Why doctors aren’t good at predicting how long patients have left

    6 October 2016

    Not every patient who is dying wants to know how long they have left to live, but many patients and families do, in fact, ask for such information. For these individuals, better information about survival may help to improve their experience of end-of-life care. It may help to inform meaningful conversations with family and with healthcare professionals.

    As the illness progresses, this information may help to give the dying person and their family a sense of autonomy to make key decisions about where they wish to be cared for in their final days. Many patients currently die in hospital who would rather die at home, but organising appropriate care can take time and planning. Better prognostic information may also give patients and families adequate time to deal with unfinished business.

    Predicting survival is notoriously difficult to do. Our recent review of research conducted over the last four decades has demonstrated the extent of prognostic inaccuracy in all healthcare professionals. It is important to stress that these findings only relate to professionals’ predictions about how long patients with a known terminal illness are likely to survive. It does not imply that patients are being misdiagnosed or that they are being incorrectly advised that they are terminally ill when they are in fact curable (or vice versa).

    Moreover, even for patients with a known terminal illness it is not the case that clinicians’ estimates are completely wide of the mark. There is a correlation between healthcare professionals’ predictions and the length of time that patients actually survive. Patients whom clinicians estimate will do badly do not live as long as patients whom clinicians expect to do well. However, these predictions are not well calibrated.

    Thus, although a doctor or a nurse may be able to say which patient will live longer than another, they are not very accurate at predicting how long any particular patient will actually survive. Some studies suggest that they are more likely to over-estimate than to underestimate survival, although the reasons for this are unclear. Why do health care professionals struggle to get these predictions right?

    In order to answer this, it helps to look at prognostication outside of the context of medicine. Predictions are common in many fields; weather forecasting, seismology, economics, gambling. In all these circumstances predictions are never 100 per cent accurate and only after the event has or hasn’t occurred can we look to learn from them.

    The example of weather forecasting would seem to suggest that doctors ought to be better at predicting short-term rather than longer-term survival. It is apparent that it is easier to predict what the weather will be like tomorrow than it is to predict what it will be like in two weeks’ time. This intuition is borne out by research published in Nature, which shows that while the accuracy of the three-day forecast is over 90 per cent, the accuracy of the 10-day forecast is less than 50 per cent.

    For similar reasons one might therefore expect that doctors would be more accurate at identifying patients who are likely to die tomorrow rather than trying to predict whether someone will die next week or next month. However, very few studies have assessed the accuracy of doctors’ predictions over such short timescales and the limited evidence that there is in this area suggests that doctors are not much better at predicting imminent death than they are at predicting death within a few days or weeks.

    Nonetheless, if we want to improve the performance of doctors’ overall it makes sense to start by trying to improve the accuracy of their short-term predictions first. After all, we would not expect meteorologists to be able to provide us with an accurate long-term weather forecast before they had demonstrated their accuracy at predicting the weather over the next few days.

    Currently the most common means by which a doctor makes a prediction about when a terminally ill patient is likely to die is by using their clinical intuition. Through years of medical training and experience, each clinician has developed particular methods of judging clinical situations. Sometimes these thought processes are conscious and deliberative and sometimes they are unconscious and the clinicians themselves may be unable to articulate how they ‘just know’ the answer to a clinical problem.

    The first step towards improving the accuracy of clinicians’ estimates is to investigate exactly which conscious and unconscious mental processes the experts are using and to distinguish between those approaches which are helpful and which improve accuracy and those processes which lead doctors into predictive errors.

    For example, some clinicians may attach greater importance (either consciously or unconsciously) to the previous treatments that patients have received, whereas others may focus more on the severity of their symptoms or their nutritional status.

    We are currently undertaking a study to try to understand which factors expert doctors use to arrive at their decisions. If we can find out how the experts do it then it may be possible to use this information to try to teach the novice and less experienced doctors how to become more proficient at this clinical skill.

    Rather than trying to understand the clinical intuition of experts, an alternative approach to improving prognostic accuracy is to develop computer models. However, at the moment the best prognostic models are only slightly more accurate than an expert doctor or nurse, although as more elaborate models are being developed and validated it is likely that predictive accuracy may improve in the future.

    Whatever the likely future improvements in prognostic accuracy, it will always be important for doctors to acknowledge the inherent uncertainty in any attempt to predict the future. Uncertainty can provide a welcome space for hope. However, many patients and families would probably agree that more accurate information (whether good news or bad news) would allow them to be better prepared for what the future has to hold. If the forecast is for rain, it is probably wise to be prepared with an umbrella. But that doesn’t stop one hoping for sunshine.

    Nicola White is a PhD researcher and Paddy Stone a professor of palliative and end-of-life care at University College London