H029

Can doctors tell how long a terminally ill patient has left?

Many people believe patients live about as long as their doctor predicts. When researchers checked 468 predictions for hospice patients, only 1 in 5 was close, and most were far too hopeful.

468 doctors' predictions, checked against how long each patient actually lived
63% of predictions were too optimistic, 20% were close, and 17% were too pessimistic
“Close” means the patient lived within a third of the predicted time, either way
Across 8 studies, 1,563 predictions
Median predicted survival 42 days; median actual survival 29 days
Within a week only 25% of the time
The surprising twist
The longer a doctor had known the patient, and the more recently they had seen them, the less accurate the prediction.
Closeness made forecasts worse

Doctors' predictions lean hopeful, and closeness makes it worse. In a Chicago study of 468 hospice patients, 63% of predictions were too optimistic and only 20% were within a third of the truth; on average, doctors predicted survival 5.3 times longer than it turned out to be. A review of eight studies found the same pattern: doctors predicted a median of 42 days, and patients lived a median of 29. Yet the predictions weren't random. Patients doctors expected to die sooner usually did; the doctors were simply too hopeful about everyone.

So why does “they lived just as long as the doctor said” feel true? Because the predictions that come true are the ones we remember and repeat. When we care about how something turns out, our forecasts drift toward what we hope, even when we're experts.

Try it somewhere else

A friend is sure their new business will turn a profit in the first year.

When people predict something they care about, which way are they usually wrong, and how would you check?

Can a prediction make itself come true?

Sometimes, in intensive care. Ethicist Dominic Wilkinson has described how a prediction of a poor outcome can become self-fulfilling: if life support is withdrawn because of the prediction, the patient may die partly because the treatment stopped, and no one can know what would have happened otherwise. Researchers studying brain bleeds have raised the same concern about early do-not-resuscitate orders.

That's a real effect, but it's different from the hospice studies above, where patients were already near the end and doctors mostly predicted too long, not too short. If predictions were steering outcomes, they'd likely match far more often than 1 in 5.

Why does it matter?

Overly hopeful predictions can delay hospice referrals and the conversations families want to have while there's still time. In the Chicago study, the median patient lived 24 days after being referred to hospice. A 2016 review found that doctors do slightly better when asked for a probability, such as the chance a patient lives another month, than when asked for a number of days.

None of this means doctors are careless. Predicting a single person's remaining time is very hard, and a hopeful lean is a very human error. It means a prognosis is an estimate, best treated as a range.

Related exhibits

Sources: Nicholas A. Christakis and Elizabeth B. Lamont, “Extent and determinants of error in doctors' prognoses in terminally ill patients: prospective cohort study,” BMJ (2000): 343 doctors, 468 patients at five Chicago hospice programs in 1996; accuracy defined as actual survival between 0.67 and 1.33 times the prediction; median survival 24 days. Paul Glare et al., “A systematic review of physicians' survival predictions in terminally ill cancer patients,” BMJ (2003): 8 studies, 1,563 predictions. Paddy White et al., PLOS ONE (2016), on probabilistic versus time-based estimates. Dominic Wilkinson, “The self-fulfilling prophecy in intensive care,” Theoretical Medicine and Bioethics (2009).