Modeling therapeutic response to radioiodine in metastatic thyroid cancer: a proof-of-concept study for individualized medicine

Abstract

Purpose: Radioiodine therapy (RAI) has traditionally been used as treatment for metastatic thyroid cancer, based on its ability to concentrate iodine. Propositions to maximize tumor response with minimizing toxicity, must recognize the infinite possibilities of empirical tests. Therefore, an approach of this study was to build a mathematical model describing tumor growth with the kinetics of thyroglobulin (Tg) concentrations over time, following RAI for metastatic thyroid cancer.

Experimental design: Data from 50 patients with metastatic papillary thyroid carcinoma treated within eight French institutions, followed over 3 years after initial RAI treatments, were included in the model. A semi-mechanistic mathematical model that describes the tumor growth under RAI treatment was designed.

Results: Our model was able to separate patients who responded to RAI from those who did not, concordant with the physicians' determination of therapeutic response. The estimated tumor doubling-time (Td was found to be the most informative parameter for the distinction between responders and non-responders. The model was also able to reclassify particular patients in early treatment stages.

Conclusions: The results of the model present classification criteria that could indicate whether patients will respond or not to RAI treatment, and provide the opportunity to perform personalized management plans.

Keywords: mathematical model; metastatic thyroid cancer; personalized medicine; radioactive iodine therapy; therapeutic nuclear medicine.

Figure 1. Observed Stimulated Tg Classification of Patients

Observed stimulated Tg values of patients. This figure represents the group classification visualized by the mixture algorithm by the model, which separates the patients according to the pace of the thyroglobulin evolution curve.

Figure 2. Tumor Doubling Time Group Classifications

The boxplot above separates the groups via the tumor doubling, from non-responder Td average 9.8 months, and responder Td average 66.6 months.

Figure 3. Visual Predicted Check Scheme

A visual predicted check scheme was done that showed all empirical percentiles were within the corresponding 90% pharmacodynamic confidence intervals.

Figure 4. Patient Example Predictions

Model fits for two patients classified as non-responder (P31) and responder (P33). The top 2 subfigures represent the full patient stimulated Tg data and model fits. The bottom two represents the same patients, while using less stimulated Tg data points. The importance of this figure is to demonstrate that while having limited data, the model can still classify patients to whither they will respond or not to treatment.

Figure 5. Schematic Model Diagram

A represents the activity of radioiodine, N, represents the metastatic thyroid cancer cell count and Tg, represents the thyroglobulin concentration. Solid lines depict model flow between compartments. Dashed lines depict interactions between compartments.