Withdrawing Life Support from Mechanically Ventilated Recipients of Bone Marrow Transplants: A Case for Evidence-Based Guidelines

Methods

Design

We used a nested case–control study design. Because marrow transplant recipients rarely survive after receiving mechanical ventilation, we defined “cases” as all survivors from the cohort and “controls” as a sample of the patients who died to conclude that no one with that combination of risk factors would survive. The control group of patients who did not survive allowed us to estimate the number of patients in the entire cohort who died with any given combination of risk factors [16]. From this, we estimated probability of survival with a specified combination of risk factors.

Study Population

The Fred Hutchinson Cancer Research Center (Seattle, Washington) does bone marrow transplantation in a high-risk, referral population. All patients are cared for by a multidisciplinary team led by an oncologist who is experienced with transplantation. Attending physicians and fellows specializing in pulmonary and critical care, infectious disease, gastroenterology, and nephrology are available 24 hours per day for in-house consultation. Patients are followed closely in Seattle for a minimum of 100 days after they receive their transplant. The Center maintains a computerized registry that includes demographic and long-term follow-up information for transplant recipients.

All patients who required mechanical ventilation after receiving their first transplant between January 1980 and July 1992 were eligible for inclusion in the study. Patients who were ventilated for less than 24 hours after a procedure (open lung biopsy, bronchoscopy, or computed tomography) were excluded; the remaining patients form the study cohort. Patients who were intubated but only received continuous positive airway pressure and patients who were not intubated but who received noninvasive, mask ventilation were not defined as having mechanical ventilation. All patients who met the definition of a survivor were selected as cases. All other patients in the cohort, including those who died while receiving mechanical ventilation, those who died after extubation with prolonged hospital survival, and those who were discharged and died less than 30 days after extubation, were considered “nonsurvivors” and thus potential controls. To control for changes in medical care during the 12 years of the study and to maximize statistical power, two nonsurvivors (controls) were matched by year of transplantation to each survivor (case) [16].

Data Abstraction

Four experienced critical care chart abstractors (including the two authors and a nurse-clinician specializing in bone marrow transplantation) recorded laboratory, hemodynamic, and treatment variables for each day of mechanical ventilation. Survivor and nonsurvivor charts were evenly distributed among abstractors, who were blinded to mortality as much as possible. Fractional inspired oxygen concentration and positive airway pressure requirements were recorded every 8 hours during mechanical ventilation. Scores on APACHE III were calculated using the published weights for the physiologic variables [17]. All patients were assigned chronic health points for metastatic cancer, leukemia, or lymphoma according to diagnosis. Because most patients were heavily sedated and many received neuromuscular blocking agents, Glasgow Coma Scores could not reliably be ascertained and the APACHE III scores did not include a contribution from this variable. Scores on APACHE III calculated without a Glasgow Coma Score range from 0 to a theoretical maximum of 272. Patient demographics and long-term survival were taken from the computerized database of the Fred Hutchinson Cancer Research Center.

Variable Definitions

The definition of survival used to identify cases was chosen as a compromise between meaningful long-term survival, which may result from many factors (including the success of transplantation), and short-term survival, which is more directly related to the outcome of mechanical ventilation. Survival for 30 days after extubation with subsequent hospital discharge may not be as meaningful an end point as 6- or 12-month survival, but death after hospital discharge is difficult to relate to the efficacy of mechanical ventilation or intensive care.

Definitions of risk factors were developed before data abstraction and were intentionally selected to reflect relatively minor physiologic impairment. Hepatic and renal failure was defined as a total bilirubin level greater than 68 mmol/L (4 mg/dL) and a serum creatinine level greater than 177 mmol/L (2 mg/dL) during the first 3 days of mechanical ventilation. Vasopressor use, which we used as a proxy for hemodynamic instability, was defined as dopamine, more than 5 µg/kg·min⁻¹, or norepinephrine, epinephrine, or phenylephrine for more than 4 hours in any 24-hour period. Dopamine administered to increase splanchnic blood flow (≤ 5 µg/kg·min⁻¹) and drugs used to increase cardiac output or heart rate (dobutamine, amrinone, or isoproterenol) were not included as vasopressors. Lung injury was defined as a fraction of inspired oxygen concentration greater than 0.6 or positive end-expiratory pressure greater than 5 cm H₂O at any of the three daily recorded values after the first 24 hours of mechanical ventilation. We evaluated this variable after the initial 24 hours of mechanical ventilation because many patients who receive high inspired oxygen concentrations during the initial phase of critical illness are rapidly weaned from this level of support. Patients who met these criteria at any time during the period of data collection were counted as having the given risk factor. Thus, patients who received vasopressors for 12 hours on the first day of mechanical ventilation and who required an inspired oxygen concentration of 0.8 after another 3 days were counted as having vasopressor and lung injury risk factors. Such patients were considered to meet the combined criterion of lung injury and vasopressors on day 3 after mechanical ventilation.

Statistical Analysis

Continuous variables were compared by using the Student t-test or the Wilcoxon rank-sum test, depending on distribution. Categorical variables in matched cases and controls were compared using the Mantel-Haenszel chi-square test [18]. Trends over time in the continuous variables of timing of mechanical ventilation, age, and APACHE III score were evaluated using linear regression. Trends over time in the categorical variables of fractions of patients intubated, patient survival, and patients receiving an unmatched transplant were evaluated by using the chi-square test for trend. Statistically significant trends were indicated by a regression coefficient for year of transplant or a chi-square test for trend with a P value less than 0.05. Exact CIs for ratios were calculated from the binomial distribution. Analyses were done on a personal computer using STATA software (Stata Corp., College Station, Texas).

The probability for survival of patients with one or more of the study risk factors can be calculated with the following equation: Equation 1

We know the numerator of this ratio exactly because we have data for risk factors for all 53 patients who survived from the entire cohort of study patients. We can estimate the denominator (that is, the total number of patients with risk factors in the cohort) from data we have on the cases and controls. The probability for survival of a patient who has one of the study risk factors can be estimated with the following equation: Equation 2

Results

Outcome

From 1980 to 1992, 3635 patients received a first bone marrow transplant, and 909 (25%) of these patients were mechanically ventilated. Of patients who were mechanically ventilated, 44 were ventilated for less than 24 hours after a procedure. All 53 (6.1%) of the 865 remaining patients who met the criteria for survival were analyzed as cases. One hundred six controls were randomly selected from among the 812 nonsurvivors (2 for each survivor matched by year of transplantation). Charts were found for all of the patients who were selected for the study.

Eighty-seven (82%) of the 106 controls died while receiving mechanical ventilation. The remaining 19 (18%) died in the hospital a median of 18 days (range, 1 to 80 days) after extubation. Median survival after extubation for the 53 survivors was 634 days (range, 54 days to 12 years). Eighteen of the survivors (34%) lived for 6 months or less after extubation, 5 (9%) died within a year of extubation, and 30 (57%) lived for more than a year after extubation. We found no statistically significant differences between survivors and nonsurvivors in sex, underlying diagnosis, preparative regimen, relapse status at time of transplantation, donor-recipient histocompatibility, or duration of mechanical ventilation (Table 2). Survival was statistically associated with younger age, lower APACHE III score on day 1, and intubation earlier in the course of transplantation (Table 2). Despite the statistical significance, examination of the ranges showed considerable overlap between survivors and nonsurvivors. Post hoc cutoff points for these variables, which were derived from direct inspection of the data, did not yield useful criteria. For example, although the oldest survivor was 58 years of age, only 1 (1%) of the nonsurvivors was older than 58 years of age. No one intubated more than 149 days after the bone marrow transplantation survived, but only 2 (2%) nonsurvivors were intubated beyond this cutoff point. No one survived if they had an APACHE III score greater than 114 on day 1, but only 10 (9%) of the nonsurvivors had scores higher than 114.

The risk factors of vasopressors, hepatic and renal failure, and lung injury occurred frequently in this group of critically ill patients. Half of the controls required vasopressors or had sustained hepatic and renal failure, and 67% had lung injury (Table 3). The absence of these factors was strongly associated with survival (P < 0.01 for all risk factors).

Probability of Survival

The estimated numbers of mechanically ventilated bone marrow recipients who had lung injury, had hepatic and renal failure, or received vasopressors were 557, 391, and 416, respectively. The estimated probabilities of 30-day survival with hospital discharge associated with these risk factors were 2.3%, 2.0%, and 0.48%, respectively (Table 4). Forty-nine percent (52 of 106) of the nonsurvivors reviewed had lung injury and had either received vasopressors or had hepatic and renal failure. Using this proportion, we estimated that 398 transplant recipients in the cohort had either of these combined exposures and that none of these (0%) survived (95% CI, 0% to 0.9%). The proportion of nonsurvivors with combined exposures increased to 58% when the hepatic and renal failure plus vasopressor criterion was included, yielding an estimated 476 patients with any two of the three combined exposures. Of these, only 1 patient (0.2%) survived (CI, 0.005% to 1.2%) (Table 4). Forty percent of the reviewed nonsurvivors (or an estimated 377 in the cohort) did not develop any two of the three combination exposures. Fifty-two (13.5%) of the patients who did not have any of the three combined risk factors survived.

Survival Trends

From 1988 through the first half of 1992, survival increased from 5% to 16% (Figure 1). This improvement during the study period (1980 to 1992) is statistically significant (P = 0.008 by the chi-square test). In an attempt to understand which factors may have contributed to this improved survival, we evaluated the following variables for trends over time: proportion of patients receiving mechanical ventilation, patient age, timing of intubation after transplantation, and proportion of patients receiving unmatched or unrelated transplants (Table 5). These data were available from the Fred Hutchinson Cancer Research Center database for all patients in the cohort. To further evaluate changes in severity of illness, we analyzed the APACHE III scores on our dataset of 159 patients. No statistically significant change was seen in the proportion of transplant recipients who received mechanical ventilation or in the timing of intubation during the study period. Age and proportion of unmatched or unrelated transplant recipients increased significantly over time (P < 0.001 for each). Scores on APACHE III also increased over time both in the selected nonsurvivors (P = 0.003) and when the survivors and nonsurvivors were analyzed together (P = 0.002). When survival of mechanically ventilated patients increased, age, number of mismatched transplants, and severity of illness as measured by APACHE III also increased.

Timing of Predictors of Death

Accurate prediction of death immediately before its occurrence is not clinically useful. We analyzed our data to see when death could have been predicted using the risk factors in this study. Of the 60% of patients who developed any two of the three risk factors, half had done so by day 2 and 90% by day 4. On average, risk factors developed 6 ventilator days and 9 hospital days before death. If life support had been withdrawn on the first day that two of the three risk factors were met and if death had followed swiftly, we estimate that more than 7300 hospital days and 4800 ventilator days could have been avoided for the 812 patients who died.

Discussion

We show that no patients in a large group of mechanically ventilated marrow transplant recipients with lung injury and either hepatic and renal insufficiency or hemodynamic instability (defined by a requirement for vasopressors) survived. Only one patient survived any combination of lung injury, hemodynamic instability, and hepatic and renal failure as defined in this paper. We studied a group of patients from a 12-year cohort whose numbers equalled 90% of the total number of previously reported mechanically ventilated bone marrow transplant recipients and exceeded the total number of previously reported survivors. We also found that survival after mechanical ventilation in bone marrow transplant recipients appears to be increasing. We believe it is appropriate, even after study limitations are taken into account, to establish clinical guidelines from these data.

Most physicians, ethicists, and jurists agree that futile, inappropriate, experimental, or unreasonable medical care should not be provided even if it is requested [19, 20]. However, there is no consensus about how these terms should be defined and applied in practice. Clinicians vary considerably when deciding on the appropriate level of care for similar patients [21, 22], and they may be unconsciously influenced by such arbitrary factors as a patient's ethnicity, an iatrogenic complication, or the city in which they practice [22-24]. Several authors [25, 26] have suggested that available outcome data should be used to develop community, institutional, and professional guidelines to inform these difficult decisions.

Our study has several limitations. First, outcomes were limited to the first episode of mechanical ventilation after a first bone marrow transplantation and should be applied cautiously in other settings. The Fred Hutchinson Cancer Research Center releases stable patients approximately 100 days after transplantation. Therefore, our cohort does not accurately reflect the outcome of intubation and intensive care beyond this time point.

Second, such therapeutic interventions as the use of vasopressors or the level of positive end-expiratory pressure have been criticized as unreliable predictors because of institutional variations in practice style [27]. If clinicians institute vasopressor use or manage oxygenation failure differently, it is possible that these variables would not reliably predict outcome at other institutions. Although the decision to use increased oxygen supplementation instead of increased levels of positive end-expiratory pressure to maintain arterial oxygen saturation is often debated, the need for either alternative suggests oxygenation failure. The cutoff points that we selected represent treatment thresholds used by many intensivists [28]. The threshold at which individual clinicians institute vasopressor therapy may vary in a similar manner, but it is unlikely to be a purely institution-specific phenomenon.

Third, we may only be validating a local decision bias in this trial; that is, it may be standard practice at the Fred Hutchinson Cancer Research Center to be less aggressive when treating patients who fulfilled our study's risk factors. However, this is unlikely for several reasons. Our cohort showed outcomes that were improved compared with the 0% 30-day survival rate for 51 patients older than 40 years of age or the 0% 100-day survival rate in 118 patients who were intubated before day 90 in a series of 191 patients from Minnesota [13]. Furthermore, the fact that most patients received mechanical ventilation and intensive life support for many days after they met the study criteria argues that physicians were not implicitly using these criteria as reasons to withdraw life support.

Fourth, clinicians may claim that care is superior or that patients are less sick at their institutions and thus that predictors developed at the Fred Hutchinson Cancer Research Center would not apply. Although comparisons between studies are difficult, survival after mechanical ventilation among high-risk patients referred to the Fred Hutchinson Cancer Research Center for bone marrow transplantation equals or exceeds that seen in other published series, allowing for the CIs around small samples (Table 1). Some may argue that a low-risk subset of transplant recipients is not well represented in this series. Because we have data on every survivor, we know that no patient from any putative low-risk group who had the proposed criteria survived. Because they are less inclined to require mechanical ventilation, children, autologous transplant recipients, and patients in remission make up a small proportion of this or any other cohort of bone marrow transplant recipients who received mechanical ventilation.

Finally, the estimated denominators may be incorrect. The selected controls may not have represented the exposures in the entire cohort. If this is true, then the column in Table 4 that shows the estimated number of patients with a risk factor may not give accurate counts. Differences in the estimated denominator would change the CIs presented in Table 4. However, a true value for survival with the risk factors equal to 0 of 250 with an upper 95% confidence limit of 1.4%, rather than the reported 0 of 398 with an estimated upper 95% confidence limit of 0.9%, would not alter our conclusions.

Forty percent of nonsurvivors did not develop any of the identified risk factors. We did not expect any risk factor to predict death in all patients. Patients who died without fulfilling the risk factors may have done so because of withdrawal of life support at the request of the patient or surrogate or diseases not captured by the predictors, or they may have had a late death occurring during hospitalization that was not related to mechanical ventilation. Clinical situations with poor prognoses that may account for some of the unpredictable deaths include neurologic trauma from intracranial hemorrhage, tumor relapse despite transplantation, and fungal infection with progressive graft-versus-host disease requiring immunosuppression.

The finding of improved survival over time was unexpected. Similar improvements in survival have been reported for patients with the adult respiratory distress syndrome who did not receive transplants [29]. Transplantation techniques have improved considerably during the assembly of this cohort. Advances in the prevention of infectious complications after bone marrow transplantation have nearly eliminated Pneumocystis carinii and cytomegalovirus pneumonias. The introduction of marrow-stimulating cytokines may have improved the outcome of critically ill patients with complications related to infections.

Changes in intubation practice, severity of illness, or another confounding variable may account for the improvement seen in survival. If patients at high risk for death no longer received mechanical ventilation or if less-sick patients were being intubated, survival would appear to improve. However, analysis of the available data does not support this hypothesis. During the study period, we detected no statistically significant change in the percentage of transplant recipients who received mechanical ventilation. We evaluated age, transplant matching, diagnosis, and timing of intubation after transplantation in all 865 patients. This analysis showed either that no statistically significant change in these factors occurred over time or that the ventilated population was at higher risk for complications caused by increasing age and mismatched transplants. An analysis of the APACHE III scores, which was available only for the 159 cases and controls, corroborated these findings and showed that the mechanically ventilated patients had an increasing severity of illness (reflected in higher APACHE III scores) over time. If these factors suggest anything, it is that the trend in improved survival underestimates true improvement because of the confounding influence of worsening case mix and severity.

Clinical prediction tools cannot easily be applied until they have been validated in other settings. However, even a wide search that yielded one or more patients who survived the proposed prediction criteria would not invalidate decisions made on the basis of these data. We do not claim that survival is impossible for an intubated bone marrow transplant recipient with lung injury who requires more than 4 hours of vasopressors; absolute certainty is not attainable from any empiric outcome study and is an unrealistic goal for medical decision making [30]. There is a precedent for acting decisively on clinical experience that is smaller than that for mechanically ventilated bone marrow transplant recipients. Criteria for clinical brain death and proposals to limit treatment in these cases were based on less empiric evidence [31].

We believe that establishing guidelines on the basis of outcome data to provide some framework for decision making in these difficult cases is superior to a process limited by arbitrary factors and cognitive biases [32]. The elective nature of the transplantation and the maturity of the prognostic data permit the development of rational guidelines that may be presented explicitly to the patient and family members before acute illness. An example of such a guideline is presented in the Appendix.

Because a prediction can be made for most patients within 4 days of intubation and survival in the remaining patients is 13.5%, it is reasonable to give mechanical ventilation to all who request it. Mechanical ventilation and other life-support measures might temporarily reverse the physiologic abnormalities in patients who fulfill the study predictors and are thus not futile in a strictly physiologic sense [33]. Our study highlights both the advances in care and the continued need for research into the prevention and treatment of the complications of bone marrow transplantation. However, in a world of unavoidable clinical uncertainty, finite resources, and competing demands, allocation decisions must be made in health care. It is difficult to specify limits beyond which treatment should be withheld when there is any chance that a life can be saved [34]. However, if we cannot agree that treating 400 patients with prolonged intensive care without producing a single survivor is beyond such a limit, then it is unlikely that we can reach a consensus about limiting care in any clinical situation.

Appendix

Proposed Guidelines for Mechanical Ventilation after Bone Marrow Transplantation

I. The goal of bone marrow transplantation is to cure the underlying condition and return the patient to an acceptable quality of life. When these goals are no longer attainable or at the request of a suitably informed patient or surrogate, further intensive life support should cease.

II. All bone marrow transplant recipients, their surrogates, and involved physicians and nurses should participate in the informed consent for transplantation. As part of this informed consent, an estimate of the patient's risk for requiring mechanical ventilation and developing hepatic veno-occlusive disease should be conveyed in simple language [3, 35].

III. The outcomes for mechanically ventilated patients should be presented. Individual institutions that have formally collected prognostic data from samples of similar size may substitute their experience in this section.

A. Approximately 6% of patients survive for 30 days after extubation and are discharged from the hospital. Half of these survivors live for more than 2 years.

B. Patients who are mechanically ventilated, develop lung injury, and either receive vasopressors or develop hepatic and renal insufficiency (as previously defined) do not survive (as estimated in 398 similar patients at the Fred Hutchinson Cancer Research Center).

C. Patients without this combination of risk factors have a survival rate of about 13%.

IV. The following conditions make the goals of bone marrow transplantation, specified in (I), unattainable: massive intracranial hemorrhage, tumor relapse despite transplantation, and fungal infection with progressive graft-versus-host disease requiring immunosuppression.

V. The presumption is that patients who fulfill the criteria in (IIIB) or (IV) will not receive prolonged life support because the goals of transplantation specified in (I) would no longer be attainable. Patients and surrogates who do not agree with this standard of care should be encouraged to discuss their concerns at the time of informed consent for the transplantation.

VI. The guidelines are not meant to be rigid. Patients who enter an approved experimental trial to improve the outcome of critical illness in bone marrow transplant recipients may be exempted. When care deviates from this guideline, review by an institutional committee should be initiated. The committee should review the case in a timely fashion and ensure that those involved have communicated the outcome data fairly and heard all opinions. Most cases should be able to be resolved by discussion, appeal to the data, and referral to the informed consent. Intensive care may be continued if the reasons to do so are compelling (for example, rapid clinical improvement during the review period), although the expectation of survival for 30 days after extubation and hospital discharge is unchanged.

Dr. Crawford: Fred Hutchinson Cancer Research Center, 1124 Columbia Street, Box 358080, Seattle, WA 98104.