Abstract
BACKGROUND: In here, we constructed personalized models for predicting breast cancer (BC) recurrence according to timing of recurrence (as early and late recurrence). METHODS: An efficient algorithm called group LASSO was used for simultaneous variable selection and risk factor prediction in a logistic regression model. RESULTS: For recurrence < 5 years, age (OR 0.96, 95% CI = 0.95-0.97), number of pregnancies (OR 0.94, 95% CI = 0.89-0.99), family history of other cancers (OR 0.73, 95% CI = 0.60-0.89), hormone therapy (OR 0.76, 95% CI = 0.61-0.96), dissected lymph nodes (OR 0.98, 95% CI = 0.97-0.99), right-sided BC (OR 0.87, 95% CI = 0.77-0.99), diabetes (OR 0.77, 95% CI = 0.60-0.98), history of breast operations (OR 0.38, 95% CI = 0.17-0.88), smoking (OR 5.72, 95% CI = 2.11-15.55), history of breast disease (OR 3.32, 95% CI = 1.92-5.76), in situ component (OR 1.58, 95% CI = 1.35-1.84), tumor necrosis (OR 1.87, 95% CI = 1.57-2.22), sentinel lymph node biopsy (SLNB) (OR 2.90, 95% CI = 2.05-4.11) and SLNB+axillary node dissection (OR 3.50, 95% CI = 2.26-5.42), grade 3 (OR 1.79, 95% CI = 1.46-2.21), stage 2 (OR 2.71, 95% CI = 2.18-3.35), stages 3 and 4 (OR 5.01, 95% CI = 3.52-7.13), and mastectomy+radiotherapy (OR 2.97, 95% CI = 2.39-3.68) were predictors of recurrence < 5 years. Moreover, relative to mastectomy without radiotherapy (as reference for comparison), quadrantectomy without radiotherapy had a noticeably higher odds ratio compared to quadranectomy with radiotherapy for recurrence < 5 years. (OR 17.58, 95% CI = 6.70-46.10 vs. OR: 2.50, 95% CI = 2-3.12). Accuracy, sensitivity, and specificity of the model were 82%, 75.6%, and 74.9%, respectively. For recurrence > 5 years, stage 2 cancer (OR 1.67, 95% CI = 1.31-2.14) and radiotherapy+mastectomy (OR 2.45, 95% CI = 1.81-3.32) were significant predictors; furthermore, relative to mastectomy without radiotherapy (as reference for comparison), quadranectomy without radiotherapy had a noticeably higher odds ratio compared to quadranectomy with radiotherapy for recurrence > 5 years (OR 7.62, 95% CI = 1.52-38.15 vs. OR 1.75, 95% CI = 1.32-2.32). Accuracy, sensitivity, and specificity of the model were 71%, 78.8%, and 55.8%, respectively. CONCLUSION: For the first time, we constructed models for estimating recurrence based on timing of recurrence which are among the most applicable models with excellent accuracy (> 80%).