Sick To Death > Chapter 1 > "Dying" and the Problem of Prognostication
After the writings of Elizabeth Kübler-Ross (1969) and the ensuing wave of attention popularized the term, "dying" came to connote a period of rapid, progressive illness and disability after treatment failed. Family and caregivers - indeed, patients themselves - learned to expect that "dying" individuals would put their affairs in order and be dead within a very short time, no more than a month or two. A physician could label a patient with weight loss, weakness, and metastatic cancer, for example, as "dying." That label triggered a pattern of behaviors for patient, family, and clinicians. Friends expected the "dying" person to go through "stages" of dying (denial, anger, bargaining, and acceptance) in about that order (Kübler-Ross 1969). Family and friends assumed that "dying" people would not make long-term plans or pursue long-shot therapies but would instead make peace with the ending of their lives and say their good-byes to loved ones.
Among the major causes of death, only cancer routinely has a confined period of time, usually less than eight weeks, in which the patient loses weight, energy, and ability to carry on daily tasks (McCarthy et al. 2000; Morris et al. 1986). Such a patient discontinues most activities, often asks for increased comfort care, and ordinarily dies at a rather predictable time. But only about one-quarter of all deaths are from cancer (National Center for Health Statistics 2001), and most serious chronic illnesses follow a very different course. The relatively quick and predictable course of dying with cancer is no longer the only useful model around which to organize medical care or social supports.
Most Americans die with failure of a major organ (heart, lungs, kidneys, or liver), dementia, stroke, or general frailty of old age (Murphy 2000; Lunney 2000; Lunney et al. 2003). Unlike cancer, these conditions lead to long periods of diminished function and involve multiple unpredictable and serious exacerbations of symptoms. The timing of death for individuals with these diseases is ordinarily not predictable. And unless their course mimics that of terminal cancer - with a discernable period of rapidly declining weight, strength, and function - people with these less predictable diseases cannot be classified as "dying." If they never have such a phase, then death comes unexpectedly and without preparations. They do not qualify for hospice or have plans to avoid resuscitation. In short, while patients die from heart failure or emphysema, for example, they often do so with no discernable phase that can be labeled "dying".
I have often heard a family member of one of my patient say something like, "If I had only known Mother was dying, I would have come home to be with her." The mother had been on oxygen for many months and was over ninety years old. Surely, I would think, this seventy-yearold person could not have expected his or her mother to go on forever. Yet the survivor assumed that there would be a time of "dying," a time for good-byes, and regretted missing that opportunity.
The degree to which the timing of death is not predictable is surprising to the public. Somehow, most of us have the sense that physicians could tell us how much time we have left, if they wanted to (and if we asked). One project studied the medical records of nearly ten thousand seriously ill patients; interviewed the patients, family members, and physicians; and then developed a set of mathematical formulas that predict the timing of death (Knaus et al. 1995). For one hundred people with a particular disease, the predictive formulas were quite accurate. If the predictions said that half would die within six months, about half actually died by then. Asking physicians to make predictions about their own patients yielded about the same good average accuracy. But neither approach worked very well to predict the timing of death for a specific patient. Substantially erroneous predictions in optimistic or pessimistic directions average out for groups but scramble the reliability for individuals.
Figure 1 reflects the striking unreliability of predictions for persons who turn out to be very near death. Even in the last weeks of life, many patients have good odds of living another two months. Lung cancer (here, "non-small-cell" inoperable cancer that started in the lung) is one of the most predictable among fatal conditions. Once the person is seriously ill and losing weight, survival for more than a few weeks becomes quite unlikely. Nevertheless, just seven days before dying, the median patient with lung cancer still has almost a 50–50 chance of living for two months. How can that be? It turns out that a substantial number of lung cancer patients have their course cut short by a complication - an infection, a bad reaction to treatment, a heart attack, or something else that crops up unpredictably and shortens life. In each case, the physicians and the patients and families all knew that the patients had fatal lung cancer, but just a few days before death, no one could know that life would end this week. All involved might say, "But he didn't really die of his cancer; this complication was what caused his death." Without his cancer, of course, the complication would have been unlikely to prove fatal.
Heart failure death is much less predictable than lung cancer death. All of the heart failure patients in this study were very ill. They had been in the hospital with shortness of breath, they were on multiple medications to boost circulation and protect the heart, and they were unable to walk far or to climb stairs. They clearly would die of their heart failure, unless something even worse came along. They would have long periods of stable function, too. They would come home from the hospital and get settled, figure out how to avoid having to climb steps or carry groceries, and live fairly well for some time. Then some complication would arise, often from a relatively small thing like having a cold or eating too much salt; and the delicate balance would tip over into a serious struggle to breathe.
Hospital care would generally pull the person through, more than once. Eventually, though, one of these episodes would prove fatal, or the damaged heart would simply lose its regular rhythm and fail to pump. Either way, the death most often happened within a week of the patient having been stable, though living with serious constraints on activity.
When asked to predict timing of death for persons who are very sick, physicians usually give estimates that are a bit generous (Christakis 1999). The reasons for this are unclear, but wishing to be kind or to avoid anger from patients and families who make plans around the prediction may well play a role. One study asked physicians to state a likelihood of the patient surviving for two or for six months; and that question, on average, got accurate answers (Zhong and Lynn 1999). So the answers physicians give may depend somewhat on the question asked.
Since the timing of death for individual patients is usually unpredictable, in order to provide reliably good care in the last phase of life, the care system will have to be able to deliver good care to some people for years, not just for the last weeks or months. Building a set of desirable services that is available only to those who are sure to die soon is not a strategy that would actually include most dying people.
Nevertheless, that is exactly what the Medicare hospice benefit aimed to do. Hospice (which I discuss in chapter 3) is an effective provider of comprehensive end-of-life services, mostly to persons dying at home. The Medicare hospice benefit requires the attending physician to certify that the patient has a prognosis of less than six months. Remarkably, the federal regulation has never specified whether that prognosis should be "virtually certain" or just "more likely than not" (Lynn 2001). Obviously, it is a very different thing to predict anything with certainty than it is to bet the odds. On the one hand, allowing the "more likely than not" definition acknowledges the uncertainty of individual predictions but also means that many people will qualify for hospice care throughout many months, even years. Those will be the people whose luck held out; they walked a long time on the high wire before encountering their final complication. On the other hand, adopting a "virtually certain to die" definition means that hospice will be available to a much smaller cohort, mostly very near death, and dying of conditions that allow prediction, at least when death is coming very soon. Hospice enrollment patterns and the government's audits for fraud seem to use mostly a definition of prognostic eligibility that is closer to "virtually certain to die" than to "more likely than not." Hence most people die from serious chronic disease and never have the opportunity to use hospice, since they went from an ambiguous prognosis to death within a few days.
The impact of these observations is just beginning to shape the thinking of professionals and the public. Whereas once the family doctor told the anguished parents of a child with diphtheria or whooping cough that the child would get better or die within a few days, now patients ask, "How long do I have?" and doctors are often inarticulate about why they cannot readily answer the question. It turns out that most of us now die with unpredictable timing from predictably fatal chronic disease. The fact that the condition will eventually be fatal is clear, and the survival sta tistics for a large group of affected patients may likewise be clear, but a reliable prediction for one patient is not possible. Two patients with the same amount of heart damage and the same array of concurrent illnesses might have very different life spans for largely unpredictable reasons. Dying "suddenly" in the midst of long-term chronic conditions is a common pattern for the end of life. But we have not yet learned a set of metaphors and expressions that gives voice to that experience, and we certainly have not yet learned how best to live and die in this situation.