Palliative Care Perspectives : Chapter 2: Dying Trajectories and Prognostication : Prognostication

Prognostication, as an art, refers to prediction and communication about future health. Prognostication relates not only to predicting death, but other outcome states such as what percentage of patients with a cancer of a certain stage on initial presentation will eventually develop metastatic disease. As Nicholas Christakis points out, of three traditional domains of medicine (diagnosis, therapy, and prognosis), prognosis has received relatively little attention in modern medical training and research.¹ It is easier to predict when death will occur for patients with some illnesses than for others. Proper prognosis at the end of life enables better decision making about care options and planning for patients and families. The definitive text on prognosis as it relates to palliative care is Death Foretold: Prophecy and Prognosis in Medical Care by Christakis.¹

Predicting Death: The Search for the Holy Grail of Prognosis

We all desire and fear certainty. A desire for certainty arises, I think, in response to apparent chaos in the world.^2,3 However, fear arises because not all that is certain is good. Certainty also negates ambiguity and possibility, wherein people find hope that they can alter a problematic future. Therefore, an intense desire arises for some magic formula that will erase such uncertainty. While people may lament their lack of control over "bad" outcomes such as death, the ability to predict and know the future represents a form of control if the future unfolds as predicted. Many studies have been devoted to a search for certainty in predicting death, often with little to show for it. It is very disturbing to many that dying is a process largely beyond mortal control.

Prediction of death is not linear. That is, we are not necessarily better at predicting death in an ultrashort time frame (seconds to hours) than we are in a long time frame (months to years). Rather, it is like predicting the weather. In California we are good at long-range predictions for a dry summer and wet winter. We are good at predicting that rain will fall on a certain day two to five days beforehand but cannot make such a prediction a month beforehand. In the ultrashort range it defies our abilities to predict exactly when the next raindrop will hit a finger. Similarly, we are fairly good at predicting that a patient is at a high risk of dying over a matter of several months. For certain diseases, especially cancer, we are reasonably good at predicting death over a matter of weeks. It is usually impossible to predict the exact moment of death.

Most "holy grail" approaches to predicting death use a grouping of diagnostic criteria and apply them at a certain point in time to establish a probability of dying at some time in the future.⁴ The major problem with these approaches is that they are usually discrete, one-time predictions. Real clinical prognostication is more iterative, a form of "fuzzy logic."⁵ That is, the most valuable prognostic tool is to note the magnitude of change observed since the last prediction and incorporate this change into a new prediction. For most serious and chronic disease processes the earlier trend of an illness is the best predictor of the future trend. Patients whose clinical decline is rapidly accelerating will likely die sooner than those with otherwise identical clinical parameters but who decline more slowly.

Recent studies have pointed out physician deficits in terms of the ability to predict time of death.^6-8 The general bias in physicians is to be overly optimistic about prognosis by a factor of twofold to fivefold, although errors at the other extreme have also been observed. Christakis and Lamont, in a study of 343 physicians, studied their ability to predict time of death for 468 patients referred to hospice. They found that only 20% of predictions were accurate (within 33% of actual survival), with 63% of predictions overly optimistic and 17% overly pessimistic. Accuracy was calculated by dividing predicted by observed life expectancy. Values falling between 0.67 and 1.33 were considered accurate. Median life expectancy was 24 days from referral to hospice. On average physicians overestimated life expectancy by a factor of 5. It is interesting to note that 67% of the patients had cancer, usually considered to have a relatively predictable dying trajectory. In this study cancer patients were the most likely to have overly optimistic predictions.⁷

There are many theoretical reasons why physicians are as inaccurate as they appear to be. For a more detailed discussion of these issues the reader is referred to Christakis's book on prognosis. I highlight here only a few key points:

1. Poor predictions may reflect educational deficiencies in clinicians. Because prognosis as a medical art, especially as it relates to dying, has been devalued, better training about what is predictive would likely improve things.
2. Optimistic predictions may reflect a socially sanctioned selection bias for physicians. After all, who wants a pessimist for a doctor? We may select and reward those clinicians who have a rosy view of things.
3. What teaching does exist about prognosis is often gleaned from controlled trials, which tend to carefully select "ideal" patients with isolated diseases, otherwise good health statuses, and thus better prognoses than in patients usually encountered in practice. In cases in which patients have multiple diseases (comorbidities) clinicians may prognosisticate based on each individual disease as a discrete entity, rather than consider how diseases interact in the person to create a particular illness with a poorer prognosis in sum than is implied by consideration of isolated diseases.

Next >>> [ Go Up ]

Palliative Care Perspectives James L. Hallenbeck, M.D. Order hardback from Amazon.com Order paperback from Amazon.com Copyright © 2003 by Oxford University Press, Inc. The online version of this book is used with permission of the publisher and author on web sites affiliated with the Inter-Institutional Collaborating Network on End-of-life Care (IICN), sponsored by Growth House, Inc.