PCN283 A MICRO-SIMULATION MODEL FOR THE NATURAL HISTORY PROGRESSION OF BENIGN AND MALIGNANT THYROID NODULES: COMPARISON OF CALIBRATION METHODS

Abstract

Thyroid nodules and cancer are leading causes of morbidity and health care utilization. Substantial over-diagnosis and concern for over-treatment of indolent subtypes have led to less aggressive management strategies. There is a critical need for risk-stratified approaches in order to identify and treat patients with aggressive subtypes of thyroid cancer while minimizing harm. We utilize a micro-simulation model that simulates the natural progression of thyroid nodule growth, including benign nodules and four subtypes of thyroid cancer (papillary, follicular, medullary, and anaplastic). We compare the following calibrations methods for the 34 unknown parameters of the model: Latin Hypercubic Sampling (LHS), Simulated Annealing (SA), random sets, and greedy algorithm. The calibration targets are thyroid cancer incidence by age, size of nodule, and stage of disease for each cancer sub-type, derived from SEER database. 95% confidence intervals (Cis) are computed from SEER dataset from 2010 to 2016, and the outputs of simulation model are compared to the targets’ intervals accordingly. Targets for benign nodule prevalence were based on population and autopsy studies. We use Mean Square Error (MSE) to calculate the goodness of fit for each of the calibration methods. If the model output is within the target confidence interval, the error is considered zero. Otherwise, the MSE is calculated as squared distance from the output to the closest target bound. The average(95% CI) MSE after 1000 simulation runs with 5 million population for SA, greedy, random, and LHS algorithms is 3.69E+05 (±1.17E+04), 7.43E+11 (±5.22E+10), 2.30E+12 (±4.97E+11), 2.35E+12 (±3.22E+11), respectively. The best MSE is achieved from SA (MSE=1.42E+05), followed by LHS (2.17E+07), random(7.07E+07), and greedy algorithm(1.27E+11). The best performance of calibration process is achieved by using SA. The validated model provides a dynamic platform to assess various clinical management strategies to improve patient outcomes and optimize resource utilization for patients with thyroid nodules.