Tree-based model for thyroid cancer prognostication

Abstract

Background: Death is uncommon in thyroid cancer patients, and the factors important in predicting survival remain inadequately studied. The objective of this study was to assess prognostic effects of patient, tumor, and treatment factors and to determine prognostic groups for thyroid cancer survival.

Methods: Using data from the Surveillance, Epidemiology, and End Results Program (SEER), we evaluated overall and disease-specific survival (DSS) in 43 392 well-differentiated thyroid cancer patients diagnosed from 1998 through 2005. Multivariable analyses were performed using Cox proportional hazards regression, survival trees, and random survival forest. Similar analyses were performed using National Cancer Data Base data, with overall survival (OS) evaluated in 131 484 thyroid cancer patients diagnosed from 1998 through 2005. Relative importance of factors important to survival was assessed based on the random survival forest analyses.

Results: Using survival tree analyses, we identified 4 distinct prognostic groups based on DSS (P < .0001). The 5-year DSS of these prognostic groups was 100%, 98%, 91%, 64%, whereas the 10-year survival was 100%, 96%, 85%, and 50%. Based on random survival forest analyses, the most important factors for DSS were SEER stage and age at diagnosis. For OS, important prognostic factors were similar, except age at diagnosis demonstrated marked importance relative to SEER stage. Similar results for OS were found using National Cancer Data Base data.

Conclusion: This study identifies distinct prognostic groups for thyroid cancer and illustrates the importance of patient age to both disease-specific and OS. These findings have implications for patient education and thyroid cancer treatment.

Figure 1.

Survival tree for DSS based on the SEER cohort.

Figure 2.

Survival tree for OS based on the SEER cohort.

Figure 3.

A, Kaplan-Meier curves for the 4 distinct prognostic groups based on DSS. Group I (red) contains patients ≤63 years old with localized disease or patients ≤48 years old with regional disease. Group II (blue) contains patients 49 to 63 years old with regional disease or ≥64 years old with localized disease. Group III (green) includes patients ≥64 years old with regional disease or ≤51 years old with distant disease. Group IV (gold) contains patients ≥52 years old with distant disease. B, Kaplan-Meier curves for the 4 distinct prognostic groups based on OS. Group I (red) consists of female patients ≤49 years old. Group II (blue) includes male patients ≤49 years old, patients 50 to 61 years old with localized/regional disease, or female patients 62 to 70 years old with localized/regional disease. Group III (green) includes female patients 71 to 75 years old with localized/regional disease and male patients 62 to 75 years old with localized/regional disease. Group IV (gold) includes patients 50 to 61 years old with distant disease, patients 62 to 75 years old with distant disease, patients ≥76 years old with localized/regional disease, or patients age ≥76 years old with distant disease.

Figure 4.

A, Relative importance of each of the patient, tumor, and treatment variables to DSS based on the random survival forest analysis of the SEER cohort. B, Relative importance of each of the patient, tumor, and treatment variables to OS based on the random survival forest analysis of the SEER cohort. C, Relative importance of each of the patient, tumor, and treatment variables to OS based on the random survival forest analysis of the NCDB cohort.