Impact of correlated PBPK parameters on intra‐patient global sensitivity analysis in molecular radiotherapy
Elham Yousefzadeh‐Nowshahr, Deni Hardiansyah, Valentina Vasic, Anna‐Lena Theisen, Johannes Tran‐Gia, Philipp E. Hartrampf, Ambros J. Beer, Michael Lassmann, Gerhard GlattingAbstract
Background
Physiologically based pharmacokinetic (PBPK) models integrated into digital twins provide a robust framework for optimizing individualized molecular radiotherapy (MRT). Nevertheless, uncertainties in PBPK‐model‐derived dosimetry can compromise the reliability of dosimetry‐guided treatment planning and individualized therapy decisions.
Purpose
This study employs intra‐patient global sensitivity analysis (GSA) to assess the influence of uncertainty and correlations among PBPK model parameters on estimated kidney and tumor absorbed doses (ADs) in prostate‐specific membrane antigen (PSMA)‐targeted theragnostics.
Methods
Data from 10 patients with metastatic castration‐resistant prostate cancer who underwent pre‐therapeutic [ 68 Ga]Ga‐PSMA‐I&T PET/CT and a cycle of [ 177 Lu]Lu‐PSMA‐I&T therapy, followed by post‐therapeutic SPECT/CT scans at Days 1 and 4 post‐injection, were analyzed. A previously established whole‐body PBPK model for PSMA‐targeted radionuclide therapy was used as the baseline model. Fifteen model parameters, their standard errors, and inter‐parameter correlations were estimated by fitting to measured time‐activity data and subsequently evaluated using intra‐patient GSA based on the Kucherenko method implemented in UQLab. In one approach, correlations were ignored to identify the direct influences of uncertainty in input parameters on ADs. In a second approach, correlations were included to reflect realistic parameter distributions and to capture indirect effects arising from parameter interdependencies. Pharmacokinetics were simulated to estimate therapeutic kidney and tumor ADs as well as main‐ and total‐effect indices (. Sensitivity indices, ADs, and their associated uncertainty metrics were compared between scenarios with and without parameter correlations.
Results
Including parameter correlations in the GSA yielded kidney ADs of 4.2 Gy [3.0–7.2 Gy] (median [min–max]) with standard deviations (SDs) of 0.5 Gy [0.4–0.9 Gy] and tumor ADs of 10.1 Gy [2.0–26.8 Gy] with SDs of 1.7 Gy [0.6–3.4 Gy]. Neglecting correlations resulted in significantly higher uncertainties in ADs, by a factor of 1.2 [1.0–1.8] for both kidneys and tumors. Kidney ADs were most sensitive to kidney receptor density ( ) and to the filtered fraction of renal blood flow . Tumor ADs were most affected by tumor receptor density ( ). The involvement of parameter correlations did not alter the identification of the parameters most influential on AD estimates and further reduced computational runtime.
Conclusions
Incorporating parameter correlations into intra‐patient GSA reduced the estimated uncertainty associated with ADs, underscoring the importance of preserving parameter dependencies when propagating uncertainty in PBPK‐based individualized MRT. The identification of key PBPK parameters governing kidney and tumor ADs remained stable when including correlation, confirming the robustness of the underlying model structure.