Machine learning prediction of metabolic-associated fatty liver disease in type 2 diabetes: Emphasizing data
Zahra Khosravi1, Farnaz Barzinpour1, Soghra Rabizadeh2
1School of Industrial Engineering, Iran University of Science and Technology, Tehran, Iran.
Abstract:
Metabolic-Associated Fatty Liver Disease (MAFLD) is common among Type 2 Diabetes (T2DM) patients. The coexistence of these conditions increases the risk of MAFLD progression and diabetes complications. Detecting MAFLD early is challenging due to its asymptomatic initial stages. In this study, we aimed to develop a machine learning model to predict MAFLD in T2DM patients. We conducted a cross-sectional study on 3,654 Iranian T2DM patients using their demographic and lab data. This study involved thorough data preprocessing, including evaluating various imputation methods on simulated missingness in a complete subset of the dataset. Additionally, four feature selection methods were applied to eight machine learning models to identify the most effective predictive model. The XGBoost classifier without feature selection achieved the best performance, with an accuracy of 80.6% and an area under the receiver operating characteristic curve (AUC) of 88.9%.Notably, certain features, such as alanine aminotransferase (ALT), platelet count (PLT) and Vitamin D(VitD) influenced the predictive performance.
Insights
Machine learning effectively predicts Metabolic-Associated Fatty Liver Disease (MAFLD) in Type 2 Diabetes (T2DM) patients. An XGBoost model achieved 80.6% accuracy, identifying key predictors like ALT, PLT, and Vitamin D levels.
Area of Science:
- Medical Informatics
- Hepatology
- Endocrinology
Background:
- Metabolic-Associated Fatty Liver Disease (MAFLD) frequently coexists with Type 2 Diabetes Mellitus (T2DM).
- This comorbidity accelerates MAFLD progression and exacerbates diabetes-related complications.
- Early MAFLD detection is difficult due to its often asymptomatic nature in initial stages.
Purpose of the Study:
- To develop and evaluate a machine learning (ML) model for predicting MAFLD in T2DM patients.
- To identify key demographic and laboratory features predictive of MAFLD in this population.
Main Methods:
- A cross-sectional study involving 3,654 Iranian T2DM patients.
- Comprehensive data preprocessing, including imputation method evaluation.
- Application of four feature selection methods across eight ML models, including XGBoost.
Main Results:
- The XGBoost classifier, without feature selection, demonstrated superior predictive performance.
- Achieved an accuracy of 80.6% and an Area Under the Receiver Operating Characteristic Curve (AUC) of 88.9%.
- Key predictive features identified include alanine aminotransferase (ALT), platelet count (PLT), and Vitamin D (VitD) levels.
Conclusions:
- Machine learning models, particularly XGBoost, show significant potential for early MAFLD detection in T2DM patients.
- Specific biomarkers like ALT, PLT, and VitD are crucial for accurate MAFLD prediction.
- This approach can aid in timely intervention to prevent MAFLD progression and diabetes complications.
