To explore the prognostic value of serum albumin（ALB）level at admission in de novo non-M3 acute myeloid leukemia（AML）.

A retrospective cohort study was conducted on the medical records of newly diagnosed non-M3 AML patients admitted to our center from January 2012 to July 2019. They were divided into a normal ALB group（n=196）and a low ALB group（n=52）based on the optimal cut-off value of 30.4 g/L generated by the X-Tile software. The clinical data were compared between the 2 groups, such as gender, age, white blood cell（WBC）count, hemoglobin（HB）level, platelet（PLT）count, lactate dehydrogenase（LDH）level, French American British（FAB）type, percentage of bone marrow blasts（BM blasts）, risk stratification of the European Leukemia Network（ELN）, and treatment regimens. All patients were followed up for their survival status. Kaplan-Meier survival curve and Cox regression model were used to compare the survival between the 2 groups. A prediction model was constructed based on the results of multivariate Cox regression analysis. Calibration curve and receiver operating characteristic（ROC）curve were plotted to evaluate the model, and Harrell's concordance index（C-index）and decision curve analysis（DCA）were performed to evaluate the incremental value of ALB's prediction model.

The low ALB group had a significantly lower complete response（CR）rate after induction chemotherapy（28.8% vs 62.8%, P<0.001）and a higher mortality rate（63.4% vs 30.6%, P<0.001）than the normal ALB group. Survival analysis found that the overall survival time（OS）and disease-free survival time（DFS）were significantly shorter in the low ALB group than the normal ALB group（P<0.001）. Multivariate analysis showed that ALB was an independent predictor of OS in AML prognosis（HR=4.352, 95%CI: 2.760~6.863, P<0.001）. The AUC value of the model constructed based on Cox analysis results was 0.800（95%CI: 0.734~0.867）and 0.890（95%CI: 0.795~0.984）, respectively, for predicting 1-year and 3-year survival rate. The calibration curve performed well between predicted survival and actual survival. Compared with traditional risk stratification and the model without ALB factor, our model that included ALB had a higher C-index（0.564 vs 0.671 vs 0.730）. DCA also showed its higher clinical benefits.

ALB is an independent predictor of prognosis in non-M3 AML patients, which can significantly improve the clinical predictive ability and the attendant benefits of survival prediction models.