Elevated atherogenic index of plasma is associated with severe acute kidney injury in critically ill patients with cardiovascular disease – a MIMIC-IV analysis with external validation in the eICU-CRD database Authors: Peng Liu, Qiwei Shen, Yang Liu, Kui Hong Published: 2026 ABSTRACT AbstractThe atherogenic index of plasma (AIP) is a lipid-based marker of cardiovascular risk, yet its prognostic value for severe acute kidney injury (AKI) in critically ill patients with cardiovascular disease (CVD) remains uncertain. This retrospective study analyzed 5,872 patients from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, with external validation in the eICU Collaborative Research Database (eICU-CRD). The primary outcome was stage 3 AKI, and secondary outcomes included renal replacement therapy (RRT) and length of stay. Stage 3 AKI was observed in 20.7% of patients, with the incidence increasing across AIP quartiles (Q1–Q4: 15.9–26.8%), and 5.3% of patients requiring RRT (Q1–Q4: 2.6–10.9%). High AIP was associated with stage 3 AKI (adjusted OR = 1.35; 95% CI: 1.11–1.64), RRT (OR = 1.95; 95% CI: 1.33–2.87), longer hospital LOS (β = 3.58; p < 0.001) and ICU LOS (β = 1.14; p < 0.001). Restricted cubic splines revealed linear associations between AIP and both renal outcomes (stage 3 AKI and RRT). Mediation analysis indicated that stage 3 AKI mediated 12.6% of the association between the AIP and in-hospital mortality. The predictive models achieved AUCs of 0.916 for RRT and 0.727 for stage 3 AKI. These associations with renal outcomes were robust, as confirmed by entropy balancing and external validation in the eICU-CRD. Elevated AIP is a robust, externally validated predictor of stage 3 AKI and RRT, with substantial predictive accuracy for RRT. Furthermore, it appears to influence in-hospital mortality partially through renal injury, warranting confirmation in future prospective studies. Introduction Critically ill patients with cardiovascular diseases (CVDs) often have unstable circulation, metabolic disorders, and multiorgan dysfunction, all of which increase the risk of death [1,2]. Acute kidney injury (AKI) is among the most severe complications and is strongly associated with increased mortality [3,4]. Early detection and accurate risk stratification of AKI are critical for improving outcomes and reducing mortality in high-risk populations. However, most currently available biomarkers lack sensitivity for early prediction and fail to reflect a patient’s overall metabolic status [5,6]. The atherogenic index of plasma (AIP), defined as the logarithmic ratio of triglyceride level to high-density lipoprotein cholesterol level, has emerged as a simple and accessible marker of lipid metabolism imbalance and atherosclerotic risk [7,8]. Beyond cardiovascular outcomes, recent consensus statements and the results of some studies have underscored the growing importance of metabolic–renal risk assessment, which emphasizes the close interplay between lipid metabolism abnormalities, systemic inflammation, and renal dysfunction [9–12]. Nevertheless, the potential role of AIP within this framework has not been systematically investigated, particularly in the prediction of AKI. The present study used data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database to explore the association between the AIP and the risk of stage 3 AKI, the need for renal replacement therapy (RRT), and hospital and ICU LOS. To strengthen the robustness and generalizability of the results, the findings were externally validated in the multicenter eICU Collaborative Research Database (eICU-CRD). These findings might provide new insight into metabolic–renal risk assessment and support the use of the AIP as a potential early warning biomarker in critically ill cardiovascular patients. Methods Study design and data source This retrospective study was conducted using data from the MIMIC-IV database (version 3.1) [13]. The MIMIC-IV database contains comprehensive clinical data from patients admitted to the emergency department and intensive care unit (ICU) at Beth Israel Deaconess Medical Center (BIDMC) from 2008 to 2022. The dataset includes records of 364,627 patients, with 223,452 having their first hospital admissions documented and 65,366 of those representing initial ICU admissions. Data extraction and study population For the present study, only patients who were diagnosed with CVD, defined as having at least one of the following conditions: hypertension, coronary artery disease (CAD), atrial fibrillation (AF), or heart failure (HF), were included. The detailed inclusion and exclusion criteria are presented in Figure 1A. Data for this study were extracted by an approved researcher, Qiwei Shen (certification number: 59,548,768). The code for data querying and extraction is available from the MIMIC Code Repository. Demographic information (age, sex, and race), body mass index (BMI), hospital length of stay (LOS), ICU LOS, vital signs (systolic blood pressure (BP), diastolic BP, mean arterial pressure, heart rate), laboratory values (potassium levels, sodium, cardiac troponin T (cTnT), creatine kinase-myocardial band (CK-MB), N-terminal pro-B-type natriuretic peptide (NT-proBNP), blood urea nitrogen (BUN), creatinine, aspartate aminotransferase (AST), alanine aminotransferase (ALT), glucose, triglyceride (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), white blood cell (WBC) count, red blood cell (RBC) count, hemoglobin), comorbidities (hypertension, diabetes, CAD, AF, HF, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG) were extracted using Structured Query Language (SQL) and Navicat Premium version 15.0. Comorbidities were identified through ICD-9 or ICD-10 discharge diagnosis codes. Vital signs and laboratory values were both extracted from the first recorded measurements after patient admission. Treatment data (vasoactive agents, nephrotoxic agents, PCI, CABG) were restricted to those administered within the first 24 h of ICU admission to minimize potential bias. Specifically, vasoactive agents included dopamine, epinephrine, norepinephrine, phenylephrine, vasopressin, dobutamine, and milrinone; nephrotoxic agents included vancomycin, aminoglycosides (gentamicin, tobramycin, amikacin), and ketorolac. Flow chart of the study population inclusion process. (A) MIMIC-IV database. (B) eICU-CRD database. For missing data, we adopted a rigorous two-step strategy to ensure the validity of the estimates. First, variables with missing data rates exceeding 25% were excluded from the final multivariable regression models. This exclusion was chosen over the missing indicator method (dummy variables) because the latter can introduce spurious associations and lead to residual confounding, even under the assumption that data are missing completely at random [14–17]. Consequently, variables such as BMI, NT-proBNP, CK-MB, and cTnT were not included in the adjustment. Second, for the remaining covariates with missing data rates less than 25%, we utilized the multiple imputation method implemented via the mice package in R to minimize information loss. Exposure and outcomes The primary exposure variable was the AIP, defined as the base-10 logarithm of the ratio of triglyceride levels to high-density lipoprotein cholesterol levels (log10[TG/HDL-C]), both measured in mmol/L [18,19]. AIP is considered a surrogate marker for atherosclerotic risk and lipid-related cardiovascular dysfunction [9,20]. The primary outcome was stage 3 AKI, defined according to the Kidney Disease: Improving Global Outcomes (KDIGO) criteria as any of the following: an increase in serum creatinine to ≥3.0 times baseline, an increase in serum creatinine to ≥4.0 mg/dL (≥353.6 μmol/L), initiation of RRT, or a decrease in urine output to <0.3 mL/kg/h for ≥24 h or anuria for ≥12 h [21,22]. Secondary outcomes included the need for RRT, hospital LOS, and ICU LOS. Statistical analyses For normally distributed continuous variables, the data are presented as the means ± standard deviations (means ± SDs) and were compared across groups using one-way ANOVA. For nonnormally distributed continuous variables, the data are summarized as medians (interquartile ranges), and group differences are assessed with the Kruskal–Wallis test. Categorical variables are presented as frequencies and percentages, with differences evaluated by the chi-square test. Collinearity among all baseline variables was first assessed, and variables with high collinearity were selectively excluded. Variables with a variance inflation factor (VIF) > 5 or a tolerance < 0.2 were considered highly collinear and removed from the analysis. Specifically, systolic BP, diastolic BP, AST, hemoglobin, and TC were excluded. The results of the collinearity assessment for all baseline variables are shown in Supplementary Table S1, and the results of the collinearity analysis for the covariates included in the final models are presented in Supplementary Table S2. The association between the AIP and clinical outcomes was assessed using multivariable regression models. Logistic regression was used for binary outcomes, including stage 3 AKI and RRT, whereas linear regression was used for continuous outcomes, including hospital LOS and ICU LOS. AIP was analyzed both as a continuous variable and as a categorical variable based on quartiles: Q1 (< −0.160), Q2 (−0.160 to 0.058), Q3 (0.058 to 0.287), and Q4 (> 0.287). The median AIP was 0.058, and the interquartile range (IQR) was (−0.160, 0.287). Three models were constructed: Model I was unadjusted; Model II was adjusted for age, sex, and race; and Model III was further adjusted for vital signs (heart rate, mean arterial pressure), laboratory examinations (potassium levels, sodium levels, creatinine levels, BUN, WBC count, RBC count, ALT levels, LDL-C, glucose), comorbidities (hypertension, diabetes, CAD, AF, HF, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). Additionally, univariable analyses were performed for all the covariates (Supplementary Table S3). Age, sex, race, and variables with a P value < 0.1 were included in the final multivariable analysis (Supplementary Table S4). Finally, to address baseline imbalances across AIP quartiles, we employed the entropy balancing method to generate weights and conducted a weighted regression analysis (doubly robust estimation) to verify the robustness of our findings [23–25] (Supplementary Table S5). Subgroup analyses and restricted cubic splines (RCSs) were conducted only for the binary outcomes of stage 3 AKI and RRT. Subgroup analyses were conducted by age (<65 years), sex, race, hypertension status, CAD status, AF status, HF status, and diabetes status. RCSs, with four knots placed at the 5th, 35th, 65th, and 95th percentiles of the AIP distribution, were employed to evaluate potential nonlinear associations between AIP and these clinical outcomes. To evaluate the predictive value of the AIP for the primary outcome and RRT, multivariable predictive models were constructed that incorporated age, sex, race, and variables with a P value < 0.1 in the univariable analysis. Notably, creatinine and BUN were excluded from these models to avoid data leakage, as they are inherent components of the AKI diagnostic criteria. On the basis of the final model, receiver operating characteristic (ROC) curves were generated, and the area under the curve (AUC) was calculated. Furthermore, a nomogram was constructed to facilitate individual risk assessment. Additionally, we examined whether stage 3 AKI mediates the association between AIP and in-hospital all-cause mortality using the KHB (Karlson–Holm–Breen) method, which decomposes the total effect into direct and indirect effects while adjusting for confounders [26]. All analyses were performed with R version 4.5.1, except for the KHB mediation analysis, which was conducted in Stata 17.0. Statistical significance was set at p < 0.05. External validation in the eICU-CRD Independent external validation was performed using the eICU-CRD as a separate cohort. The database, which includes clinical records from 208 ICUs across the United States between 2014 and 2015, contains extensive biochemical and physiological monitoring data [27]. Notably, B-type natriuretic peptide (BNP) was extracted in the eICU-CRD as a surrogate for NT-proBNP due to data availability. The same inclusion and exclusion criteria (Figure 1B), variable definitions, and statistical methods as those applied in the MIMIC-IV analysis were adopted. Regarding missing data, the same handling strategy was applied: variables with missing rates exceeding 25% (BNP, CK-MB, and cTnT in this cohort) were excluded from the multivariable adjustment to ensure robustness, while remaining missing values were imputed using the mice package. Through this independent cohort analysis, the robustness, generalizability, and consistency of the findings were further evaluated. Results A total of 5,872 patients with CVD, including CAD, hypertension, AF and HF, were included in the study, with a median age of 70.5 years. Among them, 41.7% were female, and 60.3% were white. The prevalence of comorbidities was as follows: hypertension 84.9%, CAD 49.2%, AF 36.0%, and HF 32.7%. Overall, 20.7% of patients developed stage 3 AKI. Notably, the incidence of stage 3 AKI increased across AIP quartiles, reaching 26.8% in the highest quartile (Q1–Q4: 15.9%, 18.3%, 19.5%, and 26.8%, respectively) (Figure 2A). Additionally, 5.3% of patients required RRT, with the incidence increasing progressively across AIP quartiles (Q1–Q4: 2.6%, 3.6%, 4.2%, and 10.9%, respectively) (Figure 2B). Patients in the highest AIP quartile (Q4) also experienced longer ICU LOSs (Q1–Q4: 2.8, 2.9, 2.8, and 3.3 days) and hospital LOSs (Q1–Q4: 7.7, 8.2, 9.0, and 10.8 days) (Figure 2C–D). Baseline characteristics by AIP quartile are summarized in Table 1. Clinical outcomes stratified by AIP quartiles. (A) Incidence of stage 3 AKI. (B) Incidence of RRT. (C) Length of ICU stay. (D) Length of hospital stay. Table 1.Baseline characteristics of CVD patients according to AIP quartile.CharacteristicsTotal (n = 5,872)Q1, (n = 1,468)Q2, (n = 1,468)Q3, (n = 1,468)Q4, (n = 1,468)P valueAge, years70.5 (60.4–80.5)76.1 (65.6–84.4)72.6 (62.9–81.8)69.3 (60.1–79.2)64.7 (54.9–73.8)<0.001Sex, female, n (%)2447 (41.7)755 (51.4)631 (43.0)532 (36.2)529 (36.0)<0.001BMI, kg/m228.1 (24.5–32.3)26.2 (22.9–30.0)27.3 (23.9–31.4)28.5 (25.1–32.3)30.1 (26.1–34.7)<0.001ICU stay time, days2.3 (1.3–4.2)3.1 (2.1–5.0)2.2 (1.3–4.0)2.1 (1.3–4.0)2.3 (1.3–4.3)<0.001Race, n (%) White3543 (60.3)874 (59.5)907 (61.8)863 (58.8)899 (61.2)0.302 Other/unknown2329 (39.7)594 (40.5)561 (38.2)605 (41.2)569 (38.8)Vital signs Heart rate, beats/min82.0 (72.0–95.0)81.0 (70.0–94.0)82.0 (72.0–94.0)82.0 (72.0–95.0)85.0 (75.0–99.0)<0.001 Systolic BP, mmHg131.0 (113.0–150.0)136.0 (118.0–153.0)132.0 (114.0–150.0)130.0 (112.0–149.0)127.0 (109.0–145.0)<0.001 Diastolic BP, mmHg72.0 (60.0–86.0)74.0 (62.0–87.0)73.0 (61.0–86.0)71.0 (59.0–84.0)70.0 (58.8–84.0)<0.001 Mean arterial pressure, mmHg89.0 (77.0–102.0)92.0 (80.0–103.0)90.0 (78.0–103.0)88.0 (76.0–100.2)87.0 (75.0–100.0)<0.001Laboratory examinations Potassium, mEq/L4.1 (3.7–4.5)4.1 (3.7–4.4)4.1 (3.7–4.5)4.1 (3.8–4.5)4.1 (3.8–4.6)0.033 Sodium, mEq/L139.0 (136.0–141.0)139.0 (136.0–141.0)139.0 (137.0–141.0)139.0 (137.0–141.0)138.0 (136.0–141.0)<0.001 cTnT, µg/L0.1 (0.0–0.7)0.1 (0.0–0.8)0.2 (0.0–0.7)0.1 (0.0–0.8)0.1 (0.0–0.6)0.913 CK-MB, IU/L5.0 (3.0–13.0)5.0 (3.0–14.0)5.0 (2.0–13.0)4.0 (2.0–13.0)5.0 (3.0–12.0)0.448 NT-proBNP, pg/mL3340.0 (1063.2–9696.2)3319.5 (1049.2–10910.5)3862.0 (1283.0–8517.5)3506.0 (1402.0–11325.0)2897.0 (737.0–9322.0)0.306 Creatinine, mg/dL1.0 (0.8–1.3)0.9 (0.7–1.2)1.0 (0.8–1.3)1.0 (0.8–1.3)1.1 (0.8–1.6)<0.001 BUN, mg/dL18.0 (14.0–27.0)17.0 (13.0–24.0)18.0 (13.0–26.0)18.0 (14.0–27.0)20.0 (14.0–33.0)<0.001 AST, IU/L28.0 (20.0–52.0)26.0 (20.0–45.0)26.0 (19.0–45.0)28.0 (19.0–53.0)32.0 (20.0–73.0)<0.001ALT, IU/L22.0 (15.0–39.0)20.0 (14.0–33.0)21.0 (14.0–35.0)23.0 (15.0–40.0)25.0 (16.0–51.0)<0.001 Glucose, mg/dL125.0 (103.0–164.0)120.5 (100.0–150.0)121.0 (102.0–156.0)125.5 (103.0–162.0)136.0 (106.0–188.0)<0.001 TG, mg/dL108.0 (78.0–154.0)66.0 (54.0–79.0)95.0 (82.0–112.0)127.0 (107.0–150.0)202.0 (156.0–264.0)<0.001 TC, mg/dL150.0 (119.0–184.0)157.0 (129.0–186.0)149.0 (120.0–179.0)146.0 (115.0–179.0)149.0 (112.0–192.2)<0.001 HDL-C, mg/dL42.0 (33.0–54.0)60.0 (51.0–72.0)46.0 (40.0–54.0)38.0 (32.0–44.0)29.0 (21.0–36.0)<0.001 LDL-C, mg/dL79.0 (55.0–107.0)78.0 (57.0–104.0)80.0 (58.0–107.0)81.0 (55.0–109.0)75.0 (50.0–108.0)0.004 WBC, K/uL9.6 (7.3–12.7)9.0 (7.0–11.7)9.5 (7.2–12.2)9.6 (7.3–12.8)10.4 (7.6–14.1)<0.001 RBC, K/uL4.1 (3.6–4.6)4.1 (3.6–4.5)4.1 (3.7–4.6)4.2 (3.6–4.6)4.1 (3.4–4.6)0.007 Hemoglobin, g/dL12.3 (10.7–13.7)12.3 (10.9–13.6)12.4 (10.9–13.7)12.4 (10.7–13.8)12.1 (10.2–13.8)0.043Comorbidities Hypertension, n (%)4983 (84.9)1245 (84.8)1259 (85.8)1235 (84.1)1244 (84.7)0.668 CAD, n (%)2887 (49.2)622 (42.4)759 (51.7)762 (51.9)744 (50.7)<0.001 AF, n (%)2116 (36.0)608 (41.4)559 (38.1)501 (34.1)448 (30.5)<0.001 HF, n (%)1921 (32.7)470 (32.0)500 (34.1)480 (32.7)471 (32.1)0.616 Diabetes mellitus, n (%)2041 (34.8)349 (23.8)463 (31.5)543 (37.0)686 (46.7)<0.001 Cerebrovascular disease, n (%)3130 (53.3)924 (62.9)821 (55.9)762 (51.9)623 (42.4)<0.001 Chronic pulmonary disease, n (%)1135 (19.3)273 (18.6)293 (20.0)285 (19.4)284 (19.3)0.829 Renal disease, n (%)1211 (20.6)252 (17.2)290 (19.8)308 (21.0)361 (24.6)<0.001 Sepsis, n (%)599 (10.2)65 (4.4)105 (7.2)147 (10.0)282 (19.2)<0.001Treatments Vasoactive agents, n (%)1231 (21.0)215 (14.6)270 (18.4)322 (21.9)424 (28.9)<0.001 Nephrotoxic agents, n (%)1562 (26.6)245 (16.7)330 (22.5)425 (29.0)562 (38.3)<0.001 PCI, n (%)124 (2.1)21 (1.4)34 (2.3)35 (2.4)34 (2.3)0.220 CABG, n (%)602 (10.3)115 (7.8)160 (10.9)174 (11.9)153 (10.4)0.003Outcomes Stage 3 AKI, n (%)1182 (20.1)234 (15.9)268 (18.3)286 (19.5)394 (26.8)<0.001 RRT, n (%)313 (5.3)38 (2.6)53 (3.6)62 (4.2)160 (10.9)<0.001 Hospital LOS, days8.9 (5.0–15.7)7.7 (4.6–12.8)8.2 (4.9–14.2)9.0 (5.3–15.8)10.8 (5.8–21.0)<0.001 ICU LOS, days2.9 (1.8–5.7)2.8 (1.8–5.1)2.9 (1.7–5.2)2.8 (1.7–5.5)3.3 (1.8–7.3)<0.001ALT: alanine aminotransferase; AST: aspartate aminotransferase; BMI: body mass index; BP: blood pressure; BUN: blood urea nitrogen; CK-MB: creatine kinase-myocardial band; cTnT: cardiac troponin t; HDL-C: high-density lipoprotein cholesterol; ICU: intensive care unit; LDL-C: low-density lipoprotein cholesterol; NT-proBNP: N-terminal pro-B-type natriuretic peptide; RBC: red blood cell count; RRT: renal replacement therapy; TC: total cholesterol; TG: triglyceride; WBC: white blood cell count. Baseline characteristics of CVD patients according to AIP quartile. ALT: alanine aminotransferase; AST: aspartate aminotransferase; BMI: body mass index; BP: blood pressure; BUN: blood urea nitrogen; CK-MB: creatine kinase-myocardial band; cTnT: cardiac troponin t; HDL-C: high-density lipoprotein cholesterol; ICU: intensive care unit; LDL-C: low-density lipoprotein cholesterol; NT-proBNP: N-terminal pro-B-type natriuretic peptide; RBC: red blood cell count; RRT: renal replacement therapy; TC: total cholesterol; TG: triglyceride; WBC: white blood cell count. Elevated AIP is linearly associated with increased risk of stage 3 AKI The primary endpoint of this study was the occurrence of stage 3 AKI. When AIP was treated as a continuous variable, each increase was associated with 35–107% higher odds of stage 3 AKI across all the models (Model I: OR = 1.92, 95% CI 1.63–2.27; Model II: OR = 2.07, 95% CI 1.74–2.47; Model III: OR = 1.35, 95% 1.11–1.64; all p < 0.05) (Table 2). Subgroup analyses revealed no significant interactions across all the predefined subgroups (P for interaction > 0.05), indicating that the association between the AIP and stage 3 AKI was consistent and robust across different clinical conditions (Figure 3A). Association between AIP and stage 3 AKI. (A) Forest plot of subgroup analyses. (B) RCS. The panels represent Model I (top), Model II (middle), and Model III (bottom). Table 2.Associations between AIP and stage 3 AKI in CVD patients.ModelModel IAdjusted Model IIAdjusted Model IIIOR (95% CI)P valueOR (95% CI)P valueOR (95% CI)P valueContinuous variable1.92 (1.63, 2.27)<0.0012.07 (1.74, 2.47)<0.0011.35 (1.11, 1.64)0.003Quartile Q11 1 1 Q21.18 (0.97, 1.43)0.0961.20 (0.99, 1.46)0.0601.08 (0.88, 1.33)0.452Q31.28 (1.05, 1.54)0.0121.32 (1.09, 1.60)0.0051.13 (0.92, 1.39)0.257Q41.93 (1.61, 2.32)<0.0012.06 (1.71, 2.49)<0.0011.34 (1.09, 1.66)0.006P for Trend1.24 (1.17, 1.31)<0.0011.26 (1.19, 1.34)<0.0011.10 (1.03, 1.18)0.006Model I: unadjusted.Model II: adjusted for age, sex, race.Model III: adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). Associations between AIP and stage 3 AKI in CVD patients. Model I: unadjusted. Model II: adjusted for age, sex, race. Model III: adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). When analyzed as quartiles, a graded positive association was observed in all the models. Compared with patients in the lowest quartile (Q1), those in Q2, Q3, and Q4 had progressively higher odds of having stage 3 AKI (Model I: ORs 1.18, 1.28, and 1.93; Model II: ORs 1.20, 1.32, and 2.06; Model III: ORs 1.08, 1.13, and 1.34; P for Trend < 0.05) (Table 2). RCS analysis was conducted to examine the potential linear relationship between the AIP and stage 3 AKI (Figure 3B). In all three models, the overall association between the AIP and stage 3 AKI was statistically significant (P overall < 0.05). However, no evidence of nonlinearity was observed (P for nonlinearity > 0.05), suggesting an approximately linear relationship. Elevated AIP is associated with increased need for RRT and prolonged hospitalization The secondary outcomes included the associations between AIP and renal RRT as well as length of stay. When AIP was treated as a continuous variable, each increase was associated with 95–299% higher odds of RRT across all the models (Model I: OR = 3.99, 95% CI 3.09–5.16; Model II: OR = 3.60; 95% CI 2.74–4.72; Model III: OR = 1.95; 95% CI 1.33–2.87; all p < 0.001) (Table 3). The association between AIP and RRT appeared stronger in patients without HF (P for interaction = 0.035), whereas no significant interactions were detected across other clinical subgroups (P for interaction > 0.05) (Figure 4A). When analyzed as quartiles, a graded positive association was observed in all the models. Relative to patients in the lowest quartile (Q1), those in Q2, Q3, and Q4 had progressively higher odds of RRT (Model I: ORs 1.41, 1.66, and 4.60; Model II: ORs 1.38, 1.56, and 4.13; Model III: ORs 0.99, 1.00, and 1.68; P for Trend < 0.05) (Table 3). Association between AIP and stage RRT. (A) Forest plot of subgroup analyses. (B) RCS. The panels represent Model I (top), Model II (middle), and Model III (bottom). Table 3.Associations between AIP and RRT in CVD patients.ModelModel 1Adjusted Model IIAdjusted Model IIIOR (95% CI)P valueOR (95% CI)P valueOR (95% CI)P valueContinuous variable3.99 (3.09, 5.16)<0.0013.60 (2.74, 4.72)<0.0011.95 (1.33, 2.87)<0.001Quartile Q11 1 1 Q21.41 (0.92, 2.15)0.1121.38 (0.90, 2.11)0.1360.99 (0.58, 1.70)0.975Q31.66 (1.10, 2.50)0.0161.56 (1.03, 2.36)0.0351.00 (0.59, 1.70)0.996Q44.60 (3.21, 6.61)<0.0014.13 (2.84, 5.98)<0.0011.68 (1.04, 2.73)0.035P for Trend1.72 (1.54, 1.93)<0.0011.65 (1.47, 1.86)<0.0011.22 (1.05, 1.42)0.008Model I: unadjusted.Model II: adjusted for age, sex, race.Model III: adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). Associations between AIP and RRT in CVD patients. Model I: unadjusted. Model II: adjusted for age, sex, race. Model III: adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). In addition, the AIP was positively associated with both hospital LOS and ICU LOS. In the fully adjusted model (Model III), the AIP was associated with increases of 3.58 days in the hospital and 1.14 days in the ICU (p < 0.001 for both) (Table 4). Table 4.Association between the AIP score and length of stay in the ICU and hospital.OutcomeModel 1Adjusted Model IIAdjusted Model IIIβ (95% CI)P valueβ (95% CI)P valueβ (95% CI)P valueHospital stay time6.88 (5.86, 7.90)<0.0016.34 (5.28, 7.40)<0.0013.58 (2.54, 4.61)<0.001ICU stay time2.14 (1.72, 2.56)<0.0011.96 (1.52, 2.40)<0.0011.14 (0.72, 1.55)<0.001Model I: unadjusted.Model II: adjusted for age, sex, race.Model III: adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). Association between the AIP score and length of stay in the ICU and hospital. Model I: unadjusted. Model II: adjusted for age, sex, race. Model III: adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). RCS analysis demonstrated a significant overall association between AIP and RRT in all three models (P overall < 0.05). Evidence of nonlinearity was observed in the unadjusted model (Model I, P for nonlinearity = 0.041), whereas no significant nonlinearity was found after covariate adjustment (P for nonlinearity = 0.053 and 0.911 in Models II and III, respectively), suggesting an approximately linear positive association in the fully adjusted models (Figure 4B). Elevated AIP partially influences in-hospital mortality through stage 3 AKI High AIP values were found to be associated with increased mortality, suggesting that metabolic dysregulation reflected by the AIP might contribute to poor prognosis in critically ill cardiovascular patients. Given the strong association between the AIP and stage 3 AKI, a mediation analysis was performed to determine whether stage 3 AKI partially explained this relationship. KHB mediation analysis was employed to decompose the total effect of AIP on in-hospital all-cause mortality into direct and indirect effects. The total effect of AIP on mortality was statistically significant (coefficient = 0.266; 95% CI: 0.005–0.527; p = 0.046). Decomposing this relationship revealed a significant indirect effect mediated through stage 3 AKI (coefficient = 0.034; 95% CI: 0.011–0.057; p = 0.004). In contrast, the direct effect of AIP on mortality was not statistically significant (coefficient = 0.232; 95% CI: −0.029–0.493; p = 0.081). The analysis revealed that 12.6% of the total effect of AIP on in-hospital mortality was mediated by stage 3 AKI. These findings suggest that the adverse prognostic impact of metabolic dysregulation (AIP) on mortality, to a significant extent, is mediated by the development of severe kidney injury (Table 5). Table 5.Mediation analysis of the association between AIP and in hospital mortality in patients with stage 3 AKI.Effect TypeCoefficient (95% CI)P valueDirect Effect AIP -> in-hospital all-cause mortality0.232 (−0.029, 0.493)0.081Indirect Effect AIP -> in-hospital all-cause mortality (via AKI)0.034 (0.011, 0.057)0.004Total Effect0.266 (0.005, 0.527)0.046Proportion of Mediated = 12.6%.adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). Mediation analysis of the association between AIP and in hospital mortality in patients with stage 3 AKI. Proportion of Mediated = 12.6%. adjusted for age, sex, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). External validation in eICU CRD consistent with MIMIC findings In the external validation cohort derived from the eICU-CRD, the associations between the AIP and adverse clinical outcomes in patients with CVD were consistent with the trends observed in the MIMIC-IV database. When analyzed as a continuous variable, higher AIP levels were significantly associated with an increased risk of stage 3 AKI (OR 1.41; 95% CI 1.20–1.66; p < 0.001) and the initiation of RRT (OR 2.29; 95% CI 1.56–3.36; p < 0.001). Quartile-based analysis further corroborated these findings. Compared with patients in the lowest quartile (Q1), patients in the higher quartiles had greater risks of stage 3 AKI (Q2: OR = 1.31; 95% CI 1.12–1.53; Q3: OR = 1.27; 95% CI 1.09–1.50; Q4: OR = 1.36; 95% CI 1.15–1.60; all p < 0.05). Similarly, the risk of needing RRT was significantly greater in the top quartile (Q4: OR = 1.80; 95% CI 1.20–2.71; p = 0.005). Tests for trend across quartiles were statistically significant for both outcomes (stage 3 AKI: OR = 1.09; 95% CI 1.04–1.15, p < 0.001; RRT: OR = 1.24; 95% CI 1.09–1.42, p = 0.001), suggesting a dose–response relationship (Table 6). Table 6.External validation of the association between AIP and renal outcomes in the eICU-CRD database.Modelstage 3 AKIRRTOR (95% CI)P valueOR (95% CI)P valueContinuous variable1.41 (1.20, 1.66)<0.0012.29 (1.56, 3.36)<0.001Quartile Q11 1 Q21.31 (1.12, 1.53)<0.0010.91 (0.58, 1.43)0.683Q31.27 (1.09, 1.50)0.0031.21 (0.78, 1.85)0.393Q41.36 (1.15, 1.60)<0.0011.80 (1.20, 2.71)0.005P for Trend1.09 (1.04, 1.15)<0.0011.24 (1.09, 1.42)0.001adjusted for age, sex, BMI, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). External validation of the association between AIP and renal outcomes in the eICU-CRD database. adjusted for age, sex, BMI, race, vital signs (heart rate, mean arterial pressure), and laboratory examinations (potassium, sodium, creatinine, BUN, LDL-C, WBC, RBC, ALT, glucose), comorbidities (hypertension, diabetes, atrial fibrillation, coronary artery disease, heart failure, cerebrovascular disease, chronic pulmonary disease, renal disease, sepsis) and treatments (vasoactive agents, nephrotoxic agents, PCI, CABG). Robustness validation and predictive value of AIP Univariable analysis revealed potential confounders (p < 0.1) associated with renal outcomes, including heart rate, mean arterial pressure, potassium levels, sodium levels, creatinine levels, BUN levels, LDL-C levels, WBC count, RBC count, ALT levels, glucose levels, diabetes, AF, CAD, HF, renal disease, sepsis, use of vasoactive agents, use of nephrotoxic agents, and use of CABG (Supplementary Table S3). After adjusting for age, sex, race, and these significant covariates, the AIP remained significantly associated with adverse renal outcomes. As a continuous variable, each unit increase in AIP was associated with a greater risk of stage 3 AKI (OR = 1.33; 95% CI 1.09–1.61; p = 0.005) and the need for RRT (OR = 2.01; 95% CI 1.37–2.94; p < 0.001). Consistent results were observed when AIP was analyzed in quartiles, with patients in Q4 exhibiting the highest risk compared with those in Q1 (P for trend < 0.05 for both outcomes) (Supplementary Table S4). To verify the robustness of our findings, we employed entropy balancing. As shown in Supplementary Figure S1, the covariate balance significantly improved after weighting, with standardized mean differences (SMDs, blue points) converging toward zero. In the subsequent weighted regression analysis of the MIMIC-IV cohort, continuous AIP remained a risk factor for stage 3 AKI (OR = 1.25; p = 0.041) and RRT (OR = 1.92; p = 0.003). While the trend analysis for categorical AIP showed marginal significance (P for trend = 0.071 for stage 3 AKI, P for trend = 0.077 for RRT), likely because the strict weighting process reduced the effective sample size, the direction of association remained consistent. Importantly, in the external validation using the larger eICU-CRD database, the results of the entropy-weighted analysis confirmed a robust and statistically significant association. Both continuous AIP and categorical AIP (Q4 vs. Q1) were significantly associated with stage 3 AKI and RRT, with highly significant trends observed across quartiles (P for trend = 0.005 and 0.002, respectively), reinforcing the validity of the AIP as a prognostic marker (Supplementary Table S5). Predictive models were constructed incorporating age, sex, race, and significant covariates identified in the univariable analysis (p < 0.1), while creatinine levels and BUN levels were explicitly excluded to prevent data leakage. The model for stage 3 AKI demonstrated moderate discriminative ability, with an AUC of 0.727 (sensitivity: 0.654; specificity: 0.691) (Supplementary Figure S2A). Notably, the model exhibited excellent predictive performance for RRT, with an AUC of 0.916 (sensitivity: 0.882; specificity: 0.810) (Supplementary Figure S2B). To facilitate clinical application, nomograms based on these multivariable models were constructed to visualize individual risk prediction (Supplementary Figure S2C–D). Discussion In this large retrospective analysis of 5,872 critically ill patients with CVD, we observed that elevated AIP was associated with adverse renal and clinical outcomes. Specifically, a linear dose–response relationship was evident: for every 1-unit increase in AIP, the likelihood of stage 3 AKI increased by 35% (OR = 1.35), and the odds of requiring RRT nearly doubled (OR = 1.95). Risk stratification analysis further revealed that patients in the highest AIP quartile (> 0.287) had the most severe prognosis, with 1.34-fold and 1.68-fold greater associations with stage 3 AKI and RRT, respectively, than with the lowest quartile (< −0.160). Beyond these associations, our predictive models incorporating AIP demonstrated substantial clinical utility, achieving an AUC of 0.727 for stage 3 AKI and an exceptional AUC of 0.916 for RRT prediction. This metabolic burden also coincided with significant resource utilization, extending hospital and ICU lengths of stay by approximately 3.6 days and 1.1 days, respectively. Mechanistically, a mediation analysis revealed that 12.6% of the AIP-related excess mortality risk was mediated specifically through the development of stage 3 AKI. The specific associations with severe renal outcomes (stage 3 AKI and RRT) were consistently validated in the eICU-CRD database, suggesting that the AIP may serve as a valuable prognostic indicator for renal risk stratification in this high-risk population. Early risk stratification is urgently needed because conventional diagnostic tools have significant limitations. Traditional markers, such as serum creatinine levels and urine output, often lag behind actual renal injury. Furthermore, they fail to capture the complex metabolic status of critically ill patients in the ICU [28,29]. Crucially, while AKI has a broad clinical spectrum, stage 3 AKI represents the most devastating phenotype associated with the highest mortality and RRT dependency. However, to our knowledge, previous studies linking AIP to renal outcomes have largely overlooked this specific severe endpoint, leaving the metabolic-renal association in this critical subgroup unexplored. To address this unmet need, our findings identify the AIP, a logarithmic transformation of the triglyceride-to-HDL-C ratio, as a readily available and robust prognostic marker for severe renal outcomes. While prior work has extensively linked AIP to atherosclerosis, coronary artery disease, and metabolic disorders [9,19,30], our study extends these insights into the domain of critical care nephrology. We observed a distinct linear dose–response relationship between the AIP and the risk of stage 3 AKI, suggesting that the AIP reflects a continuous gradient of metabolic–renal stress rather than a simple threshold effect. Translating this into practice, patients with elevated AIP levels may benefit from targeted surveillance strategies, including closer hemodynamic monitoring, cautious fluid stewardship, restricted use of nephrotoxic agents, and earlier nephrology consultation. Therefore, incorporating the AIP into routine admission panels or electronic risk prediction algorithms could represent a cost-effective strategy to optimize metabolic-renal management in this vulnerable population. Potential mechanisms might explain the observed association between AIP and kidney injury. A lipid profile characterized by high triglyceride levels and low HDL-C levels could contribute to endothelial dysfunction, impaired renal microcirculation, oxidative stress, and systemic inflammation, all of which may damage the kidney [31–33]. Because renal structures such as nephrons are nonregenerative, persistent lip toxicity could further impair mitochondrial function and cellular homeostasis [34,35]. In critically ill patients, these detrimental effects might be amplified by low cardiac output, the use of vasopressors, systemic inflammatory responses, and ischemia-reperfusion injury, thereby increasing the risk of severe AKI. Our mediation analysis supports this explanation, suggesting that part of the link between AIP and mortality may be mediated through kidney injury. This study has several limitations. First, although the primary analysis was based on the single-center MIMIC-IV database and was validated in the eICU-CRD database, which partly mitigated single-center bias, both datasets were derived from U.S. ICU settings. Therefore, given the potential differences in geography, population characteristics, and clinical practices, the generalizability of our findings requires verification in other countries. Second, the AIP was measured only at ICU admission; thus, variations in fasting status could have influenced lipid profiles. Third, despite the use of multiple imputations to address missing data, residual bias cannot be excluded. Notably, key echocardiographic parameters, specifically left ventricular ejection fraction, were excluded because of excessive missingness, which precluded a more detailed adjustment for baseline cardiac function. Fourth, owing to the retrospective observational design, inherent limitations such as selection bias and unmeasured confounding cannot be fully eliminated. Despite rigorous multivariable adjustment, our findings establish statistical associations rather than causality. Finally, although cross-database external validation was performed, we were unable to validate the prediction model using primary clinical data from our own center due to sample size constraints. This may limit the generalizability of our findings across different geographic regions and clinical practices. Therefore, future multicenter prospective studies are essential to verify the model’s performance and clinical utility in diverse healthcare settings. Conclusion In conclusion, our findings demonstrate that elevated AIP is significantly associated with stage 3 AKI and RRT, indicating a notably high predictive value for RRT. While entropy balancing and external validation strengthen these results, further prospective studies are needed to confirm the clinical utility of the AIP and its potential role in guiding preventative strategies for critically ill patients with CVD. Supplementary Material Supplementary Table 3.docx Supplementary Table 5.docx Supplementary Table 1.docx Supplementary Table 4.docx Supplementary Table 2.docx Supplementary Figure 2.tif Supplementary Figure 1.tif