Developing and validating a predictive model for all-cause mortality in patients with metabolic dysfunction-associated steatotic liver disease

Abstract

Objective: This study aimed to construct a scientific, accurate, and readily applicable clinical all-cause mortality prediction model for patients with metabolic dysfunction-associated steatotic liver disease (MASLD) to enhance the efficiency of disease management and improve patient prognosis.

Methods: This study was a retrospective cohort study based on the National Health and Nutrition Examination Survey database. The 17,861 participants diagnosed with MASLD were randomly assigned to either a training cohort (n = 12,503) or a validation cohort (n = 5358). Potential predictors were subjected to LASSO regression analysis, and independent risk factors were subsequently identified through multivariate Cox regression analysis. An all-cause mortality prediction model was constructed based on the significant predictors, and a nomogram was generated to illustrate the survival probability of patients at various time points. The model's performance was evaluated using receiver operating characteristic (ROC), calibration, and decision curve analysis (DCA) curves.

Results: A multiple Cox regression analysis identified several independent predictors significantly influencing all-cause mortality in patients with MASLD. These included gender, age, smoking status, hypertension, red blood cell count, albumin, glutamyl transpeptidase, glycosylated hemoglobin, and creatinine. The constructed predictive model demonstrated high accuracy in the training and validation cohorts, with AUC values approaching 0.85 at 3, 5, and 10 years, respectively. Calibration and DCA curves were employed to verify the stability and generalizability of the model.

Conclusions: We successfully constructed and validated an all-cause mortality prediction model for MASLD patients. This model provides a powerful tool for clinical risk assessment and treatment decision-making.

Keywords: All-cause mortality; Cox regression analysis; Metabolic dysfunction-associated steatotic liver disease; NHANES; Predictive modeling.

Fig. 1

Participant screening flowchart. FLI fatty liver index, MASLD metabolic dysfunction-associated steatotic liver disease

Fig. 2

LASSO regression for variable selection in MASLD patients to reduce multicollinearity. A The coefficient profiles of predictors showcase the LASSO coefficient profiles for the 30 evaluated predictors. A coefficient profile plot against the logarithmic λ sequence [log(λ)] is displayed. This plot helps to understand how each predictor’s influence varies with changes in the λ value, illustrating the dynamic nature of variable selection in the LASSO model. B Tuning parameter (λ) selection using tenfold cross-validation to show the process of λ optimization in the LASSO model to balance model complexity and predictive accuracy. The dotted vertical line indicates the chosen λ value, which resulted in the selection of 19 significant predictors with non-zero coefficients

Fig. 3

Nomogram for predicting the probability of survival for patients with MASLD at specific time points, namely 3, 5, and 10 years. Values for each variable are individually plotted and correspond to point values assigned from the point scale (top). A total score was obtained from the values of each index and plotted on the total point scale (bottom), which is used to assign a corresponding value for the predicted rate of the nomogram. RBC red blood cell count, GGT gamma-glutamyl transferase

Fig. 4

ROC curves of nomogram in training and validation cohorts. A ROC curves of the training cohort at 3, 5, and 10 years; B ROC curves of the validation cohort at 3, 5, and 10 years. ROC receiver operating characteristic, AUC area under the receiver operating characteristic curve

Fig. 5

Calibration curve of the nomogram. This figure presents the calibration curve of the nomogram using data from A–C the training cohort and D–F the validation cohort

Fig. 6

Evaluation of the clinical benefit of the nomogram. This figure displays the assessment of the clinical benefit of the predictive model using data from A–C the training cohort and D–F the validation cohort

Fig. 7

Kaplan‒Meier curves for all-cause mortality stratified by gender (A), smoking status (B), and hypertension status (C) in the training cohort. Kaplan‒Meier curves for all-cause mortality stratified by gender (D), smoking status (E), and hypertension status (F) in the validation cohort. In the Kaplan–Meier curves, the population is stratified into two groups and statistical analysis is conducted using the log-rank test