Lu Turković 1, Iztok Grabnar 2, Matea Baković 3, Tajana Silovski 4,5, Maja Ortner Hadžiabdić 1, Andrijana Ščavničar 4, Miranda Sertić 1
1 University of Zagreb Faculty of Pharmacy and Biochemistry (Zagreb, Croatia), 2 University of Ljubljana Faculty of Pharmacy (Ljubljana, Slovenia), 3 PrimeVigilance (Zagreb, Croatia), 4 University Hospital Centre Zagreb (Zagreb, Croatia), 5 University of Zagreb School of Medicine (Zagreb, Croatia)
Objectives: Ribociclib, a cyclin-dependent kinase 4/6 inhibitor, is an established therapy for hormone receptor-positive, human epidermal growth factor receptor 2-negative breast cancer. It is typically administered at doses of up to 600 mg once daily, in 28-day cycles (21 days on treatment followed by 7 days off). Dose reductions are commonly implemented to mitigate dose-dependent adverse effects, including neutropenia, nausea, and fatigue. As adherence to oral anticancer medications is essential to achieve optimal clinical outcomes, robust methods for monitoring adherence are required. Our group has previously evaluated adherence to ribociclib using indirect assessment methods employing validated questionnaires [1]. Direct measurement of plasma drug concentrations, when integrated with population pharmacokinetic (popPK) simulations, may provide a more objective and clinically informative assessment of adherence [2]. This study aimed to evaluate published popPK models of ribociclib, focusing on covariate effects and between-subject variability, to propose plasma concentration cut-off values indicative of non-adherence. The applicability of this approach was examined in a small pilot cohort.
Methods: A previously published popPK model by Lu et al. [3] was used for the analysis. This model is a two-compartment structure with first-order absorption and absorption lag time, developed from pooled data encompassing phase I-III studies (1059 patients, 7960 concentration measurements).
Steady-state plasma concentrations were simulated at 1-hour intervals over a 24-hour period using NONMEM (v. 7.5.0, ICON) and Perl-speaks-NONMEM v. 5.5.0 (Uppsala Pharmacometrics). Simulations (2000 virtual patients per scenario) were performed for six covariate combinations (dose: 200 or 600 mg; body weight: 50, 70, or 120 kg) and eight adherence scenarios reflecting different patterns of missed last three doses. Statistical analyses of model‑predicted concentrations were carried out using R (v. 4.3.3) and RStudio (v. 2025.09.2, Posit).
Receiver operating characteristic (ROC) curves were generated for each timepoint, covariate combination, and adherence scenario. Youden indices and areas under the ROC curve (AUROC) were calculated, and the maximum Youden index for each configuration was used to determine optimal concentration thresholds for discriminating full adherence from non-adherence with the highest achievable sensitivity and specificity.
A total of 94 steady-state plasma samples from 22 patients were collected at various timepoints after dose administration. Ribociclib concentrations were quantified using a validated liquid chromatography-tandem mass spectrometry assay (linear range 250–5000 ng/mL; relative standard deviation <6%, bias within ±5%).
Adherence in real patients was evaluated by comparing measured concentrations to model‑derived threshold values specific to each timepoint and dose/body‑weight combination, using a posterior probability cut‑off of 50%.
Results: Two non-adherence subgroups were identified based on whether the last dose was taken: Group 1 included scenarios in which the last dose was administered, and Group 0 included scenarios in which the last dose was omitted. The highest Youden indices reached 0.5330 in Group 1 and 0.9375 in Group 0.
When examining overall discriminative performance using AUROC values, Group 1 showed its maximum performance at time = 0 h (AUROC 0.6290–0.8498). In Group 0, the maximum AUROC was at 3–4 h after dosing (around Tmax), ranging from 0.9737–0.9966. These findings indicate that these timepoints provide the greatest ability to differentiate between full adherence and non‑adherence. Furthermore, classification accuracy improved when the omitted dose was closer to the sampling time, consistent with the expected pharmacokinetic behaviour of ribociclib. Among real patient samples, 9 measurements from 4 patients were substantially below the predicted threshold concentrations, corresponding to a posterior probability of non‑adherence of up to 60.7%.
Conclusions: Adherence assessment using a popPK approach was successfully implemented. When combined with traditional adherence monitoring tools, this method offers valuable insight into patient‑specific adherence patterns and can support clinical decision‑making. Ensuring adequate plasma concentrations not only promotes sustained anticancer efficacy but may also reduce the risk of toxicity through more informed individual dose adjustments.
This work was fully supported by the Croatian Science Foundation through projects IP-2025-02-1885 and MOBDOK-2023-3041, as well as by the European Regional Development Fund (project Farminova, KK.01.1.1.02.0021).
References:
[1] Baković M et al. Acta Pharmaceutica (2023) 73, 633–654.
[2] Ding J et al. CPT:PSP (2024) 13, 795–811.
[3] Lu Y et al. J Clin Pharmacol (2021) 61, 1054–1068.
Reference: PAGE 34 (2026) Abstr 12242 [www.page-meeting.org/?abstract=12242]
Poster: Clinical Applications