When BERT Fails -- The Limits of EHR Classification
arXiv:2208.10245v11 citationsh-index: 8
Originality Synthesis-oriented
AI Analysis
This work addresses reliability issues in clinical decision support systems, but it is incremental as it focuses on specific failure patterns rather than a new solution.
The paper investigated failure cases of BERT in electronic health record (EHR) classification for readmission prediction, identifying patterns that lead to inferior performance.
Transformers are powerful text representation learners, useful for all kinds of clinical decision support tasks. Although they outperform baselines on readmission prediction, they are not infallible. Here, we look into one such failure case, and report patterns that lead to inferior predictive performance.