Competing-Risk Nomogram for Predicting Cancer-Specific Survival in Multiple Primary Colorectal Cancer Patients after Surgery

Our research focuses on analyzing cancer specific survival of patients with multiple primary colorectal cancers. We explore the factors influencing cancer specific survival in these patients, and then develop a model to support clinical decision-making.

In our research, we utilize a competing ReX model rather than a Cox model to investigate the factors influencing cancer specific survival. This approach allow us to minimize the impact of competing events and achieve more precise results. In the future, we plan to focus on genomics research in multiple primary colorectal cancers, aiming to uncover their distinct genetic characteristics.

[Instructor] To begin, download the SEERStat 8.4.3 software from the SEER database website. Register and log in to SEERStat 8.4.3 to obtain the relevant patient data. After logging in, click on case listing session and data. Then select the incident SEER research data, 17 registries, November 2022 sub, 2000 to 2020 database. Click on selection, then edit and choose race, sex, year dx, year of diagnosis, 2004 to 2015, and site and morphology, site recode ICD03 and WHO 2008 as colon and rectum. Choose site and morphology diagnostic confirmation as positive histology, and multiple primary fields sequence number is not one primary only. Now click on okay and save the selection. Then click on table, and in the available variables interface, select age recode with single ages and 85 plus, sex, site recode ICD03 and WHO 2008 CS tumor size, grade recode through 2017, derived AJCCT sixth edition, 2004 to 2015, derived AJCCN sixth edition, 2004 to 2015, derived AJCCM sixth edition, 2004 to 2015, radiation recode, chemotherapy recode, yes, no or unknown. SEER cause specific death classification, SEER other cause of death classification, Survival months, patient ID, and RX summary surgery, prim site 1,998 plus. After that, click on column. Then click on output. Name the data and click on execute to output and save the data. After completing the data download, use the patient ID to filter out patients diagnosed with two or more occurrences, indicating multiple primary colorectal cancers. Calculate the interval between tumor diagnoses based on their survival times. Then classify patients into synchronous, multiple primary colorectal cancer and metachronous multiple primary colorectal cancer, using six months as the cutoff. Based on the age at the time of initial diagnosis, categorize patients' ages as 65 years or younger, and over 65 years. Classify tumor grades as grade one for good differentiation, grade two for moderate differentiation, grade three for low differentiation, and grade four for un-differentiation. Then classify tumor location based on the distribution of multiple tumors, such as right colon, left colorectum, or entire colorectum. Determine tumor size by selecting the largest tumor diameter among multiple tumors in the same patient. Randomly divide all 8,931 patients into a training cohort and a validation cohort in a seven to three ratio. Download RStudio and RSoftware. Open RStudio to run RSoftware. Then click on new file and select RScript to create a new RProgramming interface. Enter the relevant code in the code editor and click on run to execute the code. Use the code to perform univariate analysis and plot the CIF curve. After running the code, click on export, then click save as image, and finally click save to save the image. Replace sex in the code with other factors to perform univariate analysis for all factors. Next, use the code to perform best subset regression, multivariate analysis, and visualization. Then save the image as shown earlier. Use the code to plot the nomogram, ROC curve, calibration curve, and DCA curve. Next, use the GGSCIDCA package to plot the DCA curve before saving the image. The ROC curves demonstrated that the model achieved an AUC of 0.762, 0.742, and 0.734 for one year, three year, and five year predictions, respectively, in the training cohort. And 0.801, 0.740, and 0.743 in the verification cohort. The calibration curves indicated a high agreement between the predicted probabilities and the actual outcomes in both the training and verification cohorts. DCA showed that the model provided a good net benefit across various threshold probabilities.