Development of a predictive model for severe hyperlipidemic acute pancreatitis based on LASSO regression: a retrospective study

Introduction

Hyperlipidemic acute pancreatitis (HLAP), defined as acute pancreatitis resulting from hyperlipidemia, is characterized by significantly increased triglyceride levels, moderately elevated amylase levels, pseudohyponatremia, and a tendency to occur at a younger age (Expert Consensus on the Diagnosis and Treatment of Acute Pancreatitis with Hypertriglyceridemia (Emergency Department) (Expert Group), 2021; Zhou et al., 2024; Fan, Li & Wu, 2025). In recent years, the incidence of HLAP has been increasing, making it the second most common cause of acute pancreatitis (Chinese Pancreatic Surgery Association et al., 2021). HLAP is associated with high morbidity and mortality rates, imposing a significant medical burden (Deng et al., 2025). Therefore, it is of great significance to study the risk factors for HLAP-related severe cases and develop a prediction model, in order to identify severe cases early and initiate early and active intervention, thereby reducing the incidence of severe cases, mortality rate, and alleviating the disease burden.

A number of scholars have studied the risk factors related to the severe progression of HLAP, such as serum calcium, D-dimer, neutrophil percentage, C-reactive protein (CRP), and blood urea nitrogen (BUN). The research methods have mostly included single-factor analysis and logistic regression analysis (Thong, Mong Trinh & Phat, 2021). Some scholars have studied the predictive performance of a single indicator or combined indicators for severe HLAP (Zhu et al., 2020), while others studied the predictive performance of commonly used acute pancreatitis scoring systems for severe HLAP (Qiu et al., 2015; Yang et al., 2016). However, there are few studies that specifically target the construction of a model for predicting severe HLAP (Dong et al., 2024; Lin et al., 2024). No prediction models or scoring systems have been recommended for clinical use by guidelines or consensus statements so far. Therefore, developing early diagnostic prediction models with excellent performance that are specifically designed for the severe progression of HLAP in the early stage is urgently needed. LASSO regression is an effective variable selection tool that can solve the problems of collinearity and overfitting, which is particularly important in the era of big data (Emmert-Streib & Dehmer, 2019). In this context, in this study, we collected case data of HLAP patients treated at Taixing People’s Hospital Affiliated to Yangzhou University, used LASSO regression to screen and analyze risk factors, and constructed a logistic model to help clinicians accurately identify severe patients in the early stage.

Materials & Methods

Results

Variables selection of LASSO regression

The 50 variables identified through the univariate analysis were further analyzed using 10-fold cross-validation LASSO regression (Figs. 1 and 2), with Lambda.1se = 0.076 selected as the optimal solution. At this time, the screened variables were D-dimer, serum calcium, cholesterol, SB, total carbon dioxide, and CAR (Table 2).

Table 2:

Partial regression coefficients of non-return-to-zero variable of lasso regression (Lambda= Lambda.1se).

Variable	Partial regression coefficients
D-dimer	0.0002
Ca	−1.2003
Total cholesterol, TC	0.0409
Standard bicarbonate, SB	−0.0131
Total CO2	−0.0275
C-reactive protein/albumin	0.0037

DOI: 10.7717/peerj.20471/table-2

Logistic regression model

The abovementioned variables were included in the multivariate logistic regression analysis, and the results showed that D-dimer, serum calcium, and cholesterol were the risk factors for severe HLAP (AUC = 0.8341 (95% CI [0.7724–0.8958]); Table 3). The ROC curve graph is presented in Fig. 3, showing that the model has good discriminatory ability. The Hosmer–Lemeshow goodness-of-fit test was conducted (χ² = 6.8383, P = 0.5542). The calibration curve graph is shown in Fig. 4, where the curve is close to the ideal curve, indicating that the model is consistent.

Table 3:

Logistic regression model.

	Coefficients	Standard error	Wald Z	P	OR	95%CI
Constant	2.7796	1.5088	1.84	0.065	16.1118	0.8372–310.0716
D-dimer	0.0009	0.0003	3.59	<0.001	1.0009	1.0004–1.0015
Ca	−3.4379	0.6803	−5.05	<0.001	0.0321	0.0085–0.1219
Total cholesterol, TC	0.1744	0.0326	5.35	<0.001	1.1905	1.1168–1.2691

DOI: 10.7717/peerj.20471/table-3

Notes:

The significance of coefficients was assessed using the Wald test. P < 0.05 was considered statistically significant.

Discussion

This study utilized LASSO regression for variable selection and logistic regression modeling, identifying elevated D-dimer levels, decreased serum calcium, and increased cholesterol as significant risk factors for severe HLAP. These findings are consistent with many previous studies (Zhou et al., 2024; Zhu et al., 2020; Dong et al., 2024; Lin et al., 2024; Wu et al., 2025; Liu et al., 2023; Xu et al., 2022). Current research suggests that the pathogenesis of HLAP primarily involves free fatty acids, impaired pancreatic microcirculation, inflammatory response, oxidative stress, and intrapancreatic calcium overload (Qiu et al., 2023). D-dimer is a degradation product of fibrin, and an elevated D-dimer level indicates the presence of a state of hypercoagulability and secondary fibrinolysis activation. This study identified D-dimer as a risk factor for severe HLAP, which may be because HLAP releases large amounts of inflammatory mediators, activates the coagulation and fibrinolysis system, forms microvascular thrombi, and impairs pancreatic microcirculation, thereby exacerbating the condition (Dong et al., 2024; Wu et al., 2025). Previous studies have shown that hypocalcemia often indicates a severe condition, which may be related to the binding of fatty acids to calcium ions and intrapancreatic calcium overload (Lin et al., 2024). Therefore, monitoring serum calcium is of great significance. This study also revealed cholesterol’s predictive role for severe HLAP, which is consistent with the research of Liu et al. (2023), and may be related to the mechanisms underlying intrapancreatic calcium overload and directly damaging endothelial cells in pancreatic cells (Xu et al., 2022). Many previous studies have shown that the ratio of neutrophils to lymphocytes, CRP, PCT, blood glucose, and BUN are risk factors. Our univariate analysis also showed that these variables were significantly different between the mild/moderate severe group and the severe group, but did not pass the LASSO regression and logistic regression screening. It is possible that these variables contribute less significantly to prediction, or that the sample size is small, or there is case bias.

LASSO regression, a regularization technique, effectively screens variables and simplifies models by imposing a shrinkage penalty term, thereby mitigating overfitting issues arising from multicollinearity (Zhou, Li & Zhang, 2021). It is highly efficient and is particularly suitable for high-dimensional data processing (Emmert-Streib & Dehmer, 2019). In this study, univariate analysis, LASSO regression, and logistic regression were combined to select three variables from 107 variables, with good model performance, which can help clinicians judge the severity of HLAP. The AUC value of the model built in this study is smaller than that of some studies, but the variables in this study are all commonly used clinical indicators, do not include various complex scoring systems, and have the advantages of simplicity and practicality. Moreover, the model does not include pleural effusion, peritoneal effusion pancreatic pseudocyst and other variables occured in the middle and late stages. Therefore, it can be used for early diagnosis.

There are limitations to this study. First, we were unable to collect data on some potential risk factors, such as body mass index, previous surgeries, gene mutations, pain score, narcotic usage, alcohol or smoking habits, endocrine functions, and abdominal fat content and distribution. Second, in this study, a relatively large dataset was collected based merely on empirical judgment, without quantitative approaches (such as estimating learning curves) to assess the sample size. Finally, while our model demonstrated good performance, there is still room for improvement.

Conclusions

In this study, risk factors for severe HLAP were identified, and a logistic regression model was developed based on LASSO regression. The results showed that the risk factors for severe cases of HLAP included increased D-dimer, decreased serum calcium, and increased cholesterol. The logistic regression prediction model demonstrated robust performance, with an AUC of 0.8341 (95% CI [0.7724–0.8958]), indicating good discriminatory ability. The model outperformed several existing models in terms of simplicity and practicality, thus demonstrating it to be a valuable tool for early identification and intervention by clinicians. Future research should aim to collect more comprehensive data, including additional risk factors, and conduct multicenter external validation to enhance the model’s generalizability and accuracy.

Supplemental Information