Albumin-Corrected Anion Gap And Mortality In Septic Intensive Care Patients: An Analysis Using Medical Information Mart For Intensive Care IV

Summary

This retrospective study of 8,286 intensive care unit (ICU) sepsis patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database found that a higher albumin-corrected anion gap (ACAG) was significantly associated with increased 30-day hospital and ICU mortality, showing a nonlinear relationship with an inflection point at 23.

Abstract

Early prognostic assessment is crucial for sepsis patients in intensive care units (ICU). This study investigated the relationship between the albumin-corrected anion gap (ACAG) and clinical outcomes in ICU patients with sepsis. This retrospective cohort study analyzed 8,286 sepsis patients from the MIMIC-IV (version 2.2) database. Primary outcomes were 30-day in-hospital mortality and 30-day ICU mortality. The authors used Kaplan-Meier analysis, log-rank tests, multivariable Cox regression, and restricted cubic spline (RCS) regression to evaluate associations between ACAG and the outcome. Subgroup analyses assessed consistency across different factors. Higher ACAG levels were significantly associated with increased mortality risks. As a continuous variable, ACAG independently predicted higher in-hospital mortality in fully adjusted models (HR 1.01, 95% CI 1.01–1.02, p < 0.001). Quartile analysis revealed that Q4 patients (ACAG > 26.62) had 41% higher in-hospital mortality risk than Q1 in fully adjusted models (HR 1.41, 95% CI 1.24–1.49, p < 0.001). For ICU mortality, Q4 showed 54% higher risk than Q1 (HR 1.54, 95% CI 1.41–1.75, p < 0.001), although continuous ACAG was not significant in the fully adjusted model (p = 0.099), suggesting a threshold effect. RCS analysis identified a nonlinear association with an inflection point at ACAG = 23. Subgroup analyses confirmed consistent effects, with significant interactions for body mass index, acute kidney injury, continuous renal replacement therapy, and mechanical ventilation. Higher ACAG levels are significantly associated with increased 30-day mortality in ICU sepsis patients, showing a nonlinear pattern with an inflection point at 23. Monitoring ACAG may aid risk assessment in critically ill sepsis patients, particularly when values exceed this threshold. Further research is needed to elucidate mechanisms and validate clinical utility.

Introduction

Sepsis is a heterogeneous syndrome resulting from an inappropriate body response to infection. Its incidence and fatality rates remain high across various populations and age groups, imposing a significant burden on the global healthcare system¹. Sepsis and septic shock are critical conditions characterized by inadequate tissue perfusion and hypoxia, leading to multiple organ dysfunction and posing life-threatening risks^2,3. Despite advancements in the diagnosis and treatment of sepsis, it continues to be a major cause of high patient mortality, second only to severe cardiac conditions⁴. By applying minimal clinical data at the patient’s arrival for diagnosis, healthcare providers can promptly and effectively evaluate prognosis, identify potentially critical cases, and administer targeted early treatment. This approach can reduce mortality among sepsis patients. A key focus is identifying biomarkers to predict sepsis mortality.

In clinical practice, multiple parameters are employed to predict sepsis, including C-reactive protein (CRP), procalcitonin (PCT), interleukin-6 (IL-6), white blood cell count (WBC), and the neutrophil-to-lymphocyte ratio (NLR), among others⁵. However, some of these indicators or scoring systems exhibit certain limitations, such as a lack of specificity or low sensitivity⁶. Additionally, some parameters are too complex for easy application in clinical settings. Compared to these traditional biomarkers, the albumin-corrected anion gap (ACAG) offers the potential advantage of integrating both acid-base status (via the anion gap) and nutritional/inflammatory status (via albumin), while being easily calculable from routine laboratory tests. Hence, there remains a need for sensitive, specific, and easily obtainable parameters to predict sepsis severity and prognosis. This is crucial for clinicians to develop effective diagnosis and treatment plans, efficiently allocate medical resources, and alleviate the financial burden on patients and families.

Metabolic acidosis is commonly observed in critically ill patients due to various causes⁷. Studies have indicated that the type of metabolic acidosis at ICU admission correlates with patient mortality^8,9. Clinically, the serum anion gap (AG) is typically used to assess the type of metabolic acidosis and the degree of acid-base imbalance^10,11. Current research has shown that AG is a promising predictor for various diseases, including heart failure¹², myocardial infarction¹³, acute kidney injury¹⁴, and cerebral infarction¹⁵. Conversely, a study¹⁶ reported that AG’s predictive ability for sepsis mortality is limited. However, other researchers¹⁷ have suggested that this might be due to the influence of albumin charge, causing AG to exhibit “pseudo-normal” behavior^18,19,20. The albumin-corrected anion gap (ACAG) may better reflect the acid-base balance in patients with liver²⁰ and kidney diseases²¹, thereby guiding clinical treatment.

Recent studies^14,15 have explored ACAG in specific sepsis subgroups, such as those with acute kidney injury²² or cirrhosis²³, and in older adults²⁴, but a comprehensive evaluation of ACAG in a large, unselected ICU sepsis population—particularly the identification of a nonlinear relationship and a clinically actionable threshold—has not been performed. Therefore, this study aims to investigate the independent association between ACAG and 30-day mortality (both in-hospital and ICU) in a large cohort of ICU patients with sepsis using the MIMIC-IV database, explore the potential nonlinear relationship between ACAG and mortality using restricted cubic spline (RCS) regression, specifically testing for the presence of an inflection point; and assess the consistency of this association across clinically relevant subgroups defined by demographics, comorbidities, and organ support therapies. It is hypothesized that elevated albumin-corrected anion gap levels are independently associated with increased 30-day mortality, with a nonlinear relationship above a specific threshold, beyond which mortality risk rises sharply.

Protocol

Data source and study design

A retrospective cohort study was performed using data sourced from version 2.2 of the Medical Information Mart for Intensive Care IV database. This database comprises two in-house systems: a comprehensive hospital-wide electronic health record (EHR) and an ICU-specific clinical information system, covering data from 2008 to 2024¹⁸. This study was performed in compliance with institutional guidelines. The MIMIC-IV database received approval from the Massachusetts Institute of Technology and the Beth Israel Deaconess Medical Center. This study was reviewed and approved by the Human Research Ethics Committee of Ningbo No. 2 Hospital (approval number: YJ-NBEY-KY-2022-107-01, dated September 7, 2022). Due to the retrospective design and the anonymized nature of the patient health information, the same ethics committee waived the requirement for informed consent under the same approval (YJ-NBEY-KY-2022-107-01). Access to the database was secured by one of the authors (JQ L), who received the necessary authentication and completed the Collaborative Institutional Training Initiative examination (authentication number 60691748).

Given the retrospective nature of this study, traditional sample size calculations or power analyses were not performed. However, the volume of available records in the database was deemed adequate to meet the study’s objectives. This assessment is based on investigating the relationship between the albumin-corrected anion gap (ACAG) and mortality outcomes within a large and diverse patient population. The convenience samples drawn from the database effectively represent the intensive care population, thus supporting the statistical analyses employed in this study.

Participants

The study encompassed all sepsis patients from the MIMIC-IV v2.2 database. Sepsis was defined according to the Sepsis-3.0 criteria, implemented using structured query language (SQL) queries to identify patients with suspected infection and a sequential organ failure assessment (SOFA) score change of ≥2 within 24 h¹⁹. All patients in this cohort were admitted to the ICU as the first point of hospital entry; patients admitted to general wards and later transferred to the ICU were excluded to ensure a homogeneous population with early ICU admission. Data extraction was performed using SQL via a database management tool²⁰, with all utilized software, databases, and analytical tools listed in an alphabetically sorted Table of Materials. Inclusion criteria specified sepsis patients aged over 18 years who were admitted to the ICU for the first time. The exclusion criteria included: (1) patients under 18 years old; (2) ICU stays of less than 48 h; (3) multiple ICU admissions due to sepsis; and (4) insufficient data, such as missing records for albumin or anion gap (Figure 1).

Research procedures and definitions

Data extraction from MIMIC-IV was performed using SQL queries. The extracted data encompassed patient demographics (age, height, weight, gender, insurance status, race, and marital status), medical history (including hypertension, type 2 diabetes, heart failure, myocardial infarction, malignant tumors, chronic kidney disease, cirrhosis, hepatitis, pneumonia, chronic obstructive pulmonary disease, and hyperlipidemia), and initial laboratory test results. These results included white and red blood cell counts, sodium, potassium, calcium, chloride, glucose, anion gap, blood pH, thrombin time, fibrinogen, partial thromboplastin time, international normalized ratio, total cholesterol, aspartate aminotransferase, and alanine aminotransferase. Additionally, information on special treatments (such as continuous renal replacement therapy and mechanical ventilation), clinical scores (SOFA, acute physiology and chronic health evaluation III [APACHE III], simplified acute physiology score II [SAPS II], Oxford acute severity of illness score [Oasis], Charlson comorbidity index [Charlson], and Glasgow coma scale [GCS] scores), and clinical outcomes (length of hospital stay, in-hospital mortality, ICU stay, and ICU mortality) was collected. During data cleaning, predictors with >30% missing data were excluded.

The albumin-corrected anion gap (ACAG) was calculated using the first available measurements of serum anion gap and serum albumin obtained within 24 h of ICU admission. If multiple measurements were available within the first 24 h, the values closest to the admission time were used. The formula for ACAG is:

Albumin - corrected anion gap = Anion gap + [40 - Serum albumin (g/dL)] × 0.25²¹ (1)

Outcomes and Measures

The primary outcomes of this study were 30-day in-hospital mortality and 30-day ICU mortality. The index date for survival analysis was the time of ICU admission. Patients were followed for 30 days from this index date, regardless of hospital discharge status. Mortality events occurring after discharge were captured using the social security death index (SSDI) records integrated within the MIMIC-IV database, ensuring accurate ascertainment of out-of-hospital deaths.

Statistical analysis

Continuous variables were expressed as mean (standard deviation) or median (interquartile range), while categorical variables were presented as frequency and percentage. For data adhering to a normal distribution, t-tests or analysis of variance (ANOVA) were utilized. Data not following a normal distribution were analyzed using the Mann-Whitney U test or Kruskal-Wallis test. Kaplan-Meier survival analysis was employed to assess the incidence of endpoint events across different levels of ACAG, with differences evaluated using the log-rank test (Figure 2).

The Cox proportional hazards model was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between ACAG and mortality. ACAG was analyzed both as a continuous variable and by quartiles. Model adjustments were made as follows: model 1 involved univariate analysis; model 2 adjusted for age, gender, height, weight, and race; and model 3 further adjusted for insurance status, marital status, WBC and RBC counts, RDW, albumin, chloride, ALT, AST, SOFA score, APACHE III score, SAPS II score, Oasis score, Charlson score, and a range of comorbidities including hypertension, type 2 diabetes mellitus, heart failure, myocardial infarction, malignant tumors, chronic kidney disease, acute renal failure, cirrhosis, hepatitis, pneumonia, stroke, hyperlipidemia, acute kidney injury, and chronic obstructive pulmonary disease. Collinearity among covariates in model 3 was assessed using the variance inflation factor (VIF), with VIF > 5 indicating significant collinearity. No significant collinearity was detected (all VIF < 3).

A restricted cubic spline (RCS) regression model with four knots was applied to examine the nonlinear relationship between baseline ACAG and mortality in both hospital and ICU settings. The RCS analysis specifically tested for an inflection point where the slope of the mortality risk curve changes. Subgroup analyses were conducted to explore potential differences across strata based on age (≤70 years vs. >70 years), sex, body mass index (BMI; <21.4, 21.4–31.2, ≥31.2 kg/m²), presence of hypertension, type 2 diabetes, myocardial infarction, chronic kidney disease, stroke, acute kidney injury, as well as use of continuous renal replacement therapy (CRRT) and mechanical ventilation. These analyses evaluated the consistency of ACAG’s prognostic value for the primary outcomes. Cox models were also used in subgroup analyses to adjust for all baseline variables. Interaction p-values were calculated to assess effect modification. Data processing and analysis were performed using R version 4.3.0 (RRID: SCR_001905), with statistical significance set at two-tailed p < 0.05. For handling missing values, multiple imputation by chained equations (MICE) was applied using the R package “mice” under the missing at random assumption. Variables with missing values exceeding 30% were excluded prior to imputation.

Results

Among adult patients in the MIMIC-IV database, 22,517 subjects met the eligibility criteria. Initially, 152 prognostic factors were extracted from the database. Following data cleaning, 105 predictors with over 30% missing data were excluded. Ultimately, 47 prognostic factors were included in the model for analysis.

Characteristics of included patients

A total of 8,286 sepsis patients who met the study’s inclusion criteria were included. The screeni...

Discussion

This study demonstrates that a higher albumin-corrected anion gap (ACAG) is significantly associated with increased 30-day mortality in patients with severe sepsis. A key novel finding is the nonlinear relationship between ACAG and mortality, with an inflection point at a value of 23, above which the risk escalates disproportionately. Subgroup analyses revealed significant effect modification by body mass index (BMI), acute kidney injury (AKI), continuous renal replacement therapy (CRRT) use, and mechanical ventilation, ...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Database management tool	Navicat Premium (version 16)	https://www.navicat.com	Used for executing SQL queries and extracting data from the MIMIC-IV database.
MIMIC-IV database (version 2.2)	MIT Laboratory for Computational Physiology	https://mimic.mit.edu	Source of de-identified electronic health records for critically ill patients; access requires completion of CITI training.
PostgreSQL	PostgreSQL Global Development Group	https://www.postgresql.org	Relational database management system used for querying the MIMIC-IV database.
R package: ggplot2	R Foundation for Statistical Computing	https://cran.r-project.org/package=ggplot2	Used for generating publication-quality figures, including Kaplan-Meier curves and RCS plots.
R package: mice	R Foundation for Statistical Computing	https://cran.r-project.org/package=mice	Used for multiple imputation of missing data under the missing at random assumption.
R package: rms	R Foundation for Statistical Computing	https://cran.r-project.org/package=rms	Used for restricted cubic spline (RCS) regression to model nonlinear relationships.
R package: survival	R Foundation for Statistical Computing	https://cran.r-project.org/package=survival	Provides functions for Kaplan-Meier survival analysis and Cox proportional hazards regression.
R software (version 4.3.0)	R Foundation for Statistical Computing	RRID: SCR_001905	Statistical computing environment for all data analyses, including Cox regression, RCS, and subgroup analyses.