Does Smoked Cannabis Cause Aversive Effects in People at Psychosis Risk? What a 12-Person Placebo Study Found

Short answer: In a small double-blind, placebo-controlled laboratory study, six regular cannabis users at clinical high risk for psychosis reported a stronger bad drug effect after active smoked cannabis than after placebo. They also copied fewer digit strings correctly in a computer task. Six psychiatrically healthy cannabis users instead showed an increase in ratings of a good drug effect after active cannabis. But several other cognitive outcomes were nonsignificant, and the researchers analyzed the two groups separately. The experiment therefore does not establish a statistically tested difference in the magnitude of THC’s effects between groups, a personal psychosis diagnosis, or an individual driving-impairment threshold.

The underlying primary study is Madden and colleagues’ Smoked Cannabis Effects in Cannabis Users at Clinical High-Risk for Psychosis: A Further Investigation of Cognition and Reward (2022), DOI: 10.26828/cannabis/2022.01.003. Its central forensic lesson concerns measurement and inference: a statistically significant result on a particular rating or laboratory task does not automatically generalize to every aspect of cognition, psychiatric illness, public-road driving, or the behavior of another person.

What question did the original experiment answer?

The investigators studied whether active smoked cannabis, compared with a cannabis cigarette containing no THC, altered certain subjective drug experiences and computer-based cognitive outcomes among regular cannabis users. One group met research criteria for clinical high risk (CHR) for psychosis; the other was psychiatrically healthy. The experiment was not designed as a roadside validation study, a THC blood-concentration study, or a trial of long-term psychiatric outcomes.

CHR is a research and clinical-risk designation, not a finding that the participants already had schizophrenia. The investigators assessed risk using structured interviews and required the risk symptoms not to occur exclusively in the context of cannabis use. The six CHR participants met the study’s attenuated positive symptom syndrome criteria. Participants with a prior psychotic or bipolar disorder were excluded. This definition matters when interpreting commentary that casually substitutes “people with psychosis” for “people at elevated risk.” Those populations are not interchangeable.

The study was a follow-up investigation using participants and methods connected to an earlier report. This particular paper concentrates on response inhibition, digit tasks, a drug-word attention task, and rewarding versus aversive subjective drug ratings. Findings attributed to the earlier paper should not be mistaken for new outcomes independently tested here.

Who participated, and how much statistical information is available?

There were 12 distinct non-treatment-seeking adult regular cannabis users: six CHR participants and six psychiatrically healthy controls. Participants were between 18 and 30 years old. The reported mean age was 23.2 years (SD 4.0) in the CHR group and 24.3 (SD 3.0) in the controls. The CHR group included five men and one woman; the control group included four men and two women. Group mean estimated IQs were 105.7 and 102.2, respectively. These are descriptions of twelve people, not large normative datasets.

Both groups reported beginning cannabis use at approximately age 16 and using it approximately four times weekly, on average. Eligibility required recent repeated cannabis use, supported by cannabinoid-metabolite-positive urine tests on two different days. That urine screening supported an exposure-history criterion; it did not quantify a THC concentration at the moment of the cognitive tasks. Two CHR participants, and no controls, were taking stable antidepressant medication. The authors could not adequately adjust for that difference with so few participants.

Not every analysis necessarily had complete observations from all six people in a group. Some reported ANOVA degrees of freedom are smaller than would be expected from six complete datasets. Thus, “six CHR participants” is the recruitment denominator, not a guarantee that each outcome had six fully analyzable matched measurements. The main report also directs readers to supplementary tables for more complete raw data and test results.

Was the cannabis dose controlled?

Researchers used National Institute on Drug Abuse cannabis cigarettes in a randomized, double-blind session sequence. The participants smoked one-half cigarette according to a paced-puff procedure. The sessions included placebo plant material labeled as 0.0% THC and active cannabis containing approximately 5% THC. The original methods list 5.05%, while the paper’s abstract, figure labeling and portions of the narrative use 5.5%; the percentage discrepancy should not be quietly resolved into a precisely known delivered dose.

The protocol contemplated a third, lower-potency cannabis session. Because two CHR participants did not complete that session, the authors formally analyzed the placebo and higher-potency sessions here. The test therefore does not furnish a complete three-level dose–response relationship. Reported average puff counts were not significantly different between the active and placebo sessions within either group: for CHR, approximately 4.0 versus 3.5 puffs, F(1,5) = 1.4, p = .30; for controls, approximately 4.7 versus 4.2 puffs, also F(1,5) = 1.4, p = .30.

The percentage of THC in plant material is not the milligrams actually absorbed, a serum THC value, a venous whole-blood THC concentration, or a measurement of intoxication severity. The experiment did not present a participant-level pharmacokinetic series correlating THC in blood with each task outcome. Puff count alone also does not establish pulmonary absorption. This distinction limits retrospective reconstruction of another person’s exposure.

What specimens and measurements were actually used?

The investigators collected urine toxicology evidence during screening and assessed recent use at sessions. They also used breath-alcohol and breath-carbon-monoxide checks, self-report, and a pre-administration field sobriety assessment. None of those procedures was a quantitative breath-THC measurement, a confirmatory blood-THC assay, or a post-dose on-road driving test. The reported research outcomes were primarily computerized responses and participants’ self-ratings, not chemical concentrations.

For additional context on the meaning of a tested physical specimen, see the existing guide to biological matrices in toxicology. For why a qualitative screen should not be presented as a chemically specific quantitative result, see confirmatory drug testing. Those general references provide interpretive context; neither is a substitute for the complete methods and results of this primary study.

How were the task and rating times organized?

The principal computerized cognitive battery was given before smoking, during the approximately 45-minute pre-administration interval, and again about 20–80 minutes after smoking. The Drug Stroop task was administered at approximately 130 minutes. Digit Recall and Digit Symbol tasks were administered again around three hours. The subjective ratings extended over multiple post-smoking assessments through approximately 245 minutes. The different timing means a Go/No-Go result, a Stroop effect and a late self-rating cannot be treated as if they all describe the same simultaneous functional state.

Participants were directed to abstain from cannabis that morning and from other psychoactive drugs apart from usual caffeine and permitted prescribed medicines. Screening and field checks were used before administration. Absence of observed pretreatment intoxication under those checks does not verify a cannabinoid-free biological baseline; regular users can have residual cannabinoid metabolites. Testing was conducted in a controlled laboratory and did not expose participants to road traffic, unexpected hazards, or real driving demands.

What happened to the aversive and rewarding ratings?

The CHR group rated the bad drug effect higher during the active cannabis condition than during placebo. The source reports a drug-condition main effect, F(1,5) = 14.0, p < .013, partial η² = .74. The corresponding test in the psychiatrically healthy control group was nonsignificant, F(1,5) = 1.0, p > .013. This is a within-CHR association with the experimentally administered condition, not proof of psychiatric deterioration or an enduring symptom disorder.

The positive-effect result had a different pattern. For good drug effect ratings, the control group showed a drug-condition-by-time interaction, F(5,25) = 3.8, p < .013, partial η² = .43, associated with greater ratings after active cannabis on selected assessments. The CHR interaction was reported as nonsignificant, F(2.0,10.2) = 2.0, p > .013. The overall main drug-condition effects on good-effect ratings were not significant in either group at the paper’s corrected threshold.

It would be incorrect to conclude that CHR cannabis users categorically never experience positive effects. Their plotted good-effect ratings sometimes rose after active cannabis, but the specified interaction did not meet the adjusted statistical criterion. Likewise, a nonsignificant control-group bad-effect test is not affirmative proof that every control person felt no discomfort. These are small samples analyzed as group-level comparisons of ratings.

Did Go/No-Go testing actually show worse post-smoking inhibition?

The paper describes the Go/No-Go measure as response inhibition and reports a significant CHR drug-condition-by-time interaction for commission errors: F(1,4) = 16.7, p < .025, partial η² = .81. The figure and prose require a critical qualification. On the active-cannabis visit, the CHR participants had fewer errors before smoking than on their placebo visit; that baseline difference met a simple-effect test (p < .05). After smoking, the study reports no significant active-versus-placebo difference in errors (p > .05). Neither group showed a significant Go/No-Go main drug-condition effect at the paper’s prespecified corrected threshold.

Consequently, the significant interaction must not be paraphrased as “the active group demonstrably made more Go/No-Go errors than placebo after THC.” A significant change in relative pattern can coexist with a nonsignificant post-dose condition contrast. The control-group Go/No-Go interaction was also nonsignificant, F(1,4) = 1.5, p > .025. The original Figure 2 is particularly important because it exposes the baseline imbalance obscured by a broad statement that response inhibition worsened.

This statistical distinction matters in expert testimony. The observed interaction answers a question about the shape of repeated measurements; the post-dose simple effect answers a different question about the comparison at the alleged exposure time. A lawyer or analyst should identify precisely which estimand a quoted p value concerns before using it to describe any person’s functional behavior.

Which digit and working-memory results were significant?

On the Digit Recall Task, the number of digit strings copied correctly while visible had a CHR drug-condition main effect, F(1,4) = 32.7, p < .013, partial η² = .89. CHR participants copied fewer strings correctly during active-cannabis sessions than placebo sessions. The corresponding control-group effect was nonsignificant, reported as F(1,5) = 0.0, p > .017. Copying while digits appear on the screen is not the same endpoint as remembering them after disappearance.

Indeed, the paper reports no statistically significant main drug-condition effect or drug-condition-by-time interaction on the Digit Recall immediate-recall percentage for either group (p > .017). The N-Back working-memory sensitivity measure, which combined 2-Back and 3-Back performance, also showed no statistically significant main drug effect or condition-by-time interaction in either group (p > .025). A descriptive narrative that says simply “memory was impaired” without naming the tested task conflates a positive digit-copy finding with distinct null immediate-recall and N-Back findings.

Digit-symbol substitution accuracy was also nonsignificant on both the main effect and interaction tests for either group at the corrected criterion (p > .017). This limits generalizations about uniform psychomotor slowing. The source discussion refers to psychomotor speed in its broader interpretation, but the particular digit-symbol percent-correct statistic presented in this paper did not meet the stated significance threshold.

Did the Drug Stroop establish greater attentional impairment?

The Drug Stroop task compared the interference associated with cannabis-related and other drug-related words. In the CHR group, the authors report a drug-condition-by-word-type interaction, F(1,4) = 7.9, p < .05, partial η² = .67. Their description is that interference from cannabis-related words, relative to mixed drug words, was reduced during active cannabis rather than placebo. Neither group showed a significant main effect of drug condition on interference reaction time (p > .05), and the corresponding word-type interaction in the control group was nonsignificant, F(1,5) = 0.1, p > .05.

A smaller interference score is not automatically global improvement or worsening of attention. The researchers speculated that the shift could reflect appetitive attentional processing. That mechanistic explanation was not independently validated as a measure of reward-system activity, and the task was administered later than the main 20–80-minute cognitive battery. For forensic use, its observed association should be identified by exact task, word contrast and timing.

Which positive and null results belong together?

The investigators applied different corrected alpha criteria across outcome families, sometimes .013, .017, .025 or .05. A nominal p below .05 is not necessarily significant when the applicable family-specific threshold is more stringent. The table follows the numerical test labels and denominators provided in the primary main text, without manufacturing confidence intervals that the paper did not report. Some test-specific details and missing-data procedures reside in the original supplementary tables.

Can these results prove that THC affects the two groups differently?

Not by a formal between-group test in this report. Because each six-person group was analyzed separately, one significant within-group result and one nonsignificant within-group result do not mathematically establish that the treatment effects differ significantly across groups. That would require an appropriate direct drug-by-group (and often time-by-drug-by-group) interaction analysis with sufficient power and an uncertainty interval for the between-group contrast. The authors expressly acknowledge that their small sample precluded their preferred mixed within- and between-group examination.

Readers can accurately say that the observed patterns differed across these two groups in this sample. They should not quote the within-group p values as if those numbers proved a statistically significant effect modification by CHR status. The distinction is especially important when applying group research to testimony about a particular defendant, witness, patient or employee.

Can a urine screening result stand in for blood THC or impairment?

No. These participants had a cannabinoid-positive urine exposure history, but this publication provides no paired quantitative whole-blood THC result at the exact cognitive assessment time. A metabolite-positive urine screen cannot establish the person’s contemporaneous parent-THC concentration in venous whole blood, a last-use interval, or a functional deficit on any particular trial. Likewise, pretreatment breath-alcohol and carbon-monoxide checks did not measure exhaled THC or quantify inhaled cannabis dose.

Assay identity is not a semantic detail. Urine can contain metabolites arising after use; separated plasma, serum and whole blood have different distributions and reported concentrations; oral fluid and exhaled breath have still other collection and interpretation problems. None of these values can be substituted for an experimental self-rating, task reaction time, or person-specific legal impairment decision without separate validation. The general pharmacokinetics guide addresses why dose, sampling time and specimen cannot be assumed interchangeable.

What could confound an individual interpretation?

Several features constrain external validity. Twelve participants provide imprecise estimates and limited information about the distribution of responses. Two CHR participants were receiving antidepressants whereas none of the controls were; CHR participants also reported more mood and anxiety conditions and more negative cannabis-effect expectancies. These factors are features of the selected groups, not randomized independent manipulations. The protocol excluded patients with prior psychotic disorders and those reporting serious previous cannabis reactions, which narrows generalizability.

All participants were experienced users rather than cannabis-naive individuals. Product THC potency was modest compared with some contemporary products, and the active dose was delivered as half a paced-smoked research cigarette rather than vape concentrate or oral edible. The study did not directly calculate actual absorbed dose or test a range of blood levels. Laboratory tasks and short observation periods are not equivalent to emergency assessment, police observations, workplace performance, a crash investigation, or psychosis incidence over years.

In addition, the active/placebo sessions were repeated under a crossover arrangement with potential baseline variability and task practice. The Go/No-Go baseline difference is an especially clear demonstration. The report gives repeated-measures results but does not provide a validated model for predicting any future individual’s performance from CHR diagnosis, a THC potency label, or a urine finding. Absence of a significant result in one small endpoint is not proof that an effect cannot exist; neither is a significant result proof of a universal effect.

How should an expert use this study in a forensic dispute?

First identify the proposed proposition. A narrow proposition—”a small placebo-controlled experiment documented higher aversive ratings within six CHR users after active smoked cannabis”—is supported. A stronger proposition—”all people at risk for psychosis become dangerously impaired after cannabis”—is not. Another unsupported shortcut is to infer that an individual’s urine cannabinoid finding proves the specific Go/No-Go, digit copying or subjective rating changes observed here.

Second match the outcome. If a dispute concerns a person’s alleged disorientation, a subjective bad-drug rating can illustrate possible subjective responses, not diagnose the person’s psychiatric condition. If it concerns reaction to a hazard, copied digits or a drug-word Stroop interference contrast is not a validated road-hazard or collision-probability measure. If it concerns a dated biological specimen, request the authentic analytical result with its matrix, sampling time, target analyte, calibrators, cutoff, uncertainty and chain of custody rather than borrowing a percentage THC figure from this cigarette protocol.

Third disclose the comparison actually tested. State when a result was a main drug effect, a condition-by-time interaction, a particular post hoc comparison, or a null. Do not convert a significant within-CHR p value into a formally significant CHR-versus-control differential, and do not convert the Go/No-Go interaction into a statistically significant post-smoking active-versus-placebo commission-error difference when the publication says that contrast was nonsignificant.

Finally, preserve competing explanations. CHR classification, medication, expectancy, baseline task performance, residual exposure and laboratory timing can all matter. Whether any of them caused a particular participant’s response was not separately randomized in this paper. A case-specific evaluation would need contemporaneous observations, records, medical context and appropriately validated measurement before reaching an individualized opinion.

Frequently asked questions about the study

1. Was this a human randomized, placebo-controlled cannabis experiment?

Yes. Twelve adult regular users completed active and placebo half-cigarette sessions in randomized, double-blind order. The principal analysis contained six CHR individuals and six psychiatric controls, but the paper analyzed the groups separately and some task-specific datasets were smaller. This makes it an informative pilot, not a definitive estimate of individual risk.

2. Did the study directly diagnose cannabis-induced psychosis?

No. CHR was a pre-enrollment clinical-risk designation, and people with a prior psychotic or bipolar disorder were excluded. The outcome was a short-term set of self-ratings and tasks. The experiment did not establish a new chronic psychotic illness or measure future conversion to psychosis.

3. Does the Go/No-Go result prove greater post-THC errors?

No. A CHR condition-by-time interaction was significant, but the paper reports a difference between visit baselines and no significant active-versus-placebo post-administration difference. Those are different statistical comparisons. Present the baseline imbalance and the nonsignificant postdose contrast together.

4. Were all memory outcomes impaired by active cannabis?

No. CHR participants copied fewer numbers correctly while visible on one task, but the immediate-recall percentage and N-Back d-prime drug comparisons were nonsignificant in the groups at the stated corrected criteria. Copying accuracy, immediate recall and working-memory sensitivity should not be collapsed into one result.

5. Was a positive urine test a measured THC concentration during the task?

No. Urine cannabinoid testing was used to corroborate recent use or to screen before sessions. The primary report does not supply a quantitative paired whole-blood THC result during each task. A metabolite screen cannot independently specify parent-THC concentration, time of use, behavior or impairment.

6. Does a significant CHR result and a nonsignificant control result prove susceptibility differs between groups?

Not without a direct, adequately powered between-group interaction test. Separate within-group p values alone do not establish a difference between their effects. The reported patterns are hypothesis-generating, especially with six people per group and differences in psychiatric history, medication and expectancies.

7. Can a court use these findings as a numeric DUI or witness-credibility cutoff?

No. The study measured ratings and laboratory cognitive tasks under controlled smoking, not validated on-road impairment or the truthfulness of testimony. It provides no sensitivity, specificity, likelihood ratios or blood-THC boundary for a person-specific legal conclusion. Such inferences require other validated evidence and the facts of the actual case.

Bottom line and original source

In the studied six-person psychosis-risk group, active smoked cannabis was associated with higher aversive ratings and fewer correctly copied digit strings. The paper also reports a Go/No-Go interaction affected by baseline differences, a specific Drug Stroop interaction, and important nonsignificant working-memory, immediate-recall, digit-symbol and desire-to-repeat comparisons. The healthy-user group showed a time-varying favorable self-rating pattern. These are selected experimental results in twelve experienced users; they are not a clinical diagnostic rule, a direct between-group differential test, a measured whole-blood THC–impairment relation, or proof of anyone else’s causal event.

Primary original research: Madden SP, Keilp JG, Wu O, Corcoran CM, Girgis RR, Vadhan NP. Smoked Cannabis Effects in Cannabis Users at Clinical High-Risk for Psychosis: A Further Investigation of Cognition and Reward. Cannabis. 2022;5(1):18–29. https://doi.org/10.26828/cannabis/2022.01.003. Interpretive statements here follow the specific experiment and its reported analysis rather than extending laboratory measurements to individual legal or clinical determinations.