Evaluation of Cannabis Per Se Laws: A Semi‐Mechanistic Pharmacometrics Model for Quantitative Characterization of THC and Metabolites in Oral Users

Abstract

Recreational cannabis use has increased notably in the United States in the past decade, with a recent surge in oral consumption. This trend has raised concerns about driving under the influence. Current cannabis‐impaired driving laws lack standardization, with some states implementing blood Δ9‐tetrahydrocannabinol (THC) per se limits (1, 2, and 5 ng/mL). However, these limits have been criticized for their inaccuracy and unreliability, highlighting the need for legal refinement. Addressing this issue requires understanding the complex pharmacokinetics (PK) and pharmacodynamics (PD) of THC, cannabis's primary psychoactive component, which can be characterized using a population PK model. However, existing PK models mainly focus on inhalation data and do not account for the growing number of oral cannabis users. To bridge this gap, a semi‐mechanistic population PK model was developed using data from 10 published studies following intravenous or oral administration of cannabis to characterize THC and its metabolites in oral users. Simulated THC plasma concentrations for doses from 2.5 mg to 100 mg in frequent and occasional users were used to evaluate the effectiveness of existing per se limits. Results showed that the 1 ng/mL limit was least effective due to a high risk of false positives, while the 2 and 5 ng/mL limits remain inconclusive due to limited PD data linking blood THC levels to impairment. These findings suggest that the existing per se laws may not fully address the complexity of cannabis impairment, underscoring the need for further research and refinement of cannabis‐impaired driving laws.

Introduction

Since the relaxation of legal restrictions on cannabis in the United States beginning in 1996, there has been a marked increase in its social acceptance, accessibility, and consumption. ¹ This shift has contributed to a notable rise in recreational cannabis use. ² Traditionally, cannabis has been predominantly consumed through inhalation, such as smoking and vaping. However, in recent years, oral administration has gained considerable popularity, becoming the second most utilized route after inhalation, largely due to its more pronounced pharmacological effects and the discreet nature of its administration. ³ , ⁴ , ⁵

Regardless of the route of administration, the primary psychoactive constituent in cannabis is delta‐9‐tetrahydrocannabinol (THC). Following a cannabis intake, THC undergoes hepatic metabolism via cytochrome P450 (CYP) enzymes, ⁶ , ⁷ resulting in the formation of its equipotent psychoactive metabolite, ⁸ 11‐hydroxy‐Δ9‐tetrahydrocannabinol (11‐OH‐THC). This metabolite is subsequently converted into its inactive form, 11‐nor‐9‐carboxy‐Δ9‐tetrahydrocannabinol (THC‐COOH). The chemical properties of THC and its metabolites are summarized in Table S1.

Despite the potential therapeutic applications of THC, ⁹ , ¹⁰ a significant number of individuals consume cannabis primarily for recreational purposes. THC exerts physiological effects by binding to cannabinoid receptors (primarily CB1 and CB2) in the brain and body. ¹¹ , ¹² Consequently, while THC can induce euphoria or the sensation of being “high,” its impact on brain chemistry can result in side effects such as reduced coordination, impaired sensory processing, and delayed reaction time—all critical skills required for safe driving. This has contributed to the growing incidence of cannabis‐related driving under the influence (DUI) cases.

A report by Pearlson et al suggested a potential association between the rising incidence of cannabis use and the increase in cannabis DUI. ¹³ Notably, cannabis is the second most frequently detected substance in the bodies of drivers involved in fatal motor vehicle accidents following alcohol. ¹⁴ However, unlike alcohol, there is no standardized test equivalent to the blood alcohol concentration (BAC) for detecting cannabis impairment. Although both alcohol and cannabis contain psychoactive substances—ethanol and THC, respectively—their pharmacokinetic properties differ markedly. Ethanol demonstrates zero‐order elimination with a clear and direct correlation between blood concentration and intoxication. ¹⁵ In contrast, THC presents several unique pharmacokinetics (PK) and pharmacodynamics (PD) challenges, such as route‐dependent PK, usage‐pattern‐dependent PK, and PK–PD disconnection. These three complex PK and PD behaviors make it difficult to rely solely on THC blood concentration to assess impairment.

Largely due to these challenges, there is no standardized cannabis DUI law across the United States as of October 2024. While some states enforce zero‐tolerance policies, six states—Illinois, Montana, Nevada, Ohio, Pennsylvania, and Washington—have implemented per se cannabis DUI laws, setting THC whole blood concentration limits of 1 ng/mL (Pennsylvania), 2 ng/mL (Nevada and Ohio), or 5 ng/ mL (Illinois, Montana, and Washington) as thresholds for impairment. ¹⁶ Under the per se laws, exceeding a specified THC blood concentration serves as sufficient legal evidence for a cannabis DUI charge, regardless of demonstrable impairment. ¹⁶ Therefore, ensuring the accuracy and validity of the existing THC per se limits is critical from a legal perspective. Multiple studies, however, suggested that these per se limits lack scientific basis and perform poorly in correlating THC blood levels with actual impairment. ¹⁷ , ¹⁸ Thus, there is a pressing need to refine current cannabis DUI regulations, which cannot be accomplished without addressing the THC PK and PD complexities. This requires a detailed analysis that considers various scenarios, including different routes of administration, frequencies of use, and types of biomarkers being measured.

One promising approach to address these challenges is population‐based pharmacokinetic modeling. However, most current PK models focus on inhaled THC and often rely on limited sampling periods, ¹⁹ , ²⁰ which may not fully reflect the broader patterns of cannabis use, especially among the growing number of oral users. In this study, we present a semi‐mechanistic pharmacometrics model developed using data from 10 published human cannabis studies to quantitatively characterize the disposition of THC and its metabolites in oral cannabis users, incorporating an extended sampling period. The primary objective of this study was to assess the effectiveness of the existing per se limits for occasional and frequent oral cannabis users, where the effectiveness of per se limits is evaluated based on the assumptions of the “Impairment Window” and “Minimal Impairment Dose”, which are thoroughly explained in the Methods section. Ultimately, this study aimed to provide data‐driven insights that could inform the refinement of cannabis‐impaired driving regulations.

Methods

Data Source

The data used for population PK model development was derived from 10 published THC human studies. ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ These studies that were granted IRB or ethics approvals and obtained informed consent have documented this information in their respective articles. A detailed summary of these studies is provided in Table S2. The included studies comprised seven single oral dose studies (ID 1a–8) and three single IV dose studies (ID 1b, 9, and 10), covering a broad range of doses: oral doses ranged from 5 to 90 mg, and IV doses ranged from 1.6 to 5 mg. The sampling times in the final dataset spanned from 4 to 54 h.

In studies ID 4 and 5, data from both occasional and frequent cannabis users were reported, but only the occasional user data were included. Study ID 6 evaluated both fed and fasted conditions, only the data from the fasted condition were selected. Most studies reported concentrations for THC, 11‐OH‐THC, and THC‐COOH, while studies ID 3 and 6 reported only THC and 11‐OH‐THC, and study ID 1 reported only THC concentrations. Four studies (ID 2, 4, 5, and 7) reported concentrations in whole blood, and these were converted to plasma concentrations using conversion factors of 0.71 for THC, 0.73 for 11‐OH‐THC, and 0.65 for THC‐COOH based on the experimental blood‐to‐plasma ratios reported in the literature. ³¹ As shown in Table S1, due to the minimal differences in molecular weight observed between THC and its metabolites (less than 5% per metabolism step), the final dataset opted to use weight‐based concentrations rather than molarity. All plasma or whole blood concentration–time curves were digitized using Engauge Digitizer 12.1 (https://markummitchell.github.io/engauge‐digitizer/).

Population Pharmacokinetic Modeling

All PK data for THC and its metabolites from the published human studies were analyzed simultaneously using the nonlinear mixed‐effects modeling with NONMEM (version 7.4.3; Icon Development Solutions, Ellicott City, MD) interfaced with Pirana (version 2.9.9, http://www.pirana‐software.com/). The first‐order conditional estimation method with interaction (FOCEI) with a user‐defined subroutine ADVAN14 was employed for model parameterization. As the data from each study were reported as mean values, inter‐individual variability could not be assessed. However, inter‐study variability was incorporated to account for differences across studies. The residual variability (RV) was evaluated. SigmaPlot 14.0 (Systat Software, San Jose, CA) was used for data handling and graphical analyses.

Structural Model Evaluated

Various pharmacokinetic model structures were explored during the model‐building process, including two‐, three‐, and four‐compartment models with first‐order absorption and elimination. A base structure of a THC parent‐metabolite model was found to appropriately characterize THC and its metabolites’ disposition in oral route. In this base structure, THC and 11‐OH‐THC were each represented by three compartments, while THC‐COOH was modeled with two compartments. This base model was further refined to incorporate THC‐specific PK properties for the best model fitting and parameterization. A key model development history can be found in Table S3. The following are several representative models, and these model structures are presented in Figure S1.

Model Featuring Pre‐Systemic Metabolism Rate Constant (Figure S1A )

In addition to the base structure, this model emphasized THC's first‐pass metabolism via a pre‐systemic metabolism rate constant (Kgut). Orally consumed THC first reached the gastrointestinal tract, represented by THC depot compartment. From the depot, a portion of THC underwent the first‐pass metabolism before reaching the systemic circulation where THC was metabolized to 11‐OH‐THC. While the remainder of the THC was absorbed (Ka) to the system and followed the same metabolic pathway. After 11‐OH‐THC was formed, it was further metabolized into THC‐COOH and subsequently eliminated from the body.

Model Featuring mPBPK (Figure S1B )

A minimal physiologically based pharmacokinetic (mPBPK) model was used to describe the disposition of THC, while a compartmental model was applied to capture the PK of 11‐OH‐THC and THC‐COOH. The THC mPBPK structure consisted of five compartments: (i) the central THC compartment, where oral THC was absorbed, (ii) a high‐perfused tissue compartment representing highly vascularized organs such as the brain, liver, and kidneys, (iii) a low‐perfused tissue compartment representing less vascularized organs like skin and resting muscles, (iv) an interstitial fluid compartment linking adipose tissue, and (v) an adipose tissue compartment where THC was stored and slowly released. The compartmental model for 11‐OH‐THC included three compartments, with one peripheral compartment linked to adipose tissue, while THC‐COOH was modeled using a two‐compartment structure.

Model Featuring Adipose Tissue Disposition (Final Model) (Figure  1 )

Figure 1.

Final population PK model describing the pharmacokinetics of THC and its metabolites. Oral THC was absorbed from the depot compartment through first‐order absorption rate constant (Ka). From THC central compartment (C_THC, V_THC), THC could undergo distribution to two peripheral compartments (C_THC,peri1, VTH_C,peri1 & C_THC,peri2, V_THC,peri2) with distribution flows (Q1 and Q2) and further distributed to adipose tissue (C_THC,adipose, V_THC,adipose) via permeability surface area product (PS). Then, THC molecules were metabolized (CL_THC) to 11‐OH‐THC (C_{11‐OH‐THC}, V_{11‐OH‐THC}), where 11‐OH‐THC molecules also distributed to its peripheral compartments (C_{11‐OH‐THC,peri1}, V_{11‐OH‐THC, peri1} & C_{11‐OH‐THC,peri2}, V_{11‐OH‐THC, peri2}) via distribution flows Q3 and Q4. The 11‐OH‐THC molecules could also distribute to adipose tissue (C_{11‐OH‐THC,adipose}, V_{11‐OH‐THC,adipose}) by its permeability surface area product (PS2). The 11‐OH‐THC molecules were then metabolized (CL_{11‐OH‐THC}*Fm) to THC‐COOH and excreted (CL_{11‐OH‐THC}*(1‐Fm)) unchanged renally. THC‐COOH (C_THC‐COOH, V_THC‐COOH) molecules could distribute to its peripheral compartment (C_{THC‐COOH,peri}, V_{THC‐COOH,peri}) by distribution flow Q5 and were cleared (CL_THC‐COOH) out of human body.

In this model, the compartment model structure was employed. This model highlighted the adipose tissue disposition of both THC and 11‐OH‐THC by incorporating an adipose tissue compartment into each peripheral compartment of THC and 11‐OH‐THC. In this final model, the sequential metabolism of THC and 11‐OH‐THC was represented through their respective clearances (CL_THC, CL_{11‐OH‐THC}). THC‐COOH, the final metabolite, was formed hepatically from 11‐OH‐THC at a rate determined by its clearance multiplied by the fraction metabolized (CL_{11‐OH‐THC} *Fm). Each of THC and 11‐OH‐THC had two peripheral compartments, one of which connected to the adipose compartment. The tissue‐to‐plasma partition coefficient (Kp) was utilized to model the partitioning of drug molecules between plasma and adipose tissue. THC‐COOH was represented with two compartments and its own clearance (CL_THC‐COOH). This represented the final model. The differential equations used to characterize this comprehensive parent‐metabolite population PK model are as follows:

$$\begin{array}{cc}& {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC},\mathrm{depot}}}{\mathrm{dt}}=-{\mathrm{k}}_{\mathrm{a}}\times {\mathrm{A}}_{\mathrm{THC},\mathrm{depot}}}\\ & {\mathrm{A}}_{\mathrm{THC},\mathrm{depot}}\left(0\right)=\mathrm{Dose}\end{array}$$ (1)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC},\mathrm{cen}}}{\mathrm{dt}}={\mathrm{k}}_{\mathrm{a}}\times {\mathrm{A}}_{\mathrm{THC},\mathrm{depot}}-\left({\mathrm{CL}}_{\mathrm{THC}}+{\mathrm{Q}}_1+{\mathrm{Q}}_2\right)}\hfill \\ & & \times {\mathrm{C}}_{\mathrm{THC},\mathrm{cen}}+{\mathrm{Q}}_1\times {\mathrm{C}}_{\mathrm{THC},\mathrm{peri}1}+{\mathrm{Q}}_2\times {\mathrm{C}}_{\mathrm{THC},\mathrm{peri}2}\hfill \\ & & {\mathrm{A}}_{\mathrm{THC},\mathrm{cen}}\left(0\right)=0\hfill \end{array}$$ (2)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{cen}}}{\mathrm{dt}}={\mathrm{CL}}_{\mathrm{THC}}\times {\mathrm{C}}_{\mathrm{THC},\mathrm{cen}}}\hfill \\ & & -\left({\mathrm{CL}}_{11-\mathrm{OH}-\mathrm{THC}}+{\mathrm{Q}}_3+{\mathrm{Q}}_4\right)\times {\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{cen}}\hfill \\ & & +{\mathrm{Q}}_3\times {\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}1}+{\mathrm{Q}}_4\times {\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}2}\hfill \\ & & {\mathrm{A}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{cen}}\left(0\right)=0\hfill \end{array}$$ (3)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{cen}}}{\mathrm{dt}}={\mathrm{CL}}_{11-\mathrm{OH}-\mathrm{THC}}\times \mathrm{Fm}}\hfill \\ & & \times {\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{cen}}-\left({\mathrm{CL}}_{\mathrm{THC}-\mathrm{COOH}}+{\mathrm{Q}}_5\right)\hfill \\ & & \times {\mathrm{C}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{cen}}+{\mathrm{Q}}_5\times {\mathrm{C}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{peri}}\hfill \\ & & {\mathrm{A}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{cen}}\left(0\right)=0\hfill \end{array}$$ (4)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC},\mathrm{peri}1}}{\mathrm{dt}}={\mathrm{Q}}_1\times \left({\mathrm{C}}_{\mathrm{THC},\mathrm{cen}}-{\mathrm{C}}_{\mathrm{THC},\mathrm{peri}1}\right)}\hfill \\ & & {\mathrm{A}}_{\mathrm{THC},\mathrm{peri}1}\left(0\right)=0\hfill \end{array}$$ (5)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC},\mathrm{peri}2}}{\mathrm{dt}}={\mathrm{Q}}_2\times \left({\mathrm{C}}_{\mathrm{THC},\mathrm{cen}}-{\mathrm{C}}_{\mathrm{THC},\mathrm{peri}2}\right)}\hfill \\ & & -\mathrm{PS}\times \left({\mathrm{C}}_{\mathrm{THC},\mathrm{peri}2}-\frac{{\mathrm{C}}_{\mathrm{THC},\mathrm{adipose}}}{{\mathrm{K}}_{\mathrm{P}}}\right)\hfill \\ & & {\mathrm{A}}_{\mathrm{THC},\mathrm{peri}2}\left(0\right)=0\hfill \end{array}$$ (6)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC},\mathrm{adipose}}}{\mathrm{dt}}=\mathrm{PS}\times \left({\mathrm{C}}_{\mathrm{THC},\mathrm{peri}2}-\frac{{\mathrm{C}}_{\mathrm{THC},\mathrm{adipose}}}{{\mathrm{K}}_{\mathrm{P}}}\right)}\hfill \\ & & {\mathrm{A}}_{\mathrm{THC},\mathrm{adipose}}\left(0\right)=0\hfill \end{array}$$ (7)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}1}}{\mathrm{dt}}={\mathrm{Q}}_3}\hfill \\ & & \times \left({\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{cen}}-{\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}1}\right)\hfill \\ & & {\mathrm{A}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}1}\left(0\right)=0\hfill \end{array}$$ (8)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}2}}{\mathrm{dt}}={\mathrm{Q}}_4}\hfill \\ & & \times \left({\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{cen}}-{\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}2}\right)\hfill \\ & & -{\mathrm{PS}}_2\times \left({\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}2}-\frac{{\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{adipose}}}{{\mathrm{K}}_{\mathrm{P}2}}\right)\hfill \\ & & {\mathrm{A}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}2}\left(0\right)=0\hfill \end{array}$$ (9)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{adipose}}}{\mathrm{dt}}={\mathrm{PS}}_2}\hfill \\ & & \times \left({\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{peri}2}-\frac{{\mathrm{C}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{adipose}}}{{\mathrm{K}}_{\mathrm{P}2}}\right)\hfill \\ & & {\mathrm{A}}_{11-\mathrm{OH}-\mathrm{THC},\mathrm{adipose}}\left(0\right)=0\hfill \end{array}$$ (10)

$$\begin{array}{ccc}& & {\displaystyle \frac{{\mathrm{dA}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{peri}}}{\mathrm{dt}}={\mathrm{Q}}_5}\hfill \\ & & \times \left({\mathrm{C}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{cen}}-{\mathrm{C}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{peri}}\right)\hfill \\ & & {\mathrm{A}}_{\mathrm{THC}-\mathrm{COOH},\mathrm{peri}}\left(0\right)=0\hfill \end{array}$$ (11)

In our model structure, state variables are denoted using a general format: ${\mathrm{A}}_{\mathrm{drug},\mathrm{compartment}}$, where “A” denotes the amount. For compartments, “cen” is presented for central compartment, and “peri” is presented for peripheral compartment. The PK parameters follow a similar naming convention, with detailed definitions provided in Table 1.

Table 1.

Estimated Parameters from the Final Population PK Model

Parameters	Unit	Definition	Estimates	RSE%	Shrinkage%
THC
F		Oral bioavailability of THC	0.144	1
Ka	h−1	Absorption rate constant	1.46	0.1
VTHC	L	Volume of distribution of THC	204	186.3
VTHC,peri1	L	Volume of distribution of THC peripheral compartment 1	30.2	0.3
VTHC,peri2	L	Volume of distribution of THC peripheral compartment 2	88.3	8.2
VTHC,adipose	L	Volume of distribution of THC adipose pool	91.8	<0.1
CLTHC	L/h	Clearance of THC	87	2.6
Q1	L/h	Distribution flow from THC central to peripheral compartment 1	59.9	0.9
Q2	L/h	Distribution flow from THC central to peripheral compartment 2	29.7	<0.1
PS	L/h	Permeability surface area product of THC	5.5	<0.1
KP		Tissue to plasma partition coefficient for THC	14	0.1
ω2 VTHC		Variance of inter‐study variability on VTHC	1.19	<0.1	0.1
ω2 CLTHC		Variance of inter‐study variability on CLTHC	0.506	<0.1	7.2
σ2 prop,THC		Proportional variance of residual variability of THC	0.375	<0.1	12.6
11‐OH‐THC
V11‐OH‐THC	L	Volume of distribution of 11‐OH‐THC	49.1	1.2
V11‐OH‐THC,peri1	L	Volume of distribution of 11‐OH‐THC peripheral compartment 1	9.95	<0.1
V11‐OH‐THC,peri2	L	Volume of distribution of 11‐OH‐THC peripheral compartment 2	84.3	1.8
V11‐OH‐THC,adipose	L	Volume of distribution of 11‐OH‐THC adipose pool	98.7	<0.1
CL11‐OH‐THC	L/h	Clearance of 11‐OH‐THC	80.4	0.1
Q3	L/h	Distribution flow from 11‐OH‐THC central to peripheral compartment 1	19.9	<0.1
Q4	L/h	Distribution flow from 11‐OH‐THC central to peripheral compartment 2	47.3	0.7
PS2	L/h	Permeability surface area product of 11‐OH‐THC	5.11	<0.1
KP2		Tissue to plasma partition coefficient for 11‐OH‐THC	9.87	<0.1
ω2 V11‐OH‐THC		Variance of inter‐study variability on V11‐OH‐THC	1.08	<0.1	0.1
σ2 prop,11‐OH‐THC		Proportional variance of residual variability of 11‐OH‐THC	0.31	<0.1	9.1
THC‐COOH
VTHC‐COOH	L	Volume of distribution of THC‐COOH	0.986	0.1
VTHC‐COOH,peri	L	Volume of distribution of THC‐COOH peripheral compartment	27.6	0.1
CLTHC‐COOH	L/h	Clearance of THC‐COOH	4.34	0.3
Q5	L/h	Distribution flow from THC‐COOH central to peripheral compartment	6.35	<0.1
Fm		Fraction of 11‐OH‐THC metabolized to THC‐COOH	0.904	0.6
ω2 VTHC‐COOH		Variance of inter‐study variability on VTHC‐COOH	0.846	<0.1	37.5
σ2 prop,THC‐COOH		Proportional variance of residual variability of THC‐COOH	0.144	<0.1	0.1

Stochastic Model Evaluated

Inter‐study variability was assessed using an exponential model:

$${\mathrm{P}}_{\mathrm{i}}=\mathrm{TVP}\times \exp \left({\eta }_{\mathrm{i}}\right)$$ (12)

where ${\mathrm{P}}_{\mathrm{i}}$ represented the PK parameter for the ith individual, $\mathrm{TVP}$ denoted the population mean value of the parameter, and ${\eta }_{\mathrm{i}}$ represented the inter‐study variability. The ${\eta }_{\mathrm{i}}$ terms were assumed to follow a normal distribution with a mean of 0 and a variance of ω².

Residual variability was evaluated using a combined proportional and additive error model:

$${\mathrm{C}}_{\mathrm{ij}}={\overline{\mathrm{C}}}_{\mathrm{ij}}\times \left(1+\mathrm{E}{1}_{\mathrm{ij}}\right)+\mathrm{E}{2}_{\mathrm{ij}}$$ (13)

where ${\mathrm{C}}_{\mathrm{ij}}$ denoted the observed concentration of THC and its metabolites for the ith individual at time j, ${\overline{\mathrm{C}}}_{\mathrm{ij}}$ represented the predicted concentration, $\mathrm{E}{1}_{\mathrm{ij}}$ was the proportional error, and $\mathrm{E}{2}_{\mathrm{ij}}$ was the additive error. Both $\mathrm{E}{1}_{\mathrm{ij}}$ and $\mathrm{E}{2}_{\mathrm{ij}}$ were assumed to follow a normal distribution with a mean of 0 and a variance of ω².

Model Evaluation

The final model selection was based on the listed criteria: (i) successful model convergence with reasonable estimation of the parameter precision, (ii) physiologically plausible parameter and error estimates, (iii) concordance between population‐predicted and observed concentrations, and (iv) improvement in goodness‐of‐fit plots. The likelihood ratio test was utilized for comparing nested models, with a statistically significant improvement defined as a decrease in the NONMEM objective function value (OFV = −2 log‐likelihood) of 3.84 (α = 0.05). For non‐nested model comparisons, model selection was based on a reduction of at least two points in the Akaike information criterion (AIC = −2 log‐likelihood + number of parameters).

Model Simulation

THC Multiple Oral Dose Simulation for Model Validation

A cannabis human clinical study by Goodwin et al ³² was used for the final population PK model validation. Their study evaluates four THC oral doses: 0.13, 0.157, 2.5, and 4.93 mg administered three times daily (TID) over 5 days. In the 2.5 mg group, only two doses are given on the first day. The sampling time for plasma concentrations of THC and its metabolites in their study are 1.5, 4.5, 6.5, 9, 11.5, 23.5, 47.5, 71.5, 95.5, 107, 121, 130.5, 145, and 154.5 h following first dose administration. Our final model was used to simulate THC and metabolites’ plasma concentrations with the exact same dose regimen from the Goodwin study for a head‐to‐head comparison. In the dose groups 0.13 and 0.157 mg, observed THC and 11‐OH‐THC concentrations were below the limit of quantification (LOQ), so only THC‐COOH simulations were compared with observations. For the remainder doses, the simulated and observed PK curves for all compounds were compared.

The population PK model simulation was performed with NONMEM (version 7.4.3; Icon Development Solutions, Ellicott City, MD). Data handling and graphical analyses was carried out by SigmaPlot 14.0 (Systat Software, San Jose, CA).

THC Oral Dose Simulation for Evaluating Per Se Limits and Cannabis Impairment

The objective of this study was to leverage a population PK model to evaluate the effectiveness of current per se limits in differentiating cannabis‐impaired drivers. To accomplish this, the final PK model was used to simulate the concentration–time profiles of THC and its metabolites at varying oral doses in both frequent and occasional cannabis users.

A standard THC dose of 5 mg has been widely recognized by various institutions as a baseline for dosing. ³³ Building on this concept, six oral dose levels were selected for the study: 2.5, 5, 10, 20, 50, and 100 mg. These doses represent a broad range of typical recreational cannabis use.

The simulations were conducted for two usage patterns: occasional and frequent use. Occasional users were defined as individuals using cannabis once or more per month but less than three times per week, while frequent users were defined as those using cannabis daily or nearly daily (at least 5 days per week). ²⁸ , ³⁴ For occasional users, a single dose was administered, while for frequent users, each dose was administered twice daily (BID) over a period of 14 days.

Overall, the simulations were performed across the six different dose levels for both user groups for 48 h from the last given dose. A total of 1000 virtual subjects were simulated per dose group.

Two key assumptions were made for simulated PK data interpretation:

Assumption 1

(—Impairment Window) One of the major challenges with THC is the disconnect between PK and PD, making it difficult to pinpoint a precise time frame for impairment after THC administration. ³⁵ For oral consumption, studies reported that impairment can occur and last anywhere from 1 to 8 h, with the duration being potentially dose dependent. ⁵ , ²⁴ , ³⁶ To account for the full range of potential impairment detection time, an “impairment window” of 1 to 8 h after oral cannabis administration was defined in the present study to characterize acute cannabis impairment.

Assumption 2

(—Minimal Impairment Dose) Prior research showed that the THC dose of 10 mg can produce minimal impairment in psychomotor performance. ³ , ⁵ , ³⁷ , ³⁸ In countries where recreational THC is legal, such as Canada, the maximum legal THC dose per edible product is 10 mg. ³⁶ Based on these findings, it was hypothesized that doses at or below 10 mg result in minimal driving impairment.

The currently enforced per se laws in various states typically set THC levels at 1, 2, or 5 ng/mL in whole blood. A conversion factor of 0.71 was applied ³¹ to convert these whole blood concentrations to plasma levels, resulting in THC plasma concentrations of 1.4, 2.8, and 7 ng/mL. With the two assumptions we made, these three plasma concentration levels were assessed in relation to model‐simulated THC plasma concentrations and the predefined impairment window. The effectiveness of each per se limit was quantitatively measured by the probability of impairment detection within the impairment time window (1 to 8 h) and from 8 to 24 h after last given dose.

$$\begin{array}{ccc}& & \mathrm{Probability}\mathrm{of}\mathrm{Impairment}{\mathrm{Detection}}_{\left(1\mathrm{-}8\mathrm{h}\mathrm{post}-\mathrm{dose}\right)}\hfill \\ & & {\displaystyle =\frac{\#\mathrm{of}\mathrm{Concentrations}>\mathrm{Each}perse\mathrm{Limit}}{\mathrm{Total}\#\mathrm{of}\mathrm{Concentrations}\mathrm{Within}1\mathrm{-}8\mathrm{h}}}\hfill \end{array}$$ (14)

$$\begin{array}{ccc}& & \mathrm{Probability}\mathrm{of}\mathrm{Impairment}{\mathrm{Detection}}_{\left(8\mathrm{-}24\mathrm{h}\mathrm{post}-\mathrm{dose}\right)}\hfill \\ & & {\displaystyle =\frac{\#\mathrm{of}\mathrm{Concentrations}>\mathrm{Each}perse\mathrm{Limit}}{\mathrm{Total}\#\mathrm{of}\mathrm{Concentrations}\mathrm{Within}8\mathrm{-}24\mathrm{h}}}\hfill \end{array}$$ (15)

Results

Population PK Modeling

Goodness of Model Fitting

A comprehensive population PK model was developed to characterize the sequential metabolism of THC and the adipose tissue disposition of both THC and 11‐OH‐THC in oral cannabis users (Figure 1). The observed versus population‐predicted plasma concentration–time profiles for THC, 11‐OH‐THC, and THC‐COOH for representative oral dose levels are presented in Figure 2. The model fittings for all po and iv dose groups are provided in Figure S2. As shown in Figure 2, the proposed population PK model adequately captured the concentration–time profiles for THC and its metabolites across most dose groups. Some suboptimal fitting was observed such as in the 75 and 90 mg groups (Figure S2), which was likely due to limited data availability (n = 1 with sparse sampling). Despite it, the overall close agreement between observed data and model‐predicted PK curves demonstrates the effectiveness of our model in characterizing the PK of THC and its metabolites.

Figure 2.

Concentration time courses of population predicted concentrations (solid lines) versus literature reported concentrations (symbols) from studies for a) THC, b) 11‐OH‐THC, c) THC‐COOH following representative single oral doses (10 mg, 20 mg, 45 mg, and 50 mg). The literature reported concentrations are digitized from various studies, which are summarized in Table S2.

The goodness‐of‐fit plots are displayed in Figure 3. Figure 3a and 3b show that the individual and population‐predicted concentrations align well with the observed data, with points distributed symmetrically around the line of identity. This distribution indicates that the model accurately captures the PK of THC and its metabolites at both the individual and population levels. Additionally, Figures 3c and 3d display conditional weighted residuals (CWRES) plotted against time and population‐predicted concentrations, respectively, with the residuals uniformly scattered around the zero line. This pattern confirms that the final population PK model provided a robust characterization of THC and its metabolites' disposition.

Figure 3.

Goodness‐of‐fit plots for the final model of THC and metabolites. (a) Observed concentration versus population‐predicted concentration. (b) Observed concentration versus individual predicted concentration. (c) Conditional weighted residuals versus time. (d) Conditional weighted residuals versus population‐predicted concentration. The solid line represents the identity line for (a) and (b) or zero line for (c) and (d).

Parameter Estimation

The final model estimated parameters of THC, 11‐OH‐THC, and THC‐COOH are presented in Table 1. The clearance (CL) values were estimated at 87 L/h for THC, 80.4 L/h for 11‐OH‐THC, and 4.34 L/h for THC‐COOH. Central volumes of distribution (Vd) were 204 L for THC, 49.1 L for 11‐OH‐THC, and 0.986 L for THC‐COOH. Notably, the partition coefficients (Kp) for adipose tissue were estimated at 14 for THC and 9.87 for 11‐OH‐THC, signifying extensive adipose tissue distribution. The permeability surface area product (PS) was estimated at 5.5 L/h for THC and 5.11 L/h for 11‐OH‐THC, considerably lower than the respective distribution flows (Q₂ = 29.7 L/h, Q₄ = 47.3 L/h) that connected the adipose compartments to the central compartments of THC and 11‐OH‐THC, further validating their adipose tissue disposition. The fraction of 11‐OH‐THC metabolized to THC‐COOH (Fm) was estimated at 0.904, indicating that approximately 90.4% of 11‐OH‐THC underwent metabolic conversion to THC‐COOH, while about 10% was excreted unchanged via renal clearance. This finding was consistent with previously reported elimination pathways of 11‐OH‐THC, ³⁹ further supporting the validity of the model in accurately capturing the PK of THC and its metabolites.

Inter‐study variability was assessed for key parameters such as CL, V2, V3, and V4 due to the large number of estimated parameters (n = 25). The inter‐study variability and residual variability shrinkage were generally below 30%, except for inter‐study variability on V4, which reached 37.5%. However, removing the inter‐study variability on V4 negatively affected the model stability.

THC Multiple Oral Dose Simulation for Model Validation

The final model was validated by comparing simulated plasma concentrations of THC and its metabolites with observed data from a published multiple‐dose study (0.13, 0.157, 2.5, and 4.93 mg TID over 5 days). ³² The time‐course profiles of observed versus simulated plasma concentrations for THC and its metabolites are illustrated in Figure 4. The simulated PK curves aligned well with the observed data from the clinical study, supporting the accuracy and applicability of the final model.

Figure 4.

Model simulated (solid lines) versus literature reported (symbols) plasma concentration time courses for THC and metabolites in human adults following a) oral dose of 0.13 mg THC TID for 5 days, b) oral dose of 0.157 mg THC TID for 5 days, c) oral dose of 2.5 mg THC TID for 5 days with only two doses given on the first day, d) oral dose of 4.93 mg THC TID for 5 days. The literature reported concentrations are from Goodwin et al, 2006 (Ref. 32).

THC Oral Dose Simulation for Evaluating Per Se Limits and Cannabis Impairment

The final population PK model simulated various scenarios to mimic real‐world oral cannabis use, generating time–concentration profiles for 1000 virtual subjects across doses of 2.5 to 100 mg for both occasional and frequent users. Using an impairment window of 1 to 8 h, we evaluated the effectiveness of current per se limits—1.4, 2.8, and 7 ng/mL—in capturing true positive impairment cases. The model‐simulated THC plasma time–concentration profiles for different doses and user frequencies are illustrated in Figure 5. The probability of impairment detection of each dose group and frequency type is summarized in Table 2.

Figure 5.

Model‐simulated THC plasma concentration–time data following (a) single oral THC doses (2.5 to 100 mg) representing occasional users and (b) multiple oral THC doses (2.5 to 100 mg) representing frequent users. Dashed lines indicate current per se limits (1.4 ng/mL—dark red, 2.8 ng/mL—blue, and 7 ng/mL—purple) The shaded areas denote specific time intervals: the predefined impairment window (1–8 h post‐dose) is represented by the grey shading in (a) and the dark orange shading in (b), while the light pink in (a) and light orange in (b) areas show simulated THC concentrations up to 48 h post‐dose. The per se limits are analyzed in conjunction with the simulated THC pharmacokinetic (PK) data and their overlap with the impairment window.

Table 2.

Probability of Impairment Detection for THC Per Se Limits in Occasional and Frequent Users

		Occasional Oral Cannabis Users (Single Dose)		Frequent Oral Cannabis Users (BID Dose for 14 days)
Per Se Limit in Plasma (ng/mL)	Dose (mg)	Probability of Impairment Detection 1‐8 h After Dose (%)	Probability of Impairment Detection 8‐24 h After Dose (%)	Probability of Impairment Detection 1‐8 h After Dose (%)	Probability of Impairment Detection 8‐24 h After Dose (%)
1.4 (corresponds to 1 ng/mL in whole blood)	2.5	2.0	0	4.9	0.6
	5	12.6	0.1	21.6	4.3
	10	32.8	2.0	43.9	13.6
	20	53.7	11.5	58.5	27.7
	50	67.9	34.2	69.3	46.5
	100	73.0	48.7	74.0	60.7
2.8 (corresponds to 2 ng/mL in whole blood)	2.5	0	0	0.2	0
	5	2.0	0	4.9	0.6
	10	12.6	0.1	21.6	4.3
	20	32.8	2.0	43.9	13.6
	50	58.3	16.6	61.7	32.3
	100	67.9	34.2	69.3	46.5
7 (corresponds to 5 ng/mL in whole blood)	2.5	0	0	0	0
	5	0	0	0	0
	10	0.7	0	2.3	0.2
	20	7.7	0	14.8	2.6
	50	32.8	2.0	43.9	13.6
	100	53.7	11.5	58.5	27.7

Analysis of Current Per Se Limits in Occasional Users

As presented in Table 2, the probability of detecting impairment using the 1.4 ng/mL THC threshold ranged from 2.0% to 32.8% for doses between 2.5 and 10 mg during the impairment window. Beyond the impairment window and up to 24 h post‐dose, this threshold resulted within 2.0% probability of detecting potential false positives in these dose groups. For higher doses between 20 and 100 mg, the probability of impairment detection increased to 53.7% to 73% within the impairment window. In addition, the detection probabilities for those doses in 8 to 24 h were 11.5% to 48.7%, indicating a higher risk of detecting false positives.

At the 2.8 ng/mL threshold, the impairment detection probability was below 12.6% for doses between 2.5 and 10 mg during the impairment window. In the 8 to 24 h period, only the 10 mg dose group had a 0.1% probability of detection. For higher doses (20 to 100 mg), the probability of impairment detection ranged from 32.8% to 67.9% during the impairment window. Even after the impairment window and up to 24 h, the detection probabilities remained at 2.0% to 34.2% for doses from 20 to 100 mg, respectively, showing ongoing detection beyond the impairment window.

The 7 ng/mL threshold resulted in no detection for doses up to 10 mg, except for the 10 mg dose within the impairment window, where the probability was 0.7%. For doses between 20 and 100 mg, the detection probability ranged from 7.7% to 53.7% during the impairment window. In the 8 to 24 h period, these probabilities dropped to under 11.5%. This per se threshold showed very low detection probabilities for doses at or below 10 mg, which was assumed by our study to have minimal impairment, and moderate detection for higher doses within the impairment window.

Analysis of Current Per Se Limits in Frequent Users

As shown in Table 2, the 1.4 ng/mL THC threshold resulted in a detection probability of 4.9% to 43.9% during the impairment window for doses ranging from 2.5 to 10 mg. For these same doses, the detection probability at 8 to 24 h was 0.6% to 13.6%. Higher doses (20 to 100 mg) showed increased detection probabilities of 58.5% to 74% during the impairment window, with probabilities of 27.7% to 60.7% persisting in the 8–24 h period.

For the 2.8 ng/mL threshold, the detection probability for 2.5 to 10 mg doses ranged from 0.2% to 21.6% within the impairment window, dropping to below 4.3% in the 8 to 24 h period. For doses from 20 to 100 mg, the impairment detection probabilities during the impairment window ranged from 43.9% to 69.3%, while at 8 to 24 h, the probabilities remained at 13.6% to 46.5%.

At the 7 ng/mL threshold, detection probabilities for lower doses (2.5 to 10 mg) were minimal, remaining within 2.3% within the impairment window, with only a 0.2% chance of detection at 8–24 h for the 10 mg dose. Higher doses (20 to 100 mg) showed detection probabilities of 14.8% to 58.5% during the impairment window, decreasing to 2.6% to 27.7% in the 8 to 24 h period.

Overall, none of the evaluated per se limits provided consistent effectiveness across all doses and user types. The 1.4 ng/mL threshold produced a high rate of false positives, particularly in doses at or below 10 mg where they were assumed to cause the minimal impairment, making it the least effective. The 2.8 ng/mL threshold showed a higher likelihood of detecting impairment across all doses, but it also carried an elevated risk of false positives, particularly for higher doses. The 7 ng/mL threshold was most effective at reducing false positives but exhibited lower detection rates for impairment in higher doses during the impairment window.

Discussion

Cannabis‐impaired driving remains an area that lacks uniform legislation, with per se laws implemented for cannabis DUI across a few states yet heavily criticized. ¹⁷ , ¹⁸ This controversy, coupled with the complex PK and PD of THC, underscores the need for a thorough analysis and potential refinement of existing laws. To address this issue, we built a comprehensive population pharmacokinetic model with the data from oral cannabis users to provide a mechanistic framework for understanding the complex PK of THC and its metabolites by accounting for their unique disposition.

The primary objective of this study was to evaluate the existing per se limits for cannabis DUI laws—whole blood THC levels at 1, 2, and 5 ng/mL—to determine their effectiveness in identifying cannabis‐related impairment in the hope to find the most reliable per se limit. However, after a thorough exploration using our population PK model, it was clear that finding a definitive threshold to meet this goal was challenging. A key takeaway from our model simulations was that the effectiveness of using any current per se threshold to indicate impairment varied significantly based on both the THC dose and the user frequency (occasional vs frequent users).

As shown in Table 2, the 1.4 ng/mL plasma limit (equivalent to 1 ng/mL in whole blood) exhibited the highest probability of detection across all doses and in both user groups, regardless of whether the time was within or beyond the impairment window. Given our assumptions—that doses at or below 10 mg and times outside the impairment window represent minimal impairment—this high detection probability suggested a considerable risk of false positives. The 2.8 ng/mL threshold (equivalent to 2 ng/mL in whole blood) showed moderate detection across all doses in both groups but maintained a high detection probability beyond the impairment window (8 to 24 h) for higher doses (20 to 100 mg). In contrast, the 7 ng/mL threshold (equivalent to 5 ng/mL in whole blood) minimized detection for doses at or below 10 mg but also demonstrated reduced effectiveness in detecting impairment for doses at or above 20 mg within the impairment window. Overall, the 1.4 ng/mL threshold was found to be the least effective due to its high risk of generating false positives. The effectiveness of the 2.8 and 7 ng/mL thresholds remains inconclusive, primarily due to the lack of PD data that specifically correlates different oral THC doses to their driving impairment effects across various time periods and user frequencies.

This outcome highlighted a long‐standing issue: the complexity of using blood concentrations of THC and metabolites to assess impairment. Our findings reinforced that a “one‐size‐fits‐all” per se limit for THC, similar to BAC limits for alcohol, may not be well suited for cannabis DUI offenses among oral users. A study by Arkell et al reported similar findings after evaluating simulated driving performance following an average smoked dose of 13.75 mg in 14 occasional users. ¹⁷ Their research on inhalation users aligned with our findings, highlighting concerns about the fairness and reliability of existing per se laws for cannabis DUI. In the United States, DUI is a serious criminal offense requiring a high standard of evidence, typically relying on reliable impairment measures like BAC for alcohol. However, THC's unique PK properties and poor correlation with driving impairment challenge the fairness of per se limits, where cannabis DUI convictions are based solely on exceeding predefined blood THC levels, regardless of actual impairment. ¹⁶

Nevertheless, the need for standardized cannabis‐impaired driving laws remains critical. There may still be value in adopting some concentration limits, provided they are interpreted appropriately. For instance, Colorado's cannabis inference law, which sets a THC limit of 5 ng/mL in the blood, allows the judge or jury to use the driver's blood THC levels to suggest impairment in court, but this does not guarantee a conviction. The driver can still present evidence to challenge the presumption of impairment based on THC concentration alone. ⁴⁰ This system is less rigid than a strict per se law and offers a more nuanced way to incorporate THC concentration into legal decisions.

In light of this, and as part of our study's goals, this research aims to contribute to the ongoing effort to refine cannabis‐impaired driving regulations and inform potential legal adjustments. To this end, current cannabis per se limits could be reconsidered as part of a broader framework for assessing impairment rather than serving as definitive evidence for DUI charges. For example, it may be worth exploring the use of per se limits within inference laws, as seen in Colorado, or in the context of driving while ability impaired (DWAI) laws. DWAI laws provide a more nuanced approach by focusing on evidence that a driver's ability to operate a vehicle is impaired, rather than relying solely on predefined substance concentration thresholds. This framework may be particularly relevant for cannabis, where the relationship between blood THC levels and actual impairment is complex and not well defined. Incorporating per se limits into inference laws or DWAI frameworks could offer a starting point for refining legal approaches to cannabis‐impaired driving.

While the less strict inference or DWAI laws offer some flexibility, the challenge of accurately identifying cannabis‐impaired drivers remains significant. Our findings highlight the disconnect between THC blood levels and impairment, evidenced by substantial THC accumulation in adipose tissue. This disconnect underscores the limitation of relying solely on blood THC concentrations to capture cannabis impairment accurately. Therefore, to provide a more comprehensive understanding of driving impairment, incorporating additional methods of evaluation could be helpful. For examples, standard deviation of lateral position (SDLP) and functional near‐infrared spectroscopy (fNIRS) provide more direct assessments of impairment. ⁴¹ , ⁴² The SDLP specifically measures driving performance by assessing lane deviations, giving a direct evaluation of a driver's ability to stay centered. Meanwhile, fNIRS is a more advanced method that measures brain activity to detect physiological signs of impairment, providing insight into cognitive changes that may not be outwardly visible but still affect driving abilities. A combined approach involving at least one biofluid test (e.g., blood THC levels) alongside other methods of evaluation could offer a more balanced way to assess cannabis driving impairment. This framework might help address THC PK variability and provide a potential direction for future refinement of legal assessments.

Current per se limits for cannabis impairment primarily emphasize THC, the main psychoactive component. However, it is important to note that 11‐OH‐THC, THC's primary metabolite, also exhibits psychoactive properties. Due to extensive first‐pass metabolism in oral route, a nearly 1:1 ratio of 11‐OH‐THC to THC is produced, making 11‐OH‐THC an important factor when evaluating cannabis impairment. ⁴³ Given both THC and 11‐OH‐THC exhibit similar biochemical and pharmacokinetic properties, ⁴⁴ several studies have reported that the psychoactivity of 11‐OH‐THC is similar to or greater than that of THC, ⁸ , ⁴⁵ , ⁴⁶ , ⁴⁷ , ⁴⁸ indicating that 11‐OH‐THC may play a role in cannabis‐related impairment. Despite minor differences in their molecular weights (THC = 314.5 Da, 11‐OH‐THC = 330.5 Da, 5% difference), incorporating both THC and 11‐OH‐THC plasma concentrations provides a more comprehensive marker for assessing acute impairment following oral cannabis use. A study by Ménétrey et al discovered a better correlation with cannabis impairment when the sum of blood THC and 11‐OH‐THC concentrations was used, rather than THC alone. ²⁴ Therefore, the combined blood concentrations of THC and 11‐OH‐THC should be considered for a more accurate assessment of impairment, particularly for oral cannabis users, which could potentially lead to the establishment of more reasonable per se limits.

Our study had several limitations. The population PK model was constructed using digitized mean PK profiles from various published studies, which introduced variability from data sources to the analyzed parameters. Additionally, the oral THC doses were administered in varying formulations. While some studies used synthetic THC medications like dronabinol, others involved self‐prepared cannabis products such as brownies or hemp oil, where these products may introduce further variability. ⁵ , ²¹ , ²⁴ , ³⁰ Also, due to limited availability of extended sampling data, the final model could not fully capture the accumulation characteristics of THC as expected. Moreover, the simulation for the frequent user group involved administering oral THC doses twice daily for 14 days, which was a regimen that may not accurately reflect real‐world cannabis use patterns. Another limitation was the reliance on a fixed impairment window of 1–8 h post‐dose to determine impairment. Due to the lack of PD data, the present study could not directly analyze or correlate THC plasma concentrations to the impairment status. However, given the unpredictability of cannabis impairment, a fixed impairment window may also not accurately reflect true acute impairment as it is known that the impairment duration is dose dependent. ⁵ , ²⁴ , ³⁶

Conclusions

The current study successfully built a semi‐mechanistic pharmacokinetic model to characterize THC, 11‐OH‐THC, and THC‐COOH disposition following oral cannabis consumption. Our findings challenged the effectiveness of current per se laws for cannabis DUI, as these thresholds do not adequately account for the variability in how cannabis affects individuals based on dose, frequency of use, and individual physiology. Model simulations indicated that the 1.4 ng/mL threshold (equivalent to 1 ng/mL in whole blood) was the least effective due to its high likelihood of producing false positives. The effectiveness of the 2.8 and 7 ng/mL thresholds (equivalent to 2 and 5 ng/mL in whole blood, respectively) for detecting cannabis impairment remains unclear due to insufficient PD data. Thus, further research linking THC blood concentrations and driving impairment with specific doses, time intervals, and user types is essential for a more accurate assessment of these per se limits in detecting cannabis‐related impairment.

With the present findings, the existing cannabis per se laws are not scientifically robust enough to serve as definitive evidence for DUI charges, given the disconnect between THC blood concentrations and impairment. This highlights the need for refining cannabis‐impaired driving laws to ensure fairness and reliability. We suggest interpreting THC blood levels as part of a broader framework, such as inference laws or DWAI laws, rather than relying on them as sole evidence for DUI. Additionally, incorporating methods that directly assess impairment could complement the use of blood THC concentrations. While promising, these approaches require further studies to validate their effectiveness. In future studies, we plan to explore the combined use of THC and 11‐OH‐THC blood concentrations to improve accuracy in detecting cannabis impairment. Next, we plan to incorporate cannabis smoking PK and PD data to thoroughly assess the feasibility of per se limits and explore potential alternative thresholds that better capture cannabis‐related impairment across both oral and inhalation routes. Overall, this study serves as a preliminary step toward refining cannabis impairment regulations by accounting for differences in routes of administration and usage frequency, paving the way for more accurate interpretations of blood THC levels relative to impairment.

Author Contributions

Concept of the work: Guohua An. Model development: Peizhi Li and Guohua An. Manuscript Preparation: Peizhi Li and Guohua An.

Conflicts of Interest

The authors declare no conflicts of interest.

Funding

This project was partially supported by Innovative Analytics.

Supporting information

Supporting Information

Data Availability Statement

N/A.