Predictive modelling of overall survival in adult patients with primary diffuse large B-cell lymphoma of the breast using the Surveillance, Epidemiology, and End Results (SEER) database.

Abstract

We aimed to identify critical clinical features to develop an accurate web-based prediction model for estimating the overall survival (OS) of primary breast diffuse large B-cell lymphoma (PB-DLBCL) adult patients. We first included all PB-DLBCL cases with available covariates retrieved from the Surveillance, Epidemiology, and End Results database. We sequentially performed univariate and multivariate Cox regression approaches to identify the predictors independently associated with prognosis, and all the predictors that passed these tests were then constructed to build a nomogram for predicting 3-, 5-, and 10-year survival rates of patients. The C-index and the receiver operating characteristic curve (ROC) were used to evaluate the prediction discrimination, and the calibration curve was applied to estimate the calibration. A total of PB-DLBCL adult patients were included (median age was 69 with the interquartile range [IQR] of 57-79 years), of which 466 (70%) were randomly allocated to the development cohort, and the remaining cases were collected for validation. Using three identified independent predictors (i.e., age, stage, and radiation), an accurate nomogram for predicting OS was developed and validated. The C-indices of our nomogram were both relatively acceptable, with 0.74 (95% CI: 0.71-0.78) and 0.72 (95% CI: 0.70-0.75) for the development and validation cohorts, respectively. The calibration curves also accurately predicted the prognosis of PB-DLBCL in all cases. In addition, ROC curves showed our nomogram to possess superior predictive ability compared to any single variable. To visually present this prediction model, a convenient web-based tool was implemented based on our prognostic nomogram. For patients with PB-DLBCL, a more convenient and accurate web-based prediction model was developed and validated, which showed relatively good performances in both discrimination and calibration during model development and validation. External evaluation and validation are warranted by further independent studies.