Pharmacometric Modeling of Drug Adverse Effects: An Application of Mixture Models in Schizophrenia Spectrum Disorder Patients Treated with Clozapine
Orwa Albitar (1), Siti Maisharah Sheikh Ghadzi (1)*, Sabariah Noor Harun (1), Siti Nor Aizah Ahmad (3), Maria C. Kjellsson (2)
(1) School of Pharmaceutical Sciences, Universiti Sains Malaysia, 11800 Pulau Pinang, Malaysia. (2) Pharmacometrics Research Group, Department of Pharmacy, Uppsala University, Uppsala, Sweden. (3) Psychiatric Department, Hospital Pulau Pinang, Ministry of Health Malaysia, Jalan Residensi, Georgetown, 10460, Penang, Malaysia
Objectives: Clozapine is the drug of choice for treatment-resistant schizophrenia [1]. However, its use is limited to only 5% of eligible patients with a significant delay in initiation, treatment discontinuation, and poor compliance mainly due to fear of its adverse drug reactions (ADRs) such as neutropenia, weight gain, and tachycardia [2]. The current study aimed to extend the applications of disease progression and mixture models to describe the Dose relatedness, Timing, and patient Susceptibility (DoTS) classification of ADRs [3] with a real data example of clozapine-induced absolute neutrophils count (ANC) decrease, weight gain, and tachycardia in schizophrenia spectrum disorder (SSD) patients.
Methods:
Data: Data were extracted from medical records of SSD patients in the Psychiatric Clinic, Penang General Hospital, Malaysia, and retrospectively analyzed.
Base models: ADRs change the normal baseline physiology as a function of time and/or dose. Additionally, the mixture model describes the individual susceptibility to the ADRs. Different time courses of ADRs were explored, such as the simple offset, linear, piecewise linear, exponential, and transient (inverse Bateman [4] or surge [5] functions). Total daily dose, dose frequency, and other covariates were explored.
Combined model: The individual ADR models were combined to evaluate the probability of individuals belonging to eight subpopulations representing all possible combinations of ADRs. Removal was then carried out in a stepwise manner to exclude subpopulations with no impact on objective function value (OFV) and high relative standard error (RSE). This was followed by forward addition to test the removed sub-populations individually.
Model evaluation: The model structure and parameters were estimated using nonlinear mixed-effect modeling NONMEM 7.4.4 software [6]. Mixture and prediction corrected visual predictive check (mix-pVPC) and sampling-importance resampling (SIR) method were performed [7–9]. Furthermore, the incidence percentages of ADRs were compared between the observed data and 100 simulated datasets based on the final model.
Results:
Data: More than 7000 observations from 116 SSD patients receiving clozapine followed up for a mean of 6 years were included.
Base models: Tachycardia, weight gain, and ANC decrease were best described by an offset, a piecewise linear, and a surge functions, respectively.
Combined model: The sub-populations retained in the combined model were those with all the ADRs (42.9%) and tachycardia (32.5%). Body mass index (BMI) increased by 0.01 Kg/m2 every week over an estimated baseline of 24.7 Kg/m2 until reaching a plateau after 279 weeks post-therapy initiation. ANC decreased by 20% from the estimated baseline of 4540 cells/µL between week 12 and 20.8 after clozapine treatment. Tachycardia was represented as a 14% consistent PR increase, over the estimated baseline of 87.9 bpm, with a clozapine maintenance dose of 450 mg. The percentage increased to 20% with a maintenance dose of 600 mg. The interindividual variabilities in the ANC baseline, pulse rate baseline, BMI baseline, and BMI slope were 26.5%, 8.2%, 23.7%, and 142.5%, respectively.
Model evaluation: The model showed good predictive performance for all three ADRs based on the mix-pVPC and SIR. Furthermore, based on 100 simulated datasets of realized design, a mean percentage (90% confidence intervals) of 75.5% (70.7-81.0%) of patients were predicted to have PR > 100 bpm, 1.9% (0-4.3%) patients had ANC ≤ 500 cells/µL, and 57.2% (51.7-63.8%) patients had weight gain > 2.5%; comparable with the corresponding observed percentages of 78.4% PR > 100 bpm, 0.9% ANC ≤ 500 cells/µL and 56.9% weight gain > 2.5%.
Conclusions: In the current study, a new pharmacometrics approach was proposed to describe the DoTS classification with continuous, longitudinal ADRs data as a function of time, using the mixture feature to assess susceptibility. The mixture model was also used to investigate the co-existence of ADRs in the population. The methodology was applied to model clozapine-induced weight gain, ANC decrease, and tachycardia. Applications of the ADR models are not only limited to existing drugs with well-known adverse effects but can be extended to new drug development by monitoring changes in physiological parameters and their distributions to test the possible ADRs.
References:
[1] Albitar O, Harun SN, Ahmad SNA, Ghadzi SMS. A Repeated Time-to-Positive Symptoms Improvement among Malaysian Patients with Schizophrenia Spectrum Disorders Treated with Clozapine. Pharm 2021, Vol 13, Page 1121. 2021;13(8):1121.
[2] Martini F, Spangaro M, Buonocore M, et al. Clozapine tolerability in Treatment Resistant Schizophrenia: exploring the role of sex. Psychiatry Res. 2021;297:113698.
[3] Aronson JK, Ferner RE. Joining the DoTS: New approach to classifying adverse drug reactions. Br Med J. 2003;327(7425):1222-1225.
[4] Holford N. Clinical pharmacology = disease progression + drug action. Br J Clin Pharmacol. 2015;79(1):18-27.
[5] Jauslin PM, Frey N, Karlsson MO. Modeling of 24-hour glucose and insulin profiles of patients with type 2 diabetes. J Clin Pharmacol. 2011;51(2):153-164.
[6] Beal SL, Shiener LB, Boeckman AJ. NONMEM Users Guides (1989-2008). Icon Development Solutions, Ellicott City, MD; 2008.
[7] Bergstrand M, Hooker AC, Wallin JE, Karlsson MO. Prediction-Corrected Visual Predictive Checks for Diagnosing Nonlinear Mixed-Effects Models. AAPS J. 2011;13(2):143-151.
[8] Arshad U, Chasseloup E, Nordgren R, Karlsson MO. Development of visual predictive checks accounting for multimodal parameter distributions in mixture models. J Pharmacokinet Pharmacodyn. 2019;46(3):241-250.
[9] Dosne AG, Bergstrand M, Harling K, Karlsson MO. Improving the estimation of parameter uncertainty distributions in nonlinear mixed effects models using sampling importance resampling. J Pharmacokinet Pharmacodyn. 2016;43(6):583-596.
