Association Between Albumin-Corrected Anion Gap and In-Hospital Mortality and Sepsis-Associated Acute Kidney Injury

Background

Sepsis, caused by a maladaptive host response to infection, precipitates acute organ dysfunction, prominently impacting the kidneys [1], with acute kidney injury (AKI) manifesting in 40–50% of septic patient, which has a very high mortality rate of up to 60% [2]. Sepsis-associated acute kidney injury (S-AKI) is typically defined as AKI concomitant with sepsis, devoid of other significant causative factors, or as the co-existence of both Kidney Disease: Improving Global Outcomes (KDIGO) criteria and Sepsis-3 criteria [3]. Due to its intricate pathophysiology, the therapeutic arsenal for S-AKI remains limited and lacks consensus [4]. Current interventions – fluid resuscitation, vasopressor medications, and renal replacement therapy (RRT) – are commonly performed but are not very effective in reducing patient mortality [3]. Its global incidence and fatality rates continue to rise, making it a critical public health issue and potentially increasing the risk of mortality in the Intensive Care Unit (ICU) setting. Despite its significance, the effective prediction of mortality rates in S-AKI patients presents substantial challenges. The complexity lies in identifying precise indicators that can accurately predict mortality [5–7].

The anion gap (AG) serves as a crucial measurement in assessing acid-base balance, maintaining its function within a standard range under normal circumstances, and thus providing valuable insight into the acid-base status [8]. In instances of severe sepsis, however, the scenario becomes complicated due to the onset of hyperlactatemia, which triggers a high anion gap metabolic acidosis [9]. Elevated anion gap has been observed in patients with acute kidney injury [10]. Notably, when AG reaches or exceeds 14 mmol/L, there is a heightened risk of short-term mortality during ICU stay for patients with S-AKI [11]. Furthermore, serum albumin (ALB), endowed with a critical role as a negatively charged protein, contributes significantly to the plasma’s buffering capacity by binding to cations such as sodium. This binding propensity, in turn, impacts the calculated anion gap. In patients with AKI, diminished serum albumin levels are frequently observed [12]. It is noteworthy, however, that hypoalbuminemia can lead to a reduction in the anion gap, adding another layer of complexity to the assessment [9]. Hence, to facilitate a more precise evaluation of acid-base balance, particularly when confronted with conditions of hypoalbuminemia, the concept of serum ACAG has been introduced [13].

Despite these advancements, the mechanism underlying ACAG’s influence on S-AKI patients remains largely uncharted. In the present study, we hypothesized that there is an association between ACAG and ICU mortality rates within this patient population. To test this hypothesis, we used the MIMIC database as our investigative tool, seeking to increase understanding of sepsis-associated AKI and to provide insights potentially helpful in clinical practice. Confirmation of this hypothesis would further underscore the importance of routinely considering ACAG in clinical assessments of S-AKI patients.

Material and Methods

DATABASE:

The MIMIC-IV (Medical Information Mart for Intensive Care IV, version 2.2) database, is a contemporary, freely available repository of critical care information. It offers data that previously received institutional review board consent. Capturing records from the period 2008–2019, this database aggregates a wealth of clinical data from patients treated at the Beth Israel Deaconess Medical Center [14–16]. After successfully navigating our online training modules and passing the Protecting Human Research Participants assessment, we were granted permission to extract data from the MIMIC-IV databases. Our research gained ethical clearance from a department linked to the Massachusetts Institute of Technology (Approval Code 56616826). All the information about patients in the database is rendered anonymous, thus eliminating the requirement for explicit informed consent from individual patients. We structured and presented our research in alignment with the guidelines from the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement, ensuring that every step met the ethical benchmarks established in the Declaration of Helsinki.

STUDY POPULATION:

The inclusion criteria were as follows: (1) adults aged 18 years and older; (2) with acute kidney injury (AKI) secondary to sepsis. The diagnostic criteria for sepsis followed Sepsis-3 – for patients with infections or suspected infections in the ICU, the diagnosis of sepsis was made when the SOFA score was ≥2 [17]. Exclusion criteria encompassed the following scenarios: (1) instances where the duration of AKI was less than 48 h; (2) when the patient’s survival period was less than 7 days; (3) patients with stage 5 chronic kidney disease (CKD) or end-stage kidney disease, defined as an estimated glomerular filtration rate of less than 15 ml/min/1.73 m2; and (4) cases with missing key data, such as demographic details or variables needed to calculate traditional severity scores. These criteria were meticulously designed to ensure a robust and relevant study.

DATA EXTRACTION:

Information was sourced from the MIMIC-IV database (v2.2). We selected critical data points which comprised demographic details like age, sex, ethnicity, weight, height, and length of hospital/ICU stay, along with coexisting comorbidities (hypertension, diabetes, chronic pulmonary disease, and myocardial infarction), laboratory tests (hemoglobin, platelets, white blood cell, pH, glucose, calcium, blood urea nitrogen, creatinine, sodium, potassium, chloride, bicarbonate, albumin, and anion gap), vital signs (mean blood pressure, respiratory rate, and heart rate), urine output, Sequential Organ Failure Assessment (SOFA) score, and therapeutic interventions like mechanical ventilation and renal replacement therapy. For metrics that were recorded several times within the first day of admission, we determined an average to maintain data uniformity. The AG was calculated using the formula:

AG (mmol/l)=(sodium+potassium)-(chloride+bicarbonate).

Subsequently, the ACAG was determined using the formula:

ACAG (mmol/l)=[4.4-{albumin(g/dl)}]×2.5+AG [13].

STATISTICAL ANALYSIS:

To assess the distribution of continuous variables, we employed the Kolmogorov-Smirnov test. Variables conforming to a normal distribution are presented as mean±SD and were compared using the independent sample t test. Conversely, non-normally distributed data were denoted by the median and interquartile range (IQR), with the Wilcoxon rank-sum test facilitating comparative evaluations. Additionally, categorical variables were delineated by their counts and respective percentages and analyzed using the χ² test.

We investigated the association between ACAG and in-hospital mortality rates in ICU-admitted patients afflicted with S-AKI. Binomial logistic regression analysis was performed to identify whether ACAG was an independent risk factor for in-hospital mortality for the patients (variables with P values less than 0.1 in the univariable regression analysis were included in the multivariable regression analysis), and receiver operating characteristic (ROC) curves were plotted to clarify its efficacy in predicting in-hospital mortality. The area under the ROC (AUC) curves were compared using the Z-test. We also performed a decision curve analysis (DCA) to clarify whether there were net clinical benefits for patients when ACAG was used to predict in-hospital mortality [18].

Our statistical evaluations were carried out utilizing a combination of IBM SPSS Statistics (v27.0), MedCalc Statistical Software (v19.6.1), and R software (v4.0.3, CRAN). We adopted the approach by Delong et al for the Z-test, facilitated through the MedCalc platform [19]. The DCA was conducted using the R environment, leveraging the “rmda” package. Throughout our analyses, we deemed a P value below 0.05 as indicative of statistical significance.

Results

BASELINE CHARACTERISTICS:

The foundation of our research is built on the MIMIC-IV dataset, which includes an extensive set of 73 181 ICU admissions. From this considerable data pool, we meticulously selected a group of 7750 subjects, separated into 1909 individuals who unfortunately died in the ICU and 5591 who made a successful recovery. Through rigorous analysis, we noted a considerable decrease in both hospital and ICU duration in the death group compared to the survival group (P<0.001). An array of health indicators showed significantly higher values in the death group, including age, SOFA score, levels of blood urea nitrogen, white blood cell count, urea, creatinine, potassium, and AG. Additionally, we noticed a higher frequency of interventions (mechanical ventilation and renal replacement therapy) in the death group. In contrast, the survival group we identified higher levels of hemoglobin, platelets, chloride, bicarbonate, and ALB, as well as increased urine output and a higher pH value, each demonstrating statistical significance (P<0.001). Despite these differences, the 2 groups showed consistency in sex distribution, ethnicity, sodium levels, and the occurrence of chronic pulmonary disease. Table 1 shows the fundamental attributes of the research cohort.

ACAG IS ASSOCIATED WITH S-AKI IN-HOSPITAL MORTALITY:

We discerned marked multicollinearity between AG and ACAG, prompting us to omit AG from subsequent binomial logistic regression evaluations. As ACAG was the subject of the study, the variables we included in the regression analyses were predominantly electrolytes, and due to the multivariable regression variable limitations (which would result in an unstable model if there were too many variables), we included as many variables as possible that were related to the death event. The outcomes of this analysis are detailed in Table 2. ACAG was identified as an independent risk factor for in-hospital mortality in the patients after adjusting for serial confounders (OR=1.129, 95% CI 1.109–1.150, P<0.001).

Through ROC curve evaluations, we gauged the prognostic efficacy of ACAG, AG, and ALB in predicting in-hospital mortality for ICU-admitted S-AKI patients (Table 3). The AUC metrics for ALB, AG, and ACAG were 0.648, 0.598, and 0.675, respectively, as illustrated in Figure 1. In comparing these AUCs, distinct disparities emerged between ALB versus AG (Z=4.449, P=0.0001), ALB versus ACAG (Z=8.702, P<0.0001), and AG versus ACAG (Z=8.171, P<0.0001). Of these metrics, ACAG exhibited the highest sensitivity (56.00%) and the highest Youden’s index (0.2675). In fact, it is difficult to reconcile sensitivity and specificity, as setting high specificity usually leads to a decrease in sensitivity. Therefore, we need a composite indicator (Youden ‘s index) to judge the predictive efficacy of an indicator in a comprehensive way. In contrast, ALB displayed the highest specificity, registering at 72.24%.

COMPARISON OF DECISION CURVES:

In the DCA, ACAG, denoted by the solid red line, consistently surpassed AG, represented by the solid blue line. Moreover, ACAG consistently outperformed ALB, indicated by the solid green line. The descending order of net benefits was observed as ACAG, AG, and ALB, suggesting ACAG is the optimal predictor among the 3 prognostic indicators (Figure 2).

Discussion

Our study draws attention to the importance of the ACAG as a predictor of mortality among patients with S-AKI. Importantly, when ACAG levels surpass the threshold of 20.25 mmol/L, they serve as a potent predictor for in-hospital mortality among intensive care patients with S-AKI. This metric offers healthcare providers a concrete, actionable criterion for patient surveillance, potentially guiding therapeutic interventions and adjustments in treatment strategy. While ALB demonstrated a slightly higher specificity, ACAG consistently showed a higher Youden’s index and sensitivity, thus underscoring the potential clinical utility of ACAG. Relying solely on a single indicator to predict mortality in S-AKI is inappropriate because the predictive value of a single indicator is inevitably not high enough. We demonstrated in this study that ACAG has a better predictive value than AG in the prognostic prediction of S-AKI, which implies that ACAG may be an excellent alternative variable in constructing a scoring system. Meanwhile, the advantage of ACAG over AG may not be an isolated case, but may also be a pattern in other diseases (some studies have already confirmed this, including sepsis [20], acute pancreatitis [21], and acute myocardial infarction [22]), and our study provides more evidence for subsequent studies. Furthermore, this advantage of ACAG over AG has some value in the clinic, and perhaps, in the future, ACAG could be a direct substitute for AG in certain diseases.

The DCA in our study offers a nuanced approach to evaluating the ACAG as a prognostic factor for S-AKI. By quantifying the net benefit, a delicate balance between true positives and false positives, DCA provides a robust method to assess the overall advantages and risks of ACAG’s clinical application [18]. The versatility of DCA lies in its threshold flexibility, allowing comparative assessments across a spectrum of values, and imparting relevant insights across varied clinical contexts and risk profiles. This results in an intelligible interpretation of the predictive capability of ACAG, facilitating a parallel evaluation against other predictive parameters. Further, DCA’s ability to offer quantifiable net benefits at divergent thresholds underpins a tailored decision-making process. It empowers clinicians and patients to arrive at optimal choices aligned with individual circumstances and preferences. Complementing the receiver operating characteristic (ROC) curves, DCA adds valuable dimensions concerning decision thresholds, thereby elevating its practical relevance in clinical practice. The integrated use of DCA and ROC curves in our analysis culminates in a comprehensive appraisal of ACAG’s predictive efficacy for S-AKI mortality. In summary, the incorporation of DCA in our research not only substantiates the significance of ACAG in predicting S-AKI mortality but also highlights the multidimensional strengths of DCA as an evaluative instrument. The synergistic application of DCA with ACAG furnishes refined insights into S-AKI management, contributing to the broader conversation surrounding personalized healthcare strategies.

The relationship between the AG and S-AKI is well established, as AG serves as a crucial clinical indicator for assessing the acid-base balance within the body [23]. In patients with sepsis, underlying mechanisms such as insufficient tissue perfusion, inflammation, and immunosuppression can alter the metabolic profile, causing AG elevation [24]. This rise in AG has been observed to correlate with higher in-hospital mortality rates [25]. Hypoalbuminemia is a common complication in sepsis and has been correlated with mortality rates in ICU patients with sepsis [26,27], and ALB infusion can improve the short-term mortality rates in patients with AKI and septic shock [28]. Specific research demonstrated that with a decrease of 10 mg/L in serum ALB, there is a corresponding drop of 2.5 mmol/L in AG. This relationship is attributable to the charged nature of ALB, leading to circumstances where AG might present as falsely “normal” [29]. Such an understanding emphasizes the need for careful consideration of ALB levels when evaluating AG as a clinical metric, especially in the context of sepsis-related complications. Enter the ACAG, which enhances AG by incorporating the ALB level. ACAG presents a more nuanced perspective on AG, refining our understanding of a patient’s acid-base status. By reflecting the complex interplay between nutritional condition, immune function, inflammation, and tissue injury, ACAG stands out as a more comprehensive tool for evaluating patients with S-AKI. ACAG’s potential superiority over AG and ALB levels as a prognostic factor stems from its multidimensional approach. It not only corrects the AG’s potential discrepancies due to hypoalbuminemia but also provides insights into the overall health status of the patient, thereby yielding a more accurate prediction of mortality risk. While AG remains an invaluable metric for assessing acid-base balance, its prognostic efficacy may be limited by variations in ALB levels. ACAG, by integrating ALB, offers a more holistic patient assessment, underscoring its potential to predict patient outcomes more precisely in the context of S-AKI. This insight into ACAG’s potential advantage calls for more extensive research to validate its application in clinical settings and emphasizes its potential role as a more effective prognostic tool in managing S-AKI [30,31].

Our study introduces a nuanced perspective to the existing body of knowledge by carefully evaluating the potential of ACAG in predicting outcomes for patients with S-AKI. While recognizing the complexity of the condition, we found that the relatively simple and cost-effective measurement of ACAG may have merit, especially in resource-constrained settings. We also noted the ongoing relevance of both AG and ALB in the context of S-AKI, reaffirming their importance in prognostic models. Our findings suggest that ACAG’s predictive efficacy needs further exploration, rather than regarding it as an absolute superior metric. The strength of ACAG, as observed in our study, could potentially contribute to earlier identification of patients at increased risk of mortality, facilitating timely and resource-efficient interventions. However, the integration of ACAG into routine practice for S-AKI must be approached with caution and further substantiated through ongoing research and clinical trials. In this way, our research offers a step toward understanding, rather than a definitive conclusion, highlighting a path that could eventually lead to improved patient care.

While our study provides valuable insights, it is important to recognize its inherent limitations. First, the retrospective nature of the model could introduce potential biases. Second, as the data were sourced from an American population, questions arise regarding the applicability of our findings to other ethnic or regional groups. Moreover, we were not able to identify a causal relationship between ACAG and in-hospital mortality – did elevated ACAG lead to increased mortality, or did the deterioration of patients predisposed to death for a variety of reasons ultimately lead to elevated ACAG? Lastly, the absence of dynamic tracking of ACAG might result in an under- or overestimation of its prognostic significance. However, the robustness of our large sample size lends credence to our conclusions and strengthens our study. Despite these limitations, the scope of our study underscores the reliability of our findings and encourages further exploration in this field. Our research opens the door to understanding the central role ACAG may play in the management of S-AKI, providing a foundation for future work. Nevertheless, we strongly advocate for additional prospective studies and research across diverse populations to firmly establish ACAG’s role as a consistent prognostic tool in the context of S-AKI.

Conclusions

We observed that elevated ACAG levels were correlated with higher risk of mortality for S-AKI patients during their in-hospital period. Furthermore, ACAG appears to be a more valuable predictor in this context than ALB and AG, enhancing its potential clinical utility.