Can Plasma THC and Its Metabolites Predict Recognition-Memory Errors? A 31-User Mobile-Lab Study

Short answer: A 2020 observational experiment with legal-market cannabis found that higher combined plasma concentrations of THC and two THC metabolites were associated with poorer performance on a short word-recognition test. In the analyzed sample of 31 regular users, recognition accuracy declined after cannabis use in the primarily THC-product group but did not significantly decline in the THC-plus-CBD-product group. The association does not turn a measured plasma concentration into a validated prediction of an individual person’s memory performance, driving impairment, or accuracy as an eyewitness.

The original research is Curran, Devillez, YorkWilliams and Bidwell, Acute effects of naturalistic THC vs. CBD use on recognition memory: a preliminary study, Journal of Cannabis Research (2020), DOI 10.1186/s42238-020-00034-0. The decisive design feature is the pairing of a controlled verbal-recognition task with directly assayed separated plasma cannabinoids around a naturalistic use episode. The findings concern group-level associations, not an impairment cutoff.

What was the precise question?

The useful forensic question is whether an objective laboratory cannabinoid measurement tracks the change in a participant’s ability to distinguish previously presented words from unfamiliar ones. That differs from determining whether a drug was present, whether somebody used cannabis recently, or whether the person’s account of a real event is true. A numerical concentration may be analytically sound while remaining incapable of answering a separate functional or legal question.

The investigators also compared commercially available THC-dominant products with cannabis flower containing both THC and CBD. That comparison helps generate hypotheses about the roles of different cannabinoids. It was not a randomized, matched-dose test of whether adding isolated CBD reverses THC impairment. The product categories differed in THC exposure, habitual use patterns and administration methods. Those differences materially limit a protective-effect interpretation.

How many independent people were studied?

Thirty-two adults who used cannabis regularly completed the recognition-memory feasibility study. The prespecified THC-dominant category initially included 15 participants, while the THC-plus-CBD category included 17. One participant from the latter category was excluded from the primary analyses because the pre-use combination of cannabinoid biomarkers exceeded a statistical outlier rule. Consequently, the analyzed comparison comprised 15 THC-dominant participants and 16 THC-plus-CBD participants: 31 independent people.

The participants were ages 21 to 66 and recruited in the Boulder-Denver area. Eligibility required cannabis use at least four times in the prior month and experience with potent products; other substance use, heavy alcohol patterns and certain psychiatric histories were among the screening exclusions. These were experienced users, not a representative random sample of all drivers, patients or first-time cannabis consumers.

Every person completed a pre-use and post-use recognition task. Forty tested words per task, several blood samples, or multiple analytes from one specimen do not multiply the study’s independent human sample size. The authors’ models used the paired observations and group assignments, not a claim that dozens of word trials were independent volunteers.

How were cannabis products selected and consumed?

The THC-dominant arm used concentrates labeled 70% or 90% THC with negligible CBD, or a flower labeled 24% THC and approximately 1% CBD. The THC-plus-CBD arm used several flower products spanning markedly different labeled cannabinoid percentages, including 14% THC/9% CBD, 6% THC/9% CBD, 9% THC/10% CBD and approximately 1% THC/23% CBD. These are product concentrations by composition, not measured blood concentrations and not equivalent inhaled doses.

Participants purchased the assigned legal-market product, used it during the preceding five days, and were instructed not to use cannabis before the study appointment on the experimental day. They supplied a pre-use sample and took the recognition test in the mobile laboratory, returned home, consumed their product as they normally would, and returned for another blood draw and later cognitive testing. The study deliberately accommodated several inhalation devices and personal usage practices, rather than administering a standardized laboratory inhalation dose.

Importantly, nearly every THC-dominant participant used a concentrate or a high-THC flower, whereas the THC-plus-CBD group used flower. The amount of product actually inhaled and absorbed could not be measured from the label alone. The researchers weighed the product before and after use, then estimated milligrams of constituent cannabinoids from weight and label percentages. That estimate is distinct from a directly administered or biologically absorbed dose.

What biological specimen did the laboratory actually measure?

A phlebotomist drew venous blood from the arm. Investigators then centrifuged the specimen at 400 times gravity for 15 minutes, separated the plasma from red blood cells, froze it at −80 °C and sent it for high-performance liquid chromatography–tandem mass spectrometry. The API5500 assay quantified four analytes: parent Δ9-THC, 11-hydroxy-THC, 11-carboxy-THC and CBD. Values in the source are reported in ng/mL of plasma.

This is not a study of directly quantified forensic whole-blood THC, urinary THC-COOH, saliva, exhaled breath or hair. Plasma and whole blood are distinct specimen matrices. There is no justified universal conversion factor here for translating these concentrations into a particular driver’s legal whole-blood concentration. A venous blood draw does not make its separated plasma result a whole-blood result.

The investigators additionally summed the numerical concentrations of THC, 11-hydroxy-THC and THC-COOH into a composite they termed THC plus metabolites. This arithmetic sum of unlike molecules is an analytical exposure index in the study, not a chemical measurement of pure active THC and not a validated pharmacologic equivalence. In particular, THC-COOH is a different analyte from psychoactive parent THC. Reporting the composite simply as “blood THC” would misidentify the measurement.

What did the recognition-memory task measure?

The participants studied 20 ordinary concrete nouns. Following a three-minute intervening Sudoku exercise, they classified 40 test words as familiar or new: 20 were from the study list and 20 were previously unseen. The pre-use and post-use versions employed different, matched word sets, with response-button assignments counterbalanced. This approach attempts to reduce trivial list and button effects while still measuring performance on a constrained laboratory task.

The main outcome was the signal-detection discrimination statistic d-prime (d′). A higher d′ means better separation of old from new words, integrating the proportion of correct recognition of studied words (hits) and incorrect recognition of unstudied words (false alarms). The paper also reported response bias and button-press reaction time. These measures provide different information and should not be collapsed into a generic assertion that “memory failed.”

Calling a word “old” when it was new is a laboratory false alarm, not a statement that a person fabricated an eyewitness account. The participant faced brief, standardized word lists under a laboratory decision rule. Real events involve perception, attention, stress, encoding, delay, suggestion, and questioning, none of which was directly tested by this task.

What were the observed plasma concentrations and product estimates?

The following evidence table reproduces the source’s group-level means, but compresses the reported information to the values needed for interpretation. The analyzed THC group was n = 15 and the THC-plus-CBD group n = 16, except that product-weight estimates in the THC group used n = 14. Plasma entries are ng/mL; product-content estimates are mg. They are not interchangeable units or specimen types.

These are group means, not the level each individual achieved. The original table also provides 95% confidence intervals, revealing considerable variability. For example, the mean post-use parent-THC concentration in the THC-dominant arm was accompanied by an unusually wide interval, approximately 145–684 ng/mL. It would be misleading to present the group means as predictions for an unknown individual at a known number of minutes after smoking.

A participant’s weighed product does not tell how much smoke or vapor was generated, drawn into the lungs, retained, absorbed or distributed. The very large difference in nominal THC consumed also prevents the THC-versus-THC-plus-CBD product comparison from isolating CBD as a causal mechanism. The observed contrast is a comparison of product-and-use conditions, not CBD added to an identical administered THC challenge.

What happened to recognition accuracy and false alarms?

The second evidence table summarizes the source’s original signal-detection results. Values are group means at pre-use and post-use testing. A higher d′ reflects better discrimination, while a higher false-alarm fraction indicates more incorrect “old” responses to words that were not studied. Each measure addresses an aspect of a laboratory memory decision rather than broad real-world functional capacity.

The most relevant finding is the interaction, not just the raw direction of either group mean. The THC-dominant arm showed worse post-use discrimination; the combination-product arm did not show a statistically significant pre/post change. Adding the log-transformed composite plasma THC-plus-metabolites variable as a covariate left the session-by-group interaction significant, F(1,28) = 6.18, p < 0.05. That persistence is informative, but it cannot eliminate nonrandom product preferences or unmeasured confounding.

In the paper’s separately reported analyses, post-use hit rates differed in a simple t test, yet the fuller repeated-measures ANOVA did not produce a significant hit-rate effect. A careful account should describe the test being cited rather than selecting whichever significance label seems most dramatic. Reaction time also did not show a significant corresponding change despite the strong d′ pattern.

Did higher plasma cannabinoids statistically predict poorer memory?

In the authors’ regression across analyzed participants, higher log-transformed THC plus metabolites was associated with lower post-use recognition d′ (p < 0.05; reported partial η² about 0.28). Neither the CBD term nor the THC-by-CBD interaction was significant. The article also states that cannabinoid concentrations were not significantly associated with d′ within each product group. These are three distinct findings and should all be retained.

The first shows a group-pooled association between a composite plasma marker and the task outcome. The second means this dataset did not detect an independent CBD concentration association or a statistically established interaction with that composite. The third warns against presenting a reliable within-group concentration-to-score mapping. In a small sample where product types and cannabinoid exposures differ markedly, pooling can emphasize between-group structure rather than individual predictive calibration.

Neither an effect-size statistic nor a regression p value supplies a validated individual diagnostic equation. There is no reported prospective sensitivity, specificity, positive predictive value, legally defensible decision threshold, measurement error for classifying an unknown driver’s memory state, or external validation on new individuals. A statistical association can justify further investigation without justifying categorical testimony that an unknown person with a given plasma value must have made memory errors.

Does this establish that CBD protects memory?

No causal protective effect can be isolated from these conditions. The THC-plus-CBD participants used markedly less estimated THC and more CBD, and they used flower instead of the concentrates preferred by most THC-dominant participants. The THC-dominant participants also reported more days of cannabis use in the preceding month: 25.15 versus 18.94 days on average, a significant group difference. Their pre-use parent-THC plasma concentrations were already unequal, while pre-use d′ differences were not significant.

Moreover, this was a feasibility analysis embedded in larger studies, not a purpose-built, adequately powered, THC-dose-matched, randomized CBD-addition experiment. Blinding study staff to assigned product did not remove differences between participants’ usual products or completely randomize users to comparable exposure routes. The study warrants the narrower statement that one naturally occurring THC-plus-CBD product condition did not show the group-average decrease observed under another condition.

That result is neither evidence that all CBD products prevent memory impairment nor proof that CBD has no influence. A nonsignificant change can reflect a small effect, insufficient power or heterogeneity. Conversely, a significant difference between groups cannot be allocated exclusively to CBD when the THC doses and modes of use were unequal. The authors themselves called the work preliminary and emphasized replication.

Why do the time points matter in a forensic interpretation?

The immediate post-consumption venous sample and the word-recognition test were not obtained at an identical instant. Participants left the mobile laboratory to consume at home, then returned for phlebotomy; the recognition task occurred approximately 35 minutes after returning to the van, or roughly 50 minutes after use in the study’s broader description. Thus the reported cannabinoid biomarker concentration is a temporally proximate exposure measurement, not necessarily the concentration exactly while each word decision was made.

The investigators deliberately prioritized realistic product use over precise control of every puff, heating device, inhalation volume, last inhalation time and absorption profile. The plasma sampling schedule cannot establish a pharmacokinetic elimination curve for a specific individual, a retrospectively reconstructed use time, or a simultaneously measured whole-blood concentration at a hypothetical traffic stop.

Different cannabis-use histories further complicate baseline residual concentrations. For an experienced user, a measurable predose metabolite and even parent THC do not by themselves document use immediately before an event. The paper demonstrates why specimen timing and task timing belong in the factual record, not why a solitary laboratory number should override them.

What does the study say about an alleged eyewitness memory error?

It supports a limited general proposition: under these naturalistic conditions, performance on a verbal old/new recognition task varied with cannabis product category and a composite plasma cannabinoid exposure index. It does not test whether a person accurately perceived a face, identified a suspect, remembered a conversation, reported a sequence of actions, or testified truthfully. Laboratory false alarms to unrelated nouns are not a validated rate of mistaken eyewitness identifications.

For a witness-specific evaluation, relevant matters would include exactly what was witnessed, when the witness encoded it, when cannabis was used, whether cannabis concentrations were measured in a documented specimen, whether there was contemporaneous functional evidence, elapsed delay, intervening information and the quality of the identification procedure. This single pilot study cannot fill missing facts. It is especially inappropriate to use it to declare a witness credible or incredible merely from a measured cannabinoid metabolite.

The absence of a significant reaction-time result also warns against treating one intact cognitive measure as proof every other measure is intact. Here recognition discrimination changed in one product group without a corresponding significant group-level reaction-time effect. Cognitive functions are related but are not interchangeable forensic endpoints.

Can the results establish drug-impaired driving?

No. There was no on-road driving assessment, closed-course driving task, validated simulator outcome or individual traffic-safety endpoint in this feasibility report. The administered exposure was self-selected and not matched to a driver’s known dose. Separated plasma was assayed, not a forensic whole-blood specimen, and the cognitive endpoint was recognition of presented words rather than lane control, braking or divided-attention performance under driving conditions.

It would therefore be unsupported to name a numerical plasma THC-plus-metabolite value as a general impairment cutoff, translate it mechanically into whole-blood THC, or assert that a specific crash was caused by cannabis because of this paper. Conversely, a nonsignificant within-group association does not demonstrate that cannabis cannot impair individual function. The study’s statistical scope is narrower than either absolute conclusion.

What should an evidence-bound expert report preserve?

A reliable report can identify the actual source, the specimen matrix, the analytes, the observed testing schedule, the numbers of independent participants, the experimental contrast, and the exact functional measure. For this paper the core distinction is between direct plasma LC–MS/MS, a computed sum of different cannabinoids, and observed responses on a short word-recognition test. These are separate kinds of evidence, linked statistically at the group level but not interchangeable in a case.

The report should also present the negative and qualifying findings: CBD plasma was not independently related to d′ in the reported regression; the tested THC-by-CBD interaction term was nonsignificant; within-product-group d′ concentration associations were nonsignificant; hit rate and reaction-time mixed-model effects were not significant; and product groups differed in estimated THC exposure and prior-use history. Those results materially constrain broad causal or diagnostic claims.

Finally, source fidelity matters. A result printed as ng/mL of separated plasma must not be relabeled ng/mL of forensic whole blood. Milligrams inferred from consumed product labels must not be relabeled milligrams absorbed. False-alarm proportions on word lists must not be mislabeled eyewitness false-identification probabilities. These distinctions are not minor editorial preferences: they control what the original experiment can legitimately prove.

Frequently asked questions

1. Did the experiment enroll 32 or 31 people?

Thirty-two adults participated in the feasibility study; one extreme baseline-biomarker observation from the THC-plus-CBD arm was excluded from the primary analyses. The analyzed group comparison is 15 plus 16, or 31 independent people. The authors state that analyses retaining the outlier were broadly similar, but the principal tables describe the exclusion.

2. Was the reported 278 ng/mL value whole-blood THC?

No. It is the group mean for parent THC in separated plasma after use in the THC-dominant arm. The analysis also used a separate composite of parent THC plus two metabolites. Neither number is a direct forensic whole-blood measurement.

3. Did a CBD-containing product eliminate recognition mistakes?

No. Participants in that product group still made some false alarms, and the mean false-alarm proportion increased from 0.17 to 0.23. Their change in discrimination d′ was not statistically significant. This is not evidence that individual users make no errors or that CBD completely blocks THC effects.

4. Were the THC and THC-plus-CBD conditions dose-matched?

No. Product weight and label-based estimates indicate substantially greater nominal THC use in the THC-dominant arm. Cannabis form, usual route of administration and prior cannabis-use frequency also differed. These make isolated causal attribution to CBD uncertain.

5. Can a plasma THC result predict whether one witness misremembered an event?

Not from this study. The regression was exploratory, not externally validated for individual classification, and the task tested immediate word recognition rather than accurate event testimony. A person’s memory of an actual event requires case-specific evidence rather than extrapolation from a group mean.

6. Why did the investigators report THC plus metabolites as a sum?

They constructed a quantitative exposure index by adding measured plasma THC, 11-OH-THC and THC-COOH concentrations. The molecules have different biological roles. Their numerical sum should not be equated with an active-THC dose, a blood-alcohol equivalent, or a clinical impairment score.

7. Does unchanged reaction time mean the participant was unimpaired?

No. Mixed-model reaction-time effects were nonsignificant, whereas the recognition-discrimination interaction was significant. The two tasks index different aspects of performance. Neither result alone determines real-world driving safety or individual functional capacity.

Primary science and related context

Primary research: Curran T, Devillez H, YorkWilliams SL, Bidwell LC. Acute effects of naturalistic THC vs. CBD use on recognition memory: a preliminary study. Journal of Cannabis Research. 2020;2:28. https://doi.org/10.1186/s42238-020-00034-0.

For related explanations of specimen interpretation, laboratory methods and pharmacokinetics, see biological matrices in forensic toxicology, confirmatory drug testing and pharmacokinetics in forensic toxicology. Those pages provide background context; they are not additional experiments in this source-specific analysis.