Prediction of Conditional Probability of Survival after Surgery for Gastric Cancer: A Study Based on Eastern and Western Large Data Sets
Qing Zhong,Qi-Yue Chen,Ping Li,Jian-Wei Xie,Jia-Bin Wang,Jian-Xian Lin,Jun Lu,Long-Long Cao,Mi Lin,Ru-Hong Tu,Chao-Hui Zheng,Chang-Ming Huang
DOI: https://doi.org/10.1016/j.surg.2018.02.011
IF: 4.348
2018-01-01
Surgery
Abstract:BACKGROUND:The dynamic prognosis of patients who have undergone curative surgery for gastric cancer has yet to be reported. Our objective was to devise an accurate tool for predicting the conditional probability of survival for these patients.METHODS:We analyzed 11,551 gastric cancer patients from the Surveillance, Epidemiology, and End Results database. Two-thirds of the patients were selected randomly for the development set and one-third for the validation set. Two nomograms were constructed to predict the conditional probability of overall survival and the conditional probability of disease-specific survival, using conditional survival methods. We then applied these nomograms to the 4,001 patients in the database from Fujian Medical University Union Hospital, Fuzhou, China, one of the most active Chinese institutes.RESULTS:The 5-year conditional probability of overall survival of the patients was 41.6% immediately after resection and increased to 52.8%, 68.2%, and 80.4% at 1, 2, and 3 years after gastrectomy. The 5-year conditional probability of disease-specific survival "increased" from 48.9% at the time of gastrectomy to 59.8%, 74.7%, and 85.5% for patients surviving 1, 2, and 3 years, respectively. Sex; race; age; depth of tumor invasion; lymph node metastasis; and tumor size, site, and grade were associated with overall survival and disease-specific survival (P <.05). Within the Surveillance, Epidemiology, and End Results validation set, the accuracy of the conditional probability of overall survival nomogram was 0.77, 0.81, 0.82, and 0.82 at 1, 3, 5, and 10 years after gastrectomy, respectively. Within the other validation set from the Fujian Medical University Union Hospital (n = 4,001), the accuracy of the conditional probability of overall survival nomogram was 0.76, 0.79, 0.77, and 0.77 at 1, 3, 5, and 10 years, respectively. The accuracy of the conditional probability of disease-specific survival model was also favorable. The calibration curve demonstrated good agreement between the predicted and observed survival rates.CONCLUSION:Based on the large Eastern and Western data sets, we developed and validated the first conditional nomogram for prediction of conditional probability of survival for patients with gastric cancer to allow consideration of the duration of survivorship.