Nonlinear association between hematocrit levels and short-term all-cause mortality in ICU patients with acute pancreatitis: insights from a retrospective cohort study

Abstract

Objectives: The purpose of this study was to investigate the relationship between hematocrit levels and the mortality of patients with acute pancreatitis (AP), since limited research has examined this association in intensive care unit (ICU).

Methods: In this study, clinical data were retrieved from Medical Information Mart for Intensive Care database for patients diagnosed with AP. Nonlinear relationships between hematocrit and prognosis were examined through Locally Estimated Scatterplot Smoothing (LOESS) regression, restricted cubic splines (RCS), and U-test analyses. The impact of hematocrit on prognosis was further explored using with a binomial generalized linear model with a logit link, while adjusting for potential confounding factors.

Results: The study encompassed 1,914 patients with AP, revealing a significant difference in hematocrit levels between survivors and non-survivors (33.6 (29.5, 38.1) vs. 32.1 (28.1, 37.4), P < 0.001). Hematocrit emerged as an independent prognostic indicator for mortality in both univariate and multivariate logistic regression analyses (all P < 0.05). Findings from LOESS regression, RCS regression, and the U-test indicated a U-shaped correlation between hematocrit levels and 28-day mortality, with both elevated and decreased hematocrit levels leading to increased mortality risk (P for overall < 0.001). Tertile grouping revealed that lower hematocrit levels (< 30.8%) were associated with heightened 28-day mortality risk (Crude model: Odds ratio (OR) (95%Confidence Interval (CI)) = 1.665 (1.198-2.314); fully adjusted model: adjusted OR = 1.474 (1.005-2.161), all P < 0.05). Survival analyses further supported the adverse prognosis associated with low hematocrit levels.

Conclusions: The findings of this study indicate that in AP patients in the intensive care unit, only low HCT levels were identified as a risk factor for 28-day mortality, despite the presence of a U-shaped correlation between HCT levels and 28-day all-cause mortality.

Keywords: Cohort study; Critical care; Erythrocyte pressure; Pancreatitis; Risk factors.

Fig. 1

Flow chart of the study design

Fig. 2

Correlation analysis between HCT and critical care score in patients with acute pancreatitis

Fig. 3

Association between HCT and 28-days mortality in Patients with Acute Pancreatitis (a) LOWESS regression; (b) RCS regression; (C) Tertile grouping. Note: the study cohort is stratified into thirds according to the tertile values of HCT, and the 28-day mortality rates of patients within each interval are tallied. Logistic regression analysis was conducted to compute the P-value for trend using the HCT tertiles as categorical variables

Fig. 4

The risk score based on HCT levels effectively predicts the 28-day mortality in patients with acute pancreatitis. (a) Kaplan-Meier Survival Analysis of 28-Day Mortality in Patients with Acute Pancreatitis Stratified by HCT Tertiles; (b) ROC curves predicting 90-day mortality rates for various machine learning models: GBC model AUC (95% CI) = 0.852 (0.809–0.895); DTC model AUC (95% CI) = 0.812 (0.756–0.868); RF model AUC (95% CI) = 0.878 (0.840–0.917); GNB model AUC (95% CI) = 0.821 (0.772–0.869); (c) Decision Curve Analysis curves for various machine learning models indicate performance metrics, where curves above represent better model performance; (d) Feature variable importance ranking in the RF model; (e) Visualization of SHapley Additive Explanations illustrating the impact of individual features on RF prediction model outcomes. The color spectrum denotes the degree of influence, with warmer hues indicating higher relevance to 28-day mortality risk and cooler tones indicating lower relevance to 28-day mortality risk