Machine learning and nomogram novel prognostic models based on LODDS for elderly breast cancer

Abstract Purpose Breast cancer (BC) is a common malignant tumor in women worldwide. This study is based on three machine learning (RF, XgBoost and GBM) and nomogram to explore the survival prognosis of positive lymph node logarithmic ratio (LODDS), lymph node ratio (LNR) in elderly BC. Methods Information on elderly BC from 2010 to 2015 was collected from the SEER database, and clinical and pathological information on 23,893 elderly BC patients were included through screening criteria and randomized 7:3 into a training set and a test set. Univariate, multivariate cox regression and LASSO regression analyses were used to determine the prognostic factors, on the basis of which nomogram and machine learning models were constructed. The predictive efficacy of the models was evaluated by c-index and AUC. Results 14 indicators Age, Marital, Grade, Subtype, Estrogen receptor (ER), Progesterone receptor (PR), Stage, T, N, Radiation, LODDS, bone metastasis, brain metastasis and liver metastasis as independent prognostic factors affecting CSS in older BC. The prognostic optimal cutoff values for age and LODDS were determined based on ROC, respectively 75 and 0.06. Fourteen variables were included in the model to construct Cox, RF, XgBoost and GBM models. The best predictive efficacy of the RF prognostic model was found by calculating C-index and AUC (C-index=0.811, AUC=0.881). Conclusion LODDS staging has a better survival prognosis in older BC. Three machine learning and nomogram models are constructed based on the SEER database of elderly BC patients, which can intuitively predict the survival probability of elderly BC patients.