PURPOSE/OBJECTIVE(S): Staging for metastatic disease is unreliable; there is usually only one category of Stage IV in the American Joint Commission on Cancer (AJCC) 8th edition, but survival can range from weeks to years. We hypothesize that prognostication can be improved for metastatic disease by using common clinical covariates in a pan-cancer phenotype staging system. MATERIALS/METHODS: Latent class analysis was performed to explore the phenotypes using nationally representative patient data from the National Cancer Database (NCDB; n = 461,357, years 2010-2013, for the training cohort) and the Surveillance, Epidemiology, and End Results database (SEER; n = 106,595, years 2014-2015, for the validation cohort). The optimal number of clusters was determined based on Bayesian information criterion (BIC) and sample-size-adjusted BIC. Kappa coefficients were used to assess external validation. In addition, we used a Cox proportional hazards model with latent classification for survival prediction. Internal validation of the NCDB and the external validation of the training set of SEER were assessed using the concordance (C)-index. RESULTS: Of the 461,357 patients included from the NCDB (median age 65), latent class analysis identified five metastatic phenotypes: (Stage IVA) nearly-exclusive bone-only metastases, commonly seen in lung, breast, and prostate cancer (n = 59,049, 12.8%; median survival = 12.7 months); (IVB) predominant lung metastases, commonly seen in breast, stomach, kidney, ovary, uterus, thyroid, cervix, and soft tissue cancer (n = 62,491, 13.6%; median survival = 11.4 months); (IVC) predominant liver/lung metastases, commonly seen in colon, rectum, pancreas, lung, esophagus, and stomach cancer (n = 130,014, 28.2%; median survival = 7.0 months); (IVD) bone/liver/lung metastases predominant over brain, commonly seen in lung and breast cancer (n = 61,004, 13.2%; median survival = 5.9 months); (IVE) brain/lung metastases predominant over bone/liver, commonly seen in lung cancer and melanoma (n = 148,799, 32.3%; median survival = 5.7 months). These phenotypes had differences in overall survival, and long-term survivors were identified in Stages IVA-B (P < 0.0001). A pan-cancer nomogram model to predict survival (STARS; based on Site, Tumor, Age, Race, Sex) was created and validated. STARS has 13% better prognostication than AJCC: C-index at 1 month of 0.69 (95% CI 0.68-0.69) vs 0.61 (95% CI 0.60-0.61). CONCLUSION: We created and validated a novel pan-cancer staging system for metastatic disease, termed STARS, which is significantly more prognostic than AJCC staging. STARS is innovative because: (1) it takes a global approach, not just focusing on primary tumor, (2) it can be used in all communities, since it relies on readily available, inexpensive predictors; (3) it is simple and allows for rapid validation.
|Original language||English (US)|
|Journal||International journal of radiation oncology, biology, physics|
|State||Published - Nov 1 2021|
ASJC Scopus subject areas
- Radiology Nuclear Medicine and imaging
- Cancer Research