Predicting Individualized Postoperative Survival for Stage II/III Colon Cancer Using a Mobile Application Derived from the National Cancer Data Base.

BACKGROUND Prediction calculators estimate postoperative survival and assist the decision-making process for adjuvant treatment. The objective of this study was to create a postoperative overall survival (OS) calculator for patients with stage II/III colon cancer. Factors that influence OS, including comorbidity and postoperative variables, were included. STUDY DESIGN The National Cancer Data Base was queried for patients with stage II/III colon cancer, diagnosed between 2004 and 2006, who had surgical resection. Patients were randomly divided to a testing (nt) cohort comprising 80% of the dataset and a validation (nv) cohort comprising 20%. Multivariable Cox proportional hazards regression of nt was performed to identify factors associated with 5-year OS. These were used to build a prediction model. The performance was assessed using the nv cohort and translated into mobile software. RESULTS A total of 129,040 patients had surgery. After exclusion of patients with carcinoma in situ, nonadenocarcinoma histology, more than 1 malignancy, stage I or IV disease, or missing data, 34,176 patients were used in the development of the calculator. Independent predictors of OS included patient-specific characteristics, pathologic factors, and treatment options, including type of surgery and adjuvant therapy. Length of postoperative stay and unplanned readmission rates were also incorporated as surrogates for postoperative complications (1-day increase in postoperative stay, hazard ratio [HR] 1.019, 95% CI 1.018 to 1.021, p < 0.001; unplanned readmission vs no readmission HR 1.35, 95% CI 1.25 to 1.45, p < 0.001). Predicted and actual 5-year OS rates were compared in the nv cohort with 5-year area under the curve of 0.77. CONCLUSIONS An individualized, postoperative OS calculator application was developed for patients with stage II/III colon cancer. This prediction model uses nationwide data, culminating in a highly comprehensive, clinically useful tool.