Monitoring complex care pathways is a key factor for hospitals operational success: both for patient's satisfaction and proper cost management. Detecting which care pathway are likely to be complex is a challenging task for an AI model, solely considering the available information at the entry point in the hospital (e.g. age, sex, GEMSA code, CCMU code). We illustrate the AI-models weak inference capabilities when not considering text report, as it contains most of the clinical insights.
