Nonlinear mixed effects models were developed to describe the relationship between cabozantinib exposure and target lesion tumor size in a phase III study of patients with progressive metastatic medullary thyroid cancer. These models used cabozantinib exposure estimates from a previously published population pharmacokinetic model for cabozantinib in cancer patients that was updated with data from healthy-volunteer studies. Semi-mechanistic models predict well for tumors with static, increasing, or decreasing growth over time, but they were not considered adequate for predicting tumor sizes in medullary thyroid cancer patients, among whom an early reduction in tumor size was followed by a late stabilization phase in those receiving cabozantinib. A semi-empirical tumor model adequately predicted tumor profiles that were assumed to have a net growth rate constant that was piecewise continuous in the regions of 0-110 and 110-280 days. Emax models relating average concentration to average change in tumor size predicted that an average concentration of 79 and 58 ng/ml, respectively, would yield 50% of the maximum possible tumor reduction during the first 110 days of dosing and during the subsequent 110-280 days of dosing. Simulations of tumor responses showed that daily doses of 60 mg or greater are expected to provide a similar tumor reduction. Both model evaluation of observed data and simulation results suggested that the two protocol-defined cabozantinib dose reductions from 140 to 100 mg/day and from 100 to 60 mg/day are not projected to result in a marked reduction in target lesion regrowth.