Development of a mortality risk prediction score for patients with AML requiring critical care

Abstract

Background: Though patients with acute myeloid leukemia (AML) have high mortality after intensive care unit (ICU) admission, long-term survival is possible. There is no accepted model for predicting longer-term mortality after ICU admission. We examined the role of five AML and ICU related risk factors in mortality prediction in this population. Methods: This was a retrospective cohort study of 238 patients with AML admitted to the ICU in a single academic medical system who survived at least 24 hours. The risk factors examined were AML composite model score (AML-CM) from the time of diagnosis, disease status at ICU admission, history of hematopoietic cell transplant (HCT), use of invasive mechanical ventilation (IMV) within 24 hours of ICU admission, and serum creatinine ≥2.0 mg/dl within 24 hours of ICU admission. The outcome was 90-day mortality after ICU admission. Risk factors were compared using sensitivity, specificity, and AUC, and added sequentially to a multivariable logistic regression model. Sequential multivariable models were compared using AUC. A mortality score was created from the regression model with the best balance of fit and simplicity by dividing the beta coefficients in the model by 0.25. The score was the sum of those coefficients. Cut-points for the score were examined using sensitivity and specificity. Results: The final model used AML-CM, the presence of relapsed or resistant disease, HCT, and IMV within 24 hours of ICU admission to predict 90-day mortality. A term was included for interaction between relapsed/resistant disease and HCT. The AUC of the multivariable logistic regression model was 0.74 (95% CI 0.68-0.8). The AUC of the final mortality score was 0.74 (95% CI 0.68-0.8). The score ranged from 0 to 18. A cut-point of ≥4 had a sensitivity of 93% and a specificity of 30%; a cut-point of ≥10 had a sensitivity of 25% and a specificity of 94%. Conclusions: This mortality score is the first that examines 90-day mortality from the time of ICU admission for patients with AML. While it shows some promise at distinguishing patients with exceptionally high or low mortality, in a number of patients the use of the score will lead to an erroneous assessment of prognosis.