GAPPA: Enhancing prognosis prediction in primary aldosteronism post-adrenalectomy using graph-based modeling

Area of Science:

• Biomedical informatics
• Machine learning in healthcare
• Graph Neural Networks (GNNs)

Background:

• Accurate prediction of postoperative prognosis is crucial for clinical decision-making in patients undergoing adrenalectomy (ADX).
• Unilateral primary aldosteronism (UPA) patients require precise prognostic assessment post-adrenalectomy (ADX).

Purpose of the Study:

• To introduce GAPPA, a novel GNN-based approach for predicting post-adrenalectomy (ADX) outcomes in unilateral primary aldosteronism (UPA) patients.
• To leverage GNNs within a bipartite graph structure to capture complex relationships between clinico-biochemical features and clinical outcomes, enhancing prognostic prediction accuracy.

Main Methods:

• Prognostic prediction was framed as a link prediction task on a bipartite graph, incorporating patients, clinico-biochemical features, and clinical outcomes.
• The Graph Attention-based Prognostic Prediction Approach (GAPPA) employed GNNs to model feature-outcome dependencies, evaluated on 640 UPA patients undergoing unilateral ADX (uADX).
• Comparative analysis used repeated stratified five-fold cross-validation against conventional machine learning methods and prior studies.

Main Results:

• GAPPA significantly outperformed conventional methods and previous studies (p < 0.05) across multiple metrics.
• Achieved an F1-score of 71.3% ± 3.1%, accuracy of 71.1% ± 3.4%, sensitivity of 69.9% ± 4.3%, and specificity of 72.4% ± 7.2%.
• Demonstrated a robust Area Under the Curve (AUC) of 0.775 ± 0.030, with stable outcome-edge embeddings.

Conclusions:

• GAPPA enhances preoperative prognosis assessment for UPA patients undergoing adrenalectomy.
• The approach facilitates improved patient counseling and contributes to advancing GNN applications in the biomedical field.