Context: Thyroid cancers represent a conglomerate of diverse histological types with equally variable prognosis. There is no reliable prognostic model to predict the risks of relapse and death for different types of thyroid cancers.
Objective: The purpose of this study was to build prognostic nomograms to predict individualized risks of relapse and death of thyroid cancer within 10 years of diagnosis based on patients' prognostic factors.
Design: Competing risk subhazard models were used to develop prognostic nomograms based on the information on individual patients in a population-based thyroid cancer cohort followed up for a median period of 126 months. Analyses were conducted using R version 2.13.2. The R packages cmprsk10, Design, and QHScrnomo were used for modeling, developing, and validating the nomograms for prediction of patients' individualized risks of relapse and death of thyroid cancer.
Setting: This study was performed at CancerCare Manitoba, the sole comprehensive cancer center for a population of 1.2 million.
Patients: Participants were a population-based cohort of 2306 consecutive thyroid cancers observed in 2296 patients in the province of Manitoba, Canada, during 1970 to 2010.
Main outcome measures: Outcomes were discrimination (concordance index) and calibration curves of nomograms.
Results: Our cohort of 570 men and 1726 women included 2155 (93.4%) differentiated thyroid cancers. On multivariable analysis, patient's age, sex, tumor histology, T, N, and M stages, and clinically or radiologically detectable posttreatment gross residual disease were independent determinants of risk of relapse and/or death. The individualized 10-year risks of relapse and death of thyroid cancer in the nomogram were predicted by the total of the weighted scores of these determinants. The concordance indices for prediction of thyroid cancer-related deaths and relapses were 0.92 and 0.76, respectively. The calibration curves were very close to the diagonals.
Conclusions: We have successfully developed prognostic nomograms for thyroid cancer with excellent discrimination (concordance indices) and calibration.