant association between elevated ACAG and increased 60-day in-hospital mortality. The findings of this study underscore the clinical utility of ACAG as a cost-effective and accessible biomarker for monitoring disease progression and predicting outcomes in ICU-admitted shock patients. These findings suggest that ACAG could serve as a valuable prognostic tool, potentially helping clinicians to identify high-risk patients, tailor therapeutic strategies, and possibly improve survival outcomes in critical care settings. ACAG was introduced, theoretically minimizing errors due to hypoalbuminemia and enhancing the reliability of acid–base assessments. It is reasonable to hypothesize that ACAG provides a more accurate assessment of acid–base status and is therefore a superior predictor of mortality in shock patients. In this study, we analyzed the clinical and laboratory differences between survival and mortality groups, identifying for the first time a significant association between elevated ACAG and increased mortality in ICU-admitted shock patients. Notably, ACAG demonstrated greater specificity and sensitivity at its critical threshold for predicting shock mortality, offering unique prognostic value for clinicians in monitoring disease progression and intervening in high-risk populations. Our findings suggest that ACAG is a cost-effective and convenient predictor of mortality in ICU-admitted shock patients. While ACAG has been associated with adverse outcomes in conditions like acute kidney injury, cardiovascular diseases, and acute pancreatitis [9, 16, 22, 23], studies exploring its predictive value for mortality in shock patients remain limited and predominantly small scale. The Kaplan–Meier survival curve visually demonstrates that the 60-month survival probability of the high ACAG group is markedly higher than that of the low ACAG group, which is consistent with previous research by Yuan et al.[24]. The study by Yuxing Wang et al. reported a HR of 1.42 for ACAG in predicting 28-day mortality in 1508 patients with cardiogenic shock in the ICU. In comparison, our study, which had a larger sample size (10,144) and a broader spectrum of disease severity, yielded a higher HR value (1.725) [25]. An intriguing finding emerged from the assessment of the proportional hazards assumption. The global test for the ACAG variable resulted in a significant p-value (p < 0.05), initially suggesting that its effect varied over time. However, in an extended Cox model explicitly specifying a time-dependent covariate for ACAG the interaction term was not statistically significant (p = 0.080). This indicates that the hazard ratio for ACAG remained relatively stable throughout the follow-up period. We posit two plausible explanations for this apparent discrepancy. First, the significant result from the initial proportional hazards test may have been confounded by time-varying effects of other covariates included in the model. Second, given the large sample size, the test may have detected a deviation from proportionality that, while statistically significant, is negligible in magnitude and thus lacks clinical or practical importance. The Cox proportional hazards model with time-dependent covariates constructed in our study provided a more accurate dynamic risk assessment. It identified ACAG as the strongest predictor, with a relatively stable effect over time. LAC was the second strongest predictor (HR = 1.147), but its effect diminished over time; SAPS II had the smallest effect size (HR = 1.045), yet its influence decreased significantly as time progressed. These findings suggest that for shock patients upon admission, priority should be given to monitoring ACAG, lactate, and SAPS II. Monitoring of lactate and SAPS II should be primarily focused on the early stages, while ACAG can serve as a reliable indicator for long-term monitoring. Moreover, we observed a linear association between ACAG and 60-day mortality, with risk steadily increasing as ACAG levels rose. Compared to AG, ACAG demonstrated superior predictive performance. Consistent with findings by Hu T. et al. in septic shock patients [23], our ROC analysis showed that ACAG had a significantly higher area under the curve (AUC) than AG (p<0.001). These results confirm the hypothesis that ACAG outperforms AG in predicting 60-day mortality among shock patients. Various scoring systems play an essential role in predicting mortality in shock [26, 27]. In this study, we incorporated SAPS II, SOFA, OASIS, and LODS scores from the MIMIC database, all of which are used to assess disease severity and prognosis in shock patients. Combining ACAG with these scoring systems improved the predictive ability for 60-day all-cause mortality compared to ACAG alone. Future prospective studies could further evaluate the practical utility of integrating ACAG with scoring systems to enhance mortality prediction. This study has several limitations. First, as a retrospective observational study, it is inherently constrained by the lack of prospective clinical data, which limits the ability to establish causal inferences and definitively confirm the predictive utility of our findings. While the ACAG addresses inaccuracies due to hypoalbuminemia, it does not account for rare cases where abnormal endogenous or exogenous cation levels significantly influence the anion gap. In such rare cases, correcting for hypoalbuminemia may fail to rectify this error, potentially affecting the accuracy of ACAG in assessing acid–base imbalances. Finally, the study utilized data derived from a single database, which, while robust, may not fully represent diverse healthcare settings or populations. Multi-center studies encompassing heterogeneous cohorts would enhance the external validity of these findings. Conclusions This study establishes ACAG as a rapid, cost-effective biomarker for assessing acid–base imbalances and predicting 60-day mortality in ICU-admitted shock patients. With superior accuracy over traditional AG measurements and enhanced predictive value when combined with scoring systems like SAPS II, SOFA, OASIS, and LODS, ACAG offers significant potential as a complementary tool for risk stratification in critical care. Acknowledgements We would like to thank the MIMIC-IV database for providing the original data and all the medical workers who uploaded the data. Author contributions The corresponding authors Dr. Yao and Dr. Luo were responsible for reviewing the methodology, editing the article, project management, and funding acquisition. Dr. Zou and Dr. Zhong were responsible for methodology, conceptualization, and original drafting. Ms. Gong and Dr. Wang were responsible for validation and formal analysis. Mr. Zeng was responsible for completing the data collation, visualization, investigation, and supervision. Funding This study was supported by the National Natural Science Foundation of China (Grant No. 82371427, Grant No.82401656), Natural Science Foundation of Chongqing, China (Grant No. CSTB2023NSCQ-MSX0323), and Chongqing Science and Health Joint Medical Research Project (2025MSXM044). Data availability The datasets analyzed during the current study are available in [https://github.com/MIT-LCP/mimic-code/tree/main/mimic-iv]. Declarations Ethics approval and consent to participate Human ethics and consent to participate declarations: not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Yan-Ke Zou and Shi-Li Zhong are co-first authors. Contributor Information Xiu-Qing Yao, Email: dryaoxq@cqmu.edu.cn. Ya-Xi Luo, Email: luoyaxi@hospital.cqmu.edu.cn. References Associated Data Data Availability Statement The datasets analyzed during the current study are available in [https://github.com/MIT-LCP/mimic-code/tree/main/mimic-iv].