131I therapy for benign thyroid disease: flexible single-time-point dosimetry using population-based model selection with non-linear mixed-effects modelling

Abstract

Purpose: This study aimed to evaluate the accuracy of time-integrated activity (TIA) estimation using single-time-point (STP) data combined with nonlinear mixed-effects modelling (NLMEM) and population-based model selection (PBMS) in 73 patients with benign thyroid disease. Methods: Biokinetic data of 131I were collected from 73 patients with benign thyroid conditions, including Graves’ disease, toxic nodular goitre, and non-toxic goitre. Uptake measurements were taken at 2, 6, 24, 48 (73 patients), 96 h (53 patients) or 120 h (20 patients) after administration. The best sum-of-exponentials function (SOEF) with four adjustable parameters, identified in our recent study (Hardiansyah et al. EJNMMI Phys, 2025) by PBMS NLMEM, and the SOEF EANM standard operating procedure (SOP) with three adjustable parameters were then employed to conduct a NLMEM-based STP dosimetry at different time points, i.e. s1TIA and s2TIAs, respectively. In addition, STP dosimetry was performed using the EANM SOP approach (Hänscheid et al. EJNMMI, 2013) to calculate TIAs (hTIAs). The accuracy of the computed s1TIAs, s2TIAs and hTIAs was evaluated by calculating the relative deviations (RDs), mean absolute percentage errors (MAPEs), root-mean-square errors (RMSEs), and percentage of absolute RDs of sTIAs and hTIAs exceeding 5% (%RD5) and 10% (%RD10) with the reference TIAs obtained from the four parameters NLMEM fit to all time points data. Results: Of the time points included, 120 h after administration was found as the optimal time point for STP dosimetry based on the mean ± SD of RD (RMSE, MAPE, %RD5, %RD10) of s1TIA, s2TIAs, and hTIA of 2% ± 4% (4%,3%,25%,0%), 13% ± 6% (14%,13%,90%,60%) and −3% ± 5% (6%,5%,55%,0%), respectively. Conclusions: While s1TIA typically produced better 131I TIA estimates for benign thyroid disease than the hTIA recommended in the EANM SOP, the STP calculation from NLMEM using SOEF EANM SOP s2TIAs was inferior to hTIA. With the best SOEF model, NLMEM provides satisfactory and reasonable STP estimates.