Prediction system for gallbladder cancer distant metastasis a retrospective cohort study based on machine learning.

BACKGROUND

The prognosis for gallbladder cancer (GBC) patients is generally poor due to the early occurrence of distant metastasis (DM). However, research on predicting the risk of DM in GBC patients is still limited. Therefore, this study aimed to apply the Surveillance, Epidemiology, and End Results (SEER) database with machine learning (ML) methods to construct a novel model for predicting the risk of DM in GBC patients.

METHODS

The data of GBC patients from the SEER database (2000-2020) were divided into a training set and an internal test set in a 7:3 ratio. Univariate and multivariate logistic regression analyses were then applied to systematically assess the risk factors for DM development. Six ML techniques were subsequently applied to construct a predictive model based on feature selection, validated by ten-fold cross-validation on the training set. Shapley additive interpretation (SHAP) was used to explain the selected models. Additionally, based on the optimal machine learning model, an online calculator was developed to provide personalized DM risk assessment for GBC patients.

RESULTS

Seven key variables were incorporated into the developed machine learning model for analysis. The extreme gradient boosting (XGB) model demonstrated high predictive accuracy [precision = 0.968; area under the curve (AUC) = 0.885]. In the assessment of risk factors, T, N, grade and age were identified as risk factors for DM in GBC patients. Conversely, rural-urban continuum, marital status, and median household income inflation adj to 2021 were identified as protective factors. Finally, an optimal learning model-based web calculator for personalized DM risk assessment was successfully constructed.

CONCLUSION

The XGB model was the most effective for predicting DM in GBC patients. This model can assist in developing personalized treatment plans for patients at an early stage, thereby improving prognosis.