One of the most potent selective serotonin reuptake inhibitors (SSRIs), paroxetine is frequently prescribed to treat depression, anxiety, panic disorder, and obsessive-compulsive disorder (Venkatakrishnan and Obach, 2005; Nishimura et al., 2016). The recommended daily dose of paroxetine is generally 20 mg, which is gradually increased by 10 mg/day (once daily) every week until a maximally tolerated dose is reached (Nishimura et al., 2016). The guidelines from the Arbeitsgemeinschaft für Neuropsychopharmakologie and Pharmakopsychiatri (AGNP) list the therapeutic range as 20–65 ng/ml (Hiemke et al., 2018). After oral treatment, paroxetine is mostly absorbed by the digestive system and partly metabolized by CYP2D6 into inactive metabolites. Paroxetine easily accumulates in vivo after repeated administration, and the pharmacokinetic (PK) characteristics alter with dosage, leading to nonlinear PKs and increased exposure with time (Bourin et al., 2001). For healthy subjects, the overall average half-life of elimination is approximately 21 h (Nishimura et al., 2016). For different formulations, immediate-release paroxetine is rapidly absorbed compared with sustained-release tablets (T_max 5.2 versus 8 h) (Foster and Goa, 1997; DeVane, 2003; U.S. Food and Drug Administration, 2014; U.S. Food and Drug Administration, 2015). A vast volume of distribution reveals that paroxetine is extensively distributed in the body and is highly bound to plasma proteins (approximately 95%) (U.S. Food and Drug Administration, 2015). It is well acknowledged that genetic polymorphisms and nonlinear metabolism are linked to paroxetine’s PK variability. To determine the paroxetine dose, therapeutic drug monitoring (TDM) and the psychiatrist’s experience are typically applied.

Previous PK models have assumed that paroxetine is eliminated by at least two different kinetic mechanisms, including a low-affinity linear process and a high-affinity saturable one. Human paroxetine metabolism is most likely dominated by CYP2D6 (high affinity) and CYP3A4 (low affinity) (Sindrup et al., 1992; Carvalho Henriques et al., 2020). In some individuals, CYP1A2 may also be important for paroxetine metabolism (Jeppesen et al., 1996; Jornil et al., 2010; Lin et al., 2010). Some studies have suggested that serotonin (5-HT) transporter (SERT) candidate genes (such as SLC6A4) and serotonin receptor genes (such as HTRLPR) may influence paroxetine treatment response (Wilkie et al., 2009; Ishiguro et al., 2011; Marazziti et al., 2020). The CYP2D6 genetic polymorphism is widely documented as a primary predictor of paroxetine PK variability; thus, most PK studies of paroxetine have focused on the prediction of the paroxetine concentrations and CYP2D6 genotype (Feng et al., 2006; Findling et al., 2006; Mikami et al., 2013; Nishimura et al., 2016). In clinical practice, the purpose of genotyping is to correlate genotypes to individual therapy responses and to alter treatment dosages accordingly. A previous study of Korean subjects examined the influence of sex, age, body weight, diagnosis, and serum albumin level on paroxetine plasma concentrations, with age and daily dose identified as the major variables (Kim et al., 2015). Another study of older people evaluated the influence of parameters such as race, sex, age, weight, and CYP2D6 genotypes on paroxetine PKs and discovered that weight and CYP2D6 polymorphisms had an impact on maximum velocity (Vm), whereas sex had an impact on the volume of distribution (Feng et al., 2006). Furthermore, age, weight, sex, CYP2D6 genotypes, and daily dosage were included as variables in a population PK (PPK) analysis of Japanese patients, and sex and genotypes were found to be significant (Nishimura et al., 2016).

With changes in pharmacodynamics and PKs, more concurrent ailments, and the increased combination of differential situations, the variables that impact PKs in the paroxetine population are complicated, with apparent individual variances. A study showed that the concentration of paroxetine showed an exponential linear curve with increasing doses, especially at lower doses, and the variance was not reflected at high doses (Carvalho Henriques et al., 2020). Most patients experienced an unequal rise in their plasma drug levels with increasing doses. It is possible that this was due to the saturation of enzyme and its self-inhibition (Sawamura et al., 2004). Given that trough concentration is closely connected with clinical outcomes, paroxetine trough concentrations are examined to help with dose modification (Kim et al., 2015). Additionally, the relationship between paroxetine trough concentrations and area under the curve (AUC) is similar between sustained- and immediate-release paroxetine (U.S. Food and Drug Administration, 2014; U.S. Food and Drug Administration, 2015); therefore, both formulations can be monitored using the same TDM approach.

Numerous pieces of research have characterized the PKs of paroxetine with immediate-release; even so, few PPK studies have characterized the PKs of formulations with both immediate- and sustained-release. Although some paroxetine PPK studies have been established to characterize factors for paroxetine, they have not been fully characterized in the Chinese population. In addition to identifying the baseline demographic and covariate characteristics that have a substantial impact on paroxetine PKs, this modeling study was conducted to describe the PPKs of immediate- and sustained-release paroxetine at different dosages.

A one-compartmental model with FOCE-I was developed to describe the PK parameters and predict the concentrations of paroxetine in this analysis. Based on a small number of sample concentrations, we evaluated the effect of demographics and biochemical indicators on the PKs of paroxetine. Finally, based on the final model, we simulated the influence of covariates for different daily doses.

In our study, only one patient received paroxetine at a dosage of 10 mg, meaning that the sample size seemed inadequate to develop a model for the receiving 10 mg paroxetine population. Therefore, we did not initially include patients receiving 10 mg. Lacking data on absorption, with wide individual variation, which resulted in the Ka value being unreliable, we decided to fix the Ka value at 0.908/h. When covariates were filtered with high criteria (p < 0.0001), CL/F-dose, V/F-formulation, and bioavailability (F)-sex were determined to be the most important covariates in final model. The theta CL about dosage was a negative value both in the final model and bootstrap, showing that the clearance of paroxetine decreased with an increasing daily dose in this study, similar to earlier studies (Sawamura et al., 2004; Kim et al., 2015). Some previous PPK studies focused on the IR-T, with just one study involving the SR-T, but not including formulation as a covariate (Feng et al., 2006; Findling et al., 2006; Saruwatari et al., 2014; Kim et al., 2015; Nishimura et al., 2016). The understanding of whether antidepressants are susceptible to sex differences has increased along with our understanding of the role that sex differences play in the etiology of depression and PKs (Gibiino and Serretti, 2012). Previous PK analyses showed that sex affected paroxetine clearance or volume of distribution (Feng et al., 2006; Kim et al., 2015; Nishimura et al., 2016); but some studies did not show the differences (Sawamura et al., 2004; Findling et al., 2006; Yasui-Furukori et al., 2011). Age and weight were attempted to be covariates; however, neither had a significant influence on the PKs of paroxetine. Although Jung-Ryul Kim et al. demonstrated that older patients had lower CL/F than younger patients, the median age was 71 years (range: 24–90), which favored older patients (Kim et al., 2015). In our study, the age group was focused on younger patients (median 37.5, range 15–90). Additionally, only six sample patients (3.26%) were older than 71 years, which partially explains the nonsignificant effect of age observed in our study. Although the fact that paroxetine’s safety in adolescents and children has not yet been fully proven, a PK study in this age group indicated that paroxetine was generally safe and well tolerated (Findling et al., 2006; Westergren et al., 2020). Thus, the PPK of paroxetine was mainly focused on patients above 15 years in this study.

When we focused on a combination of the two significant influences of dose and sex, a marked increase in dose-corrected serum concentrations (concentration/daily dose, DRC) was observed. Typically, the dose-corrected values were constant. However, comparing the differences in serum levels of paroxetine between females and males or the concentration/daily dose (C/D) ratio between lower and higher dose patients, higher serum levels in females and high dose patients were observed. The consensus guidelines of the AGNP indicated that DRC factors range from 0.37 to 0.83. However, our study results (DRC: 0.83–4.05) markedly exceeded the reference range, when receiving more than 20 mg of paroxetine, no matter whether IR tablet, or SR tablet was administered. In a PK study of children and adolescents receiving 10–20 mg doses, the concentration in those receiving 20 mg was 6.9-fold higher (2.5–20 ng/ml, DRC: 0.125–1.02) than the concentration in those receiving 10 mg (Findling et al., 1999). Another study of healthy subjects combined 20 mg paroxetine with Fosamprenavir-Ritonavir, and the C_min was 0.017–0.03 mg/L (DRC: 0.85–1.5) (van der Lee et al., 2007). Our results were close to those in the above report in individuals receiving 20 mg, where the DRC ranged from 0.83 to 1.23. Furthermore, the steady concentration changed 25-fold (25–670 nmol/L, DRC: 0.262–7.023) in some fast extensive metabolizers and elderly poor metabolizers receiving 30 mg, and the concentration/dose ratios increased throughout the dose levels from 10 to 50 mg once daily, which was also similar to our findings (Sindrup et al., 1992). It could be considered that guideline reference ranges are appropriate at low doses, while individual differences are more obvious at high doses. Given that its initial definition was based on limited evidence that might not accurately characterize the relationship of concentration-response in patients receiving high dose paroxetine, the reference value is a contentious concept. There is no doubt that the reference ranges given for most psychotic patients are from guidelines (Hiemke et al., 2018). Some patients would benefit from concentration above the upper limit or below the lower limit of the AGNP recommended ranges. However, for most patients receiving paroxetine in our study, reference suggestions were lacking.

In addition, covariate analysis discovered that the male relative bioavailability was lower than that of female subjects. This is consistent with a prior study that reported that female patients had a higher median serum paroxetine level than male patients (Feng et al., 2006; Nishimura et al., 2016). Bioequivalence analyses revealed sex differences in relative bioavailability (Lu et al., 2019). This suggests that females need a lower paroxetine dose than males, indicating that females receiving a paroxetine dose of 20 mg/d should be 25 mg/d in males to achieve the same lower limit of the therapeutic reference range. Moreover, there is a trend here that as the dose increases, the difference between genders is pronounced, resulting in the C/D ratio of females being 1.47-fold higher than males.

One of the distinguishing features of paroxetine is nonlinear dynamics, and this may be related to a decline in inherent clearance caused by saturation of the high-affinity enzyme CYP2D6, as well as time-dependent metabolism caused by auto-inactivation of the enzyme (Sindrup et al., 1992; Feng et al., 2006; Findling et al., 2006; Kim et al., 2015). CYP2D6 is inactivated by metabolic or mechanism-based inhibition (Jornil et al., 2010; Mikami et al., 2013; Uttamsingh et al., 2015). A scaling model for mechanism-based inactivation by paroxetine from in vitro data suggested that patients receiving paroxetine at doses of 20–30 mg QD should experience a 93% inactivation of CYP2D6 (Venkatakrishnan and Obach, 2005). Additionally, as previously reported, it is also recognized that CYP2D6 genetic polymorphism is a major factor affecting the paroxetine PK variability (Feng et al., 2006; Kim et al., 2015; Nishimura et al., 2016; Chen et al., 2017). Therefore, we evaluated the effect of CYP2D6 genotypes on the PK parameters. Due to the limitation in sample sizes, when the patients were classified by genotype, we were unable to discover any appreciable influence of CYP2D6 polymorphism on the correlation between plasma levels and the daily dose. A previous study showed that the differences in plasma paroxetine concentrations resulting from different CYP2D6 genotypes were manifested at lower doses (10 mg/day) in the Japanese population but were not observed at higher doses (Sawamura et al., 2004). The same conclusion was drawn in another study that showed that genetic polymorphism was related to PKs at lower doses (Saruwatari et al., 2014). According to previous PK models, the paroxetine’s mechanism-based suppression of CYP2D6 could account for the nonlinear and accumulation properties of paroxetine (Venkatakrishnan and Obach, 2005; Johnson et al., 2009; Mikami et al., 2013). CYP2D6 inhibition was associated with higher doses of paroxetine (such as 30 mg) (Sindrup et al., 1992; Carvalho Henriques et al., 2020), and our study dose groups were focused on 20 mg or 40 mg. Furthermore, due to PK overlap amongst phenotypic groups, genetic polymorphism did not accurately reflect the change in PKs over time or the effects of drug-drug interaction. Genetically extensive metabolizers may not represent a homogeneous group, and clinical practice should take available genetic data into account (Storelli et al., 2018). Previous studies revealed that extensive metabolizers have the potential to convert to poor metabolizers during paroxetine treatment (Jeppesen et al., 1996; Brosen, 2015). A study found that a higher rate of phenotypic conversion is especially observed in heterozygotes than in homozygotes (94 vs. 56%) (Storelli et al., 2018). Thus, self-inhibition and potential phenotypic transformation of CYP2D6 may lead to nonlinear kinetics of certain genotypes at common doses (Hicks et al., 2015). Consequently, the role of paroxetine as a CYP2D6 inhibitor would rise as the paroxetine dose was increased and caused a decline in enzyme activity (Findling et al., 2006). This potentially partly explains the role of genotypes being nonsignificant at high paroxetine doses. Further study is needed with more patients to evaluate the effect of genotype in the Chinese population, combined with clinical efficacy and adverse effects. Close clinical monitoring is preferred supported by serum concentration measurements (Westin et al., 2017). It is more accurate to measure serum concentration where CYP2D6 genetic polymorphism was not identified. Combining it with the genotyping could make the consequence more credible, if easily accessible (Westin et al., 2017), and may be useful in obtaining more exact PK parameters as well as determining whether anomalous serum concentrations were caused by noncompliance.

In the final model, most PK parameters had moderate interpatient variability, and mainly the apparent distribution volume exhibited significant formulation variability. There are two approved formulations of paroxetine: 20 mg per tablet and 25 mg per tablet (enteric-coated sustained-release tablets). There are no significant differences in clearance across various formulations of the same drug. But inter-patient variability in paroxetine trough levels was different for the tablet and enteric-coated sustained-release tablets. Typically, the absorption rate constant of the paroxetine SR-T was equivalent to that of paroxetine IR-T due to the extended-release features of the sustained-release formulation, however, its relative bioavailability was 0.67 depending on the product brochures of paroxetine (GlaxoSmithKline, 2005). This means that different formulations have an effect on the volume distribution. Following absorption, the typical value of the apparent V/F was 8850 L, indicating that paroxetine is extensively distributed in the body. Compared with the IR-T, the SR-T resulted in a decrease in V/F of about 66.6% in this study. The V/F differed, which meant a higher serum concentration for paroxetine IR-T than SR-T. Our study found that the difference was nonsignificant between IR-T and SR-T at lower doses (such as 25 mg), no matter whether females or males; but, when receiving a higher dose (such as 50 mg), the difference was obvious. Notably, if female patients received more than a 40 mg dose, they easily exceeded the laboratory alert level; thus, another formulation such as SR-T could be considered. Furthermore, when females receive a high-dose paroxetine IR-T, there is a clinical need to monitor the dynamic change in serum concentration and make prompt adjustments.

Dose, sex, and formulation have a significant influence on the serum concentrations of paroxetine, and additive effects must be considered. Thus, we estimated the effects of these covariates in our study, which were dose-related (2.18-fold), formulation-related (1.28-fold), and gender-related (1.7-fold), respectively. Furthermore, rather than a set reference range to evaluate the impact of paroxetine on disease, we recommend using an individual therapeutic range considering the non-linear concentration fluctuations caused by dose increases. The tendency in antipsychotic treatment is to shift away from reference ranges to individual therapeutic ranges, which may help produce an optimal response in the patient.

Some limitations should be addressed in more research. During the concomitant drug screening, this study did not identify potential agents that significantly affected the plasma concentration of paroxetine. However, it is worth noting that, as a strong inhibitor of the CYP2D6 enzyme, paroxetine may have a strong inhibitory effect on drugs also metabolized by the CYP2D6 enzyme. In addition, the high protein binding rate of paroxetine may potentially increase the free concentration of paroxetine when administered with another drug with a higher binding rate of plasma proteins. Moreover, some classifications could not be fully analyzed because we only retrospectively got a small sample of patients’ TDM concentrations for paroxetine. Furthermore, because of the prevalence of selection bias in retrospective data gathering, it was possible to include some subjects whose paroxetine treatment was ineffective, meaning that they were tolerated despite receiving higher daily doses. Some patients’ information was not included; thus, the results of some influencing factors were not reflected, such as genotype. In summary, these factors may, in part, contribute to the bias in estimates.

This sparse data paroxetine PPK model describes the observed paroxetine PK data in Chinese psychotic patients. The model characteristically describes the non-linear increase in paroxetine concentration. The PKs of paroxetine were affected by dose, formulation, and sex among the variables assessed. PPK models may be useful in developing individual treatment regimens for patients, as well as offering insight into variables influencing patient medication variability. Under the impact of covariates, the risk of clinical concentrations over the therapeutic level was dramatically enhanced. Trough concentrations should be monitored, and dosage adjusted to keep individual patient exposure to be safe.