A question: ‘Would you be surprised if this patient died in the next month?’ — posed to elicit a clinician’s overall impression of a patient — produced a strong correlation. If a clinician answered that they would not be surprised, the patient was twice as likely to die in the next month.