Development and Verification of Virtual Population Models for Predicting Drug Pharmacokinetics in Ethnic North American Populations

ABSTRACT

Ethnic variabilities can affect the outcome of drug pharmacokinetics (PK) and drug–drug interactions (DDI). This work aimed to develop four North American (NA) sub‐populations: White, African American, Asian American, and Hispanic_Latino suitable for physiologically based pharmacokinetic (PBPK) modeling and simulations. Demographic data and tissue weight/volume, blood flows, cardiac output, plasma protein levels, hematocrit, enzyme and transporter abundances/frequencies, serum creatinine, glomerular filtration rate, and gastrointestinal transit times for the different populations were collated. Equations describing various covariate relationships for these physiological parameters were developed for each ethnicity. Some key population differences that can affect drug PK were higher CYP3A5 and OATP1B1 population mean abundances for African Americans and lower CYP3A4 and OATP1B1 population mean abundances for Asian Americans. The most common CYP2C9 alleles in the White as well as the Asian populations are the *1, *2, and *3 alleles. However, additional lower‐activity alleles (*5, *6, *8, and *11) were also shown to occur among the African Americans and Hispanic_Latino subjects. Clinical studies reporting population specific PK profiles (6 studies) or results from a mixed NA population (14 studies including 6 DDI studies) were simulated, and the simulated data were within two‐fold of the observed data. PBPK simulations using the 4 NA population files and repaglinide as a probe compound predicted significant differences in the drug exposure for the Asians, African Americans, and White subjects. In conclusion, the developed population files can aid drug development for specific subgroups of the diverse NA population by accounting for variabilities in drug kinetics and DDI consequences.

Summary

• What is the current knowledge on the topic?

  • ○
    There is a potential for inter‐ethnic differences in the pharmacokinetics (PK) and pharmacodynamics (PD) of drugs, resulting from a combination of genetic, physiological, demographic, and environmental factors.
• What question did this study address?

  • ○
    This study describes the development of four North American (NA) sub‐populations to reflect the diversity of the US population and aid drug development and clinical applications that consider this diversity.
• What does this study add to our knowledge?

  • ○
    This study compiles data on the inter‐ethnic differences among these sub‐populations and highlights how known physiological and pharmacogenetic variability may translate into differences in pharmacokinetics through physiologically based pharmacokinetic modeling and simulation.
• How might this change drug discovery, development, and/or therapeutics?

  • ○
    It demonstrates how these populations, once implemented into PBPK models, can be used to prospectively predict ethnic differences in pharmacokinetic (PK) exposure and drug–drug interaction liability.

1. Introduction

In April 2022, the Food and Drug Administration (FDA) regulatory agency released a draft guidance for industry, recommending that sponsors should include a race and ethnicity diversity plan as part of an investigational new drug (IND) application that may require a clinical study [1]. This was further updated in June 2024 to extend the plan beyond only race and ethnicity and include other demographic and non‐demographic diversity characteristics [2]. This comes on the back of evidence from the US Census Bureau showing a significant increase in the diversity of the US population over the last decade and the need for clinical trials to reflect this diversity [3].

It is well known that there is a potential for inter‐ethnic differences in the pharmacokinetics (PK) and pharmacodynamics (PD) of drugs, which can result from a combination of genetic, physiological, demographic, and environmental factors [4]. Perhaps one of the most notable examples of ethnicity‐associated pharmacokinetic differences is with Tacrolimus; a drug used to prevent allograft rejection in kidney transplant patients, where the CYP3A5*1 allele was associated with the need for a significantly higher total daily dose of Tacrolimus to maintain effective plasma concentrations [5]. Of the patients expressing CYP3A5*1, 93% were of African American ancestry; thus, underscoring the need to study ethnic differences in drug metabolism and disposition. Because of examples like Tacrolimus, the FDA requirement is that a new drug application (NDA) must present effectiveness and safety data by gender, age, and race and must identify any modifications of dose or dosing interval needed for a specific subgroup [6].

Adequate recruitment and retention of study participants have long been challenges and account for 55% of terminated trials within the Clinical Trials Database [7]. Thus, this new NDA requirement has the potential to pose an additional burden to recruiting sufficient volunteers into clinical trials, especially if a specific ethnic distribution is required. One possible solution to address questions for different populations is to supplement the NDA with information from physiologically based pharmacokinetic (PBPK) simulated drug trials. PBPK modeling is now routinely being used as part of the drug development cycle, including in the optimal design of clinical studies; as well as for potential label extensions for special populations, where recruitment of patients is often challenging [8].

In anticipation of the changes to clinical trial requirements and to address these potential ethnic differences in drug pharmacokinetics safety and efficacy, four ethnic North American (NA) PBPK population models were developed. These populations account for the largest ethnic groups in North America (White, African American, Asian American and Hispanic_Latino). Future work could extend the populations to cover other ethnicities. The intention behind creating separate population files was to allow flexibility, whereby the mix of subjects in the simulated trial can be adjusted to match the demographics in the study population of interest. This manuscript describes the development of these populations, verification of these populations against observed data, and a case study demonstrating the application of these models, whereas highlighting how known pharmacogenetic variability in the different populations might translate to differences in PK of various drugs.

2. Materials and Methods

2.1. Literature/Data Searches

An extensive literature search was conducted to identify various NA ethnic‐specific demographic and physiological data required for developing virtual PBPK models. These searches were conducted between September 2021 and July 2022 in population databases such as the National Health and Nutrition Examination Survey (NHANES), or search engines such as PubMed and Google Scholar for the retrieval of relevant publications. The different physiological parameters that were queried in these searches include: organ or tissue weight/volumes and tissue blood flows; cardiac output and/or cardiac index; human serum albumin (HSA); alpha1‐acid glycoprotein or orosomucoid (ORM); hematocrit; serum creatinine and glomerular filtration rate (GFR); cytochrome P450 (CYPs) and non‐CYP enzymes involved in drug metabolism; drug transporter abundances; distribution of poor, intermediate, extensive, and ultra‐rapid phenotypes for drug metabolizing enzymes and drug transporters; parameters that are important for large molecule drugs, for example, serum IgG concentration and lymph flow; and finally gastric, small intestine, and colon‐specific parameters for pH, residence, and/or transit times as well as parameters to describe the composition, storage, and release of bile. These parameters were searched in combination with terms to identify race/ethnic differences or specific race/ethnicity information. These terms include racial differences, race, ethnic differences, ethnicity, African American, Black, Asian, Hispanic, Latino, Mexican, Caucasian, and White. Details of the search criteria are included in the search strategy section of the Supporting Information.

After relevant citations were identified, inclusion for analysis was based on the following criteria: (1) Only studies conducted in North America; (2) When possible, only studies with healthy control groups or healthy subject data; (3) A preference was given to studies whose data were categorized according to race or ethnicity. Studies that reported mean values from multi‐racial populations were used only if the ethnic configuration of the subjects was clearly stated; (4) Only data from populations greater than 18 years of age were included; (5) Studies were excluded if race could not be determined or reasonably assumed; (6) Where possible, data from underweight and obese/morbidly obese individuals were excluded. For the purposes of this analysis, underweight was defined as having a body mass index (BMI) < 18.5 kg/m² and obese as having a BMI of > 30 kg/m².

2.2. Population Development

2.2.1. Demographics

The age and gender distribution, for the four sub‐populations, were calculated using data from the 2019 United States Census survey [9]. This dataset contained information on age frequencies from 18–95 years, with a sample size of over 270 million. Distribution for all four sub‐populations was best described using discrete age bins for each gender. Thereafter, additional demographic data from three cycles of the United States National Health and Nutrition Examination Survey (NHANES) (2013–2014; 2015–2016; and 2017–2018) were combined for Non‐Hispanic White, Non‐Hispanic Black, and Non‐Hispanic Asian to develop the White American, African American, and Asian American populations, respectively. Data for the Mexican, Hispanic White, and the Other Hispanic were combined to form the Hispanic_Latino population [10].

2.2.2. Tissue Weight/Volumes, Tissue Blood Flows, and Cardiac Output

The search strategy described above yielded observed data for tissue volumes for all organs considered in the full PBPK model, except the following: bone, gastrointestinal tract, lungs, pancreas, skin, and yellow marrow. (NB: Although yellow marrow is not a direct input in the PBPK model, this value is used to estimate adipose tissue volume, hence its inclusion in our search). Further details are in Table S1. Unfortunately, no organ‐specific blood flows were identified for any of the sub‐populations investigated. Of the four populations included in this analysis, the NA White American population had the most data, followed by African American, Asian American, and lastly, Hispanic_American populations.

The next step was to determine whether previously described equations to generate organ volumes [11] were adequate to describe the data in the new populations being developed. To assess this, the demographic information (age and sex distribution, height, weight, etc.) for each North American sub‐population and previously derived organ volume relationships were used to simulate organ volume distributions for each population. These values were compared against the observed data collated from literature, and a scalar was applied, if necessary, for organs where the observed data were not well recovered.

Cardiac output information was obtained for all four sub‐populations in North America, although the amount of observed data for each sub‐population varied. There were eight studies containing data for White Americans, six studies with data for African Americans, three studies for Asian Americans, and only two for Hispanic_Latino Americans (Table S1). Similar to the organ volumes analysis, population demographics were used to generate cardiac output distribution for the different populations, using Equations (1) and (2) below. These equations were previously derived from a meta‐analysis using data from White European and White North American healthy volunteer adults. The simulated cardiac outputs were then compared to observed data.

$$\begin{array}{c}\text{Cardiac Index}=\left(0.015150\times {\mathrm{Age}}^2\right)-\left(2.712723\times \mathrm{Age}\right)\hfill \\ +261.584246\hfill \end{array}$$ (1)

$$\text{Cardiac Output}=\text{Cardiac Index}\times \text{Body Surface Area}$$ (2)

2.2.3. Plasma Proteins and Hematocrit

Individual subject data for HSA and hematocrit were available from the NHANES database. Thereafter, sex‐specific equations that describe changes in HSA as a function of age and BMI for each ethnic population were derived. For hematocrit, weighted mean and coefficient of variation (CV) values were derived separately for males and females from the observed data for each population. Ethnic‐specific data to describe levels of alpha‐1 acid glycoprotein (AAG) for the four different populations were scarce. Only two studies that evaluated the differences in AAG levels between the White American and the African American populations, comprising 27 mixed sex subjects for each sub‐population [12, 13], and another study comparing the White American and the Asian American, comprising 8 male subjects each [14] was available. These were analyzed and used to derive mean and CV values for the different populations.

2.2.4. Enzyme and Transporter Abundance and Frequency

A literature search conducted to find absolute protein abundance in North American populations for all drug metabolizing enzymes (CYPs, UGTs, CESs, etc.) and drug transporters (ABC and SLC families) that are implicated in the disposition of routinely prescribed drugs did not yield sufficient usable data to describe protein abundances in the different North American sub‐populations. Further searches were conducted to ascertain whether data describing the relative abundance of the different enzymes and transporters in the four ethnic North American populations could be found. This literature search focused on determining protein abundances specifically for the liver, GI tract, and kidney. For enzymes and proteins where there are described genotype differences, subjects that have a phenotype/genotype that is associated with increased or decreased function of the protein relative to the wild type enzymes or transporters were excluded prior to inclusion in the meta‐analysis.

Where any of these data were not available for the North American populations, the default values implemented for the Simcyp Sim‐Healthy Volunteer population, which is based on the demographics of a North European Caucasian (NEC) population were assumed for all the populations, except the North American Asian population, where the Simcyp Sim‐Chinese population values were assumed. The decision to use Chinese data as a surrogate for Asian Americans in the absence of specific data is based on the findings of a Pew Research Center study, which reported that Chinese Americans are the largest Asian origin group in the United States [15]. This is also in line with the data from the US Census bureau [9]. In addition, a previous publication has highlighted the inter‐ethnic differences that exist between the Chinese and Caucasians, thus precluding the use of the former. By integrating these ethnic specific data into a PBPK model, differences in the PK of CYP‐mediated drugs between Chinese and Caucasians were adequately predicted [4]. Further verification of this model against other drugs administered to a Chinese population has also been carried out [16, 17, 18].

Distribution of poor, intermediate, extensive, and ultra‐rapid phenotypes for enzyme and transporter frequencies was also incorporated for each North American sub‐population. This was done primarily by assessing the prevalence of the predominant genetic variants in each of the sub‐populations and translating these into extensive (normal), intermediate, poor, or ultra‐rapid metabolizer phenotype frequencies. Thereafter, population estimations of phenotype frequencies were made for intermediate, poor, and ultra‐rapid metabolizers of enzymes and transporter phenotypes. It was then assumed that extensive metabolizers and transporters made up the remaining population (i.e., frequency EM = 1−frequency (PM + IM + UM)). The main enzymes where these data were incorporated are: CYP2A6, CYP2B6, CYP2C9, CYP2C19, CYP2D6, CYP3A5, UGT1A1, UGT1A3, UGT1A9, UGT2B10, UGT2B15, UGT2B17, and CES1. Phenotype frequencies were also determined for the following transporters: OATP1B1, BCRP, and OCT1.

2.2.5. Serum Creatinine and GFR

Individual serum creatinine data were available from the NHANES database. These were used to derive sex‐dependent serum creatinine values (Mean and CV [%]) across discrete age bands for the four different populations. Within the Simcyp simulator, individual serum creatinine concentrations are utilized to obtain GFR values for individuals in a specific population through serum creatinine‐based equations for predicting GFR using Cockcroft and Gault [19] or the Modification of Diet in Renal Disease (MDRD) [20] methods.

Cockcroft and Gault equation:

$$\begin{array}{c}\mathrm{GFR}\left(\mathrm{mL}/\min /1.73{\mathrm{m}}^2\right)=\left(\frac{\left(140-\mathrm{Age}\right)\times \text{Body weight}}{72\times \frac{\mathrm{SCr}}{88.42}}\right)\hfill \\ \times \frac{1.73}{\mathrm{BSA}}\times 0.85\left(\text{if female}\right)\hfill \end{array}$$ (3)

MDRD equation:

$$\begin{array}{c}\mathrm{GFR}\left(\mathrm{mL}/\min /1.73{\mathrm{m}}^2\right)=175\times {\left(\frac{\mathrm{SCr}}{88.42}\right)}^{-1.154}\times {\mathrm{Age}}^{-0.203}\hfill \\ \times \frac{1.73}{\mathrm{BSA}}\times 0.742\left(\text{if female}\right)\hfill \end{array}$$ (4)

where age is in years, body weight in kg, SCr, serum creatinine in μM and BSA, body surface area in m².

Ethnic‐specific measured GFR values were not readily available for comparison against predicted GFR values. For the White American population, the different GFR equations were compared to the mean observed measured values from a North American study of 365 subjects, of which more than 80% of subjects were White [21]. For the African American population, these equations alongside the CKD‐EPI creatinine equations fit with and without a race component were compared to the mean measured values obtained from 579 black African American subjects [22], whereas for the Asian American population, measured values from a Chinese population of healthy subjects [23], as well as a mixed Asian population based in Singapore [24], were compared to the predicted values.

2.3. Population Verification

Clinical PK studies for small molecule drugs were collated from the Drug Interaction Solutions database (https://didb.druginteractionsolutions.org/) (DIDB) for the verification of the developed North American (NA) populations. The initial criteria for selection were to include studies performed in each sub‐population with separately reported PK profiles and parameters (area under the concentration‐time curve; AUC, maximum concentration; C _max, and/or clearance; CL). However, most studies reported only the average data for mixed NA subjects. Therefore, the inclusion criteria were widened to include these studies as a secondary phase of verification. Intravenous and/or oral administrations were considered in the comparison between simulated and predicted outputs for NA populations.

Compound files developed, verified, and implemented within the Simcyp Simulator V22 R1 to simulate the kinetics of the selected drugs in the healthy volunteer population were used. Virtual simulations differed only in the system parameters of the selected population; however, the drug files and parameters were identical in all population simulations. The simulations were matched to the collated clinical studies with respect to number of individuals, age range, and proportion of females. For each simulation, 10 trials were applied with the number of subjects in each trial matching the clinical study (Table 1). Whenever individual enzyme and/or transporter genotype or phenotype data were available, the population frequencies were modified to reflect the observed data and utilized in the simulations; otherwise, the default population frequencies for each population, as summarized in Tables S3–S7, were used.

TABLE 1.

Study design and predicted and observed exposure levels for the clinical studies used in the population verification with the mixed NA populations.

Simulated drug	Main metabolic pathway	Number of subjects (% females)	Age range (years)	Populations represented (% contribution)	References	Predicted AUC mean ± SD (μg.h/mL)	Observed AUC mean ± SD (μg.h/mL)	Predicted/Observed AUC
Warfarin(1)	CYP2C9	10 (10%)	21–52	NA White (70%), NA African American (30%)	[25]	34.47 ± 23.20	33.35 ± 9.63	1.03
Warfarin (+Fluc)(1)						62.69 ± 32.05	84.74 ± 21.97	0.74
Warfarin(2)		6 (0%)	23–29	NA White (100%)	[26]	130.40 ± 92.46	161.00 ± 60.00	0.81
Warfarin (+Fluc)(2)						274.41 ± 123.36	458.00 ± 139.00	0.60
Flurbiprofen	CYP2C9, UGT2B7	12 (25%)	19–54	NA White (66%), NA Asian (16.6%), NA African American (8.3%), NA Hispanic (8.3%)	[27]	56.79 ± 39.12	69.50 ± 11.30	0.82
Flurbiprofen (+Fluc)						101.34 ± 61.84	112.70 ± 10.30	0.90
Bupropion	CYP2B6, 3A4, 2C19	24 (0%)	18–45	NA White (75%), NA Hispanic (8.3%), NA African American (4.2%), NA Asian (12.5%)	[28]	2.14 ± 1.14	1.83 ± 0.43	1.17
Bupropion (+Cim)						2.15 ± 1.15	1.92 ± 0.45	1.12
Metformin	CYP3A4, OCT1/2	12 (50%)	23–33	NA White (25%), NA African American (50%), NA Asian (25%)	[29]	18.04 ± 6.3	11.40 ± 3.30	1.58
Midazolam	CYP3A4, UGT1A4	22 (50%)	18–50	NA White (4.5%), NA African American (50%), NA Hispanic (45.5%)	[30]	0.05 ± 0.04	0.06 ± 0.03	0.70
CBZ	CYP3A4/5, 2C8, UGT2B7	12 (30%)	20–45	NA White (66.7%), NA African American (22.2%), NA Hispanic (11.1%)	[31]	133.27 ± 78.12	139.00 ± 26.41	0.96
CBZ (+EFV)						48.87 ± 32.71	101.00 ± 21.21	0.48
EFV	CYP2B6, 1A2, 2A6	36 (30%)	20–45	NA White (66.7%), NA African American (22.2%), NA Hispanic (11.1%)	[31]	57.26 ± 41.69	71.50 ± 22.17	0.80
EFV (+CBZ)						59.12 ± 47.37	45.60 ± 18.24	1.30
S‐Mephenytoin	CYP2C19	8 (25%)	25–76	NA White (100%; 75% CYP2C9 EM, 12.5% IM, 12.5% PM)	[32]	0.53 ± 0.57	0.48 ± 0.39	1.12
Pravastatin	Renal, OATP1B, OAT3, MRP2	10 (60%)	19–55	NA White (80%), NA African American (20%)	[33]	0.17 ± 0.10	0.10 ± 0.05	1.68
Rosuvastatin*	Renal, OATP1B, OATP2B1, NTCP, MRP4, BCRP	29 (0%)	18–64	NA White (24%), NA African American (76%)	[34]	0.04 ± 0.05	0.03 ± 0.02	1.44
Ator(1)	CYP3A, CYP2C8, OATP1B, OATP2B1, P‐gp, NTCP	11 (50%)	24–48	NA White (42%), NA Asian (33%), NA African American (17%), NA Hispanic (8%)	[35]	0.07 ± 0.05	0.09 ± 0.03	0.84
Ator + RIF(1)						0.38 ± 0.23	0.72 ± 0.22	0.54
Ator(2)		10 (60%)	18–55	NA White (80%), NA African American (20%)	[33]	0.08 ± 0.04	0.09 ± 0.047	0.83
Ator* (3)		22 (55%)	18–55	> 95.7% White (Assumed to be 100% NA White)	[34]	0.06 ± 0.04	0.09 ± 0.03	0.66
Ator(4)		16 (0%)	21–54	NA White (19%), NA Hispanic (56%), NA African American (25%)	[36]	0.07 ± 0.03	0.10 ± 0.058	0.68

Abbreviations: Ator, atorvastatin; CBZ, carbamazepine; Cim, cimetidine; EFV, efavirenz; Fluc, fluconazole; NA, North American population.

^*Geometric mean values were reported.

2.4. Model Application

One of the key applications of the use of multi‐ethnic populations in PBPK modeling is in risk assessments, to identify sub‐populations or individuals that are at increased risk of an adverse drug effect or individuals that are liable to drug interactions; and would therefore require a dose adjustment. To highlight this, we explored the differences in PK for repaglinide whose disposition is dependent on enzymes and transporters that are different between the four ethnic groups.

The disposition of the antidiabetic drug repaglinide is dependent on hepatic uptake by OATP1B1 and metabolism by CYP3A4 and CYP2C8, with CYP2C8 accounting for approximately two‐thirds of its metabolism [37]. A previous study investigated the PK differences between healthy Caucasian and Japanese subjects being administered increasing doses of repaglinide and found that the Japanese cohort had higher repaglinide exposure, and also experienced more hypoglycemic reactions [38]. Thus, we did a similar comparison of repaglinide PK across the four NA populations to evaluate possible differences in PK.

Simulations of multiple doses of 0.25, 0.5, 1, and 2 mg of repaglinide were conducted using the four different populations and a default trial design of 10 trials of 20 individuals, 20–50 years old, and 50% females. The steady‐state concentrations for the four different doses across the four populations were then compared to evaluate the impact of the physiological differences on repaglinide's PK. Thereafter, a power calculation was carried out to ascertain the number of study participants from each ethnic group that would need to be recruited into a clinical study for repaglinide to be able to detect a PK difference (80% power on the basis of AUCs) between the White American population and the other sub‐populations. Finally, simulations of a clinical drug interaction between repaglinide and gemfibrozil, a potent CYP2C8 inhibitor, and between repaglinide and rifampicin, an inducer of multiple CYPs and an inhibitor of OATP transporters, were carried out using the four populations to identify subgroups that were at increased risk of the DDIs.

3. Results

3.1. Population Development and Performance Verification of Physiological Parameters

3.1.1. Demographics

The number of individual data from the NHANES database analyzed for the four different populations were 2834 subjects for the White American, 1769 subjects for the African American, 1066 subjects for the Asian American, and 2070 subjects for the Hispanic_Latino populations. Height was modeled against age using a quadratic equation, whereas weight was modeled against height using an exponential function. Assuming a normal distribution, variability was applied at each age to generate individual heights and weights. No ethnic specific data on body surface area (BSA) were available, so individual BSA was estimated using the Du Bois and Du Bois equation [39].

The predicted height and weights of virtual individuals for the four populations generated using the estimated age‐height and height‐weight equations were in good agreement with the observed data (Figure 1). Given that weight and height are used to estimate BSA, and by extension, organ volumes and cardiac output, additional verification of these equations as input to the PBPK models were performed by comparing observed versus predicted organ volumes and cardiac output. The equations for height and weight derived for the different populations are given below (Equations (5), (6), (7), (8), (9), (10), (11), (12), (13), (14), (15), (16), (17), (18), (19), (20)).

FIGURE 1.

Performance verification of the predicted heights and weights for the four North American population against observed data, on the basis of the derived age‐height and height‐weight relationships (Equations (5), (6), (7), (8), (9), (10), (11), (12), (13), (14), (15), (16), (17), (18), (19), (20)). Two thousand individuals aged 18–80 years were simulated for each sex from each population. Performance is split between males (left) and females (right) and shown for each North American sub‐population. Red dots indicate observed data collated from the NHANES database. Blue dots are Simcyp generated values for heights and weights at different ages.

African American Male

$$\text{Height}=172.46+\left(0.2401\times \mathrm{Age}\right)-\left(0.0031\times {\mathrm{Age}}^2\right)$$ (5)

$$\text{Weight}=e^{\left(2.005+0.014\times \text{Height}\right)}$$ (6)

African American Female

$$\text{Height}=159.65+\left(0.2207\times \mathrm{Age}\right)-\left(0.003\times {\mathrm{Age}}^2\right)$$ (7)

$$\text{Weight}=e^{\left(2.513+0.0117\times \text{Height}\right)}$$ (8)

Asian American Male

$$\text{Height}=173.07+\left(0.0291\times \mathrm{Age}\right)-\left(0.0018\times {\mathrm{Age}}^2\right)$$ (9)

$$\text{Weight}=e^{\left(2.012+0.0135\times \text{Height}\right)}$$ (10)

Asian American Female

$$\text{Height}=159.18+\left(0.0126\times \mathrm{Age}\right)-\left(0.0014\times {\mathrm{Age}}^2\right)$$ (11)

$$\text{Weight}=e^{\left(2.794+0.0083\times \text{Height}\right)}$$ (12)

Hispanic_Latino American Male

$$\text{Height}=173.58-\left(0.0386\times \mathrm{Age}\right)-\left(0.0006\times {\mathrm{Age}}^2\right)$$ (13)

$$\text{Weight}=e^{\left(2.406+0.0119\times \text{Height}\right)}$$ (14)

Hispanic_Latino American Female

$$\text{Height}=159.92-\left(0.0008\times \mathrm{Age}\right)-\left(0.0011\times {\mathrm{Age}}^2\right)$$ (15)

$$\text{Weight}=e^{\left(2.384+0.0121\times \text{Height}\right)}$$ (16)

White American Male

$$\text{Height}=172.89+\left(0.2746\times \mathrm{Age}\right)-\left(0.0035\times {\mathrm{Age}}^2\right)$$ (17)

$$\text{Weight}=e^{\left(2.3+0.0124\times \text{Height}\right)}$$ (18)

White American Female

$$\text{Height}=159.74+\left(0.2677\times \mathrm{Age}\right)-\left(0.0038\times {\mathrm{Age}}^2\right)$$ (19)

$$\text{Weight}=e^{\left(2.995+0.0082\times \text{Height}\right)}$$ (20)

3.1.2. Tissue Weight/Volumes, Tissue Blood Flows, and Cardiac Output

A summary of the outcome of the literature search for tissue volumes/weights and cardiac output is given in Table S1. The organs where observed data were available included blood, plasma, erythrocyte, adipose, brain, heart, kidney, liver, muscle, and spleen. Most organ volumes for the four populations were adequately recovered with the derived population‐specific demographic data and the existing equations implemented for the NEC population for estimating organ volumes (Figures S1–S7). A liver volume scalar of 0.85 for males was applied to Asian American populations, in line with what is used in the Sim‐Chinese population, as this improved the PK prediction of drugs in the Asian American population (Figure 3).

FIGURE 3.

Simulated mean plasma concentration–time profiles (data lines) and observed data points (circles) of alprazolam intravenous; IV (A) and oral (B) in healthy White American subjects (black lines and white circles) and in Asian American subjects (green lines and circles); cyclosporine IV (C) and oral (D, E) in healthy White American subjects (black lines and white circles) and in African American subjects (orange lines and circles); oral tacrolimus (F, G) in NA White (black lines and white circles) and either African Americans or Hispanic Americans, respectively (blue lines and circles); oral pravastatin (H) in NA White (black line and white circles) and African Americans (purple lines and circles); oral rosuvastatin (I) in NA White (black line) and Asian Americans (blue lines). No observed data points were provided in the clinical study for rosuvastatin. The gray and light‐colored lines represent the 5th and 95th percentiles for the total virtual population. The observed data are from [40, 41, 42, 43, 44, 45]. F, females; M, Males; P/O; predicted to observed ratio of the change in AUC.

Implementing a scalar of 0.8 to the equations for predicting plasma and erythrocyte volumes for the African American and the White American populations was required to recover both the blood, plasma, and erythrocyte volumes for these two populations, given that blood volume is a sum of plasma and erythrocyte volume. The same scalar was applied to the Hispanic_Latino and the Asian American populations, as these also recovered the blood volumes better. No plasma and erythrocyte volume data were available for these two populations (Figure S2). The observed cardiac output for the different populations was well recovered by the simulated data, which was confirmed by visual predictive checks (Figure S8). No data for blood flow to specific organs could be identified for any of the different ethnic groups; therefore, the relative blood flow to each organ was assumed to be the same as the NEC population.

3.1.3. Plasma Proteins and Hematocrit

The derived weighted mean (with CV%) of hematocrit values for the different populations from the data obtained from the NHANES database are 44.7% (7.3%) (males) and 40.5% (7.4%) (females) for the White American population; 43.3% (8.3%) (males) and 38.1% (9.6%) (females) for the African American population; 44.5% (7.3%) (males) and 39.4% (7.7%) (females) for the Asian American population; and 44.9% (7.3%) (males) and 39.4% (8.2%) (females) for the Hispanic_Latino population; whereas the HSA equations are given below (Equations (21), (22), (23), (24), (25), (26), (27), (28)).

African American

$$\text{Male}\mathrm{HSA}=48.07-0.0673\times \mathrm{Age}-0.0895\times \mathrm{BMI}$$ (21)

$$\text{Female}\mathrm{HSA}=43.53+0.0179\times \mathrm{Age}-0.1324\times \mathrm{BMI}$$ (22)

Asian American

$$\text{Male}\mathrm{HSA}=51.24-0.0735\times \mathrm{Age}-0.1545\times \mathrm{BMI}$$ (23)

$$\text{Female}\mathrm{HSA}=47.64+0.0042\times \mathrm{Age}-0.2309\times \mathrm{BMI}$$ (24)

Hispanic_Latino American

$$\text{Male}\mathrm{HSA}=50.58-0.0793\times \mathrm{Age}-0.1254\times \mathrm{BMI}$$ (25)

$$\text{Female}\mathrm{HSA}=46.18-0.0068\times \mathrm{Age}-0.1554\times \mathrm{BMI}$$ (26)

White American

$$\text{Male}\mathrm{HSA}=50.77-0.0728\times \mathrm{Age}-0.1253\times \mathrm{BMI}$$ (27)

$$\text{Female}\mathrm{HSA}=45.98-0.005\times \mathrm{Age}-0.1322\times \mathrm{BMI}$$ (28)

The average and %CV for male AAG levels of 0.77 g/L (26%) from Zhou et al., for the White American sub‐population was comparable to the male value of 0.793 g/L (23%) used in the Simcyp simulator for the NEC population. Thus, the NEC values for males as well as those for females; 0.715 g/L (24%) were assumed for the White American population, as these are based on a larger dataset of 194 males and 114 females from 7 different clinical studies. The weighted mean ratios of AAG levels in the African American to that of the White American population estimated from Johnson and Livingston [12] and Sowinski et al. [13] and that for the Asian American estimated from Zhou et al. [14] were applied to the NEC values to obtain sex‐specific mean data of 0.667 g/L (males) and 0.601 g/L (females) for the African American population; and 0.64 g/L (males) and 0.58 g/L (females) for the Asian American population. Given that no data for AAG levels in Hispanic Americans was obtained in the literature, the Caucasian data were assumed. The same sex‐specific CVs were assumed for all the populations.

3.1.4. Enzyme and Transporter Abundance and Frequency

For the CYP enzymes, no ethnic‐specific data in extensive metabolizer subjects were available for CYP1A1, 1A2, 2A6, 2C18, 2D6, 2J2, 3A4, and 3A7; thus, the NEC values were assumed for all the populations. The estimated mean liver CYP abundances from the literature search for CYP2B6, 2C8, 2C19, and 3A5 for the North American White population are given as 21.09, 26, 3.69, and 108 pmol/mg protein, respectively, which are in line with the NEC values of 21.6, 24, 4.4, and 103 (Table S3). North American White literature data were used for these enzyme abundance estimates, except for CYP2B6, which had high variability in the literature; therefore, NEC values were used instead. For CYP2C9, the mean value of 77.7 pmol/mg protein used in the NEC population was retained for the North American White population, as it was shown to give better recovery of the exposure of the CYP2C9 substrates, S‐warfarin, and celecoxib, than 55.2 pmol/mg protein, which is the mean value obtained from North American White liver samples only (Figure S9) [46, 47, 48].

For the African American population, although absolute abundance values were available for CYP2B6, 2C8, 2C9, 2C19, and 2E1, the sample size of these ethnic specific abundance data was very small (N = 2 or 3 livers) [49, 50, 51]. However, since the mean values were often in the range of those from the White American liver samples, the same input values were assumed for all the CYP enzymes except for CYP3A5, where sufficient data were available to derive a mean liver abundance value of 78.55 pmol/mg protein [52]. Ethnic specific data were lacking for the Hispanic_Latino or the Asian American populations, thus Caucasian data were assumed for Hispanic_Latinos, whereas Chinese data were assumed for Asian Americans. A summary of the CYP abundances used as input values for the North American populations and their references are given in Table S2.

Ethnic‐specific abundance data were lacking for all the non‐CYP enzymes and the transporters. However, work done by Tomita et al. showed that in addition to the differences in the allele frequencies of OATP1B1 in the Asian and non‐Asian populations, a difference in the intrinsic relative activity of OATP1B1 needed to be considered to recover the exposure levels of statins in Asian populations [53]. This relative activity scalar of 0.584 was also applied in the OATP1B1 abundance values for the North American Asian population. All other non‐CYP enzyme and transporter abundances were assumed to be the same as the NEC population.

Intestinal CYP abundances measured in the different North American sub‐populations were not available, hence a similar approach used in the development of the Chinese PBPK model [4] was applied, wherein the ratio of liver CYP abundances in the NEC population, to that in the different North American population was assumed to be the same for the intestinal CYP abundances. For the protein abundances for the other non‐CYP enzymes and transporters, there was a scarcity of ethnic‐specific data across all organs, hence the NEC population values were assumed for all the populations, except the Asian American wherein the Chinese population values were assumed.

Two important sub‐population differences in protein abundances to note between the North American White population and the other North American sub‐populations are (1) The relative decreased amount of OATP1B1 protein in Asian Americans compared to other North American populations and (2) the decreased amount of CYP3A5 protein in African American and Asian American populations.

Some other key sub‐population differences noted from the literature analysis, which have been incorporated in the PBPK models are: increased percentage of CYP3A5 extensive metabolizers in the African American population (72.5% vs. 48.2% in Asian Americans, 43.2% in Hispanic_Latino‐Americans, and 14.3% in White Americans); the increased percentage of OATP1B1 ultrarapid transporters in the African American population (63% vs. 16.7% in Asians, 15% in Hispanic_Latino‐Americans, and 6% in White Americans); and the relatively higher percentage of poor CYP2D6 metabolizers in the White population (5.2% vs. 2% in African American, 0.3% in Asian Americans, and 2.9% in Hispanic_Latino‐Americans).

The most commonly identified CYP2C9 alleles in the White American and Asian American populations are the *1, *2, and *3 alleles. However, additional alleles were also shown to occur in the African American and the Hispanic_Latino populations. These alleles include the *5, *6, *8, and *11 alleles, which also encode a decreased catalytic activity to the CYP2C9 enzyme. Importantly for the African American population, the *5, *6, *8, and *11 CYP2C9 variants were found to be more prevalent than the more widely studied *2 and *3 alleles and are the main constituents of the intermediate phenotype for that population. (A summary of the different phenotype differences for polymorphic enzymes/transporters for the North American populations and their references are summarized in Tables S3–S7).

A comparison of simulated total hepatic CYP enzymes and transporter abundance for the four populations is shown in Figure 2. A statistical analysis of sub‐population differences is included in Section 3 of the Supporting Information. CYP and transporter abundance estimates come from simulations of 1000 individuals, ages 20–50 years, 50% female with the expected phenotype frequencies and average CYP and transporter abundance for each population applied. As such, simulated differences in CYP3A5 abundance between the different populations are clearly shown, notably with all the other sub‐populations having on average a higher frequency of CY3A5 extensive metabolizers compared to Whites. Similarly, the reported higher OATP1B1 abundance in African Americans compared to both Asians and White populations is also observed in the simulations (Figure 2).

FIGURE 2.

Comparison of simulated total hepatic CYP (A) and transporter (B) abundances for African Americans (red), Asians (green), Hispanic_Latino (orange), and White (blue) populations. Plotted are the average and standard deviations of abundance derived from a simulation of 2000 individuals (ages 20–50, 50% female) for each population. Differences in population CYP and transporter abundance, phenotype frequency, and liver volume have been incorporated into the developed population models and used in all the performance verification simulations. See Section 3 of the Supporting Information for individual CYP and transporter comparisons between different North American sub‐populations.

3.1.5. Serum Creatinine and GFR

The derived serum creatinine values (Mean and CV (%)) across discrete age bands for the four different populations are summarized in Table S8. A statistical evaluation of the absolute fold difference between the predicted GFR using different GFR equations against the observed values identified the Cockcroft‐Gault equation 3 [19] for the White American, the modified diet in renal disease (MDRD) equations [20] for the Asian American population, and the CKD‐EPI creatinine equation fit without race for the African American populations respectively as the most suitable for GFR prediction [22]. Given that no observed GFR values were available for the Hispanic_Latino population, the Cockcroft and Gault equation was assumed.

3.1.6. Gastrointestinal Transit Time Parameters

Specific data or studies in which the race of the subjects was stated were less readily available for most of the gastrointestinal parameters. A clinical study [54] carried out in North America comprising a mixed population of African Americans (50%), White Americans (40%) and Asians (10%) reported a fasted gastric mean residence time (MRT) of 0.298 h (CV = 25%), which is in line with the value of 0.4 h (CV = 40%) implemented in the Simcyp simulator for the NEC population. This value was assumed for all the North American populations.

For the fed state, two different clinical studies [55, 56] carried out in Hispanic and non‐Hispanic Whites, who were administered a low‐calorie liquid meal, indicated a lower MRT in the Hispanic_Latino population. The weighted mean values of 1.35 and 1.14 h were used as input values for the White American and Hispanic_Latino populations respectively. The White American value was assumed for the Asian and African American populations. Finally, weighted mean value of 3.18 h for the small intestinal transit time (SITT) was estimated from three studies [57, 58, 59] carried out in North America. This value was used for the four North American populations. All other GIT input parameters were assumed to be the same as those of the NEC population.

3.2. Performance Verification Against Clinical Studies in NA Populations

Twenty clinical studies reporting pharmacokinetic profiles of 13 drugs (N = 110 subjects), whose PBPK models are available in the Simcyp simulator were collected from the DIDB and were identified as having sufficient ethnic‐specific demographic information to enable their inclusion in the performance verification of the developed populations. Only six of these studies reported separate pharmacokinetic profiles, parameters, or both for each of the sub‐ethnic populations making up the demographics of the study. These studies were for alprazolam, cyclosporine, tacrolimus (CYP3A4/5 substrates), pravastatin, and rosuvastatin (substrates for transporters, e.g., OATP1B1/3) (Figure 3). To replicate these studies, the number of individuals in each NA sub‐population, sex and age or age range of the study participants were used to inform the study design of the virtual population(s) using the developed NA populations.

Studies with reported average profiles in mixed populations were also simulated for other drugs, to cover other metabolic pathways such as metformin (CYP3A4 and OCT1/2 and MATE substrate), warfarin (CYP2C9 substrate), midazolam (substrate for CYP3A4/5 and UGT1A4), bupropion (CYP 2B6, CYP3A4, and CYP2C19 substrate), flurbiprofen (CYP2C9 and UGT2B7), S‐mephenytoin (CYP2C19), efavirenz (CYP2B6, CYP1A2, CYP2A6, and CYP3A4), carbamazepine (CYP3A4/5, CYP2C8, and UGT2B7), pravastatin (renal excretion with transporters), rosuvastatin (renal excretion with transporters), and atorvastatin (CYP3A, CYP2C8, and transporters). The PBPK models for all these compounds are already available within the Simcyp simulator, and their predictive performance using the developed North American populations was assessed against the observed data (Figure 4). All the predictions were within 2‐fold of the observed data. Details on the design of each study (Number of subjects, proportion of males and females, age range), the percentage contribution of each sub‐ethnic population to the overall number of subjects, and the predicted and observed mean AUC are summarized in Table 1.

FIGURE 4.

Mean and standard deviations of the area under Plasma concentration time curves for warfarin (with and without Fluc; fluconazole), flurbiprofen (with and without fluconazole), Bupropion (with and without Cim; cimetidine), metformin, midazolam, S‐mephenytoin, CBZ; carbamazepine, and EFV; Efavirenz either alone or interacting with each other, Pravastatin, Rosuvastatin, and Ator; Atorvastatin. Numbering indicates that multiple studies are available for that substrate. The simulation results using the different NA; North American Populations (open circles) were compared to observed data (closed black circles). The trial design was matched to the collated clinical studies with respect to number of individuals, age range, and proportion of females. For each simulation, 10 trials were applied with the number of subjects in each trial matching the clinical study. The references to the clinical studies where the observed data were obtained from are given in Table 1. *Refers to studies where geometric mean values are reported.

3.3. A Case Study to Demonstrate the Impact of Ethnic Differences on PK

As stated previously, proportions of OATP1B1 polymorphisms varied across the four populations, resulting in a difference in repaglinide disposition. The fraction transported via OATP1B1 ranged from 81.19% in the North American Asian sub‐population to 89.27% in the North American African American sub‐population; because of the higher proportion of OATP1B1 ultra‐rapid transporters in the latter, leading to a higher hepatic liver abundance of OATP1B1 in the African American population compared to the other NA sub‐populations. Added to this is the reduced functional activity of the OATP1B1 uptake transporter in the Asian American population. The remaining percentage represents the proportion of drug in the systemic concentration entering the liver through passive diffusion.

The differences in the total hepatic liver abundances of CYP2C8 and CYP3A4 across the four populations (Figure 2A) also resulted in variations in their in vivo fractions contributing to drug metabolism (fms) for repaglinide in the NA Asian population compared to the other three populations. The simulated repaglinide mean population fms for CYP2C8 and CYP3A4 was approximately 67% and 33% for the other three sub‐populations, whereas that for the North American Asians was 70% and 30%, respectively. This is because of the higher CYP2C8 enzyme abundance but lower CYP3A4 enzyme abundance in the Asian American sub‐population compared to the others.

These differences in total liver enzyme and transporter activity translate to differences in the oral clearance of the drug, with more than 2‐fold difference between that of the Asian and African American populations, resulting in an overall higher systemic exposure of the drug in the Asian Americans compared to the other ethnicities (Figure 5). Despite having the same in vivo fms as the NA White and Hispanic_Latino populations, the higher OATP1B1 hepatic uptake transporter activity facilitated a much higher repaglinide clearance in the African American population. Thus, simulating increasing doses of the drug showed that a dose of 2 mg repaglinide in the African American population produced an average exposure that was still lower than that of a dose of 1 mg in the NA Asian population.

FIGURE 5.

Simulated mean (±SD) steady state AUC of repaglinide after multiple doses of 0.25, 0.5, 1, and 2 mg using the four North American populations (A) Simulated AUC and C _max ratios of repaglinide at steady state, after co‐administration with gemfibrozil (B) and rifampicin (C). The trial design settings used for the simulations for the four populations are 200 virtual individuals, 50% females, 20–50 years. For the DDI simulations, 0.25 mg repaglinide was administered QID, alongside either 600 mg gemfibrozil QID or 600 mg rifampicin QID for 5 days.

The impact of the differences in PK is also reflected in the outcome of the power calculation. When the Asian American population was compared to the White American population, an inclusion of about 15 Asian subjects into a clinical study would be sufficient to detect a statistical difference (p < 0.05) between the two populations with a minimum of 80% study power. This number increased to 35 for the African American population compared to North American Whites. In contrast, it would be impossible to detect a PK difference between the Hispanic_Latino and White American populations unless the sample size is increased to approximately 5000 subjects.

After both single and multiple dose administration of repaglinide with gemfibrozil, the African American sub‐population gave the highest DDI liability, as shown by its C _max and AUC ratios of 2.48 and 5.52, and 3.23 and 7.67 after single and multiple doses respectively (Figure 5), with the Asian sub‐population having the lowest. This is because the DDI liability of gemfibrozil as a potent CYP2C8 inhibitor is directly proportional to the population with the highest amount of repaglinide available for clearance, which in this case is the African American population because of having higher repaglinide uptake into the liver. After single dose administration of rifampicin, the DDI liability is mainly driven by its OATP1B1 inhibition; thus, the African American population is that which is most affected, with a C _max and AUC ratio of 3.75 and 6.21 respectively. However, after multiple doses, rifampicin's induction of all the CYP elimination pathways becomes more important [60]; hence, there is a comparable decrease in both AUC and C _max ratios across all four ethnic populations, with the African American population still having the highest DDI liability.

4. Discussion

Ethnic differences and the physiological and socio‐cultural factors linked to these have been shown to contribute to the inter‐individual variabilities that are reported in the kinetics and therapeutic effect of various drugs [61]. Clinical trials conducted in North America constitute a major proportion of clinical studies worldwide [62]; and the huge diversity of the American population will translate to a high variability in drug kinetics and response. It is therefore important to account for biological and physiological differences among various North American sub‐populations. This work describes efforts to develop specific population PBPK models for some of the major ethnic groups in North America to facilitate drug development and possibly allow the tailoring of the dose required within this multi‐ethnic region.

Ethnicity and race are often used interchangeably, although some studies distinguish them from each other. Race typically refers to a person's biological and physical characteristics, such as their skin color, hair texture, and facial features [63]; however, it should be noted that race is a social construct. In North America, racial categories often include White, Black or African American, Asian, American Indians, Native American, or Alaska Native, and Native Hawaiian or Pacific Islander. Ethnicity, however, encompasses cultural factors such as nationality, language, religion, and customs. Thus, common ethnicities include Hispanic or Latino, Chinese, Irish, Italian, and Jewish. There are varied approaches taken in different clinical pharmacokinetic studies to classify recruited subjects on the basis of ethnicity and race. For example, some studies make a distinction of Hispanic White from non‐Hispanic populations, with further racial categorization within the non‐Hispanic group as African, White, and Asian. This, therefore, implies that Hispanic blacks or Afro‐Latino are sometimes categorized as African Americans or Blacks. Nonetheless, it is important to point out that when “race” and/or “ethnicity” are used in stratifying populations, they not only consider the genetic differences between these populations, but also incorporate other physiological, demographic and socio‐environmental factors. Genetic ancestry alone cannot disentangle socio‐environmental factors (e.g., diet, healthcare access, and environmental exposures) associated with these population groups, that can also influence PK outcomes. Therefore, accounting for race and ethnicity differences is still important.

In the current study, the North American populations were categorized on the basis of the NHANES database's classification into White, African American, Asian, and Hispanic_Latino. Those who identified as Mexican in the database were included under the Hispanic_Latino population. For clinical studies used for population development and/or performance verification, if subjects were identified as Hispanic or Mexican, they were assigned to the Hispanic_Latino population, whereas those identified as non‐Hispanic White were assigned to the North American White population. Subjects classified as African American (or Black) and Asian were assigned as such.

Unfortunately, there were few studies with reported information for the physiological parameters required for developing population PBPK models for the Hispanic_Latino population. Robust data for developing this sub‐population were mainly available for those input parameters that were obtained from the NHANES database, hence the assumption to use the North American White values as a surrogate for input parameters where Hispanic_Latino specific data were missing. This assumption is not “unjustified”, given that equations for predicting a specific physiological parameter derived using data from a given population has been shown to predict that same parameter in a completely different population, as long as the key population‐specific demographic information is used. This has been shown for the liver volume equation derived using Caucasian data [64], which adequately predicted the liver volume of all the North American populations, (Figure S1) and that of the Japanese population [65]. In addition, the same cardiac output equation derived using only healthy Caucasian data predicted well the observed cardiac output across the four sub‐populations (Figure S8).

Thus, the compilation of each population's demographics, cardiac output, organ volumes, CYP, and transporter abundance and phenotype frequencies where available provides a substantial basis for developing physiologically based pharmacokinetics models for predicting drug disposition differences (e.g., absorption, distribution, and clearance) in the four North American populations. Importantly, the database of values also allows us to simulate the expected central tendency and variability in pharmacokinetics for each population. As of the writing of this manuscript, data for the White and African American populations were the most robust among the four sub‐populations.

Further research for Asian and Hispanic_Latino population, particularly for CYP and transporter abundance, would aid in further enhancements of the population models. An additional future outlook is to extend these models to include other ethnicities in North America, depending on the availability of data, to ensure a more comprehensive understanding of pharmacokinetics across diverse populations. Our analysis was limited by what the studies reported as “race” and/or “ethnicity” for the subjects. Ideally, the additional information of genetic ancestry, especially as it applies to the Hispanic_Latino and the non‐Hispanic populations could provide a better understanding of the interplay of genetic and environmental factors on population PK variability.

The very limited number of pharmacokinetic studies which reported the individual concentration‐time profiles or the PK parameters for each ethnic group was one of the limitations of the study. This information, if available, would have been useful for sufficient verification of the developed populations, and for identifying and then guiding dose adjustments for each sub‐population, where necessary. Nonetheless, despite the limited data available, especially for the North American Asian and Hispanic_Latino populations, clinical studies in which individual concentration‐time profiles or the PK parameters for each sub‐population were reported were well recovered by the PBPK models. This includes the decrease in tacrolimus exposure in the Hispanic_Latino population and the increased exposure of alprazolam in the Asian population compared to the White sub‐population (Figure 3). The predicted changes in PK from that of the North American White population as control, including those for the African American population were in line with the observed data, with all the predicted/observed ratios within 2‐fold (Figure 3). Extending the dataset to include studies with mixed populations (Table 1), which were also adequately predicted (Figure 4) provides additional verification of the developed population models.

Modeling and simulation using PBPK models can help fill the gaps in the lack of ethnic‐specific clinical data and guide labeling on the basis of verified physiological changes and predicted PK differences across populations. This is particularly important for drugs in which a large number of subjects would need to be recruited into the clinical study before a clear difference in PK between two sub‐populations might be identified, as shown with the power calculation estimated for repaglinide between the NA White and the African American population. It is, however, important to note that the power calculation estimate is based on a comparison of the mean PK (such as the AUC) between the two populations. However, there is often considerable overlap in PK characteristics among different ethnic groups. Therefore, identifying individuals with PK values at the extremes of the population distribution may be more critical than merely comparing mean values [66]. Given that clinical studies do not often recruit individuals that constitute the extremes of the population, PBPK modeling serves as a valuable tool in this context.

Specifically, in all the population simulations of 200 subjects, there were individuals outside the 5th and 95th percentiles of the population distribution, with more than a 5‐fold difference in their steady state concentrations compared to the simulated mean value of the population. This is partly because variant genes are not necessarily restricted to specific sub‐populations or ethnicities, even though they may be more prevalent in one population than in the other. In addition, even within the same ethnicity, the variability in physiological parameters can bring about a marked PK difference.

Although genotyping provides valuable insights into genetic factors influencing drug metabolism, it alone cannot account for the full spectrum of individual PK variability. Factors such as demographic differences, variations in blood flow, and distinct metabolic pathways also play critical roles in determining drug response among individuals within and across ethnic groups. Therefore, ethnicity can serve as a useful proxy for investigating potential differences in drug response, particularly when individual genotype information is unavailable [63]. Thereafter, the use of PBPK modeling can then be used to predict these PK outcomes and not only identify at‐risk individuals within each population, but also help to inform the clinical trial design using a study power calculation.

In conclusion, the developed four population files that cover various ethnic and racial groups in North America are believed to help not only in drug development but also in personalized medicine for North American patients. For drug development, understanding the influence of various external and internal variabilities on drug pharmacokinetics and response is advocated in various regulatory guidelines. Including these changes within a PBPK model will increase the ability to capture variabilities in drug kinetics. On the other hand, a robust PBPK model that accounts for these ethnic differences can help in guiding decision makers on the best dose suitable for each individual patient and assist in avoiding drastic drug interaction consequences.

Author Contributions

U.E., J.D., and E.E.‐K. wrote the manuscript; U.E., A.B., O.H., and I.G. designed the research; U.E., J.D., E.E.‐K., A.B., and O.H. performed the research and analyzed the data. All authors reviewed the manuscript and approved the submitted version.

Conflicts of Interest

All the authors were Certara UK Limited employees at the time of the work and may hold shares in Certara.

Supporting information

Data S1

Data S2