Can Two Doses of THC Oromucosal Spray Produce Unpredictable Blood Results? A 20-Patient Pharmacokinetic Study

What exactly did the investigators test?

The cohort comprised 20 adults aged 66–94 years who had recently been hospitalized and had poor appetite. Their median age was 77 years (interquartile range 71–85), median weight 56 kg, and 14 were women. Eligibility required age at least 65 years, body mass index at most 30, and a Simplified Nutritional Appetite Questionnaire score of 14 or less. Thirteen participants had mild-to-severe xerostomia. These descriptors matter because this was not a representative sample of motorists, healthy younger recreational users, or forensic casework specimens.

The underlying clinical experiment assigned placebo and active spray days in randomized order with a minimum 14-day washout, but the pharmacokinetic analysis considered only the 20 active Sativex days. Staff administered oromucosal spray at time zero and again four hours later. Each administration comprised either two sprays (two patients in the relevant dosing scheme) or three sprays (18 patients). Each spray was labeled to provide 2.7 mg delta-9-THC and 2.5 mg cannabidiol (CBD). These are administered product amounts—not concentrations directly proven to have reached systemic circulation.

To limit uncontrolled swallowing, participants were asked to avoid talking, moving the tongue, or swallowing for 15 minutes after application. They fasted before dosing; a standardized nutritional drink followed the first dose and food followed the second. Thus investigators controlled administration more closely than ordinary real-world spray use, but did not separately quantify how much THC crossed the oral mucosa and how much was swallowed.

Was the reported THC measurement whole blood, plasma, or serum?

It was not a direct forensic whole-blood result. Researchers placed a venous catheter, took serial blood specimens in tubes without additives, allowed samples to clot for at least 45 minutes, centrifuged them, added ascorbic acid, and froze the separated fraction. This sample handling describes serum preparation. The paper nevertheless uses the term plasma in its abstract, concentration figures and model description. That source-level terminology discrepancy should be preserved rather than silently converting a published concentration into a forensic whole-blood value.

An ultra-high-performance liquid chromatographic system coupled to tandem triple-quadrupole mass spectrometry measured THC, active 11-hydroxy-THC (THC-OH), inactive 11-nor-9-carboxy-THC (THC-COOH), and CBD. The authors reported a lower limit of quantification (LLOQ 0.25 ng/mL for all four analytes) and analytical coefficient of variation no greater than 15%. Their primary fitted model described THC and THC-OH, not a simultaneously fitted full model for CBD and THC-COOH.

Serum or plasma concentrations in ng/mL cannot be translated into whole-blood THC concentrations without evidence about matrix-specific partitioning and validated conversion. The paper supplies no such conversion, nor any paired forensic whole-blood results. It also did not measure exhaled THC, driving performance, standardized field sobriety tests, or individual clinical impairment.

How many observations were available, and how many were below the assay limit?

There were 370 THC-plus-THC-OH concentration observations from the active-treatment days: 185 THC and 185 THC-OH observations. One THC value, described as an extreme outlier, was excluded from model development; the authors counted 38/185 THC measurements (20.5%) and 34/185 THC-OH measurements (18.4%) below the reporting LLOQ. The overall below-LLOQ share is reported as 19.5%. These are repeated samples from 20 people, not 370 independent patients.

The model did not simply assign every below-LLOQ measurement a value of zero. Investigators used the Beal M3 likelihood method for censored data. This is consequential in toxicology: an unquantifiable sample is not automatically evidence that no molecule was present, and the amount of censoring limits how precisely individual concentration curves can be inferred.

The original report also identifies unequal sampling schedules: two patients followed scheme A, two scheme B, and 16 scheme C. Nine to 14 sampling occasions per patient were planned across approximately eight hours. Thus any statement about the time of peak concentration depends partly on the sampling design and cannot be treated as a universally observed physiologic peak.

Did the same spray amount produce a consistent THC peak?

No. For the two-spray administration subgroup, the observed per-person THC peak concentrations averaged 1.85 µg/L (SD 0.15; n=2). For the three-spray subgroup, average peak was 4.25 µg/L (SD 1.95; n=18). One µg/L is numerically equal to one ng/mL. The different subgroup sizes, small two-spray sample and nonrandomized comparison of those schedules prevent interpreting these figures as a validated dose-response coefficient.

Across participants, the investigators reported that the observed maximum fell after the first or second dose with no consistent pattern: approximately 45% after the first and 55% after the second. That result directly undercuts a tempting shortcut—assuming a measured cannabinoid concentration necessarily identifies the more recent administration. It does not establish precisely how long a particular user took to absorb a specific fraction of an oral-mucosal dose.

The authors showed individual curves in Figure 2; the substantial separation between profiles is visible without assigning graph-derived numbers to particular patients. Those curves are descriptions of these treated patients over this brief interval, not forensic reference ranges with known error rates for reconstructing exposures.

What were the model’s main estimates?

Investigators fitted nonlinear mixed-effects models in NONMEM. Their selected structure used one central compartment for THC, two compartments for THC-OH, and three transit compartments to model delayed absorption. In the final combined model, mean transit time was 1.35 hours. The authors estimated apparent formation clearance for THC conversion to THC-OH of 765 L/h, and apparent THC clearance by other pathways of 162 L/h, giving an apparent total clearance of 927 L/h.

The qualification apparent is essential: the model estimated quantities divided by an unknown bioavailability fraction F. The 927 L/h figure is not a direct measurement of an individual’s liver eliminating 927 liters of blood per hour, nor a validated forensic THC elimination constant. Distribution may still have been occurring, because sampling ended too early to characterize the terminal phase. Similarly, an apparent distribution volume of 1,669 L does not mean the patient’s body contains that many liters of blood.

Investigators reported model-based variability of 59.1% between occasions for mean transit time and 40.2% between occasions for the formation-clearance parameter. Estimated between-person variability included 43.0% for apparent THC distribution volume, 62.5% for metabolite clearance, and approximately 152% for THC clearance by other pathways. These percentages describe fitted log-normal parameter distributions, not a measured percentage error for a particular blood result and not the percentage of people who were impaired.

How uncertain were the apparent clearances?

The original model table supplies bootstrap intervals, but these require unusually careful reading: only 338 of 1,000 bootstrap estimation runs succeeded (33.8%). The model’s apparent THC formation clearance estimate of 765 L/h had a bootstrap 95% interval of 378–1,158 L/h. For clearance through other pathways, the estimate was 162 L/h, whereas the successful-bootstrap mean was only 39.6 L/h and the interval was a wide 1.24–647 L/h. These are signs of poor parameter stability; one should not sell their point estimates as clinically precise constants.

The authors explicitly observed a tendency for the model to underpredict THC concentrations above 3 µg/L and residual trends over time. Their Figure 4 shows systematic imperfections in goodness of fit, while Figure 5 shows a visual predictive check—not a prospectively tested individualized forensic validation. A smooth fitted line does not turn a pharmacokinetic model into a reliable reconstruction of exposure in a different person.

Did age, body fat or kidney function explain the variability?

Not sufficiently for a clinically meaningful dosing covariate in this model. Investigators examined age, weight, body mass index, fat measures, muscle mass, measured and estimated glomerular filtration rate, and other patient characteristics. No covariate survived inclusion in their final parent-metabolite model. Adding allometric weight scaling did not materially improve fit (reported objective-function change −3.31). This finding is restricted to the observed covariate ranges and the small clinical cohort.

It would be wrong to generalize that kidney function or adiposity never changes cannabis pharmacokinetics. The study’s sample included 14 people with a measured GFR, and its estimated GFR range was 39–98 mL/min/1.73 m². Extreme kidney dysfunction was not comprehensively tested, and other drug-use histories, oral conditions and metabolic determinants may be relevant. Absence of a model-selected association is neither proof of no biological influence nor a license to customize a forensic dose estimate from a person’s age alone.

Does oromucosal administration avoid first-pass metabolism?

Not necessarily. The product contacts mucosal tissues, but some material can be swallowed, reach the gastrointestinal tract and undergo first-pass metabolism. The investigators did not directly measure the swallowed fraction, epithelial flux, hepatic extraction or absolute bioavailability. Their parent–metabolite model allowed formation of THC-OH, and observed THC-OH/THC-COOH profiles often rose to greater concentrations than THC. That pattern is consistent with metabolism but does not quantify exactly what percentage bypassed the liver in any given patient.

A major analytical ambiguity remains: poor or inconsistent absorption can mimic variability in apparent clearance because the model uses F-normalized values. The authors explored variability in bioavailability but encountered estimation instability. As a result, it would exceed the data to say the trial proved a precise mucosal-versus-swallowed partition or a specific individual’s first-pass metabolic rate.

Could food, dry mouth or administration technique have caused the spread?

Those mechanisms are plausible but were not isolated into randomized causal contrasts in this analysis. Patients fasted before their first administration and received structured meals later. A nutritional drink was supplied 25 minutes after the first dose, while the second dose followed several hours after that drink. The participants were instructed not to swallow or move the tongue for a period after each administration. These design controls reduced some variation but cannot establish how the same person would respond to food, xerostomia or imperfect technique at home.

Thirteen participants had some xerostomia, making oral conditions clinically relevant, but the investigators did not demonstrate that xerostomia was the cause of a particular measured THC curve. Likewise, neither CBD co-administration nor smoking habits were experimentally varied for an independent pharmacokinetic effect. An explanation that is physically possible is not the same as one established by this 20-person study.

What does the active THC-OH metabolite add to interpretation?

THC-OH is a biologically active metabolite, while THC-COOH is typically treated as a later inactive metabolite. The study measured both alongside parent THC, but only THC and THC-OH were fitted simultaneously in the final model. A higher THC-OH concentration than THC in a specimen does not by itself identify an exact ingestion time, the amount swallowed, whether an individual was intoxicated at an earlier incident, or that a specific route was used.

The study followed patients for only about eight hours after the first dose and about four hours after the second. It did not capture the full terminal elimination profile or long-term accumulation. Parent-to-metabolite concentration ratios from this setting therefore do not justify a universal clock for oral-mucosal spray exposure, especially when specimen type, formulation, tolerance, co-exposures, and sampling time may differ in litigation.

Why is this evidence useful in a forensic toxicology dispute?

The study provides a concrete counterexample to the simplistic premise that a labeled THC dose maps to one predictable circulation concentration. Under supervised administration, exact dosing intervals, controlled food timing, serial sampling and a sensitive assay, the investigators still found large unexplained variation. If a case asks a toxicologist to infer an exact administration time or absorbed dose from a single specimen, the analyst needs validation far beyond this model.

At the same time, the study does not prove that a reported THC result is meaningless. Its contribution is narrower: laboratory measurements can accurately describe analytes in a specified specimen at a recorded collection time, while the mapping back to dose, route and effect remains uncertain. Source-specific assay validation, specimen handling, timing, full medication history and appropriately comparable human studies become central.

For supporting context on those distinctions, see the site’s guides to biological matrices, pharmacokinetics in toxicology, and confirmatory drug testing. Each supplies conceptual context; none converts this population model into a person-specific impairment finding.

What cannot be concluded from the 20-patient trial?

It cannot validate a whole-blood THC threshold for impairment; determine whether a particular driver was safe to drive; identify a person’s last spray use; establish a precise THC clearance per hour; assign a patient’s metabolite ratio to one exposure route; distinguish intentional misuse from prescribed use; or adjudicate witness accuracy, criminal responsibility or collision causation. These were not trial outcomes. It also cannot establish universal prescribing recommendations for older adults or definitive clinical effects of kidney disease, since the fitted model was unstable and follow-up was limited.

The source population deserves emphasis: older patients with diminished appetite, considerable oral and physiological variability, and closely supervised spray administration. Applying any parameter to a different population, repeated chronic use, another product, or a forensic whole-blood specimen is an additional extrapolation requiring independent support.

How should an expert evaluate an individual case using this research?

A defensible review starts with the complete analytical report rather than a population-model coefficient. Identify the physically collected matrix; whether it was whole blood, separated serum, separated plasma, urine or oral fluid; the specific analyte; the analytical identification and quantitation method; calibration and uncertainty; and the actual collection and storage times. Reconcile medication administration records, formulation, nominal delivered dose, timing and co-administered compounds. Then ask whether the proposed inference was tested in a comparable population with a known error rate.

When a party relies on model clearance, insist on distinguishing apparent CL/F from absolute physiologic clearance and on disclosing the 338/1,000 bootstrap successes, wide confidence intervals, terminal-phase truncation and specimen-label ambiguity. Where facts are absent, the conclusion should remain qualified rather than filled with convenient assumptions.

Frequently asked questions

Does two or three sprays guarantee a particular THC concentration?

No. The reported subgroup mean peaks differed and individual concentration-time courses varied substantially. Those subgroup means cannot guarantee one patient’s result.

Were the study’s THC results measured in forensic whole blood?

No. Venous blood was allowed to clot and the separated fraction was analyzed; authors called the concentration data plasma despite serum-like preparation. There is no paired validated whole-blood conversion.

What does a 0.25 ng/mL LLOQ establish?

It specifies the analytical quantitation limit for the reported assay. Values below it are not reliably quantified by that method; they are not necessarily chemically absent.

Is the reported 927 L/h clearance a person’s THC elimination speed?

No. It is a model-derived apparent total clearance divided by unknown bioavailability, with incomplete terminal sampling and substantial uncertainty.

Did measured kidney function or body fat predict an individual THC level?

The fitted model did not retain a clinically meaningful covariate from the measured characteristics in these 20 patients. That is not proof of no effect in all populations or a validated individualized prediction rule.

Can the THC-to-THC-OH ratio reveal how long ago someone used the spray?

Not from this study. Parent and metabolite concentrations changed across dosing occasions, and no validated ratio-based timing classifier was constructed.

Did investigators test whether patients were impaired or driving unsafely?

No. Their endpoint was blood-fraction cannabinoid pharmacokinetics and model fit, not driving, field sobriety, crash occurrence, legal impairment or witness performance.