Clinical heterogeneity mapping and machine learning-based prediction of recurrence in gallstone disease: a
Xin Zheng1, Yunjun Yan2, Ce Bian3
1Department of Clinical Laboratory, Shandong Provincial Third Hospital, Shandong University, Jinan, China.
Background:
Gallstone disease constitutes a substantial global health burden, characterized by notable clinical heterogeneity and a complex risk of recurrence. However, large-scale studies systematically characterizing this heterogeneity and translating these findings into clinically applicable prediction tools remain limited. This study sought to systematically elucidate the clinical heterogeneity within a large cohort of gallstone patients, identify independent risk factors for postoperative recurrence, and develop a machine learning-based model for postoperative recurrence prediction.
Methods:
We conducted a retrospective analysis of 9,939 patients undergoing gallstone surgery between January 2017 and December 2023 at the Third Hospital of Shandong Province. Clinical data, including demographics, gallstone type, size, comorbidities, and recurrence status, were extracted. Statistical analyses compared subgroups based on sex, age, stone type, complication status, and recurrence. Age- and sex-matched non-recurrence controls were selected in a 1:1 case-control design. Independent recurrence-associated factors were identified using univariate and multivariable logistic regression analyses. A Random Forest model was subsequently constructed using routinely available clinical variables and evaluated in independent training and testing datasets.
Results:
Marked clinical heterogeneity was observed across sex, age groups, stone locations, and recurrence status. Distinct associations were identified between specific stone distributions and comorbidity patterns. Multivariable logistic regression identified stone size, diabetes mellitus, venous thrombosis, liver cirrhosis, and malignant tumors as independent factors associated with recurrence. The Random Forest prediction model, incorporating these factors along with sex and age, demonstrated excellent performance with an area under the curve (AUC) of 0.873 on the test set. Variable importance analysis highlighted stone size and cirrhosis as the most influential predictors.
Conclusions:
Gallstone disease is characterized by substantial clinical heterogeneity. The identified recurrence-associated factors and machine learning model provide a potential framework for individualized recurrence risk assessment and may facilitate postoperative management following further external and prospective validation.
