Can a THC Blood Cutoff Prove Impairment After an Edible? What a Ten-Study Pharmacokinetic Model Actually Shows

Short answer: No. A 2025 original pharmacometric analysis demonstrates that a single THC concentration cutoff can classify very different portions of modeled plasma concentration-time observations depending on oral dose, dosing history, and elapsed time. It does not demonstrate the sensitivity or specificity of any cutoff for real driving impairment. Its model used concentration data digitized from ten earlier human studies and then generated simulated profiles; the investigators did not conduct a new driving experiment, observe collisions, or determine which virtual participants were impaired. A legal cutoff, an analytical THC concentration, a pharmacokinetic simulation, and an individual’s functional impairment remain four separate propositions.

The primary research is Peizhi Li and Guohua An, Evaluation of Cannabis Per Se Laws: A Semi-Mechanistic Pharmacometrics Model for Quantitative Characterization of THC and Metabolites in Oral Users, Journal of Clinical Pharmacology (2025), DOI 10.1002/jcph.6181. The forensic value of this study is especially clear when its assumptions are separated from its simulated numerical outputs.

What question did this original study address?

The investigators asked how THC and two metabolites behave after oral administration and how often simulated THC concentrations exceed benchmark concentrations that correspond approximately to specified whole-blood THC thresholds. Oral products warrant separate analysis because absorption can be prolonged and variable, first-pass metabolism changes the balance between THC and 11-hydroxy-THC, and residual concentrations can remain after the selected post-dose interval. These are pharmacokinetic questions: where the substances go, how their concentrations change, and how a given sampling point should be understood.

The broader legal question is whether a numerical THC threshold provides dependable evidence of impairment. That is a pharmacodynamic and evidentiary question. To answer it directly, a study would need a specified criterion for impairment measured in actual people, paired with analyte results from the legally relevant specimen. This modeling exercise lacked that direct outcome. Consequently, its results identify weaknesses in some cutoff assumptions but do not calculate validated false-positive rates among real drivers.

Was this a new human driving trial?

No. Li and An developed an original semi-mechanistic population pharmacokinetic model by reanalyzing concentration-time curves from ten previously published human THC studies. Seven supplied single oral-dose data and three supplied intravenous-dose data, with oral doses ranging from 5 to 90 mg and intravenous doses from 1.6 to 5 mg in the development source set. Sampling across those sources extended approximately four to 54 hours. Some studies measured more analytes than others. The model was independently compared against an earlier repeated oral-dosing concentration study, but that is not the same as testing an impairment classifier.

The published curves were digitized and combined using nonlinear mixed-effects software. Importantly, the primary fitting data consisted of reported study means, not full individual raw trajectories from every participant. The model therefore accommodated differences across studies but could not estimate actual person-to-person variability directly from those mean curves. Its 1,000 virtual subjects per simulated dose group are computational draws, not 1,000 newly dosed or tested people.

Which substances and biological matrices were involved?

The three modeled compounds were parent delta-9-tetrahydrocannabinol (THC), 11-hydroxy-THC (11-OH-THC), and 11-nor-9-carboxy-THC (THC-COOH). Parent THC is psychoactive; 11-OH-THC is also pharmacologically active; THC-COOH is a metabolite generally used as an exposure marker rather than proof of current intoxicating effect. The model does not simply sum these substances into a single impairment number.

The reported simulation outputs are plasma concentrations in ng/mL. Some of the original human source papers instead presented whole-blood results. Li and An transformed such values to a common modeled plasma scale using fixed literature blood-to-plasma ratios: 0.71 for THC, 0.73 for 11-OH-THC, and 0.65 for THC-COOH. These are modeling conversion assumptions. They are not paired analytical measurements in every simulated subject, and they do not establish that a particular person’s whole-blood and plasma values follow the same exact ratio.

A forensic venous whole-blood THC result should therefore not be labeled a direct plasma measurement, and a modeled plasma threshold should not be casually compared with a laboratory whole-blood value. Breath testing, urine THC-COOH screening, saliva or oral-fluid THC, and hair analysis are still different specimen types. The distinctions matter because analytical method, matrix partitioning, specimen handling, and legal statutory wording may change the question that a reported number can answer. See the site’s biological-matrices evidence guide for the specimen-centered framework.

How did the authors translate whole-blood thresholds to plasma benchmarks?

The study examined benchmark whole-blood THC concentrations of 1, 2, and 5 ng/mL, represented in the model by approximate plasma THC benchmarks of 1.4, 2.8, and 7 ng/mL. The arithmetic is based on the assumed THC blood-to-plasma factor of 0.71; dividing 1, 2, or 5 by that factor produces approximately the reported plasma values. These are scenario thresholds chosen for the publication, not a verified statement of the controlling law in any jurisdiction today. Applicable criminal statutes, evidentiary standards, specimen definitions, and effective dates must be checked independently for the actual case.

Even if a conversion is useful for population-model comparison, it cannot make an inferred plasma value chemically identical to a directly assayed blood result. Likewise, the model’s estimated oral THC bioavailability, 0.144 in its parameter table, is a fitted pharmacokinetic parameter—not a measured percentage absorbed or metabolized by a particular defendant.

What assumptions created the study’s apparent “impairment detection” percentages?

Two choices drove the legal interpretation. First, the authors defined one to eight hours after oral dosing as an “impairment window.” Second, they assumed an oral THC dose of 10 mg or less would produce minimal impairment for purposes of their scenario classification. Neither assumption is an observed, individually validated impairment finding for the simulated subjects. An actual person might be affected for a shorter or longer interval, and effect can vary by tolerance, formulation, coexposures, dose, task, and medical context.

The computation then counted how often predicted THC concentrations exceeded each benchmark at simulated concentration-time points within hours one through eight and at points from eight through 24 hours. The paper called those proportions the probability of impairment detection. The denominator, however, was simulated concentration observations during the selected time interval—not drivers independently classified as impaired by a validated performance test. This definition is the central interpretive limit of the reported percentages.

What did the original numerical simulations report?

The authors modeled occasional users as receiving one oral dose and frequent users as receiving the selected oral dose twice daily for 14 days. The tested doses were 2.5, 5, 10, 20, 50, and 100 mg. For each group, the model generated 1,000 virtual profiles per dose. A frequent-use label here refers to the assumed administration schedule; it does not mean that every characteristic of all real daily consumers, including individual tolerance, was experimentally measured.

Table 2 below reproduces a deliberately selected subset of the original publication’s Table 2, with the metric renamed to its actual calculation: simulated plasma THC threshold-exceedance proportion. All percentages are model outputs, not counts of drivers who failed a road test. Both dosing schedules are shown because repeated dosing is central to the intended forensic comparison.

These are exact percentages reported in the original Table 2 for the selected modeled conditions. Their relative pattern is informative. At the same nominal oral dose and same plasma benchmark, frequent modeled dosing often produces more above-threshold observations than a single dose. Raising the benchmark reduces the number of observations above it. Neither mathematical fact identifies which real person was driving impaired.

Why do the modeled percentages change so much with the threshold?

A lower threshold intersects a concentration-time curve more often and at later points than a higher threshold. The phenomenon becomes more pronounced when simulated profiles include residual drug from earlier doses. For the 20 mg twice-daily scenario, the model’s concentration-exceedance proportion in hours eight through 24 was 27.7% at the 1.4 ng/mL plasma benchmark, 13.6% at 2.8 ng/mL, and 2.6% at 7 ng/mL. Those values do not measure whether any simulated driver was still unsafe to operate a vehicle; they measure how a classification rule behaves against model concentrations.

A higher cutoff may reduce delayed concentration-only classifications while potentially missing lower concentration observations during a time when an individual could still be impaired. A lower cutoff may capture more concentration observations without establishing that all correspond to functional compromise. The original article framed this as a tradeoff in per-se-law design, but its simulations cannot estimate that tradeoff’s real-world clinical error rates without a measured impairment outcome.

What does adipose tissue contribute to this model?

The semi-mechanistic model includes distribution into central, peripheral, and adipose compartments for THC and 11-OH-THC, followed by metabolite disposition for THC-COOH. Its reported fitted tissue-to-plasma partition parameters were 14 for THC and 9.87 for 11-OH-THC. These parameters help describe the fitted delayed concentration profiles; they do not directly quantify THC in a particular person’s fat tissue and do not diagnose impairment from adiposity.

The model also estimated a central THC clearance of 87 L/h and a metabolic fraction of approximately 0.904 from 11-OH-THC toward THC-COOH. Those are parameter estimates within the chosen structure, not direct organ-specific measurements in the virtual subjects. Fitted values depend on source-study coverage, assumed structure, and digitization quality. A mathematical model can fit the available mean profiles and still be unsuitable for a precise reconstruction of a single individual’s prior ingestion or mental performance.

Was the model validated against driving impairment?

No. The authors compared model-generated concentration profiles with previously published multiple-dose oral observations from a separate study involving 0.13, 0.157, 2.5, and 4.93 mg administered three times daily over five days (with a first-day variation in the 2.5 mg arm). That exercise tested how well the model represented concentration trajectories. It did not measure sensitivity for actual lane-keeping failures, psychomotor errors, reaction-time deterioration, police observations, injury, or crashes.

The original paper also displays concentration-fit and residual plots. Good agreement on those plots supports a limited modeling purpose, not clinical diagnostic calibration. The input data came from different formulations and dosing experiments; the authors noted sparse 75 mg and 90 mg source profiles and limits on extended sampling. Because the training information comprised reported means, the model lacked the individual raw trajectories needed to validate individual-specific prediction intervals. The wide spread of virtual profiles cannot be cited as an empirically established distribution of individual error.

Can the output be compared directly with a forensic THC blood report?

Only with substantial caution. The analyst must first establish the collected matrix, validated target analytes, method, calibration range, uncertainty, precise collection time, specimen integrity, and source of any conversion. A result of “5 ng/mL THC in blood” is not a blank scientific proposition: was the specimen anticoagulated venous whole blood, separated serum, or plasma? Was the result parent THC, 11-OH-THC, or THC-COOH? Was measurement specific and confirmatory, or a screening immunoassay?

That record-level scrutiny is different from the model’s harmonized plasma curves. See confirmatory drug-testing principles for the analytical question. An expert cannot replace a measured result with the model’s virtual concentration, reverse-engineer a uniquely determined oral dose, or transform the published exceedance percentages into a numerical probability that a specific driver was impaired or unimpaired.

Why are drug concentration and task performance distinct evidence?

Pharmacokinetics estimates concentration over time. Pharmacodynamics asks how exposure changes response. A pharmacokinetic curve alone lacks a measured effect variable. The present study did not administer a driving simulator task, a standardized field sobriety assessment, a psychomotor vigilance task, or a road test to its virtual subjects. Nor did it establish a lawful case-specific evidentiary rule. The source authors themselves identified inadequate pharmacodynamic information as a core limitation when assessing higher THC cutoffs.

It is equally important not to overinterpret the opposite direction. The inability of a modeled cutoff to prove impairment does not demonstrate that cannabis has no acute effect on driving-related abilities. It only means this study cannot assign those effects to a threshold with observed clinical classification accuracy. Questions of signs and symptoms, medication history, timing, co-use, observations, performance testing, and statutory requirements must be considered separately.

What should a forensic reviewer obtain before expressing a case opinion?

First obtain the actual analytical report, collection and analysis times, blood draw and preservation records, instrumental confirmation method, relevant QC and measurement uncertainty, and the full toxicology findings including parent THC and metabolites where tested. Ascertain whether blood or plasma was tested and whether any reported concentration was mathematically converted. A chain-of-custody and matrix discrepancy can be as important as the concentration itself.

Next distinguish what is known about the product route and dose from what is inferred. An oral edible taken with food does not guarantee a universal absorption curve. Unknown prior use, tolerance, co-administered drugs, individual physiology, and the specimen’s temporal relationship to alleged driving all limit reconstruction. The site’s pharmacokinetics evidence guide describes why a population concentration curve is not an individualized back-calculation.

Finally identify the precise legal question without assuming that the publication’s 2025 threshold examples remain the law today. A statute may define a concentration offense, an inference, or a broader impairment offense, and the legal significance of any laboratory result depends on the specific jurisdiction, date, and statutory text. Scientific validity and legal sufficiency are related but distinct questions. This paper is primary pharmacometric research, not a jurisdiction-specific legal opinion or substitute for the complete record.

What is the scientifically defensible conclusion?

Li and An showed that a semi-mechanistic model combining earlier human concentration studies produces markedly different simulated plasma THC exceedance percentages under selected oral dosing regimens and benchmark concentrations. The work is useful for testing the hidden assumptions behind a universal THC number. It does not measure the proportion of real drivers impaired at 1, 2, or 5 ng/mL; it does not validate an individual’s oral dose or last-use timing; and it does not determine accident causation. The best forensic use is to pose better questions about matrix, direct analysis, pharmacokinetic uncertainty, and independently observed function—not to elevate the simulation into proof of a person’s impairment.

Frequently asked questions

1. Did the study test 1,000 real drivers for every THC dose?

No. One thousand is the number of virtual simulated profiles per dose group. The source data were published means from ten earlier human pharmacokinetic studies; no new population of 1,000 drivers was given THC or evaluated on a road course.

2. Are the percentages in Table 2 true diagnostic sensitivities?

No. They are calculated proportions of model concentration-time observations above a threshold within stipulated time windows. There was no paired observed clinical impairment label for those virtual observations; diagnostic sensitivity and specificity cannot be computed from that information alone.

3. Why does the paper report 1.4 ng/mL plasma for a 1 ng/mL whole-blood cutoff?

It used an assumed THC blood-to-plasma ratio of 0.71 to put different source studies on a common plasma scale. This is an approximation for model comparison, not a patient-specific assay or universally exact legal specimen conversion.

4. Does detectable THC more than eight hours after an edible prove impairment?

No. A detectable or above-threshold concentration is chemical evidence in a specified matrix; its behavioral meaning requires additional evidence. Conversely, a result below a given concentration does not guarantee fitness to drive. The study’s eight-hour division was selected for simulation, not proven as an individual recovery boundary.

5. Does the model establish a safe oral dose of ten milligrams?

No. Treating doses of 10 mg or less as minimally impairing was one of the study’s premises used to explore threshold behavior. It was not verified by assigning real participants an impairment diagnosis in this research.

6. Can a THC-COOH urine finding be entered as a plasma THC cutoff?

No. THC-COOH is a metabolite and urine is a different matrix. Neither is a direct plasma parent-THC concentration. The paper’s per-se simulations concerned model-generated plasma parent THC, not urine immunoassay results.

7. Can this paper establish whether one named driver caused a collision?

No. It contains no crash observations, no individualized behavioral outcome, and no validated causal attribution algorithm. It may help identify interpretation errors in a concentration-only claim, but accident causation requires separate case evidence.

Primary source: Li P, An G. Evaluation of Cannabis Per Se Laws: A Semi-Mechanistic Pharmacometrics Model for Quantitative Characterization of THC and Metabolites in Oral Users. Journal of Clinical Pharmacology. 2025;65(5):535–549. doi:10.1002/jcph.6181.