TRACE: Transformer-based Risk Assessment for Clinical Evaluation

We present TRACE (Transformer-based Risk Assessment for Clinical Evaluation),a novel method for clinical risk assessment based on clinical data, leveragingthe self-attention mechanism for enhanced feature interaction and resultinterpretation. Our approach is able to handle different data modalities,including continuous, categorical and multiple-choice (checkbox) attributes.The proposed architecture features a shared representation of the clinical dataobtained by integrating specialized embeddings of each data modality, enablingthe detection of high-risk individuals using Transformer encoder layers. Toassess the effectiveness of the proposed method, a strong baseline based onnon-negative multi-layer perceptrons (MLPs) is introduced. The proposed methodoutperforms various baselines widely used in the domain of clinical riskassessment, while effectively handling missing values. In terms ofexplainability, our Transformer-based method offers easily interpretableresults via attention weights, further enhancing the clinicians'decision-making process.