How Cannabis Causes Paranoia: Using the Intravenous Administration of Δ9-Tetrahydrocannabinol (THC) to Identify Key Cognitive Mechanisms Leading to Paranoia

Daniel Freeman; Graham Dunn; Robin M. Murray; Nicole Evans; Rachel Lister; Angus Antley; Mel Slater; Beata Godlewska; Robert Cornish; Jonathan Williams; Martina Di Simplicio; Artemis Igoumenou; Rudolf Brenneisen; Elizabeth M. Tunbridge; Paul J. Harrison; Catherine J. Harmer; Philip Cowen; Paul D. Morrison


Schizophr Bull. 2015;41(2):391-399. 

In This Article



There were 121 participants. The inclusion criteria were: aged between 21 and 50, had taken cannabis at least once before, and reported a paranoid thought in the past month as assessed by the Paranoid Thoughts Scale Part B.[21] The exclusion criteria were: a history of mental illness or substance dependence, a history of major mental illness in a first-degree family member, a neurological condition, a heart condition, a history of fainting, photo-sensitive epilepsy, or being a pregnant or nursing mother. The inclusion criteria concerning age and previous use of cannabis, as well as all the exclusion criteria, were designed to minimize the chances of any significant adverse effects. The psychiatrists carrying out the administration of the vials also carried out a brief health check. The study had received approval from an NHS research ethics committee, and written informed consent was received from all participants. Participants were recruited via distribution of leaflets to local postcodes and the playing of local radio adverts in Oxfordshire, UK.[22] The study was also advertised at a number of colleges in the University of Oxford, at Oxford Brookes University, and on the website of a local newspaper. The adverts did not mention cannabis or paranoia. The wording was: "Volunteers Required for Psychological Research. We are looking for volunteers to take part in a medical research study being carried out at the university. The research would take three hours and you would be compensated for your time. If you would like to hear more about the research, then please contact us. We send detailed information about the study so that you can consider whether you would like to take part." The individuals who responded were then invited to take part in the screening stage. Depending on participant preference, the screening questionnaires were either completed online via a web-link or sent via post. Therefore, screening was self-report. A total of 1792 people were screened. The reasons for exclusion (which could be several for a person) were: outside the age range (n = 611), had not taken cannabis before (n = 1117), did not report persecutory ideation (n = 739), had a history of mental illness (n = 281), substance abuse (n = 41), or major mental illness in a first-degree family member (n = 100), had a neurological condition (n = 37), had a history of fainting (n = 53), had a heart condition (n = 83), had photo-sensitive epilepsy (n = 2), were pregnant or a nursing mother (n = 17), had other physical health problems (n = 5), or had high blood pressure prior to administration of the vial (n = 5). Totally, 106 people were suitable but declined to take part or were not contactable.


The study was a between-groups design. After assessment on the cognitive variables, participants were randomized to either placebo, THC, or THC with cognitive awareness. Randomization was carried out by a researcher independent of recruitment and testing, using randomized permuted blocks of varying size with a plan created from Participants, assessors, and the psychiatrists were blind to the placebo and THC conditions. For the cognitive awareness condition, both participant and assessor were aware the participant was to receive THC. After administration of the contents of the vials, participants completed the paranoia assessments and repeated the cognitive tests. Testing on all measures postadministration was complete in an average of 83 minutes (SD = 18). There was a debriefing at the end of testing. A taxi was provided for returning participants home at the end. Participants were contacted the following day to check for any adverse effects.

Randomization Conditions

THC (Dronabinol) was supplied by THC Pharm GmbH and prepared as vials for injection by Bichsel Laboratories according to the method of Naef and colleagues.[23] The THC had 99.5% purity. A 1.5 mg dose of THC was used. Vials of the placebo and THC were indistinguishable. A psychiatrist administered either placebo (10 ml of saline) or synthetic THC (1.5ml of THC diluted in saline to 10 ml volume). The solutions were administered via an indwelling forearm cannula in 1 ml pulses every 1 minute for 10 minutes. Blood pressure was monitored during this period, and the psychiatrist was always available during the testing. Lorazepam (oral, 1–2 mg) was available in case a person became significantly distressed but this was not used during the study.

The cognitive awareness condition, given before THC administration, involved a simple 5-minute educational module, explaining the range of effects that the drug can cause (THC was considered synonymous with cannabis for this procedure). From before to after the awareness training, ratings on 2 visual analogue scales significantly increased; these assessed the participants' potential attribution of effects ("After being given the cannabis, how much do you believe that if you have any good or bad feelings over the next 90 minutes that cannabis will be the cause?"), t [40] = −6.59, P < .001, and confidence to keep this in mind ("If you feel different after being given cannabis, how easy will it be to keep in your mind that the cannabis is the cause?"), t [40] = −5.11, P < .001.


The National Statistics Socioeconomic Classification analytic classes were used.[24] History of illicit drug use was collected using the Maudsley Addiction Profile.[25] Intellectual functioning was estimated using the Wechsler Abbreviated Scale of Intelligence.[26]

Paranoia was assessed in 4 ways. A total score (Paranoid VAS) was obtained from 6 visual analogue scales ("Right now I feel suspicious of other people," "Right now I feel that people want to harm me," "Right now I feel like people want to upset me," "Right now I feel like people are against me," "Right now I am thinking that others are trying to persecute me," and "Right now I feel like people are being hostile towards me") (Cronbach's α = .90), which were used at baseline, after administration of the vial contents, and at the end of testing. Immediately after vial administration, participants were escorted on a 5-minute walk to the virtual reality testing room. On the way, they were asked to go into the hospital canteen and purchase an item. After doing this, they completed the 6 paranoia visual analogue scales in relation to the people encountered on the walk (Social Situation Paranoia). They then completed 5 minutes in an immersive virtual reality social environment (see supplementary figure 1 The virtual underground was displayed using an NVIS SX111 head mounted display (as used in Freeman et al[27]). Because the environment is neutral, any perceived hostility is known to be unfounded. Participants' views of the computer characters were assessed using the State Social Paranoia Scale (SSPS)[28] and a visual analogue scale ("Please mark on the line how hostile you thought the people on the tube were") (VAS VR Hostile). Participants were also assessed by an interviewer on the suspiciousness item of the Positive and Negative Symptom Scale (PANSS Suspiciousness).[29] The Community Assessment of Psychic Experiences (CAPE)[30] was included, completed for the time period since vial administration.

The following putative mediation variables were assessed: anomalous experiences using the Cardiff Anomalous Perceptions Scale (CAPS);[6] anxiety using the Beck Anxiety Inventory;[31] visual analogue scales for anxiety ("how anxious you feel right now"), depression ("how miserable or sad you feel right now"), worry ("how worried you feel right now"), and self-focus ("Right now my attention is focused on my inner thoughts and feelings," "Right now my attention is focused on how I appear to others," "Right now my attention is focused on my surroundings"); catastrophizing;[32] interpersonal sensitivity;[33] negative and positive views of the self and others using the Brief Core Schema Scales;[34] threat anticipation;[35] jumping to conclusions;[36] belief flexibility assessed by responses indicating that there is a possibility that participants could be mistaken in 3 high conviction beliefs; and working memory assessed by the digit span and letter-number subtests of the WAIS III.[37] Questionnaires were adapted for state use as appropriate.

Statistical Analysis

All analyses were carried out using Stata[38] and SPSS.[39] The analyses involved data reduction (simplification) using principal components analyses (PCA), our aim being to generate orthogonal (uncorrelated) composite measures to eliminate the problems of colinearity. No attempt was made to fit a formal measurement model (eg, via confirmatory factor analysis), the latter being both unnecessary and inappropriate for an exploratory analysis of the present type. Instead of doing multiple analyses of the paranoia outcomes, a PCA was undertaken (on the correlation matrix) for all 6 measures together (Paranoid VAS post THC administration, Paranoid VAS at end of testing, Social Situation Paranoia, SSPS, VAS VR Hostile, and PANSS Suspiciousness) and the first principal component (representing 71% of the total variation) was used as the primary outcome in all future analyses. This component was scaled to have a mean (for the whole sample) of 0 and a SD of 1 (note that the scaling is arbitrary and has no effect on the findings). The following weights were used to calculate this first principal component from the above 6 standardized paranoia assessments, respectively: 0.433, 0.428, 0.350, 0.417, 0.41, and 0.398. Principal component scores were available for 119 participants.

Two binary dummy variables were then created: THC = 0 for placebo, =1 for THC or THC + Cognitive awareness; Cognitive Awareness = 0 for placebo or THC, =1 for THC+ Cognitive Awareness. When both are entered together as explanatory variables in a regression model, then the parameter corresponding to THC is the estimate of receiving THC and the parameter for Cognitive Awareness is the effect of the cognitive awareness condition on those receiving THC. Intention-to-treat (ITT) effects were then estimated using linear regression (ANCOVA). The initial visual analogue scales for paranoia and for anxiety were both used as baseline covariates to increase precision.

Similar to the data reduction exercise for the paranoia measures, all of the 27 mediation variables were first entered into a PCA, resulting in 8 principal components with an eigenvalue greater than 1.0 (the well-known Kaiser criterion—explaining 30.1%, 9.2%, 7.2%, 6.1%, 5.5%, 5.0%, 4.2%, and 3.9% of the total variation, respectively). These components were then subjected to an orthogonal (Varimax) rotation for ease of interpretation. The rotated components were then used as the mediators. The main variables contributing to these rotated components were: anomalous experiences, anxiety, worry, depression and negative beliefs about the self (component 1); internal self-focus of attention (component 2); working memory (component 3); external focus of attention (component 4); interpersonal sensitivity (component 5); belief flexibility and positive beliefs about self and others (component 6); threat anticipation (component 7); and catastrophizing and jumping to conclusions (component 8). Full details of the components are provided in supplementary table S1 Component scores were available for 114 participants.

Mediation was tested using regression methods following the logic of Baron and Kenny.[40] There were three steps to determine mediation: an ITT effect of randomization condition on the outcome (paranoia); an ITT effect of randomization condition on the putative mediator; and when the outcome (paranoia) is jointly predicted by both the randomization condition and mediator, the mediator effect is significant and the randomization condition effect is lowered. In these regressions, the binary dummy variables THC and Cognitive Awareness were entered together to test the effects of the randomization conditions, and the initial visual analogue scales for paranoia and for anxiety were used as baseline covariates.

Acknowledging that we are primarily carrying out an exploratory analysis, we did not alter significance levels for multiple testing, agreeing with the view that "simply describing what tests of significance have been performed, and why, is generally the best method of dealing with multiple comparisons."[41] Data on all outcomes was available for 114 participants and with such small amounts of missing data, this is unlikely to be a cause of any significant biases. All significance testing was two-tailed.