Conditional survival nomogram predicting real-time prognosis of locally advanced breast cancer: Analysis of population-based cohort with external validation.

BACKGROUND

Locally advanced breast cancer (LABC) is generally considered to have a relatively poor prognosis. However, with years of follow-up, what is its real-time survival and how to dynamically estimate an individualized prognosis? This study aimed to determine the conditional survival (CS) of LABC and develop a CS-nomogram to estimate overall survival (OS) in real-time.

METHODS

LABC patients were recruited from the Surveillance, Epidemiology, and End Results (SEER) database (training and validation groups, n = 32,493) and our institution (testing group, n = 119). The Kaplan-Meier method estimated OS and calculated the CS at year (x+y) after giving x years of survival according to the formula CS(y|x) = OS(y+x)/OS(x). y represented the number of years of continued survival under the condition that the patient was determined to have survived for x years. Cox regression, best subset regression, and the least absolute shrinkage and selection operator (LASSO) regression were used to screen predictors, respectively, to determine the best model to develop the CS-nomogram and its network version. Risk stratification was constructed based on this model.

RESULTS

CS analysis revealed a dynamic improvement in survival occurred with increasing follow-up time (7 year survival was adjusted from 63.0% at the time of initial diagnosis to 66.4, 72.0, 77.7, 83.5, 89.0, and 94.7% year by year [after surviving for 1-6 years, respectively]). In addition, this improvement was non-linear, with a relatively slow increase in the second year after diagnosis. The predictors identified were age, T and N status, grade, estrogen receptor (ER), progesterone receptor (PR), human epidermal growth factor receptor 2 (HER 2), surgery, radiotherapy and chemotherapy. A CS-nomogram developed by these predictors and the CS formula was used to predict OS in real-time. The model's concordance indexes (C-indexes) in the training, validation and testing groups were 0.761, 0.768 and 0.810, which were well-calibrated according to the reality. In addition, the web version was easy to use and risk stratification facilitated the identification of high-risk patients.

CONCLUSIONS

The real-time prognosis of LABC improves dynamically and non-linearly over time, and the novel CS-nomogram can provide real-time and personalized prognostic information with satisfactory clinical utility.