INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Hematopoietic stem cell transplantation (HSCT) is considered current standard of care and is potentially curative for patients with various types of malignancies (1). More recently, peripheral blood stem cell transplantation (PBSCT), both allogeneic and autologous, has become an alternative to bone marrow transplantation (BMT) (2- 4). The complications associated with BMT vary with the underlying disease and also depend on the previous treatment history, type of pretransplant conditioning therapy, and the patient's age. In general, toxicities and immunosuppression from the conditioning regimens, graft versus host disease (GVHD), infections, and relapse of the original disease are the most widely noted complications (5). The specific pulmonary problems associated with BMT have been described in the literature and include pneumonia, alveolar hemorrhage, pulmonary edema, and interstitial pneumonitis (6, 7). Several studies have indicated a very poor outcome in BMT patients with respiratory failure necessitating endotracheal intubation and mechanical ventilation ("intubation") (8). It has been difficult to identify pre- or posttransplantation characteristics that may predict respiratory failure in BMT patients, and several studies have yielded conflicting results (11, 14, 16). Rubenfeld and Crawford recently reported that, of the BMT patients who required intubation at their center, none with lung injury combined with either hemodynamic instability, or hepatic or renal failure survived (17).

It is not known whether the outcome of the PBSCT recipients with respiratory failure differs from that of BMT patients. To date, prospective analyses of the outcomes of PBSCT and BMT recipients admitted to a medical intensive care unit (MICU) have not been done. In addition, survival rates of intubated versus nonintubated patients have not been compared prospectively. The purpose of this study was to prospectively follow PBSCT and BMT patients who were admitted to the MICU, compare mortality rates among patients who were either intubated or not intubated while accounting for specific patient characteristics and subgroups, and to use logistic regression analysis to identify admission variables that were associated with hospital survival.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Patients and Setting

Data were prospectively collected on 121 consecutive blood and marrow transplant patients admitted to The University of Texas M. D. Anderson Cancer Center MICU between July 13, 1994 and February 12, 1996. This 16-bed unit admits critically ill patients who are at least 16 yr old. Five patients who received both blood and marrow for transplantation, and one who had a salvage stem cell infusion and was not on the transplant service were excluded, yielding a total sample size of 115 patients for analysis. For patients who had more than one admission to the MICU, the first admission during the study period was used for analysis. Demographic, physiologic, and clinical data were collected on admission to MICU including patient age, gender, type of malignancy, type of transplant (autologous or allogeneic), source of stem cells (bone marrow or peripheral blood), human leukocyte antigen (HLA) compatibility (match or mismatch), disease status (relapse or remission), criteria for admission to MICU, vital signs, and laboratory data. The type of malignancy was classified into one of three categories: leukemia, lymphoma/myeloma, and solid tumor. Laboratory data were obtained within 24 h of MICU admission and included white blood cell count, hematocrit, platelet count, blood urea nitrogen (BUN), creatinine, total bilirubin, and arterial blood gas values. If the laboratory value was not available within 24 h of admission, the most recent laboratory value obtained up to 72 h prior to admission was used. Pertinent contributory diagnoses were noted on admission and included acute renal failure (ARF), hepatic failure, veno-occlusive disease (VOD), disseminated intravascular coagulation (DIC), acute gastrointestinal (GI) bleeding, and a probable or documented infection. The number of days since the transplant, and the number of days the patient was in the hospital before MICU admission were also included in the analyses. It was noted if the patient was intubated within 1 h of admission or at some time during the MICU stay, and attempts were made to determine the etiology of respiratory failure through both clinical and laboratory indices. Three severity of illness scores, Mortality Probability Model II (MPM II), Simplified Acute Physiology Score II (SAPS II), and Acute Physiology and Chronic Health Evaluation II (APACHE II), were calculated for each patient within the first 24 h of admission to MICU using the appropriate formulas (18). All patients were evaluated longitudinally to determine their ultimate MICU and hospital outcome.

Patient Variables

Patient characteristics are summarized in Table 1, and diagnostic criteria for admission to MICU are given in Table 2. Table 3 provides cross-tabulations between intubation, mortality, and patient characteristics. The variables summarized in Tables 123 were consensually defined as potentially important in predicting outcome in this specific patient population in a recent multicenter outcome study of cancer patients admitted to the intensive care unit in which we participated (21). ARF was defined as acute tubular necrosis or an acute diagnosis superimposed on chronic renal failure; prerenal failure was not included. DIC was defined as a combination of a prolonged prothrombin time (PT), prolonged partial thromboplastin time (PTT), hypofibrinogenemia, thrombocytopenia, positive fibrin split products or D-dimer in the proper clinical setting. Hepatic failure was defined as a progressive rise in total serum bilirubin associated with an inability to maintain normal coagulation in the absence of documented DIC or other factor consumption in the appropriate clinical setting. VOD was diagnosed when patients exhibited any two of the following clinically manifested characteristics occurring no later than 14 d from blood or marrow infusion: hepatomegaly or right upper quadrant pain, bilirubin greater than 2 mg/dl, ascites, or unexplained weight gain. Acute GI bleeding was defined as melena, hematemesis, or obvious "coffee ground" appearance in the nasogastric tube and an associated decrease in hemoglobin and hematocrit. A probable infection was defined as a suspected infection based on clinical signs and symptoms or a documented infection from blood, urine, sputum, or other specimen sent for culture. A patient was endotracheally intubated and received mechanical ventilation if the following factors were present: respiratory failure owing to hypoxemia or hypercarbia, or airway protection for seizures or the immediate postoperative period. Mortality was defined as the outcome that a patient admitted to the MICU did not subsequently leave the hospital alive.

Surveillance Committee

The study was approved by the M. D. Anderson Cancer Center institutional review board. The investigators were exempt from obtaining informed consent because existing data were collected from bedside monitoring equipment and patient medical records and routine care were not altered. Confidentiality was maintained by coding each subject upon enrollment and recording only by study code number. No specific clinical interventions or laboratory tests were performed on any patient for the purpose of the study, and only the observation of routine care was recorded by a research nurse in the MICU. The data were reviewed and verified by the medical director of the MICU.

Statistical Methods

Associations between intubation and mortality, both overall and within patient subgroups determined by other variables, were assessed initially by Fisher exact tests (22). For these tests, each quantitative variable was dichotomized by replacing it with the indicator of whether it was above or below its median. For example, because median age was 43 the binary variable indicating whether the patient's age was  43 or > 43 was used. Logistic regression analysis (23) was used to assess the multivariate relationship between multiple patient characteristics and the probability of mortality. For these logistic regression analyses, each numerical variable was first transformed as appropriate (24, 25), based on partial residual plots from logistic regression fits with the numerical variable as a single predictor, smoothed using the locally weighted regression and smoothing scatterplot (LOWESS) method of Cleveland (26). A multivariate logistic regression model was obtained by first fitting a model including an initial set of candidate predictor variables, described in RESULTS. A stepwise backward elimination (BE) was then performed using p value cutoff 0.05. Any individual variable previously eliminated from the model in the BE was then allowed to reenter the model if its new p value was < 0.05. Terms representing interactions between transplant type, source of stem cells, or intubation and each of the other predictive variables in the model were then added, and a final BE was then performed. This procedure was carried out twice, first starting with all available candidate predictor variables and a second time with intubation excluded from the set of candidate predictors, thus resulting in two different multivariate models. Model goodness-of-fit was assessed by smoothed residual plots, deviance statistics, and Hosmer-Lemeshow chi-square tests. All computations were carried out on a Sun SPARC Station 20 (Sun Microsystems, Mountain View, CA) in Splus (27) and StatXact (Cytel Software Corporation, Cambridge, MA).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Of the 115 patients admitted to MICU who received blood or marrow transplantation and met the inclusion criteria for analysis in this study, 62 (54%) were male and 53 (46%) were female (Table 1). The median age was 43 with a range from 18 to 63 yr. The indications for transplant were leukemia (61; 53%), lymphoma or multiple myeloma (39; 34%), carcinoma of the breast (14; 12%), and sarcoma (1; 1%). Among the general categories of diagnostic criteria for admission to the MICU, summarized in Table 2, the most common was pulmonary problems (49; 43%) followed by cardiac etiologies (23; 19%) and sepsis (18; 16%).

The overall mortality rate was 53.9% (62 of 115). However, mortality among intubated patients was 81.2% (39 of 48) versus 34.3% (23 of 67) among patients not intubated (p < 0.001). Because this difference was very large, we addressed the question of whether it could be explained by other variables, especially those that may be associated with the clinical decision to intubate. Table 3 summarizes the mortality rates among subsets of patients determined by each of 33 different variables and whether or not the patient was intubated.

Mortality among intubated patients was higher than among patients who were not intubated in each of the 68 patient subgroups examined in Table 3. Each p value in the table corresponds to a Fisher exact test of association between mortality and intubation within the patient subgroup identified in that row. Notably, the higher mortality among intubated patients was statistically significant in nearly all of the subgroups, with the nonsignificant comparisons occurring in subgroups with very small numbers of patients. Furthermore, none of the patients survived if they required intubation in addition to having any of the following characteristics: HLA mismatch, matched unrelated donor transplant, comatose level of consciousness, ARF, hepatic failure, DIC, or GI bleeding. Table 4 presents cross-tabulations between mortality, source of stem cells (PBSCT versus BMT), transplant type (allogeneic versus autologous) and intubation, similarly to those given in Table 3 but presented in terms of the association between mortality and stem cell source. Notably, the only significant association in Table 4 is within the subgroup of intubated patients, where the PBSCT patients had a better survival rate than the BMT patients. While this relationship was reversed among patients who were not intubated, the effect was insignificant. Moreover, the association between stem cell source and mortality disappeared when considered in the context of a multivariate logistic regression model including intubation as a predictive variable, described subsequently. Figure 1 illustrates the type of transplant received and the type of malignancy for the 48 intubated patients in relation to source of stem cells.

View larger version (20K): [in this window] [in a new window]   Figure 1.   Hospital outcome of PBSCT and BMT patients who received mechanical ventilation organized by type of transplant and malignancy. For definition of abbreviations, see Table 1.

Among the 48 intubated patients, 12 were intubated within 1 h of the time of admission to the MICU. The mortality rate among these patients was 8 of 12 (66.7%) compared with 31 of 36 (86.1%) among the 36 remaining patients intubated subsequent to admission. Although the mortality rate among patients intubated after admission was higher, the difference is not significant (p = 0.20). Further breakdowns by transplant type or source of stem cells yielded associations consistent with those in Table 4.

The final cause of respiratory failure among the intubated patients, which was determined by laboratory and clinical indices after all reports were read, is summarized in Table 5. The number of patients having each cause of respiratory failure is listed with their subsequent MICU and hospital outcome. This breakdown indicates that patients who survived to hospital discharge were intubated for various etiologies which did not differ from those who did not survive.

Table 6 summarizes the multivariate logistic regression model obtained by performing the variable selection procedure described in STATISTICAL METHODS. The candidate predictor variables included initially were: intubation, type of transplant (allogeneic or autologous) source of stem cells (PBSCT or BMT), HLA compatibility (match or mismatch), type of donor (related or unrelated), time since transplant, type of malignancy (leukemia, lymphoma/myeloma, or solid tumor), disease status (relapse or remission), patient age, the number of days the patient was in the hospital prior to MICU admission, presence of infection, ARF, GI bleeding, systolic blood pressure, heart rate, respiratory rate, temperature, white blood cell count, hematocrit, platelet count, albumin, lactic dehydrogenase, and total bilirubin. To avoid singular or unstable model fits, variables were not included in the set of candidate predictors if they represented a very small proportion of the sample (comatose, hepatic failure, DIC), were strongly associated with more clinically relevant variables (BUN and creatinine, each highly correlated with ARF), or contained numerous missing values (arterial blood gases, which were not measured in 20 patients). These variables were not included in the logistic regression analyses. The final model contained the eight predictive variables given in Table 6. The variable days = number of days because transplant is represented by the two terms days/100 and (days/100)² in the model to properly represent its functional relationship with the probability of mortality, _MORTALITY, as determined by preliminary goodness-of-fit analyses. Similarly, HR = heart rate is represented by the two terms HR and HR², and bilirubin is represented by log_e(bilirubin). The overall model fit was quite good, with deviance = 87.3 on 104 degrees of freedom (df) and Hosmer-Lemeshow chi-square test statistic 7.49 on 8 df (p = 0.48). Because days since transplant and heart rate are represented by two terms in the model, we include the p values corresponding to the 2 degrees-of-freedom test for significance of the two variables taken together. The tabled odds ratio (OR) for each binary variable is the estimated value of OR = _MORTALITY/ (1  _MORTALITY) = e where  is the variable's estimated coefficient in the model. Values of OR larger (smaller) than 1 correspond to an increased (decreased) risk of mortality associated with the variable. The tabled odds ratio for each numerical variable is the ratio OR(y)/OR(x) for the specified values y and x of the variable, which quantifies the increased risk of death as the variable increases from x to y.

To write down the predictive model, first denote the indicator of an event E by [E] = 1 if E occurs and [E] = 0 if E does not occur. Estimated values of _MORTALITY for given numerical values of the predictive variables in the model may be obtained from the formula

(1)

where the linear predictor  is given by

Table 7 illustrates how this model may be used to predict _MORTALITY. It provides the observed and predicted values of _MORTALITY for patients who were infected and had no GI bleeding, within each of the four subgroups determined by intubation and type of transplant. These 93 patients comprise most (81%) of the sample, hence are reasonably representative of the data, while the remaining 22 patients are scattered among several much smaller subgroups. For these predictions, the numerical variables in the model were represented by their medians, specifically respiratory rate = 25, days since transplant = 62, bilirubin = 1.4, and HR = 119. There is good agreement between the observed and predicted probabilities, bearing out the overall goodness-of-fit test. The degree of uncertainty that should be attached to each estimate is quantified by its associated 95% confidence interval. Table 7 illustrates the higher mortality associated with either intubation or allogeneic transplant, as well as the substantial patient-to-patient variability that is present even after accounting for the eight covariates in the model.

We repeated the model selection procedure described previously, but now intentionally omitting the indicator of intubation from the set of candidate variables. The purpose of this analysis was to determine the manner in which the other variables predicted mortality without first accounting for intubation, since this was such a strong predictor. In particular, we were interested in determining which admission variables, if any, might enter the model if intubation were excluded. The resulting model is summarized in Table 8. The effect of forcing intubation out of the model is that source of stem cells (PBSCT favorable) and temperature (higher temperature favorable) are now significant predictors in this model while type of transplant is not, although it is marginally significant with p = 0.07. This model also gives a good fit to the data, with deviance equal to 100.38 on 104 df and Hosmer-Lemeshow statistic 5.62 on 8 df, p = 0.69, although the fit is not nearly as good as the first model summarized in Table 6.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

The purpose of this study was to examine the subgroup of critically ill patients who had undergone a blood or marrow transplant and to identify variables readily available on MICU admission that were associated with hospital mortality. Our data arise from 115 consecutive HSCT patients admitted to the MICU of our cancer center over a 19-mo period.

Many severity of illness scores such as APACHE, MPM, and SAPS have been validated in different patient populations and have been found to underestimate the risk of mortality in critically ill patients with cancer (12, 21). The authors of these scoring systems have recommended individualizing scoring systems for distinct patient subgroups. The 115 HSCT patients considered here had significantly higher actual mortality than predicted using the APACHE II, MPM II, and SAPS II probability of mortality scores (Figure 2). Similar results were found in other cancer centers (21).

View larger version (22K): [in this window] [in a new window]   Figure 2.   Actual and predicted hospital mortality of blood or marrow transplant patients admitted to MICU.

Among all 115 patients, the survival was significantly lower among intubated patients compared with those who did not require intubation. Furthermore, for virtually every subgroup of patients analyzed, mortality was still significantly higher among intubated patients. This study supports previous data showing high mortality rates among BMT recipients requiring intubation in the intensive care unit (8, 17). These data, like those of Paz and associates, are primarily limited to medical oncology patients with the exception of one patient who was admitted to MICU after surgical repair of a ruptured bladder due to severe hemorrhagic cystitis (10). In Paz's study, the survival rate for MICU patients requiring mechanical ventilation was 3.7% compared with 81.3% for the nonintubated patients. In our study, 18.8% of the HSCT patients who required intubation survived compared with 65.7% for patients who were not intubated. In contrast to Paz's study, our study did show a significantly higher overall mortality among allogeneic transplant recipients (45 of 72, 62.5%) compared with autologous (17 of 43, 39.5%; p = 0.021).

In a logistic regression analysis accounting for a large number of patient characteristics, intubation stood out as the most significant factor associated with increased mortality. Crawford and coworkers identified three pretransplant characteristics that were associated with mechanical ventilation: age  21 yr, relapsed hematologic malignancy, and a nonidentical HLA graft (14). However, there were no definable pre- or posttransplant characteristics to predict survival. In our analysis, the variables associated with an increased mortality rate included the use of intubation, allogeneic transplant, presumed or documented infection, elevated heart rate, elevated respiratory rate, elevated bilirubin, and concurrent GI bleeding. The length of time since the patient underwent the transplant also was important in predicting the in-hospital mortality risk. The effect was quadratic on the logistic scale: a longer elapsed time since transplant was predictive of an increased risk of mortality up to about a 5 yr period, with a decrease thereafter. These data confirm previous reports on a high mortality risk among intubated bone marrow transplant recipients.

In the recent report by Rubenfeld and Crawford, the survival was zero for intubated BMT recipients who had lung injury and either required more than 4 h of vasopressor support, or had sustained renal and hepatic failure (17). In our study, none of the patients survived who were intubated and had ARF, hepatic failure, DIC, GI bleeding, or a matched unrelated donor transplant. Only 3 of 10 intubated patients who received vasoactive agents on admission survived. When we excluded intubation from the set of predictors of mortality in the logistic regression model, this had the effect of bringing in source of stem cells (PBSCT favorable) and temperature (higher temperature favorable) as significant predictors and excluding type of transplant, while all other predictors remained in the model.

This is the first study to date that examines the outcome of critically ill peripheral blood progenitor cell transplant recipients. Our data indicate that these patients have a more favorable outcome than BMT patients if they require mechanical ventilation. This is in agreement with the overall better survival of PBSCT patients as compared with BMT patients reported by Przepiorka and colleagues (28). For allogeneic transplantation, blood stem cell recipients had the least regimen-related toxicity, more rapid platelet recovery, fewest early deaths, and earliest discharge. Talmadge suggested that the more rapid reconstitution of natural killer and T cells in PBSCT patients may contribute to an improved clinical outcome in these patients compared with BMT patients (2). PBSCT patients had a lower incidence of pulmonary dysfunction, liver dysfunction, protein C deficiency, and antithrombin III deficiency compared with BMT patients in another series (31). As more PBSC transplants are performed, more outcome data will likely become available.

This study is unique in that data were collected prospectively in this critically ill patient population. Those of us who provide critical care to transplant recipients have long observed, and these data confirm, that the mortality risk is high when patients require mechanical ventilation. Rubenfeld and Crawford, however, have also shown that survival after mechanical ventilation has improved at their institution from 5 to 16% over the last 5 yr (17). It is not always apparent at the time a patient's condition deteriorates whether or not ICU care and/or mechanical ventilation will result in a favorable outcome. We have shown that standard scoring systems such as APACHE, MPM, and SAPS underestimate mortality in blood and marrow transplant recipients. Our institution has recently participated in a multicenter trial specifically investigating the outcome of critically ill cancer patients. This study also found that SAPS II and MPM II consistently underestimated the mortality when cancer patients were admitted to the ICU (21). This multicenter trial has also formulated an ICU admission model that would predict outcome in cancer patients via logistic regression analysis. Our model differs from other models in that it examines the specific patient population of BMT and PBSCT recipients.

Perhaps these models and survival predictors such as ours and those by Rubenfeld will aid in the formation of more standardized guidelines to follow when deciding on appropriateness of ICU care for a particular patient. As critical care physicians, it is not possible to predict to a patient or his or her family with 100% certainty that death will or will not result after intubation and mechanical ventilation. Certain issues limit our ability to make these predictions. For example, some patients are given a do-not-resuscitate status or have limits placed on their care at various times during the ICU stay. This is difficult to follow and limits our ability to compare different groups of patients equally. However, we now have evidence that certain clinical parameters identified early in the course of a patient's admission to the MICU can give a realistic picture of that patient's ultimate chance of survival. It is imperative that the transplant recipient be educated on potentially life-threatening complications that may occur, and we must provide available outcome data to our patients early in the course of treatment. Guidelines such as those proposed by Rubenfeld and Crawford will likely be tailored to an individual institution's experience and should be presented to patients and their families to avoid futile care. Finally, in times of cost containment, we are forced to judiciously allocate limited resources, particularly in cost-intensive areas such as represented by ICUs. As more prognostically significant data become available, we will be obligated to present it to patients and families and also to use it to optimize the utilization of available resources.