Deep learning-based predictive models for assessing the impact of clinical factors and second primary malignancy on survival in patients with colorectal cancer.

In this study, we first utilized a data set of 21,522 patients with colorectal cancer (CRC) with a second primary malignancy (SPM) to develop a deep learning model for predicting 1-year, 3-year, and 5-year survival outcomes in patients with CRC who subsequently developed SPM. Our models demonstrated high performance, achieving area under the curve (AUC) values of 0.850 (95% confidence interval [CI] 0.840-0.861), 0.856 (95% CI 0.850-0.861), and 0.848 (95% CI 0.843-0.853) for the 1-year, 3-year, and 5-year survival predictions, respectively. We then examined the impact of 33 clinical factors on these predictions and found that age, radiation therapy for the SPM, and sex were the most influential factors. Age and metastatic status of the SPM emerged as the most critical predictors. Finally, using one-hot encoding, we evaluated the effects of various SPMs on survival outcomes in patients with CRC and provided clinical interpretations of these findings. Our analysis indicated that patients with second primary prostate cancer and CRC generally have better survival prospects than that of those with other SPMs. Patients with second primary pancreatic and gastric cancers have poor survival outcomes. These findings provide valuable insights into the intricate interactions among CRC, SPM, and various clinical factors, thereby improving the treatment and evaluation of patients with CRC and SPM.