Why Physicians Withdraw Ventilatory Support in the ICU

Abstract & Commentary

In this prospective, observational study, cook and associates recruited 851 patients who were likely to receive mechanical ventilation (MV) for more than 72 hours. They collected a variety of clinical data and severity scores. In addition, they recorded patient preferences for end-of-life care, physicians’ estimates of the patients’ likelihood of survival, likely functional status at discharge, plus other care variables such as dialysis, vasopressor treatment, whether weaned from ventilatory support, and so on. Some of the variables were followed daily, when appropriate. They found that out of these patients, 539 (63%) were successfully weaned, 146 (17%) died while receiving MV, and 166 (20%) had MV withdrawn in anticipation of death.

To identify a group of patients who were at risk of dying, Cook et al used logistic regression analysis on all patients in the study. The 2 groups for the logistic regression consisted of those who were weaned and those who died with MV or those who had withdrawal of MV in anticipation of death. With this analysis, they identified 300 patients who were at a relatively high risk of dying. Within this group, nearly a third (107 patients; 36%) died after withdrawal of MV, 105 (35%) died with MV in place, and the remaining 88 (29%) were successfully weaned. Among the 300 patients who had a high risk of dying, Cook et al used proportion hazard analysis, a technique used to identify determinants of a future event before the event actually occurs. They used the Cox model to identify determinants of physician-initiated withdrawal of MV.

They found that perhaps the only significant clinical determinant of withdrawal of MV was the presence of inotropic-vasopressor treatment (hazard ratio, 1.78; 95% CI, 1.2-2.66). However, physician judgment that the likelihood of survival in the ICU was less than 10% (hazard ratio, 3.49, 95% CI, 1.39-8.79) and similar likelihood of hospital survival, were significant determinants of withdrawal of MV. Similarly, physicians’ predictions of poor functional outcome and poor cognitive function after discharge were significantly associated with withdrawal of MV (hazard ratio, 2.51, 95% CI, 1.28-4.94). If physicians perceived that the patient did not prefer advance life support, those patients were likely to have withdrawal of MV (hazard ratio 4.19; 95% CI, 2.57-6.81). The study also showed that patients who were undergoing withdrawal of MV in anticipation of death also were the ones who were more likely to have dialysis stopped or vasopressors withdrawn compared to patients who died with MV in place (Cook DJ, et al—Level of Care Study Investigators and the Canadian Critical Care Trials Group. Withdrawal of mechanical ventilation in anticipation of death in the intensive care unit. N Engl J Med. 2003;349:1123-1132).

Comment by Uday B. Nanavaty, MD

In this prospective observational study, Cook et al used complex statistical techniques to determine what leads physicians to initiate withdrawal of MV. The conventional thinking was postulated to be that physicians would initiate withdrawal of care in elderly patients who had multiple chronic illnesses who were very sick. Cook et al were surprised to find that, except for presence of inotropic-vasopressor therapy, no other markers of severity of illness were significant predictors of physician decisions to initiate withdrawal of MV. Instead, physicians were more likely to initiate withdrawal of MV if they perceived that patients did not want such aggressive care. They were also more likely to initiate withdrawal of MV if they estimated that the likelihood of survival in the ICU or the hospital was less than 10%. They also were more likely to initiate withdrawal if they estimated that the functional or cognitive outcome was likely to be poor. The study included all the eligible patients and used robust techniques to come to the conclusions regarding the practice of physician- (or perhaps the ICU team-) initiated "pro-active" approach to end-of-life care.

Although Cook et al were surprised to find that severity of illness and age were not factors in withdrawal of life support, I frankly was not. I have to admit that I did not understand why Cook et al say age was not an important factor. In fact, when they looked at all the patients who had MV withdrawn in anticipation of death and compared them with all the patients who died with MV in place, their mean age was significantly different (64 yrs vs 60 yrs).

More and more, physicians have realized that the goals of ICU care should be the restoration of functional or qualitative life as opposed to the mere lower "30 day mortality." Also, physicians have accepted the limitations of critical care. As the critical care field and the physicians who provide care gain more experience, the limitations of modern technology are becoming clear. Most ICU physicians can keep the blood pressure where they want to or the pulse oximeter’s readings above a predetermined threshold number in the majority of their patients for a reasonably long period of time. Unfortunately, it is hard to decide based on these numbers, which patient will live and which patient will die. Hence physicians have turned to patients and their surrogates to gain insight into their wishes. We also know that the family members are perhaps as poor as physicians are in knowing patients’ wishes with respect to end-of-life care.

Beyond knowing what patients want, physicians have tried to have realistic expectations about the care. The problem is—how do you decide if the chances of some-one dying in next 30 days are 89% or 91%? Also unknown is the likelihood that a given patient would want to give up aggressive care. Ultimately, the decision to withdraw life support has to be individualized and has to be based on several different factors. We are learning from this large group that there are indeed multiple factors that lead to a decision, after which, not many survive (6 out of 166 patients survived withdrawal of MV in anticipation of death). It is reassuring that physicians are basing this decision on such factors as a realistic likelihood of a positive or negative outcome and patients’ perceived preferences. As long as the goals of therapy are discussed and all parties are in agreement, such practice will result in many more "good deaths"—as opposed to the long struggles that sometimes cause distress for physicians and families alike.

The study comes in line with the SUPPORT study that physicians withdraw MV if they are aware of patients’ preferences. The study highlights the changes in practice of critical care. In a survey of physicians published in 1994, 15% of physicians said they never withdrew MV. In this study, nearly half of the dying patients had MV withdrawn in anticipation of death, and all of these patients had DNR order in place. DNR orders were also written in nearly half of the patients who died with MV in place. It would be interesting to know as to what led the physicians to estimate the mortality to be greater then 90% and out of all the patients who had estimated mortality of > 90% in the first 72 hours or the first 120 hours, how many actually died. If objective criteria can be delineated to estimate such high mortality with high accuracy in a prospective manner, end-of-life care decisions—and the lives of intensivists—will be much less stressful.

Dr. Nanavaty, Pulmonary and Critical Care Medicine Rockville, Maryland