We conducted a 6.5-year retrospective matched case-control study to compare the clinical characteristics, course, and outcomes of patients with AML admitted to ICU compared with non-critically ill hospitalized AML controls and non-AML critically ill controls.
Our data have several relevant findings. First, while AML patients represent only a minority of all ICU admissions, we found that life threatening illness prompting ICU admission is not uncommon, occurring in 13% of all patients with a diagnosis of AML. Second, these critically ill AML patients had predominantly septic and/or respiratory-related diagnoses and exhibited a high acuity of illness, with the majority presenting with shock. Moreover, AML patients had greater severity of organ dysfunction. Third, AML patients were less likely to receive mechanical ventilation; yet overall they received high intensity support that included vasoactive therapy in 75.6% and RRT in 26.7%, respectively. This translated into longer, though non-significant, stays in ICU and hospital. Fourth, we also found critically ill AML patients had lower adjusted-survival when compared with controls. By univariate analysis, several factors were associated with a lower probability of survival, including older age, AML M1 subtype, higher baseline SOFA score, presence of shock, vasoactive therapy and mechanical ventilation. We also found that worsening SOFA score early after ICU admission correlated with lower survival. Alternatively, receipt of AML-specific chemotherapy while in ICU correlated with higher survival. Finally, changes in patient resuscitation status (i.e. to NFR) and withdrawal of support were both significantly associated with higher likelihood of death at 90-days.
There is a paucity of epidemiological data pertaining to patients with AML requiring life-sustaining measures in the ICU. (Additional File 1) While numerous small series have reported on the clinical outcomes for patients with all forms of hematologic malignancy admitted to ICU, very few have focused on AML, and fewer still have provided estimates of incidence[13, 14, 16, 17]. During a 5-year surveillance, Tremblay et al reported that among 163 consecutive hospitalized AML patients, with a range in AML status (including 38 having received a HSCT), only 32 were admitted to ICU and supported with mechanical ventilation (cumulative incidence 19.6%). We observed a lower cumulative incidence, with just over 1 in 10 of all newly diagnosed AML patients being supported in ICU; moreover, we observed that AML patients comprised <1% of all ICU admissions during the study period. In a 4-year retrospective study, Merz et al reported on 101 ICU admissions (n = 84) with hematologic malignancies, of which 54.4% had a diagnosis of AML. These AML admissions represented approximately 1.4% of all "emergent and medical" ICU admissions.
Among AML patients admitted to ICU, the in-hospital, 90-day and long-term adjusted-survival were observed to be lower when compared with either hospitalized AML or ICU controls. To our knowledge, this is the first study to evaluate the long-term survival experience of critically ill AML patients compared with matched controls. When compared with an unmatched convenience sample of non-AML critically ill patients, Tremblay et al found AML patients requiring mechanical ventilation to clearly have higher in-hospital mortality. Merz et al found in-hospital mortality noticeably higher for ICU admissions associated with hematologic malignancies compared with unmatched controls (33.7% vs. 10.7%, p < 0.0001). In general, the observed survival in our study is largely consistent with prior studies, where estimates of ICU, in-hospital and 1-year survival were 12-70%, 3-64%, and 0-34%, respectively [12–14, 16, 17, 20, 21, 26–34]. The median survival of critically ill AML patients in our cohort was only about three months, in contrast to approximately nine months for the control groups. While the majority of deaths in AML patients occurred early after ICU admission (median duration 5 days), the adjusted risk of death remained significantly higher during follow-up compared with either control group. These data would suggest, in general, that AML patients developing an episode of critical illness prompting ICU support have a less favorable outcome compared with hospitalized AML and non-AML ICU patients.
Prior data imply AML patients supported in ICU have greater health resource utilization. This was shown by Merz et al, who described considerably higher ICU resource use by hemato-oncological patients, as measured by the Therapeutic Interventions Scoring System (TISS) (214 vs. 95, p < 0.0001) and ICU length of stay (2.0 vs. 1.1 days, p < 0.02), compared with emergent medical ICU patients. Despite no difference in the rates of mechanical ventilation or RRT between groups, both of these interventions showed significant correlation with total ICU resource use. This also translated into greater total direct costs per ICU admission for hemato-oncological patients. While our study did not incorporate a formal cost analysis, AML patients in our study had non-significant, but longer observed stays in both ICU and hospital. Interestingly, our data found significantly fewer AML patients received mechanical ventilation, despite evidence of similar or worst lung injury. This observation may, in part, be attributable to prior data correlating less favorable prognosis for these patients requiring mechanical ventilation[17, 22, 28, 35]. These data also suggest greater health resource use for AML patients; however, there is little data on whether this is balanced by higher quality-adjusted survival of ICU admissions with hematologic malignancy. In a small series of 92 critically ill patients with hematologic malignancy, Yau et al described the quality-of-life as acceptable or good in the 7 survivors (7.6%) at 1-year, with no patient reporting significant limitations to daily activities.
Few data have explored the long-term disposition of AML patients after life-threatening critical illness. Interestingly, despite lower overall survival, 46.7% of all AML patients supported in ICU in our study survived hospitalization to be discharged home (or 84.0% of survivors). This was statistically comparable to hospitalized AML and ICU controls.
The characterization of potential factors predictive of survival for hematologic malignancy patients prior to and/or early after ICU admission has been the focus of considerable research effort[12, 13, 16, 20, 22, 26, 28, 30–32, 35]. Several factors have been suggested to predict non-survival at various time points across a range of studies including: older patient age, poor performance status, AML subtype, relapsed/refractory AML status, poor cytogenetic profile, higher illness severity or organ failure score[13, 16, 18, 22, 31, 32], leukopenia, vasopressor therapy[12, 16, 22], mechanical ventilation[13, 16, 22], acute kidney injury, absence of bloodstream infection[12, 27], and fungal sepsis. In our cohort, we found that older age, AML M1 subtype, higher organ failure score, the presence of shock, vasopressors therapy, and mechanical ventilation were all associated with lower survival. There was no association between survival and cytogenetic profile, therapeutic regimen or burden of co-morbid illness. However, these prognostic factors have been inconsistent and not readily reproducible, due largely to issues related to study design (i.e. small, single centre, retrospective, no controls, analysis) and lack of generalizability (i.e. selection bias, mixed hematologic populations). Of note, we also found receipt of active intensive chemotherapy while in ICU was associated with higher observed survival. Recently, Vandijck et al reported that active chemotherapy was associated with better ICU survival in a cohort of 77 patients admitted to ICU with severe sepsis or septic shock. These observations would suggest that prognosis for AML patients receiving active chemotherapy may be better than perceived.
Perhaps no constellation of disease or treatment-specific factors will consistently predict outcome for all AML patients developing life-threatening critical illness. As such, it may be that prognostic factors may have a limited role in the decision to admit an AML patient to ICU. Indeed, hospital survival in our study was 55.6%, with 84.0% of survivors returning home and 29.9% alive at 1-year. Instead, prognostic factors may have greater relevance for decisions regarding the withholding of life-sustaining therapies and in end-of-life care. For example, worsening illness severity and/or organ dysfunction early after ICU admission has been found to predict poor clinical outcome[11, 31, 32]. Massion et al showed an improvement in Multi-Organ Dysfunction Score (MODS) of ≤1 point between ICU admission and day 5 correlated with increased mortality (81% vs. 29%, p = 0.001). Merz et al described significant higher 28-day mortality for patients with worsening SOFA score over the first 48 hours (OR 24.4, 95% CI, 8.8-67.9, p < 0.0001). Finally, Lamia et al showed worsening Simplified Acute Physiology Score (SAPS) II score, Logistic Organ Dysfunction score (LODS), and SOFA score during the first three days after ICU admission to be the best predictors of in-hospital mortality. Our study further confirmed these observations. We found that the absence of change or an increase in organ failure score early after ICU admission correlated with lower survival.
Furthermore, we found that a change in patient resuscitation status from full resuscitation to not for resuscitation (NFR) and withdrawal of active support were common and both significantly modified the observed mortality at 90-days. In a series of 124 consecutive critically ill hematological malignancy patients, Benoit et al found 4% had a NFR order written within 24 hours of ICU admission, while 20.2% had a NFR order written after a median 7 days in ICU. In our study, only 6.7% of AML ICU patients had a NFR order at presentation, whereas an additional 46.7% had their status later changed to NFR while in ICU. While we did not specifically examine the timing of this change, decisions about withholding life-sustaining measures in our study may have related to higher or worsening organ failure score and need for escalation in treatment intensity (i.e. vasoactive therapy, mechanical ventilation, RRT). We did not find association between change in resuscitation status and age, co-morbidity score, AML subtype, poor AML cytogenetic prognosis, or duration of ICU stay. We recognize one plausible explanation for this finding may relate to a clinician perception of poor clinical outcome and/or general reluctance to pursue aggressive and/or prolonged life sustaining measures for AML patients.
There are important limitations to our study. First, our study was single-centered and retrospective in design and was therefore potentially susceptible to bias. For example, we were unable to determine whether AML patients designated as "non-ICU" had end-of-life discussions that precluded admission to ICU, despite comparable "critical illness", which would unduly influence our incidence estimate (i.e. selection bias). However, we have attempted to minimize this by showing no significant differences in the therapeutic regiments received by AML cases and controls. Second, despite over 6 years of surveillance for AML patients admitted to our ICU, our sample was small and provided limited statistical power; we therefore omitted multi-variable analysis for determination of predictors of survival. To compensate, we further matched AML patients based on age, sex and APACHE II score (ICU only). While we did not a priori match AML groups for cytogenetic profile or therapeutic regimen received, we found no statistical differences between the AML cases and controls. Third, we endeavored to match the final cohort of AML cases with non-AML ICU controls by a ratio of 1:5; however, we recognize these ICU controls still represent a heterogeneous population and, despite similar illness severity, have observed differences that may impact outcome (i.e. primary diagnosis, prevalence of co-morbid illness). Finally, we were not able to ascertain additional secondary outcomes of relevance to survivors, such as health-related quality of life or functional status and the details of ongoing care requirements, which may also have prognostic significance.