A Simple-to-Use Nomogram for Predicting Survival in Children with Acute Myeloid Leukemia

Abstract

Background: The research analyzed a group of patients to develop a statistical nomogram and a web-based survival rate predictor for the comprehensive estimate of the overall survival (OS) of children with acute myeloid leukemia.

Methods: Between 1999 to 2015, we used the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) database to evaluate and randomly divide 440 children diagnosed with AML into the population of training (n = 309) and validation (n = 131). The analysis of Lasso Cox was used to identify separate predictive variables. We have used essential forecasting considerations to construct a nomogram and a web-based calculator focused on Cox regression analysis. Nomogram validation was tested through discrimination and calibration.

Results: Compared to the multivariate training cohort models, a nomogram integrating gender, age of diagnose, WBC at diagnosis, bone marrow leukemic blast percentage, and chromosomal abnormalities [t(8; 21), inv(16)] were designed for the prediction of OS. We also developed a predictive survival nomogram and a web-based calculator. C-indexes validated internally and checked externally were 0.747 and 0.716. The calibration curves have shown that the nomogram might accurately forecast 3-year and 5-year OS.

Conclusions: A nomogram effectively predicts survival in children with AML. This prognostic model can be used in clinical practice.

Figure 1

Identification of predictive factors using the Lasso Cox regression. (a) The vertical line was plotted at the given λ, selected by cross-validation. For the optimal λ, 6 features of them with a nonzero coefficient were selected. (b) The penalization coefficient λ in the Lasso model was adjusted using cross-validation and the minimum criterion. The vertical black lines define the optimal λ.

Figure 2

Predictive nomogram for OS estimation.

Figure 3

A dynamic web-based calculator b 95%confidence interval according to the web survival rate calculator.

Figure 4

ROC curves and calibration plots of the nomogram in training and validation cohorts. (a) ROC curves for discrimination in the training set. (b) Calibration plot of observed and predicted probabilities for the nomogram in the 3-year training set. (c) Calibration plot of observed and predicted probabilities for the nomogram in the 5-year training set. (d) ROC curves for discrimination in the validation set. (e) Calibration plot of observed and predicted probabilities for the nomogram in the 3-year validation set. (f) Calibration plot of observed and predicted probabilities for the nomogram in the 5-year validation set.

Figure 5

(a) Kaplan-Meier survival curves risk score. (b) Kaplan-Meier survival curves of age. (c) Kaplan-Meier survival curves of WBC. (d) Kaplan-Meier survival curves of inv16.