AUTHORS
Majed W. Al-Nazer, MD, Osama Alzoubi, MD2, Akshay Mathavan, MD3, Akash Mathavan, MD4, Lauran Zeineddine, MD5, Alexandria Iakovidis, MD6, Miguel Castro, MD7, Dina Khalaf, CLINICAL ASSOCIATE PROFESSOR8, Hiren J Mehta, MD9
AFFILIATIONS
1 University of Florida, Gainesville, Florida, United States
2 University of Florida, Gainesville, Florida, United States
3 University of Florida, Gainesville, Florida, United States
4 University of Florida, Gainesville, Florida, United States
5 University of Florida, Gainesville, Florida, United States
6 University of Florida, Gainesville, Florida, United States
7 University of Florida, Gainesville, Florida, United States
8 University of Florida, Gainesville, Florida, United States
9 University of Florida, Gainesville, Florida, United States
ABSTRACT
Background: Hematopoietic stem cell transplant (HCT) recipients admitted to the intensive care unit (ICU) have a high mortality rate, and triage decisions about escalation of management and goals of care are often anchored to how critically ill a patient appears at the time of ICU admission. SOFA and APACHE II scores were developed in the general ICU population, where external validation shows an area under the receiver operating curve (AUC) of approximately 0.80 for each score in predicting ICU mortality. It is unclear whether these scores, or any other variables measured at admission, identify risk specifically in patients who previously underwent HCT, whose mortality is also driven by cytopenias, graft-versus-host disease (GVHD), life threatening infections, and relapse independent of admission physiology. In this retrospective cohort study, we explore how these predictive scores perform in HCT recipients admitted to the ICU.
Methods: We identified 90 adult pts with a prior history of HCT who were admitted to the UF Health medical ICU between June 2011-Oct 2024. 21 patients (25%) had multiple ICU admissions; with the most recent admission analyzed. The primary outcome was death during that admission. Candidate predictors measured at admission included the SOFA and APACHE II scores; lab values (lactate, albumin, bilirubin, creatinine, hemoglobin, platelets, absolute neutrophil count, PaO2/FiO2); performance status (ECOG, Karnofsky); age, sex, transplant type, time since HCT; and transplant complications (acute/chronic GVHD, relapse, retransplant, steroid-refractory disease). For each variable, we quantified discrimination between survivors and non-survivors using the AUC, with DeLong confidence intervals and the DeLong test for comparisons. The data was then fit using models based on Firth penalized logistic regression. We also tested all 120 possible three-variable combinations of 10 admission variables, with bootstrap correction for the selection process.
Results: The median age was 58 years (IQR 50–65), 51% were female, and 68% had received an allogeneic transplant. The median SOFA score was 9 (IQR 6–11) and the median APACHE II score was 23 (IQR 19–27). Mortality during ICU admission was 54% (49/90), and 13% were alive at the last follow-up. SOFA and APACHE II were the best admission predictors but achieved only modest discrimination, with an AUC of 0.689 (95% CI 0.576–0.802) for SOFA and 0.672 (95% CI 0.557–0.786) for APACHE II, and no detectable difference between them (p = 0.70). All other candidates showed lower discrimination. The lab value that performed the best was lactate (AUC 0.631, 95% CI 0.505–0.756), but was not statistically significant (p = 0.072). All other candidates, including the remaining lab values, performance status, age, and transplant type, had AUC of 0.50-0.58.
Transplant complications similarly showed poor discrimination, with AUC of 0.523 for acute GVHD, 0.501 for chronic GVHD, 0.587 for relapse, 0.553 for retransplant, and 0.554 for steroid-refractory disease. Nothing added significantly to SOFA, with the largest increment coming from relapse history (+0.029, p = 0.39). A search of all 120 three-variable admission models produced an apparent AUC of 0.697, which fell to 0.570 after correction for selection, worse than SOFA alone in the same pts (0.661).
Conclusions: Among HCT recipients admitted to a medical ICU, no admission variable including severity score, lab value, performance status, or transplant complication demonstrated discrimination strong enough to reliably guide prognostication for an individual patient. SOFA and APACHE II performed best, but only achieved modest discrimination, and building models directly from the data performed worse still. Notably, point estimates for both scores were numerically lower than the discrimination typically reported for these scores in general ICU populations, although our study was not designed to make this comparison formally. Nonetheless, clinicians should be cautious about anchoring triage or goals of care discussions to admission severity assessment. Our findings support the importance of continued investigation of transplant-specific prognostic markers including dynamic measures through the ICU course, immune reconstitution, and disease trajectory. This highlights a need to utilize prognostic tools developed specifically for HCT recipients admitted to the ICU, rather than reliance on scores developed for the general ICU population.
DISCLOSURES
Majed W. Al-Nazer, MD: No relevant conflicts of interest to declare.