Abstract
BACKGROUND: Advances in breast cancer treatment have prolonged survival, leading to an increased incidence of secondary primary lung cancer (SPLC) in survivors. This study aims to investigate the prognosis and treatment strategies for patients with recurrent early-stage lung cancer histories and establish predictive models to guide clinical practice. METHODS: This study analyzed clinical data from 2,775 patients (2008-2024) extracted from the SEER database and 15 patients (2008-2024) from the cancer registry of the First Affiliated Hospital of Xi'an Jiaotong University. The analysis focused on comparing clinical characteristics, prognosis, and chemotherapy benefits between early-stage second primary lung cancer (SPLC) patients with a history of breast cancer and those with primary lung cancer. The average age of patients in the SEER cohort was 69.64 ± 8.89 years(31-90), while the 15 hospital-registered patients had an average age of 67.15 ± 9.12 years(43-77). We employed neural network-based machine learning methods to develop models for predicting treatment decisions. Specifically, the COX-lung and MLP-lung models were developed, with a LOG-lung model used for comparison. RESULTS: LC patients with a prior breast cancer history had significantly poorer prognosis survival time of 93 months vs 129 months. Postoperative chemotherapy improved the prognosis for some patients; however, the population benefiting from chemotherapy exhibited specific clinical characteristics. The COX-lung and MLP-lung models accurately predicted chemotherapy beneficiaries, with the MLP-lung model achieving an AUC of 0.813 and high positive predictive value. CONCLUSION: SPLC with prior breast cancer do have a poorer prognosis than lung cancer patients, although postoperative chemotherapy can benefit some individuals, careful selection of patients to receive chemotherapy is still warranted. We developed COX-lung and MLP-lung models which can predict beneficiaries of chemotherapy, providing crucial insights for clinicians in formulating personalized treatment plans. The findings indicate that this patient population is heterogeneous, necessitating more individualized treatment strategies.