Model‐informed precision dosing using virtual twins (MIPD‐VTs) is an emerging strategy to predict target drug concentrations in clinical practice. Using a high virtualization MIPD‐VT approach (Simcyp version 21), we predicted the steady‐state clozapine concentration and clozapine dosage range to achieve a target concentration of 350 to 600 ng/mL in hospitalized patients with treatment‐resistant schizophrenia (N = 11). We confirmed that high virtualization MIPD‐VT can reasonably predict clozapine concentrations in individual patients with a coefficient of determination (R2) ranging between 0.29 and 0.60. Importantly, our approach predicted the final dosage range to achieve the desired target clozapine concentrations in 73% of patients. In two thirds of patients treated with fluvoxamine augmentation, steady‐state clozapine concentrations were overpredicted two to four‐fold. This work supports the application of a high virtualization MIPD‐VT approach to inform the titration of clozapine doses in clinical practice. However, refinement is required to improve the prediction of pharmacokinetic drug–drug interactions, particularly with fluvoxamine augmentation.

Clozapine pharmacokinetics are influenced by various genetic and environmental factors including drug interactions (CYP1A2 inhibitors or inducers, including smoking), obesity, inflammation, geriatric age, and pregnancy.

Can model‐informed precision dosing using virtual twins (MIPD‐VTs) reliably identify the clozapine dosage range to achieve target concentrations of 350–600 ng/mL?

The study verified that the MIPD‐VT can reasonably predict clozapine plasma concentrations in patients with treatment‐resistant schizophrenia and has the potential to be utilized in identifying the required clozapine dose range to achieve therapeutic concentrations.

The use of MIPD‐VT may be utilized to avoid toxicity during clinical optimization of narrow therapeutic index drugs within inpatient and outpatient settings.

Not all patients respond to the same drug equally. Variability in drug response results from differences between patients in demographic, genetic, and/or environmental factors.[[1], [3], [5]] The combined effect of these factors can significantly impact drug concentrations (pharmacokinetics [PKs]), resulting in adverse drug reactions (ADRs) or nonresponse to treatment. Model‐informed precision dosing (MIPD), in which computer modeling and simulation combines patient‐specific factors to optimize dose recommendations, is an attractive solution to achieve target drug concentrations in clinical practice. This is especially important for drugs with a well‐defined concentration‐response relationship and a narrow therapeutic index that carry an increased risk of nonresponse or life‐threatening adverse effects if drug concentrations occur outside the therapeutic range. In recent years, physiologically‐based PK (PBPK) approaches have been used to construct virtual twins (VTs) of real patients for MIPD.[[3], [7], [9]] Clozapine, the most effective antipsychotic medication in treatment‐resistant schizophrenia, is suggested to have the narrowest therapeutic index among second‐generation antipsychotic medications, and has potentially life‐threatening adverse effects, such as myocarditis, agranulocytosis, and seizures. Furthermore, its metabolism is influenced by various genetic and environmental factors, including drug interactions (CYP1A2 inhibitors or inducers, including smoking), obesity, inflammation (C‐reactive protein [CRP] elevations), geriatric age, and pregnancy.[10] Consequently, it is an ideal candidate to evaluate the potential of the MIPD‐VT approach.

In our previous study,[11] we built VTs of 42 clozapine‐treated patients using Simcyp and evaluated the impact of systematically increasing the virtualization of covariates on predicted clozapine PKs. Three evaluation scenarios were simulated: low (demographic), medium (demographic and environmental), and high (demographic, environmental, and genetic) covariate virtualization. The high covariate virtualization model provided the best prediction of the observed clozapine plasma concentrations, supporting the idea that high‐level covariate virtualization is required for clinical utilization of the MIPD‐VT approach.

In the present study, the high covariate virtualization approach is applied to a novel data set of patients with treatment‐resistant schizophrenia who had received clozapine therapy and had repeated clozapine plasma concentrations. Unlike the previous study, there was high accuracy of the timing of clozapine dosing and blood samples for clozapine concentration quantification (due to these patients being hospitalized). This new data set was used to evaluate: (i) the performance of MIPD‐VT in predicting steady‐state plasma concentrations at multiple timepoints, (ii) whether MIPD‐VT can identify lower and upper dose ranges for clozapine to achieve target concentrations (350 to 600 ng/mL) in each patient,[12] and (iii) the performance of MIPD‐VT in predicting plasma concentrations with fluvoxamine augmentation.

Eleven patients diagnosed with treatment‐resistant schizophrenia who were enrolled in the Metabolic and Genetic Explorations in Refractory Schizophrenia (MAGERS) study were selected to construct VTs in Simcyp. Briefly, MAGERS study patients were recruited and genotyped during their hospitalizations at the British Columbia Psychosis Program (BCPP), a 25‐bed tertiary inpatient resource for treatment‐resistant psychosis in the Canadian province of British Columbia. Adult patients (aged 18–65) can only be admitted to the BCPP after failing outpatient trials with two or more antipsychotics, and typically, multiple inpatient hospitalizations in general and/or regional psychosis‐specialized psychiatric units. To be eligible for the MAGERS study, BCPP inpatients had to meet Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition criteria for schizophrenia or schizoaffective disorder (with or without a history of catatonia), or catatonia associated with another mental disorder. Caffeine intake was monitored and controlled in the study patients, who consumed less than three to four cups of coffee per day. This intake of caffeine is not considered to cause clinically significant inhibition of clozapine metabolism. Informed consent, or assent plus surrogate consent, was obtained by a research assistant. The study was conducted under approval of the University of British Columbia Clinical Research Ethics Board (study H14‐02657).

Blood samples were collected at least 6 days after any clozapine or fluvoxamine/bupropion dose adjustments and ~12 h after the last clozapine dose to measure clozapine plasma concentrations.

Clozapine and N‐desmethylclozapine concentrations were quantified by an accredited clinical laboratory (British Columbia Provincial Toxicology Centre) by liquid chromatography–tandem spectrometry, using a commercial assay kit (Chromsystems, Munchen, Germany). As per the manufacturer's instructions, samples were extracted by combining 0.05 mL of patient serum with 0.025 mL of Chromsystems extraction buffer (#92005) and 0.25 mL of internal standard (ISTD) working solution. The ISTD working solution is a mix of 4 mL of internal standard mix—Chromsystems product (#92046/AN1/XT) MassTox antidepressants 1/extended and neuroleptics 1/extended internal standard set (ISTD 5 for clozapine and ISTD 22 for N‐desmethylclozapine)—and 50 mL of Chromsystems Precipitation Reagent (#92003). The high‐performance liquid chromatography column used is the Chromsystems MassTox TDM MasterColumn Series A (#92110), and the mobile phase consists of Chromsystems MassTox TDM series A Mobile Phase 1 (#92001) and Chromsystems MassTox TDM series A Mobile phase II (#92002). The samples were measured using the mass spectrometry analyzer Agilent QQQ 6470, operated in the multiple reaction monitoring (MRM) positive ionization mode. The MRMs are clozapine MRM1 327>270, clozapine MRM2 327>192, N‐desmethylclozapine MRM1 313>192, N‐desmethylclozapine MRM2 313>270, ISTD5 (clozapine ISTD) MRM1 335>275, and ISTD22 (N‐desmethylclozapine ISTD) MRM1 321>192. The limits of sensitivity are as follows: clozapine analytical measuring range 1 to 3600 nmol/L, and N‐desmethylclozapine analytical measuring range 1 to 1600 nmol/L. The precision is as follows: clozapine quality control level 1 (QC1) concentration was 964.9 nmol/L and within run standard deviation/coefficient of variation (SD/CV) 7.4/0.8 (total SD/CV 10.5/1.1), clozapine QC2 concentration was 1649.1 nmol/L and within run SD/CV 15.7/1.0 (total SD/CV 45.11/2.7), N‐desmethylclozapine QC1 concentration was 528.4 nmol/L and within run SD/CV 7.1/1.4 (total SD/CV 9.1/1.2), N‐desmethylclozapine QC2 concentration was 739.5 nmol/L and within run SD/CV 9.1/1.2 (total SD/CV 21.7/2.9), clozapine QC1 between day SD/CV 7.8/0.8, clozapine QC2 between day SD/CV 42.2/2.6, N‐desmethylclozapine QC1 between day SD/CV was not performed, and N‐desmethylclozapine QC2 between day SD/CV 16.7/2.7.

High molecular weight genomic DNA was extracted in the Fraser Health Molecular Cytogenetics Laboratory at Royal Columbian Hospital from whole blood collected in BD Vacutainer EDTA tubes. After rapid thawing, micro‐fluidic partitioned DNA libraries for each sample were created in Canada's Michael Smith Genome Sciences Centre at BC Cancer, using the Chromium system from 10× Genomics (10× Genomics). Gel beads‐in‐Emulsion (GEMs) were produced by combining DNA, Master Mix, and partitioning oil in the 10× Genomics Chromium Controller instrument with the micro‐fluidic Genome Chip (PN‐120216; 10× Genomics). The DNA in each GEM then underwent isothermic amplification with barcoding of each fragment. Barcoded fragments then underwent Illumina library construction (as described in the Chromium Genome Reagent Kits Version 2 User Guide, PN‐120229). Each resulting library was then assessed for quality using an Agilent 2100 Bioanalyzer and a DNA 1000 assay. Libraries were then sequenced at 40× average coverage (effective coverage was ~32× due to the overhead of the DNA "bar codes" tagging the gDNA reads, which were removed in software) in 150 base paired end tag lanes on an Illumina HiSeqX sequencer.[13] De‐identified FASTQ files were transferred over a secure link to the Pavlidis laboratory in UBC's Michael Smith Laboratories. Alignment of linked reads produced by whole‐genome sequencing (WGS) to the hg19 reference genome was performed using the 10× Genomics Long Ranger (version 2.2.2) pipeline.[14] Sequence variants were called with GATK (version 4.0.3.0), and filtered for quality using settings recommended by the GATK community.

Variant Call Format (VCF) and Binary Alignment Map (BAM) files were analyzed, utilizing Stargazer version 1.08[15] to call diplotypes for CYP2D6, CYP2C9, CYP2C19, CYP3A4, and CYP3A5. Stargazer uses WGS data to identify phased haplotypes comprised of single‐nucleotide polymorphisms, small indels, and large structural variants, and then maps these haplotypes to star alleles based on translation tables maintained by the Pharmacogene Variation Consortium. Genotype to phenotype translation followed procedures developed by the Pharmacogene Variation Consortium[16] and the Clinical Pharmacogenetics Implementation Consortium (CPIC).[17]

Simulations were performed blinded to the actual observed clozapine concentrations. Once the VT simulations were completed for all patients, the predicted clozapine concentrations and the clozapine dosage range to achieve the target clozapine concentrations (350–600 ng/mL) were matched to the observed clozapine concentrations and the final clinical dose of each patient. This combined data set was used to test the high covariate virtualization model in terms of its ability to (i) predict each steady‐state clozapine concentration versus the observed clozapine concentration and (ii) the predicted dosage range to achieve the nominal therapeutic range of 350 to 600 ng/mL for each patient.

All the VT predictions of clozapine plasma concentration were performed using the high covariate virtualization model described in our previous study.[11]

Individual VTs were constructed in Simcyp version 21, as previously described by Mostafa et al.[11] In short, each participant was matched for demographics, clozapine dose, CYP phenotypes, and the presence of inhibitors and inducers (see Table 1). Fluvoxamine and bupropion were the only inhibitors with validated inhibitor drug profiles in Simcyp and were accounted for when co‐prescribed with clozapine. The Simcyp population of North European Whites (Sim‐NEurCaucasian) was used for participants with a body mass index (BMI) of less than 30 and assigned as having White, Mixed White, or Middle Eastern Ancestry, as per demographic data in Table 1. The obese population in Simcyp was used for participants with a BMI between 30 and 40, whereas the morbidly obese population was used for those with a BMI greater than 40, irrespective of ancestry. The Simcyp population of Sim‐Chinese healthy volunteers was used for one participant who was assigned as having East Asian ancestry with a BMI less than 30.

1 TABLE Patient characteristics in the clinical study.

1 Abbreviations: A, fidosing period A; B, dosing period B; BPR, bupropion; C, dosing period C; C/D, cigarettes per day; D, dosing period D; E, dosing period E; EA, East Asian; F, female; FLV, fluvoxamine; Ht, height; Ind, inducer; Inh, inhibitor; M, male; MC, mixed Caucasian; ME, Middle Eastern; NAE, North American/European; NS, nonsmoker; UNK, unknown quantity; Wt, weight.

• 2 a No study participants had elevated CRP around the time of blood sample collection for clozapine measurement. CRP elevations have been associated with CYP1A2 phenoconversion.
• 3 b Participants with unknown smoking quantity were assumed to smoke one to five cigarettes/day for simulations.

The Simcyp validated drug profile for clozapine was used. Participants with known CYP1A2 genotypes (CYP1A2*1A/*1A, *1A/*1F, or *1F/*1F) were selected to assess the importance of customizing the CYP1A2 enzyme abundance based on the inducible CYP1A2*1F/*1F genotype. Participants carrying CYP1A2 alleles with unknown function (e.g., *1L or *1V) were excluded. Smokers with a CYP1A2*1F/*1F genotype were assigned a CYP1A2 enzyme abundance based on the number of cigarettes smoked per day, as described previously by Plowchalk et al.[18] The enzyme abundance values were adjusted using the following groupings of cigarette consumption: 1‐5 cigarettes/day (64 pmol/mg protein), 6‐10 cigarettes/day (79 pmol/mg protein), 11–20 cigarettes/day (90 pmol/mg protein), and greater than 20 cigarettes/day (94 pmol/mg protein). All other participants were assigned the nonsmoker CYP1A2 enzyme abundance of 52 pmol/mg microsomal protein. Smokers with the genotype CYP1A2*1A/*1F were additionally simulated using enzyme abundance values based on the number of cigarettes smoked given the potential inducibility by smoking as per available literature.[19] In two participants, the quantity of cigarettes smoked per day was unknown and thus a conservative quantity of 1‐5 cigarettes/day was assumed. The inbuilt Simcyp enzyme abundance data for the corresponding CYP2D6, CYP2C19, CYP2C9, CYP3A4, and CYP3A5 enzyme phenotypes were applied without alteration. One participant with a CYP3A4*1/*1x2 genotype was treated as a normal metabolizer (NM) in Simcyp, as little is known about CYP3A4*1 allele duplication and its effect on CYP3A4 phenotype. Carriers of the CYP3A5*1/*3 genotype were treated as NMs, as CYP3A5 is thought to play a minor role in the metabolism of clozapine and the corresponding enzyme abundance for the intermediate metabolizer phenotype is not available.

The high covariate virtualization model was also used to perform VT simulation patients to predict the lowest clozapine dose needed to reach a clozapine plasma concentration greater than or equal to 350 ng/mL and the maximum dose to reach a plasma clozapine concentration less than or equal to 600 ng/mL. Initially, each VT was simulated using a standard clozapine dose titration schedule (see Table 2) to the maximum dose of 200 mg per day. To ensure the predicted clozapine concentrations were always at steady‐state, the predicted clozapine plasma concentration was simulated at least 6 days after the last dosage adjustment (i.e., day 20). If the VT had not achieved a steady‐state concentration of greater than or equal to 350 ng/mL by the end of this dose titration period, further dose increases in increments of 25–50 mg were added to the dosing schedule. Once the minimum dose required to achieve the lower target clozapine concentration (350 ng/mL) was identified, the clozapine titration schedule was then further adjusted with additional dose titrations (25–50 mg day) to attain the target dose capable of achieving the upper clozapine concentration of 600 ng/mL. For Study IDs 3, 4, and 8, the last dose prior to the introduction of fluvoxamine was used for this analysis, as fluvoxamine augmentation is expected to result in CYP1A2 phenoconversion, reducing clozapine clearance and dose requirements.

2 TABLE Initial standard clozapine dosage titration schedule used in the initial simulation.

We have undertaken this comparison in two steps: (i) a whole population approach for the first observation only and (ii) at an individual level (i.e., a comparison between observed vs. predicted across repeated measures). The former provides a direct comparison against our previous paper,[11] whereas the latter allows better visualization and interpretation of our MIPD‐VT approach to the individual patient, which is consistent with the proposed approaches undertaken in precision medicine.

Linear regression was used to compare the first measured clozapine plasma concentration against the first predicted clozapine concentration from their VT. Given MIPD‐VT is an individual optimization approach rather than a population optimization, we have opted to display each patient individually to show their trajectory (repeated measures) with no regression approaches. GraphPad Prism (version 9.0.1 (151) for Windows; GraphPad Software, www.graph pad.com) was used to perform descriptive analysis and produce the simple linear regression graph. No statistical approach was undertaken to compare the VT dose range to achieve the nominal therapeutic range for clozapine against the final dose, as this was simple count data only. The metric for success was if the final observed clozapine dose was within the predicted lower and upper dosage range for their VT.

Figure 1a shows the predicted versus the first observed trough (dosing period A) clozapine concentrations for the full cohort of 11 patients using the high covariate virtualization model. None of the patients were receiving fluvoxamine during dosing period A. Similar to our previous work, the linear regression analysis confirmed a good relationship between the observed versus predicted plasma clozapine trough concentrations with a correlation of determination (R2) value of 0.29. When the VTs of the four smokers with CYP1A2*1A/*1F genotypes were re‐assigned to a high rather than low CYP1A2 enzyme abundance, to account for potential inducibility, predictions were visually and statistically improved, that is, R2 value increased from 0.29 to 0.60 (see Figure S1).

During the VT building process, we observed that two out of three patients who received fluvoxamine augmentation had clozapine plasma concentrations that were poorly predicted. Indeed, the clozapine concentrations were predicted to be two to four times higher than the observed clozapine concentrations. This observation was consistent on all consecutive clozapine blood concentrations taken (Figure 1b) after the co‐administration of fluvoxamine.

Figure 2 shows the predicted versus observed clozapine concentrations per patient with one to four repeated concentrations. The MIPD‐VT approach appears to predict the overall trajectory of the observed plasma clozapine concentrations and was consistent with any prescribed dose changes made by the medical team. For example, for Study ID 1, the first clozapine concentration (see

GRAPH

concentration point) was well above the nominal therapeutic range at 871 ng/mL. Following subsequent dosage reductions, the observed and predicted clozapine concentrations tracked back into the desired therapeutic range (see

GRAPH

,

GRAPH

, and

GRAPH

concentration points).

Predicted concentrations for Study IDs 3, 4, and 8 are grouped together at the bottom of Figure 2, as these were the patients who were augmented with fluvoxamine after an initial clozapine concentration taken on clozapine alone. For Study IDs 3 and 4, the predicted clozapine concentration in the presence of fluvoxamine were consistently overpredicted by a two to four‐fold increase (see

GRAPH

and

GRAPH

concentration points). In contrast, Study ID 8 predictions were more accurate. A full concentration‐time curve of the simulated clozapine plasma concentrations with and without fluvoxamine interactions is illustrated for Study IDs 4 and 8 in Figure 3a,b. When the clozapine and fluvoxamine doses were relatively low (150 mg or less for clozapine and 50 mg or less for fluvoxamine), MIPD‐VT appears to describe the observed concentrations well (see Table S1). At higher doses of clozapine, our MIPD‐VT significantly overpredicted the extent of the PK drug interaction (i.e., 2–4‐fold higher predicted clozapine concentrations).

Table 3 illustrates the clozapine dose ranges predicted by the high covariate virtualization model to get into the suggested therapeutic range of 350–600 ng/mL for efficacy. The simulation predicted dosage range captured the actual clinical dose utilized by the medical team in 8 out of 11 patients (73%). Although the dose taken by patient 5 was captured in the predicted range, the observed clozapine plasma concentration was 137.91 ng/mL, which is well below 350 ng/mL. It was noted that the patient was clinically stable at this measured concentration, thus further titration was not undertaken by the medical team. Thus, highlighting that perhaps a target concentration intervention approach rather than a TDM approach may be possible with clozapine dose optimization even in this treatment‐resistant schizophrenia population.

3 TABLE Estimation of the clozapine dosage range to reach a target plasma concentration between 350 and 600 ng/mL.

• 4 Abbreviation: VT, virtual twin.
• 5 a Target dose range to achieve a lower concentration of ≥350 ng/mL and an upper concentration of ≤600 ng/mL.
• 6 b The observed plasma concentration in these two patients were slightly above the recommended range of 600 ng/mL.
• 7 c The observed plasma concentration in these two patients were below the desired therapeutic range. Note both patients were of East Asian ancestry.
• 8 d This patient was taking bupropion conomitantly, which is a strong CYP2D6 inhibitor. VT simulations were performed with the addition of bupropion as an inhibitor drug, however, this did not make a significant impact on the predicted clozapine plasma concentration.

The performance of high virtualization MIPD‐VT was evaluated by predicting clozapine concentrations from a cohort of hospitalized patients with treatment‐resistant schizophrenia. These results confirm our previous work demonstrating that MIPD‐VT can reasonably predict clozapine steady‐state concentrations in real patients (R2 = 0.29 compared with R2 = 0.37 previously).[11] A particular strength of the study was that all individual simulations were undertaken without knowledge of the final clinical outcome and concentration (blinded).

There are several noteworthy findings of this study. First, predictions were greatly improved when the four smokers with CYP1A2*1A/*1F genotypes were assigned the induced CYP1A2 enzyme abundance in Simcyp (from R2 = 0.29 to R2 = 0.60). There is poor consensus in the literature regarding the influence of CYP1A2*1F on the inducibility of the CYP1A2 enzyme in smokers. However, our finding suggests that increased CYP1A2 abundance may not be specific to homozygous CYP1A2*1F, and that those who are heterozygous for this allele may similarly be susceptible to smoking induction.[19] Second, there was significant overprediction of clozapine concentrations in two of the three cases with fluvoxamine augmentation. Both patients were on a fluvoxamine dose of 50 mg/day and clozapine doses of 200 mg/day (Study ID 3) and 450 mg/day (Study ID 4), suggesting that extrapolation of inhibitory fluvoxamine concentrations at hepatic CYP1A2 may be overestimated in simulations with increasing clozapine dose. This highlights the need for further validation of inhibitor and inducer compound files within PBPK software to cater for phenoconversion, including weak and moderate inhibitors and inducers of CYP enzymes.[20]

It is important to note that the drug inhibitor compound file within Simcyp for fluvoxamine has been extensively validated from a PK‐drug–drug interaction (PK‐DDI) perspective for regulatory purposes. Our observations in this study suggests that PBPK compound libraries developed for early phase drug development (phase I–IIb and clinical pharmacology studies) may not always be suitable for PK‐DDI predictions in real patients. Third, we predicted the a priori dose range in 8 out of 11 patients (73%) in a difficult to treat patient population. In two patients (Study IDs 2 and 5), the therapeutic clozapine concentrations were well below the minimum target concentration of 350 ng/mL, which reinforces the importance of patient response and clinical judgment in clozapine dosing. Indeed, the target therapeutic range for clozapine is a useful guide for clinicians, however, some patients respond to clozapine concentrations outside of this range.[12]

Because of unpredictable PKs and a high burden of severe adverse effects that require resource intensive close monitoring, many psychiatrists are hesitant to prescribe clozapine and use it only as the last resort in seriously sick patients.[21] The MIPD‐VT approach could be used in clinics without rapid access to plasma clozapine concentrations or to change the titration schedule. Ideally, pharmacogenomic results should be available within the first week of commencing clozapine. Importantly, the proposed MIPD‐VT approach does not supplant TDM when it is available or clinical judgment regarding patient response. Instead, it is used to understand the factors driving the PK variability of clozapine and assist in predicting optimal starting doses. This approach also provides re‐assurance to the clinician in cases where rapid titration to a higher clozapine dose may be required, and then TDM is used to nuance this dose to achieve the desired clinical response. In other words, the MIPD‐VT approach could be used to predict optimal starting doses of clozapine, and then TDM used with standard dose proportional adjustments or Bayesian methods to nuance the dose. We therefore propose it should be possible to further reduce pseudo‐clozapine resistance caused by subtherapeutic concentrations and avoid ADRs linked to supratherapeutic clozapine concentrations by using the proposed MIPD‐VT approach.

There are several limitations of this study. First, given the retrospective real‐world nature of the clinical data, there may be some inaccuracies, such as the time of administration, patient adherence,[22] and the actual quantity of cigarettes smoked per day. However, these issues were mitigated as much as possible, as all patients were institutionalized and under direct supervision for dosing, blood sample collection, and day‐to‐day function. Second, PBPK drug profile validation is limited by access to sufficient number and quality of independent clinical data sets. This is illustrated in the overprediction of clozapine concentrations following fluvoxamine augmentation in this study, as validation of this specific DDI is difficult due to lack of adequate clinical data sets. We acknowledge that the observed overprediction with fluvoxamine augmentation in this study is limited by the small sample size of three patients. Further validation of this observed overprediction is warranted in a larger cohort of patients. Third, the Simcyp interface is not designed for clinical practice, including how diseases change physiological parameters important in PK predictions. For example, CRP elevations can result in CYP1A2 phenoconversion influencing clozapine PK.[10] Fourth, assumptions were made when making assignments of CYP1A2 enzyme abundance in smokers based on CYP1A2 genotype. Future clinical work to quantify CYP1A2 induction at varying levels of cigarette consumption by CYP1A2 genotype using CYP1A2‐selective probe substrate studies are warranted.

In summary, our previously described MIPD‐VT approach to predict clozapine PKs[11] was verified by comparing its predictions with the observations in a new cohort of patients with treatment‐resistant schizophrenia in a mental healthcare tertiary setting. Subsequently, our approach has the potential to be utilized on admission of patients with schizophrenia to proactively identify the required clozapine dose range to achieve the therapeutic threshold concentrations and help avoid toxicity during the clinical optimization of these patients. Prior to clinical implementation, however, pragmatic prospective trials will be needed to demonstrate clinical validity and utility within inpatient or outpatient settings.

S.M., R.R., T.M.P., C.M.J.K., C.B., A.R.‐H., and L.J.S. wrote the manuscript. S.M., R.R., T.M.P., C.M.J.K., R.S., and P.C. designed the research. S.M. performed the research. S.M., R.R., T.M.P., and C.M.J.K. analyzed the data.

The authors wish to acknowledge the contributions of the MAGERS study project team, study participants, clinicians at recruitment services, and staff at BC Psychosis Program (BCPP). The authors would also like to acknolwedge the staff at BC provincial toxicology centre for performing the clozapine assay and Sandrine Merette for assisting with questions on the assay. The MAGERS study received funding from the BC Schizophrenia Society and Foundation, UBC Development office (donations from Robert Baker and Elizabeth Carter), the Djavad Mowafaghian Centre for Brain Health, VGH, and UBC Hospital Foundation (donations from the James Family Foundation), UBC Neuropsychiatry ERIN Research Fund; and by in‐kind support from the UBC Institutes of Mental Health and Department of Psychiatry. Certara UK Limited (Simcyp Division) granted access to the Simcyp Simulators through a sponsored academic license. Open access publishing facilitated by Monash University, as part of the Wiley ‐ Monash University agreement via the Council of Australian University Librarians.

No funding was received for this work.

S.M. and L.S. are employees and shareholders of myDNA Inc., a pharmacogenomic testing and interpretation company. T.M.P. provides a consultancy service to Sonic Genetics for the interpretation of pharmacogenomic test results. T.M.P. and A.R.‐H. are employees of Certara, a company that provides modeling and simulation software and services to the pharmaceutical industry, including a population‐based PBPK simulator (Simcyp). C.M.J.K. was the academic lead on the Certara‐Monash Fellowship program funded by MTPConnect. C.A.B. is founder and equity holder of Sequence2Script Inc. and a member of the Clinical Pharmacogenetics Implementation Consortium and the Genetic Testing Committee of the International Society of Psychiatric Genetics. He has also received material support from Assurex, CNSDose, Genomind, and AB‐Biotics for research purposes and has ongoing research collaborations with myDNA but does not have equity, stocks, or options in these companies or any other pharmacogenetic companies. All other authors declared no competing interests for this work.

GRAPH: Table S1.