Abstract
BACKGROUND: Early death (ED) rate in acute promyelocytic leukemia (APL) remains high. Some studies have identified prognostic factors capable of predicting ED, whereas no risk rating system for ED has been reported in the literature. In this study, a risk classification system was built to identify subgroup at high risk of ED among patients with APL. METHODS: Totally, 364 consecutive APL patients who received arsenic trioxide as induction therapy were included. Ten baseline clinical characteristics were selected for analysis, and they were de novo/relapse, age, sex, white blood cell count, platelet count, serum fibrinogen, creatinine, uric acid, aspartate aminotransferase, and albumin. Using a training cohort (N=275), a multivariable logistic regression model was constructed, which was internally validated by the bootstrap method and externally validated using an independent cohort (N=89). Based on the model, a risk classification system was designed. Then, all patients were regrouped into de novo (N=285) and relapse (N=79) cohorts and the model and risk classification system were applied to both cohorts. RESULTS: The constructed model included 8 variables without platelet count and sex. The model had excellent discriminatory ability (optimism-corrected area under the receiver operator characteristic curve=0.816±0.028 in the training cohort and area under the receiver operator characteristic curve=0.798 in the independent cohort) and fit well for both the training and independent data sets (Hosmer-Lemeshow test, P=0.718 and 0.25, respectively). The optimism-corrected calibration slope was 0.817±0.12. The risk classification system could identify a subgroup comprising ~25% of patients at high risk of ED in both the training and independent cohorts (OR=0.140, P<0.001 and OR=0.224, P=0.027, respectively). The risk classification system could effectively identify patient subgroups at high risk of ED in not only de novo but also relapse cohorts (OR=0.233, P<0.001 and OR=0.105, P=0.001, respectively). CONCLUSION: All the results highlight the high practical value of the risk classification system.