The deeper problem is not just automation but ambiguity about who owns decisions, accountability and outcomes.
We show that, compared with surgeon predictions and existing risk-prediction tools, our machine-learning model can enhance ...