Age-Specific Prognostic Models for Sepsis-Associated Acute Kidney Injury: A Multicenter Cohort Study

Introduction

Sepsis-associated acute kidney injury (SA-AKI) is a common and life-threatening complication in critically ill patients, occurring in approximately 40–60% of individuals with sepsis and contributing to mortality rates exceeding 50% in severe cases [1–3]. Recent epidemiologic studies indicate that SA-AKI is responsible for up to 28% of all intensive care unit (ICU)–acquired renal failure; its incidence is rising due to population aging, increasing antimicrobial resistance, and the growing burden of healthcare-associated infections [4–6]. The global impact of SA-AKI is further aggravated by its long-term sequelae, including a 40% risk of chronic kidney disease progression within 5 years, which substantially increases healthcare costs and diminishes survivors’ quality of life [7,8].

The pathophysiology of SA-AKI involves a complex interplay among systemic inflammation, microcirculatory dysfunction, and metabolic reprogramming, which collectively exacerbate renal injury and hinder recovery. Recent advances in molecular biology have identified key mechanisms such as mitochondrial dysfunction, dysregulated autophagy, and ferroptosis, all of which contribute to tubular epithelial cell death [9–11]. Emerging evidence suggests that immune dysregulation – particularly neutrophil extracellular trap formation and inflammasome activation – plays a critical role in amplifying renal injury in SA-AKI [12]. These mechanisms not only promote acute kidney damage but also impair renal repair, resulting in persistent functional decline even after apparent clinical recovery.

Despite advances in critical care management, including early goal-directed therapy and optimized renal replacement strategies, SA-AKI remains a major determinant of adverse clinical outcomes, such as prolonged hospitalization, increased healthcare costs, and higher long-term morbidity [13]. AKI survivors frequently experience persistent renal dysfunction and elevated risks of chronic kidney disease, emphasizing the importance of improved prognostic and therapeutic approaches. Current evidence indicates that up to 40% of SA-AKI survivors progress to end-stage renal disease within a decade, underscoring the urgent need for more accurate risk stratification tools [14]. Existing prognostic systems, such as the Sequential Organ Failure Assessment (SOFA) and Acute Physiology and Chronic Health Evaluation (APACHE) scores, provide generalized risk stratification but lack adequate specificity and sensitivity to accurately predict mortality in patients with SA-AKI. Prior research has demonstrated that these scoring systems show declining predictive accuracy in older and high-risk subgroups, primarily because they fail to incorporate age-related physiological changes and comorbidities [15]. This limitation highlights the need for more precise, individualized prognostic models to guide clinical decision-making and improve patient outcomes. Recent data suggest that age plays a pivotal role in the pathophysiology and outcomes of SA-AKI: older patients exhibit distinct risk profiles, comorbidities, and treatment responses compared with younger cohorts [16,17]. Conventional “one-size-fits-all” prognostic frameworks do not adequately capture these age-related differences, which may result in suboptimal risk assessment and management. Patient stratification into age-specific subgroups (under 65, 65–80, and over 80 years) may allow identification of unique prognostic variables and facilitate the development of tailored prediction models that more accurately reflect the biological and clinical heterogeneity of SA-AKI. Such an approach aligns with the current movement toward personalized medicine in critical care and may enhance mortality prediction accuracy while informing age-specific therapeutic strategies.

In this multicenter cohort study, we aimed to develop and validate age-specific prognostic models for SA-AKI using critical care data. We hypothesized that (1) clinically modifiable factors (e.g., infection source, lactate levels, and vasopressor use) would display age-dependent associations with mortality, and (2) age-stratified models would outperform conventional severity scores in outcome prediction. By leveraging large-scale electronic health records from the Medical Information Mart for Intensive Care IV (MIMIC-IV) and the eICU Collaborative Research Database (n=3662), we sought to generate actionable insights for risk stratification and identify potential therapeutic targets across age groups. These findings may help bridge the gap between population-level prognostication and individualized care in SA-AKI.

Material and Methods

DATA SOURCE AND ETHICS:

This retrospective cohort study was conducted using data from 2 publicly available critical care databases: MIMIC-IV (version 2.2) and eICU (version 2.0) [18,19]. These databases contain comprehensive, de-identified clinical information about ICU patients, including demographic characteristics, physiological parameters, laboratory results, interventions, and outcomes. The study protocol was approved by the Institutional Review Board of the Beth Israel Deaconess Medical Center (2001-P001699/14) and the Massachusetts Institute of Technology (No. 0403000206). Given that all patient data were anonymized, the requirement for informed consent was waived.

STUDY POPULATION AND DEFINITIONS:

Patients with SA-AKI were identified according to the Sepsis-3 criteria (suspected infection plus a SOFA score ≥2) and the Kidney Disease: Improving Global Outcomes (KDIGO) guidelines (serum creatinine increase ≥0.3 mg/dL within 48 h or ≥1.5 times baseline within 7 days) [20,21]. Patients were excluded if they had end-stage renal disease, a history of kidney transplantation, missing baseline creatinine values, ICU stays shorter than 24 h (representing either mild cases requiring brief monitoring or rapid deterioration resulting in death or withdrawal of care, which could introduce survivorship bias), or age under 18 years. The final cohort (n=3662) was stratified into 3 age groups: under 65 years (n=1207), 65–80 years (n=1571), and over 80 years (n=884) for comparative analysis. This stratification reflected clinically meaningful physiological transitions: (1) the 65-year threshold corresponds to the onset of accelerated renal functional decline, and (2) the 80-year cutoff defines the “oldest-old” subgroup, characterized by substantially reduced renal reserve, higher frailty prevalence, and altered drug metabolism [22].

DATA COLLECTION:

Comprehensive clinical data were obtained from patients with SA-AKI, including demographic characteristics (age and sex), comorbidities (hypertension, diabetes, chronic heart disease, and chronic obstructive pulmonary disease), infection details (primary sites such as pulmonary, urinary tract, catheter-related, skin/soft tissue, and abdominal infections; and causative pathogens including Acinetobacter baumannii, Klebsiella pneumoniae, Escherichia coli, and Pseudomonas aeruginosa), laboratory parameters (white blood cell count, lactate level, blood pressure, and renal function markers), and treatment outcomes (vasopressor use, renal replacement therapy, ICU length of stay [LOS], and total hospital stay). Although the SOFA and Simplified Acute Physiology Score II (SAPS II) systems have recognized limitations in older populations, these widely used severity scores were included to preserve clinical interpretability, ensure comparability with previous studies, and provide baseline risk adjustment for the evaluation of novel predictors, consistent with TRIPOD guideline recommendations.

STATISTICAL ANALYSIS:

Assuming an effect size of adjusted odds ratio (OR)=1.5 (Cohen’s f²=0.04) for key predictors such as lactate, with α=0.05 (two-sided) and 80% power (1−β=0.8), calculations in PASS software indicated that at least 300 outcome events (deaths) per age stratum were required to meet events-per-variable guidelines for logistic regression. The study included 3662 patients (1152 deaths), with subgroup event counts of 387 (under 65 years), 511 (65–80 years), and 254 (over 80 years). Although the oldest group fell slightly below the target, combined stratification provided sufficient statistical power for primary analyses. Continuous variables were presented as medians with interquartile ranges (IQRs) and compared using Kruskal-Wallis tests; categorical variables were expressed as counts with percentages and analyzed using chi-square or Fisher’s exact tests, as appropriate. All statistical tests were 2-tailed, and P-values <0.05 were considered statistically significant. Independent prognostic risk factors across age groups (<65, 65–80, and >80 years) were first screened by univariate logistic regression. Variables with P-values <0.05 and ORs >1 were included in the multivariate analysis to calculate adjusted ORs with 95% confidence intervals (CIs). Factors exhibiting protective effects (OR <1) were systematically excluded to avoid masking risk associations. Predictive performances of key risk factors were evaluated through receiver operating characteristic curve analysis, with area under the curve (AUC) values calculated to assess discrimination. Optimal cutoff values were identified using Youden’s index; corresponding sensitivity, specificity, and 95% CI metrics were reported. Survival outcomes were analyzed using Kaplan-Meier curves for each age group, and differences in survival distributions were compared using the log-rank test. Hazard ratios (HRs) with 95% CIs were estimated with Cox proportional hazards models to adjust for potential confounders. The confounding variables included urinary tract infection, pulmonary infection, catheter-related infection, skin and soft tissue infection, abdominal infection, age, male sex, hemoglobin, platelet count, lactate, white blood cell count, blood urea nitrogen, creatinine, glucose, sodium, potassium, heart rate, systolic blood pressure, diastolic blood pressure, mean arterial pressure, respiratory rate, SOFA score, vasopressin use, epinephrine use, phenylephrine use, dobutamine use, dopamine use, norepinephrine use, continuous renal replacement therapy (CRRT) administration, mechanical ventilation duration, Acinetobacter baumannii infection, Klebsiella pneumoniae infection, Escherichia coli infection, Pseudomonas aeruginosa infection, Staphylococcus aureus infection, fungal infection, hypertension, diabetes, cardiovascular disease, chronic pulmonary disease, ICU LOS, hospital LOS, SAPS II score, Logistic Organ Dysfunction System (LODS) score, systemic inflammatory response syndrome (SIRS) score, and Oxford Acute Severity of Illness Score (OASIS). Variables with more than 20% missing data were excluded to minimize bias; the remaining missing values (generally below 10% for most covariates) were imputed via multiple imputation by chained equations with 5 imputed datasets. All statistical analyses were conducted using R software (version 4.3.0). Data visualization was conducted with the ggplot2 and survival packages.

Results

BASELINE PATIENT CHARACTERISTICS:

In total, 3662 patients with SA-AKI were included and stratified into 3 age groups: under 65 years (n=1207), 65–80 years (n=1571), and over 80 years (n=884). Significant differences were observed across age groups in terms of demographic, clinical, and laboratory parameters. Median ages were 56.0 (50.0–61.0) years in the youngest group, 73.0 (69.0–76.0) years in the middle-aged group, and 85.0 (83.0–88.0) years in the oldest group (P<0.001). Female predominance increased according to age (34.4% in patients under 65 years, 36.1% in those aged 65–80 years, and 44.3% in those over 80 years; P<0.001). Comorbidities significantly varied among age groups. The prevalence of hypertension decreased with age (32.7% vs 30.6% vs 26.2%; P=0.006), whereas diabetes (63.9% vs 68.9% vs 69.6%; P=0.006), chronic heart disease (37.9% vs 56.5% vs 59.3%; P<0.001), and chronic obstructive pulmonary disease (19.7% vs 28.3% vs 27.3%; P<0.001) were more common in older patients. Urinary infections occurred more frequently in older individuals (21.5% vs 24.5% vs 29.2%; P<0.001); Staphylococcus aureus infections declined with age (40.3% vs 34.0% vs 29.1%; P<0.001). Laboratory results demonstrated age-related trends. Older patients had higher platelet counts (P<0.001) and blood urea nitrogen levels (P<0.001), whereas sodium levels (P<0.001), diastolic blood pressure (P<0.001), and mean arterial pressure (P<0.001) declined with advancing age. Heart rate significantly decreased with age (P<0.001); respiratory rate showed a modest but statistically significant difference (P=0.007). Other parameters, including white blood cell count, hemoglobin, and serum creatinine, did not significantly differ among groups. These findings indicate distinct clinical profiles across age groups, suggesting that risk factors and management strategies in SA-AKI require age-specific consideration.

PATIENT OUTCOMES:

Clinical outcomes in patients with SA-AKI significantly differed among age groups. Neurological status, evaluated using the Glasgow Coma Scale, showed no pronounced differences, with median scores of 14.0 (9.0–15.0) in the under 65 and 65–80 groups and 13.0 (9.0–15.0) in the over 80 group (P=0.096). Vasopressor use varied according to age: dopamine administration increased with advancing age (8.7% in patients under 65, 10.1% in those aged 65–80 years, and 14.6% in those over 80; P<0.001), whereas dobutamine was more frequently used among patients aged 65–80 years (5.6% vs 8.9% vs 6.1%; P=0.001). Norepinephrine remained the most commonly administered vasopressor overall (66.0% in under 65 group, 62.9% in 65–80 group, and 63.2% in over 80 group; P=0.198), without significant variation among age groups (Table 1).

CRRT use significantly declined with increasing age (11.5% in under 65 group, 9.1% in 65–80 group, and 5.7% in over 80 group; P<0.001). Both hospital and ICU LOSs decreased with age. Median ICU LOS was longest in the youngest group (3.6 [1.8–8.5] days vs 3.2 [1.7–7.2] vs 3.1 [1.7–6.3]; P=0.002). Hospital LOS followed a similar pattern (13.6 [7.2–24.6] days in under 65 group vs 11.7 [6.7–20.7] in 65–80 group vs 10.2 [6.2–16.9] in over 80 group; P<0.001). These results indicate that older patients with SA-AKI received fewer intensive interventions, such as CRRT, and had shorter hospital stays, potentially reflecting differences in disease severity, therapeutic goals, or age-influenced clinical decision-making (Table 2).

Survival analysis revealed significant age-related differences in mortality among patients with SA-AKI. Twenty-eight-day mortality progressively increased according to age, such that the oldest cohort (over 80 years) demonstrated the highest mortality (log-rank test, P<0.001). A sharp reduction in survival occurred within the first week, particularly among patients older than 80 years. This divergence persisted at 90 days, with further increases in mortality, underscoring the impact of advanced age on long-term survival. In-hospital mortality also significantly differed across age groups; the oldest patients exhibited the highest death rates (P<0.001) and shortest survival durations. Conversely, the youngest group (under 65 years) showed the lowest mortality and most favorable survival outcomes across all time points, confirming that age is a critical determinant of prognosis in SA-AKI (Figure 1).

COMPARATIVE ANALYSIS OF DISEASE SEVERITY SCORES ACROSS AGE GROUPS:

When stratified by age, significant differences were observed in disease severity scores among groups. The SOFA score progressively increased with age (P<0.01), as did the SAPS II score (P<0.01). The OASIS also demonstrated an age-dependent rise (P<0.01), whereas the LODS score did not show significant variation among groups (P=0.13), suggesting relative insensitivity to age-related physiological decline in this cohort (Figure 2).

INDEPENDENT RISK FACTORS FOR SA-AKI ACROSS AGE GROUPS:

Univariate logistic regression identified significant prognostic factors for each age group (Supplementary Materials 1–3). Variables with ORs >1 and P-values <0.05 were entered into multivariate logistic regression models (Figure 3). In patients younger than 65 years (Figure 3A), catheter-related infection (OR 2.36, 95% CI 1.52–3.62), urinary tract infection (OR 1.91, 95% CI 1.32–2.75), and SOFA score (OR 1.11 per point, 95% CI 1.06–1.16), among others, were independent risk factors. In the 65–80-year cohort (Figure 3B), urinary tract infection (OR 2.17, 95% CI 1.61–2.93), lactate (OR 1.24 per mmol/L, 95% CI 1.17–1.32), and SAPS II (OR 1.02 per point, 95% CI 1.01–1.04) were independent predictors. In patients older than 80 years (Figure 3C), catheter-related infection (OR 2.87, 95% CI 1.67–4.96), lactate (OR 1.24, 95% CI 1.14–1.36), and male sex (OR 1.46, 95% CI 1.04–2.06), among other variables, were strongly associated with adverse outcomes.

PERFORMANCES OF PROGNOSTIC MODELS FOR SA-AKI ACROSS AGE STRATA:

The prognostic models demonstrated age-dependent discriminatory performance for SA-AKI, and Model 1 consistently achieved the highest AUC across all age groups: 0.753 (95% CI 0.721–0.785) in patients younger than 65 years (sensitivity 67.0%, specificity 73.1% at a threshold of 0.230), 0.769 (95% CI 0.743–0.796) in those aged 65–80 years (sensitivity 78.2%, specificity 63.2% at a threshold of 0.203), and 0.770 (95% CI 0.737–0.803) in patients older than 80 years (sensitivity 79.7%, specificity 61.5% at a threshold of 0.229). Model 1, which incorporated infection type (urinary or catheter-related), lactate, and vasopressor use (norepinephrine or dopamine), outperformed organ failure scores (SOFA, SAPS II, and LODS) in all age groups. Predictive accuracy declined with age when using severity scores alone (e.g., SOFA AUCs: 0.691 in under 65 group, 0.678 in 65–80 group, and 0.657 in over 80 group). Notably, sensitivity increased in older patients (79.7% in over 80 group vs 67.0% in under 65 group), whereas specificity decreased (61.5% vs 73.1%), indicating age-specific trade-offs in risk stratification (Figure 4, Table 3).

Discussion

LIMITATIONS AND FUTURE DIRECTIONS:

This study shares limitations inherent to electronic health record analyses, including potential biases related to documentation completeness, coding variability (particularly for comorbid conditions), and institution-specific practice patterns that may affect generalizability. Although we applied rigorous phenotyping algorithms to the MIMIC-IV and eICU datasets, 3 specific biases warrant discussion. First, the reduced CRRT utilization among older adults may reflect selection bias (e.g., preferential withholding of therapy in frail patients) rather than true clinical appropriateness, as previously reported [8]. Second, temporal bias could result from evolving CRRT practice guidelines during the study period. Third, unmeasured confounders, such as nephrologist availability, may contribute to center-level variation. Fourth, although conventional severity scores were included to maintain consistency with prior studies, their age-related calibration drift may underestimate true risk in older cohorts. Future models could address this underestimation by dynamically reweighting score components according to age decile. These limitations hinder causal inference and suggest that observed age-related disparities reflect appropriate risk-adapted care, rather than inequitable treatment. External validation using prospectively collected cohorts with standardized outcome assessment is needed to address these concerns. Future research should not only refine age-stratified prognostic models but also evaluate whether algorithmic recommendations perpetuate or mitigate existing biases in critical care resource allocation.

Conclusions

By delineating age-specific risk profiles and validating actionable predictive models, the present study advances personalized care for SA-AKI. Clinicians should prioritize infection control, lactate monitoring, and judicious vasopressor use – particularly in older adults – while applying age-tailored tools for prognostication. The development of fully age-optimized scoring systems should be a priority in future research to address existing prognostic gaps in older patients with SA-AKI.