Clinical and sociodemographic predictors of intensive care unit admission following chemotherapy in acute myeloid leukemia

Treatment-related mortality remains a major driver of early death in acute myeloid leukemia (AML). Advances in supportive care have reduced early mortality but fail to benefit patients equitably.^1-3 Our previous work had confirmed Black-White survival disparities in newly diagnosed AML and indicated that intensive care unit (ICU) admission during induction chemotherapy was a significant mediator of these disparities.⁴ Current prediction tools to identify the risk of treatment-related mortality (TRM) in AML rely on clinicopathologic variables such as marrow reserve, age, performance status, and organ function, but fail to consider the social determinants of health (SDOH) that are increasingly recognized to impact the success of AML therapy.⁵ This represents the first study to include sociodemographic characteristics along with clinicopathologic characteristics as predictors of ICU admission during induction chemotherapy.

The Chicago AML registry provides a unique resource for performing a detailed analysis of the early treatment course in a contemporary, diverse population. Data were collected in Research Electronic Data Capture (REDCap) hosted at the University of Illinois Chicago using an Institutional Review Board-approved protocol and data use agreement at each institution. We examined 950 adult patients consecutively diagnosed with non-promyelocytic AML between January 2012 and 2022 who received intensive induction chemotherapy at one of 6 academic institutions; 32 patients were excluded on the grounds of missing data on ICU admission. The objective was to identify predictors of critical illness in the early period following chemotherapy incorporating self-reported race and census tract measures. To account for missing data, multiple imputation was conducted and the multiply imputed datasets analyzed. Using the method of chained equations, we multiply imputed missing values for each of the patients and clinical risk factors (see Table 1). Binary variables were modeled in logistic regression, whereas multinomial logistic regression was used for variables with multiple categories. Allogeneic stem cell transplantation rates and survival were evaluated to determine the longterm clinical impact of critical illness during initial therapy. A total of 918 newly diagnosed AML patients were included in this analysis (Online Supplementary Table S1). The patient population comprised 65% Non-Hispanic White (NHW), 12% Non-Hispanic Black (NHB), 13% Hispanic, and 10% Other race/ethnicity. Median age was 57 years (interquartile range [IQR] 47, 65). NHB and Hispanic patients were significantly younger than NHW patients (19% NHB vs. 26% Hispanic vs. 13% NHW were <40 years). Two-thirds of NHW patients were married compared to 38% of NHB and 57% of Hispanic patients. The NHB and Hispanic patients were over-represented in moderate and high disadvantage tracts, and under-represented in high affluence tracts. Compared to NHW patients, NHB patients were more likely to have public insurance (65% vs. 44%) and Hispanic patients were the most likely to be uninsured (20% vs. 2% of NHW and 3% NHB).

Comorbities were highest in NHB patients: 39% NHB vs. 25% NHW vs. 24% Hispanic patients had a Charlson Comorbidity Index (CCI) ≥3. European LeukemiaNet (ELN) 2017 classification differed significantly based on race/ ethnicity. Hispanic patients were most likely to have a favorable ELN 2017 risk group (24% vs. 18% NHB and 21% NHW patients). Hyperleukocytosis was seen in 17% of patients, and 7% presented with extramedullary disease involvement. The majority of patients had fibrinogen level derangement; 25% had low fibrinogen (<100 mg/dL) while 38% had elevated fibrinogen (>380 mg/dL).

Hispanic patients were most likely to have elevated fibrinogen (46% vs. 34% NHW and 44% NHB). Hypoalbu-minemia, a known predictor of adverse AML outcomes,⁶ was more prevalent in NHB patients (24% vs. 15% NHW and 12% Hispanic patients). Rates of primary treatment failure were similar between groups (12-18%) although, numerically, Hispanic patients had the lowest.

A total of 161 patients representing 17.5% of the study population required ICU admission following chemotherapy. Treatment details were available for 143 patients; the majority (N=105) received standard 7+3 based regimens while 37 patients received high-dose cytarabine-containing regimens (e.g., FLAG, cytarabine-mitoxantrone). Twenty percent of ICU patients (N=29) received novel therapies (CPX-351, midostaurin, and investigational agents). For 89 patients with available data on indication for ICU admission, 30 (34%) had infectious complications, 25 (28%) respiratory failure, 16 (18%) cardiovascular complications, 13 (15%) patients had bleeding, and 5 patients had altered mental status. In the 72 patients with available data, the median number of days between start of induction and ICU admission was 22 (IQR:10-57).

We examined patient characteristics associated with ICU admission following chemotherapy. In univariate analysis (Table 1), race/ethnicity was a baseline predictor of ICU admission, with the highest rates of ICU admission in Hispanic patients at 27% compared to 19% of NHB and 14% of NHW patients (P=0.004). CCI was also associated with ICU admission; a CCI score of 0-1 was associated with an ICU admission rate of 14-16% compared to 20-21% for CCI of 2 or higher (P=0.03). Patients with high census tract disadvantage were the most likely to be admitted to the ICU compared to those from intermediate or low disadvantage tracts (61% vs. 41% and 48%, respectively, P=0.09). Disease-specific characteristics also predicted ICU admission. Patients with a high ELN 2017 genomic risk score were more likely to be admitted to the ICU compared to those with a moderate or low ELN score (21% compared to 16% and 13%, respectively, P=0.01). Patients with high disease burden as measured by significant white blood cell (WBC) count elevation (> 50K) were more likely to be admitted to the ICU. Additionally, low albumin (< 3 g/dL) and elevated fibrinogen (> 380 mg/dL) levels at diagnosis were also associated with ICU admission. Interestingly, established factors for mortality in AML, including older age, Eastern Cooperative Oncology Group (ECOG) performance status, and elevated creatinine (>1 g/dL) did not predict ICU admission.

Table 1.Sociodemographic and clinicopathologic factors associated with intensive care unit admission following high-intensity induction therapy for acute myeloid leukemia.

As previously published,⁷ elevated Body Mass Index was not predictive of adverse chemotherapy outcomes in AML. Marital status and insurance type were not associated with ICU admission.

Separate multivariable models (Table 2) were constructed for each risk factor, controlling for potential confounding variables. We did not control for variables potentially in the causal pathway from the risk factor of interest to ICU admission (see Online Supplementary Figure S1). Compared to NHW patients, Hispanic ethnicity (RR=1.87, 95%CI: 1.31, 2.67, P=0.001) and Other race (RR=1.70, 95%CI: 1.10, 2.62, P=0.02) were predictors of ICU admission after adjusting for age and gender. Patients with adverse molecular / cytogenetic characteristics defined by ELN 2017 score were more likely to require intensive care (RR 1.85, 95%CI: 1.16, 2.96, P=0.01) adjusting for race, age, gender, insurance, and social determinants. Examining lab measures at diagnosis, WBC count, serum albumin and fibrinogen were independent predictors of ICU admission adjusting for race/ethnicity, age, gender, social determinants, insurance, comorbidities, and ELN risk score.

Critical illness during intensive chemotherapy negatively impacted the ability to receive further therapies and served as a surrogate for inferior survival (HR for overall survival=1.88, 95%CI: 1.44, 2.44, P<0.0001). Patients admitted to the ICU during induction therapy were less likely to undergo curative therapy with allogeneic stem cell transplant (60% vs. 40.5%, RR=0.67, 95%CI: 0.55, 0.82). ICU admission during induction therapy was also associated with a 3.9-fold greater risk of death at 90 days (RR=3.95, 95%CI: 2.39, 6.52) and a 54% greater risk of death at one year (RR=1.54, 95%CI: 1.25, 1.90) (Figure 1).

Table 2.Multivariable adjusted predictors of intensive care unit admission following high-intensity induction therapy.

In this well-annotated contemporary cohort of AML patients treated with intensive chemotherapy, over 15% of patients developed critical complications requiring admission to the ICU, similar to previous studies reporting rates of 15-28%.^8,9 In addition to short-term morbidity and 90-day mortality, these ICU patients demonstrated compromised long-term outcomes.

We integrated validated clinical and molecular predictors of outcome in newly diagnosed AML¹⁰ with SDOH and showed for the first time that self-reported race/ ethnicity and neighborhood disadvantage are associated with increased incidence of critical illness following intensive chemotherapy. ECOG performance status was not a significant predictor in itself, although this variable was limited by a significant amount (>25%) of missing data which may have underestimated its influence. In multivariable analysis, adjusting for age and gender, Hispanic ethnicity was independently predictive of ICU admission despite younger patient age and more favorable-risk disease. Previous studies have reported inferior AML survival in Hispanic patients,¹¹ and our results suggest this may be linked to differences in treatment tolerability.

Differences in cumulative front-line organ toxicity may lead to differential treatment at relapse and drive subsequent racial disparities in survival in AML.¹² Our data in adult AML confirm that ICU admission has downstream consequences including a lower likelihood of curative allogeneic stem cell transplant and reduced 90-day and 2-year survival.

Targeted low intensity therapies in AML provide highly effective alternatives to intensive chemotherapy but the latter remains the preferred curative approach for patients under 60 years of age. Recent data demonstrate that within favorable molecular subsets of AML, the survival benefit of chemotherapy is modulated by race.¹³ The current findings suggest that the integration of race/ethnicity and allostatic load measures,¹⁴ including albumin and fibrinogen, as prognostic biomarkers into current AML treatment mortality calculators merits further study. Improved selection criteria for intensive therapy incorporating measures and biomarkers of SDOH may reduce early treatment complications and mitigate survival disparities in AML.

Figure 1.Kaplan-Meier estimates of survival in non-intensive care unit patients and intensive care unit patients with acute myeloid leukemia. HR 1.88 (95%CI: 1.44-2.44, P<0.0001)

Footnotes

• Received August 31, 2025
• Accepted December 31, 2025

Correspondence

*IEA and GHR contributed equally as first authors

I. KHAN - irum.khan@northwestern.edu

Disclosures

No conflicts of interest to disclose.

Contributions

Funding

This work was funded by a grant through the ChicagoCHEC Partnership: U54CA202995, U54CA202997, and U54CA203000 (to GHR and IK).