Identifying key features for determining the patterns of patients with functional dyspepsia using machine learning.

Area of Science:

• Gastroenterology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Pattern identification (PI) is crucial for understanding disease symptoms and signs.
• Functional dyspepsia (FD) diagnosis benefits from effective PI strategies.
• Current PI methods may not capture all patient heterogeneity.

Purpose of the Study:

• To extract features for conventional PI in FD using supervised learning.
• To identify novel PI types and patient subgroups using unsupervised learning.
• To compare feature sets derived from both supervised and unsupervised approaches for FD.

Main Methods:

• Utilized supervised learning (Support Vector Machine - SVM) on questionnaire data from 153 FD patients.
• Employed unsupervised learning (t-distributed stochastic neighbor embedding - t-SNE, k-means clustering) to discover patient subgroups.
• Applied independent-samples t-tests to identify distinguishing features between identified clusters.

Main Results:

• SVM identified loss of appetite, flank discomfort, abdominal bloating/gurgling, and complexion changes as key discriminative features.
• Unsupervised clustering revealed four distinct FD patient subgroups based on symptom profiles.
• Identified symptom clusters included systemic symptoms, upper abdominal symptoms, altered bowel movements, and nausea/vomiting.

Conclusions:

• Supervised learning effectively identified critical features for FD pattern identification.
• A novel unsupervised learning approach successfully uncovered distinct patient subgroups within FD.
• Findings support enhanced clinical decision-making for patients with functional dyspepsia.