Emilie Pilote 1,2, Nancy L. Sheehan 1,2, Marie-Élaine Métras 1,3, Jason Brophy 4, Amélie Marsot 1
1 Faculty Of Pharmacy, University Of Montreal (Montreal, Canada), 2 McGill University Health Center (MUHC) (Montreal, Canada), 3 CHU Sainte-Justine (Montreal, Canada), 4 Children's Hospital of Eastern Ontario (CHEO) (Ottawa, Canada)
Introduction: Dolutegravir (DTG) is a widely used integrase inhibitor for HIV treatment [1]. Despite its favorable pharmacokinetic profile, interindividual variability may occur and can be influenced by patient-specific characteristics. This can complicate therapeutic concentration attainment, which is critical to achieve viral suppression. In pediatrics specifically, we previously reported that 6.02% and 53.88% of DTG concentrations in Canadian pediatric patients undergoing therapeutic drug monitoring (TDM) were sub-therapeutic and supratherapeutic, respectively [2]. Considering this, new strategies are needed for TDM to ensure adequate DTG exposure in pediatric patients.
Population pharmacokinetic (PopPK) models can help optimizing DTG exposure. However, prior to clinical implementation, their predictive performance must first be evaluated. An external evaluation of published pediatric DTG PopPK models was therefore performed. Of the three identified models [3-5], none showed adequate predictive performance, with bias exceeding ±20% and imprecision greater than 30%.
Objective: We aim to perform a re-estimation with or without adjustments of these models in attempt to improve their performance in our Canadian pediatric population.
Methods: Study population: From January 1st, 2011, to March 31st, 2025, DTG pediatric TDM data were collected retrospectively from the Québec Antiretroviral Therapeutic Drug Monitoring Program. Patients less than 18 years old with at least one DTG TDM were included but were excluded if all their samples’ time post-doses were unknown and/or their TDM indication was only to define which antiretrovirals they were taking. Model selection: We selected the two models with better predictive performance [3, 5] from the external evaluation to be re-estimated. Model re-estimation: All parameters were initially re-estimated. Based on the models’ objective function value, and the residual standard error (RSE) and shrinkage of re-estimated parameters, some fixed-effect parameters were fixed to their original value, interindividual variabilities (IIVs) were added or removed, and different residual error models were tested. Model adjustment: Starting with the models’ re-estimation closest to the original models, the inclusion of new covariates was tested (i.e., ethnicity, age and co-medications with known DDIs with DTG). Covariates’ relationships with apparent clearance (CL/F) and apparent volume of distribution (Vd/F) were first explored graphically. Plausible covariates were tested with a univariate analysis, followed by a stepwise forward addition (p-value<0.05)/backward elimination (p-value<0.01) approach. If the inclusion of covariates resulted in a poor re-estimation of the other parameters, the covariates in question were removed from the analysis. Model evaluation: The predictive performance of the re-estimated models was evaluated with goodness-of-fit plots, the median predictive error (MDPE, bias) and the median absolute predictive error (MDAPE, imprecision). R (v4.3.2, https://www.r-project.org/) and NONMEM (v7.6, Icon Development) software were used for all analysis. Results: A total of 65 patients (53.8% females) and 384 samples were included. Median age and weight at baseline were 11 years (range 0-17 years) and 37.6 kg (interquartile range 23.6-50.8 kg), respectively. Following the models’ re-estimation alone, almost all parameters were re-estimated precisely (RSEs<40%), except for some in Chandasana et al.’s [3] model (i.e., Vd/F, the allometric exponent of weight on Vd/F, and the IIV on Vd/F). The a priori predictive performance of the resulting models was improved with a bias and imprecision of -1.49% and 33.44%, respectively, for Chandasana et al.’s [3] model, and of -5.90% and 32.28%, respectively, for Naidoo et al.’s [5] model. As for their a posteriori predictive performance, it was adequate with a bias of -8.17% and -4.21%, and an imprecision of 24.81% and 25.57%, respectively. Concerning model adjustments, no additional covariates could be included in the models without impacting the estimation of other model parameters or without being poorly estimated. This could be the result of the small sample size of some covariates, overparameterization of the models, or our dataset (i.e., TDM data with only one sample per dose). Conclusions: Although model adjustments were not possible, re-estimation of model parameters alone greatly improved predictive performance. Overall, their predictive performance is now adequate, except for their a priori imprecision which remained slightly above the acceptability threshold (MDAPE≤30%). Re-estimated models may better support initial dose selection compared with the original models. Both approaches appear suitable for subsequent dose adjustment. Future work should focus on directly comparing dose simulations derived from the original and re-estimated models to determine their relative impact on initial dose selection and exposure attainment in clinical practice. References: [1] World Health Organization (WHO). WHO recommends dolutegravir as preferred HIV treatment option in all populations [News release]. World Health Organization (WHO); 2019 [cited 07-01-2025]. Available from: https://www.who.int/news/item/22-07-2019-who-recommends-dolutegravir-as-preferred-hiv-treatment-option-in-all-populations. [2] Pilote E, Lim C, El Khiraoui A, Brophy J, Metras M, Karatzios C, et al. Therapeutic drug monitoring of dolutegravir in children and adolescents living with HIV: A retrospective study. (Oral presentation). 26th International Workshop on Clinical Pharmacology of HIV, Hepatitis, and Other Antiviral Drugs 2025. Amsterdam, Netherlands; September 3-4, 2025. [3] Chandasana H, Thapar M, Hayes S, Baker M, Gibb DM, Turkova A, et al. Population pharmacokinetic modeling of dolutegravir to optimize pediatric dosing in HIV-1-infected infants, children, and adolescents. Clin Pharmacokinet. 2023 Oct;62(10):1445-59. https://doi.org/10.1007/s40262-023-01289-5. [4] Waalewijn H, Wasmann RE, Bamford A, Gibb DM, McIlleron HM, Colbers A, et al. Population pharmacokinetics of dolutegravir in African children: results from the CHAPAS-4 trial. J Pediatric Infect Dis Soc. 2024 Sep 26;13(9):496-500. https://doi.org/10.1093/jpids/piae076. [5] Naidoo A, Waalewijn H, Naidoo K, Letsoalo M, Cromhout G, Sewnarain L, et al. Pharmacokinetics and safety of dolutegravir in children receiving rifampicin tuberculosis treatment in South Africa (ORCHID): a prospective cohort study. Lancet HIV. 2025 Apr;12(4):e273-e82. https://doi.org/10.1016/S2352-3018(24)00312-6.
Reference: PAGE 34 (2026) Abstr 12209 [www.page-meeting.org/?abstract=12209]
Poster: Methodology - Model Evaluation