Study design and data source

This study was a retrospective cohort analysis utilizing patient data obtained from the electronic medical database continuously maintained by the Department of Gastroenterology, the First Affiliated Hospital of Nanchang University (Jiangxi cohort, 2005–2023), and the Medical Information Mart for Intensive Care IV database (MIMIC-IV v2.2, 2008–2019). Detailed descriptions of both databases are provided in Supplementary Method 1.

Study population and ethical approval

From the Jiangxi cohort of 14,650 hospitalized AP patients, 3,094 ICU admissions were identified. After excluding patients under 18 years old, pregnant women, those with end-stage liver or kidney disease, missing baseline IAP data, or only a single IAP measurement, 1,008 patients were eligible for IAP trajectory analysis. To assess the impact of cumulative IAP burden within the first week of ICU stay, 337 patients lacking complete IAP records on days 1, 2, 3, 5, and 7 were further excluded, leaving 671 patients for CumIAP analysis (Fig. 1). Separately, 1280 AP-related ICU admissions were extracted from the MIMIC-IV v2.2 database. After excluding non-first admissions (n = 169), patients aged < 18 or > 75 years (n = 246), ICU stays < 24 h (n = 132), and those with only a single IAP record (n = 650), 83 patients remained for trajectory modeling. However, due to limited longitudinal IAP data (n = 18 with complete IAP records on days 1, 2, 3, 5, and 7), MIMIC-IV was not suitable for CumIAP analysis.

Fig. 1

Flow chart of inclusion and exclusion of study subjects

This study was approved by the Institutional Review Boards of the First Affiliated Hospital of Nanchang University (Approval No. 2011001) and the Massachusetts Institute of Technology. All procedures followed the Declaration of Helsinki and STROBE guidelines. (Supplementary Method 2) [17, 18].

Data collection and missing data handling

Multidimensional data were extracted from the structured electronic medical database of the First Affiliated Hospital of Nanchang University, encompassing demographics and anthropometrics (age, gender, weight), smoking status, drinking status, etiology, comorbidities (history of hyperlipidemia and diabetes), vital signs [temperature, pulse, respiratory, systolic blood pressure (SBP), diastolic blood pressure (DBP)], laboratory test [white blood cell count (WBC), platelet count (PLT), hematocrit (HCT), total bilirubin (TBIL), albumin (ALB), triglyceride (TG), total cholesterol (TC), creatinine (Cr), blood urea nitrogen (BUN)], medications and treatments [Insulin, albumin supplementation, low-molecular-weight heparin (LMWH), surgical, percutaneous catheter drainage (PCD), volume of fluid resuscitation (VFR)], APACHE II, and in-hospital outcomes such as length of stay (LOS), acute necrotic collection (ANC), IPN, PMOF, and mortality. Data from the MIMIC-IV v2.2 database were retrieved using SQL via Navicat Premium 15.0.12, focusing on repeated IAP measurements during ICU stay and in-hospital mortality for trajectory modeling.

Missing data were assessed by calculating the proportion of missingness for each variable (Table S1), and appropriate imputation methods were applied based on missingness degree [19, 20], as detailed in Supplementary Method 3.

Measurement of IAP and calculation of CumIAP

IAP was measured using the standardized bladder manometry method. Patients were placed in the supine position, 25 mL of sterile saline was instilled via a Foley catheter, and a pressure transducer was connected. The pubic symphysis was used as the zero reference point, and IAP was recorded at end-expiration [12, 21]. The worst IAP values on days 1, 2, 3, 5, and 7 after ICU admission were used to calculate the CumIAP over the first week. The calculation formula is CumIAP = (IAPday1 + IAPday2)/2 + (IAPday2 + IAPday3)/2 + (IAPday3 + IAPday5) + (IAPday5 + IAPday7) [22,23,24].

Diagnosis of SAP and research outcomes

According to the Revised Atlanta Classification in 2012 [2], AP can be diagnosed if the following two items are met: (1) Typical symptoms (e.g., severe acute upper abdominal pain); (2) Serum amylase or lipase levels > 3 times the upper limit of the normal; (3) CT/MRI shows pancreatic inflammatory changes (such as enlargement, exudation, necrosis). On this basis, SAP is defined as the presence of persistent organ failure (≥ 48 h, modified Marshall score ≥ 2 points, such as respiratory failure, circulatory failure, acute renal failure, etc.).

The primary outcome of this study was in-hospital mortality. Secondary outcomes included IPN and PMOF during the hospital stay. Follow-up began at ICU admission and ended at either in-hospital death or discharge, whichever occurred first.

Statistical analysis

Descriptive analysis: All subjects were grouped according to the tertiles of CumIAP to compare the baseline characteristics and clinical outcomes. Continuous variables were assessed for normality using the Kolmogorov–Smirnov test and Q-Q plots (Table S2, Fig. S1). Variables with normal distribution were presented as mean ± standard deviation (SD) and compared using one-way ANOVA, while non-normally distributed variables were expressed as median (interquartile range) and compared using the Kruskal–Wallis H test. Categorical variables were expressed as frequency (percentage), and the chi-square test was used for the comparison between groups. Kaplan–Meier curves were drawn and log-rank tests were performed to compare the differences in cumulative survival risks among the three groups of CumIAP.

Correlation analysis: Before the regression analysis, multicollinearity was assessed using multiple linear regression. Variables with a variance inflation factor value > 5 were considered to exhibit multicollinearity and were excluded from the multivariable regression models [25]. In addition, schoenfeld residual analysis was used to verify whether the association between CumIAP and the risk of in-hospital death conformed to the proportional hazards assumption. Cox regression was applied to evaluate the association between CumIAP and in-hospital mortality, while multivariate logistic regression assessed correlations with IPN and PMOF. Dose–response relationships between CumIAP and adverse outcomes were modeled using restricted cubic splines (RCS) with four knots. According to the results of the collinearity analysis and clinical relevance, in the above regression model, the demographic characteristic indicators (age, gender) of the subjects, past medical history (history of diabetes, history of hyperlipidemia), the four vital signs (temperature, pulse, respirations, SBP) upon ICU admission, and laboratory test indicators (ALB, TC, Cr, HCT, PLT) were adjusted. Additionally, the predictive performance of CumIAP for adverse outcomes was assessed using receiver operating characteristic curves, with calculation of the area under the curve (AUC), optimal cutoff values, sensitivity, and specificity.

Mediation effect analysis: To deeply analyze the potential pathological mechanism and clinical significance of CumIAP, a mediation effect analysis was further carried out. The total effect and direct effect of the association between ALB, ANC and the risks of in-hospital death, IPN, and PMOF were calculated, as well as the indirect effect of CumIAP. The mediation percentage was calculated to evaluate the mediating role of CumIAP between ALB, ANC and the poor prognosis of the subjects. The same variables as in the above regression model were adjusted in the mediation analysis model.

Subgroup and sensitivity analysis: Stratified analysis was conducted according to gender, age (60 years old), drinking status, smoking status, and the presence of comorbidities (diabetes, hyperlipidemia), with interaction tested using likelihood ratio tests. In addition, five sensitivity analyses were performed, with the specific methods detailed in Supplementary Method 4.

Analysis of latent class growth mixture model (LCGMM): To compensate for the lack of dynamic fluctuation information in CumIAP, LCGMM analysis was conducted on patients who had at least two IAP measurements within the first 7 days after ICU admission. Maximum likelihood estimation was used to model the IAP trajectories over this 7-day period. Details of the LCGMM construction are provided in Supplementary Method 5 [26,27,28,29]. Subsequently, baseline and outcome differences were compared across trajectory groups, and their prognostic value was further assessed using Cox and logistic regression. An external validation was performed using SAP patients in the MIMIC-IV cohort (n = 83), with group distributions and outcomes compared accordingly.

All statistical analyses of this study were completed using R 4.2.2, Free Statistics 2.3, and Empower 2.0, and a two-sided P < 0.05 was set as statistically significant.