Wei Zhang1, Zhigang Wang1, Thierry Dervieux2, Debby Thomas1, Marc Ferrante3,4, Séverine Vermeire3,4, Erwin Dreesen1
1Department of Pharmaceutical and Pharmacological Sciences, KU Leuven, 2Prometheus Laboratories, 3Department of Gastroenterology and Hepatology, UZ Leuven, 4Department of Chronic Diseases and Metabolism, KU Leuven
Introduction: Therapeutic drug monitoring (TDM)-guided model-informed precision dosing (MIPD) has been used to individualize dosing regimens of biologics to treat patients with inflammatory bowel diseases [1]. TDM-guided MIPD typically relies on a population pharmacokinetic (popPK) model that describes drug disposition and incorporates patient-specific factors (e.g., covariates such as body weight, serum albumin, and drug concentrations). Using these inputs, MIPD applies Bayesian forecasting to predict future drug concentrations. Notably, the concentration of biologics can be measured by different TDM assays. Ustekinumab concentrations measured by a homogeneous mobility shift assay (HMSA) are significantly higher than those measured by an enzyme-linked immunosorbent assay (ELISA) [2]. However, the impact of assay format on the predictive performance of MIPD remains unknown. Objective: To evaluate the impact of assay variation on the predictive performance of MIPD. Methods: Data from 83 patients with Crohn’s disease were repurposed for this study [3]. Patients received intravenous ustekinumab for induction therapy, followed by subcutaneous maintenance injections every eight weeks. Serum samples were collected at weeks (w)4, 8, 16, and 24, with drug concentrations measured using both HMSA and ELISA. These measurements generated two datasets: Data_HMSA and Data_ELISA. Two corresponding popPK models were previously developed: Model_HMSA (unpublished) and Model_ELISA [3]. Both models employed a two-compartment structure with body weight and serum albumin as covariates. Body weight was incorporated through allometric scaling on clearance, central and peripheral volumes of distribution, and inter-compartment clearance. Lower serum albumin levels were associated with increased clearance. Model_HMSA included interindividual variability (IIV) in clearance (29.0% coefficient of variation, CV) and peripheral volume of distribution (56.6% CV), while Model_ELISA attributed IIV to clearance (31.2% CV) and bioavailability (37.3% CV). Proportional residual error estimates were 28.8% and 13.6% for Model_HMSA and Model_ELISA, respectively, with additive errors of 0.291 mg/L and 0.272 mg/L. To investigate the impact of assay variation, Bayesian forecasting was conducted under four scenarios: (1) w8 trough concentration (TC) predicting w16 TC, (2) w8 TC predicting w24 TC, (3) w16 TC predicting w24 TC, and (4) w8 and w16 TCs predicting w24 TC. Each scenario was evaluated under two conditions: (1) model and dataset matched (Model_HMSA with Data_HMSA, Model_ELISA with Data_ELISA) and (2) model and dataset mismatched (Model_HMSA with Data_ELISA, Model_ELISA with Data_HMSA). Predictive performance was assessed using relative bias (rBias) and relative root mean square error (rRMSE). Predicted concentrations were considered clinically acceptable if rBias was within ±20% with a 95% confidence interval (CI) including zero and if rRMSE remained below 30%. Regression and Bland–Altman plots were used to assess the correlation and agreement between forecasted ustekinumab clearance estimates. Results: When datasets and models were matched, rBias was clinically acceptable across all scenarios. However, mismatched combinations yielded poorer accuracy. Model_HMSA overestimated Data_ELISA concentrations, with rBias 8.54% [95% CI 0.149%–16.9%], 18.5% [6.81%–30.1%], 27.4% [16.9%–37.8%], and 19.7% [8.88%–30.6%] for scenarios 1, 2, 3, and 4, respectively. Conversely, Model_ELISA underestimated Data_HMSA in scenarios 3 and 4, with rBias -11.2% [-17.2%– -5.13%] and -10.8% [-17.2%– -4.45%], but was clinically acceptable in scenarios 1 and 2. Precision was highest using Data_HMSA. All rRMSE values met clinical thresholds with Model_HMSA, while three scenarios (1, 2, and 4) met thresholds with Model_ELISA. Data_ELISA failed to achieve precision under any scenario, regardless of the model used. Clearance estimates were systematically lower using Model_HMSA than Model_ELISA, with high determination coefficients (R² range 0.94–1.00). Bland–Altman plots indicated poor agreement between clearance estimates from the two models. Conclusion: We assessed the impact of the assay format on the performance of MIPD. Our findings highlight the importance of matching the TDM assay format with the assay format used to build the popPK model.
[1] Bourgonje et al. Trends Pharmacol Sci. 2025;46(1):9-19. [2] Verdon et al. J Can Assoc Gastroenterol. 2020;4(2):73-77. [3] Wang et al. Br J Clin Pharmacol. 2022;88(1):323-335.
Reference: PAGE 33 (2025) Abstr 11768 [www.page-meeting.org/?abstract=11768]
Poster: Drug/Disease Modelling - Other Topics