Study design and participants
We conducted a three-session (one orientation session, two experimental sessions), within-subjects experimental laboratory study with healthy adults who were community-recruited from the Halifax Regional Municipality (Nova Scotia, Canada). We were unable to conduct an a priori power analysis for our particular analytic approach (generalized estimating equations) due to our within-subject study design parameters. To address this barrier, we conducted a power calculation after data collection using a similar, but less powerful analytic approach (repeated-measure ANOVA) in G*Power. Based on aggregated effect sizes of placebo effects for conditions that are very amenable to the placebo effect (e.g., anxiety; d = .29; Wampold et al. 2005), we are expected to have at least 85% power to detect within-factor effects with the sample size used in this study. As such, we expected to have sufficient statistical power to test our study hypotheses (i.e., main effects, two-way interactions).
Participants were required to be at least 19 years of age, as this is the age of majority in Nova Scotia. Selection criteria included ≥ 1 lifetime uses of cannabis, which was required to ensure that subjects had some experience with and knowledge about cannabis in attempt to standardize expectations to some extent. To help ensure that participants could meet the abstinence requirements, only individuals reporting cannabis use two or fewer days per week in the past month were enrolled in the study. In order to ensure cold pressor test (CPT) would be well-tolerated, participants were required to be medically healthy, and free of any serious medical conditions, or any history of fainting, seizures, circulatory disorders, heart problems, high blood pressure, diabetes, frostbite, or any current cut, sore, or fracture to their right hand/arm (Mitchell et al. 2004; Birnie et al. 2011). Subjects were also excluded if they reported current prescription medication use (except birth control in females) or any current psychiatric disorder, as diagnosed by a health care professional, including substance use disorders (American Psychiatric Association 2013). These exclusions helped prevent pre-existing neurophysiological or psychological conditions from influencing subjective and physiological stress, mood, and anxiety responses to the laboratory stressor. Participants were also required to be cannabis oil naïve (to enhance believability of the oil manipulation), and to have never previously participated in a study conducted by our group that involved deception.
Stress and anxiety induction
The Maastricht Acute Stress Test (MAST; Smeets et al. 2012) was used to induce stress and state-anxiety in our sample. The MAST was chosen since it possesses both physical and psychological features that have been demonstrated to reliably provoke subjective and physiological responses associated with stress and anxiety in laboratory settings (Smeets et al. 2012; Bali and Jaggi 2015) over multiple sessions, with little habituation (Quaedflieg et al. 2017). The physical feature is a CPT and the psychological feature mental arithmetic challenges that include a psychosocial evaluative threat (Smeets et al. 2012; Bali and Jaggi 2015).
As per the validated protocol (Smeets et al. 2012), the MAST involved a 5-min anticipation phase, in which instructions and procedures are explained to participants, followed by a 10-min acute stress phase. During the stress phase, participants engaged in trials alternating between (i) immersing their hand into ice-cold water (2 °C) (i.e., CPT), and (ii) counting backwards in steps of 17 or 13 starting at a random four-digit number. Both tasks are combined with negative social-evaluative pressure (i.e., negative feedback and videotaping).
Electrocardiogram (ECG) data was collected continuously throughout the experimental sessions using an Equivital EQ02 sensor electronic module (SEM) equipped to a fitted Life Monitor belt (ADInstruments; [ADI], Colorado Springs, USA). The EQ02 device measured ECG signal on two channels via three electrodes at a sampling rate of 1000 Hz. The raw ECG signals were later transferred to a computer and used to compute indices of heart rate and HRV, a robust, non-invasive physiological measure that has been used to assess stress and anxiety responses (Kim et al. 2018).
Demographics and CBD belief ratings
Demographic information, including age, sex, ethnicity, and level of education, were collected with a researcher-compiled self-report questionnaire. Additionally, information about participants’ baseline/a priori beliefs regarding the effects of CBD were collected using three researcher-compiled single-item questions. Participants reported on the extent to which they believed statements about the mood-, stress-, and anxiety-related properties of CBD (i.e., “improves mood,” “reduces stress “reduces anxiety”) on a 10-point scale (1- “Not at all,” 10- “Completely”). Lifetime and past month cannabis use frequency information was collected using single items via telephone screening as part of the study selection criteria.
Subjective stress, anxiety, mood, and drug effect ratings
Participants reported their current subjective state using a combination of validated measures and researcher-compiled single-item scales. Subjective stress was assessed with a single-item Numerical Rating Scale (NRS) where subjects rated the extent to which they felt “stressed” on a 10-point scale (1- “Not at all,” 10- “Extremely”). Similar single-item scales have been shown to demonstrate adequate construct validity (correlations between .45 and .66 with other validated stress measures) and discriminant validity (i.e., stressed vs. non-stressed states) (Lesage et al. 2012).
Subjective anxiety was assessed with a six-item shortened state version of the State-Trait Anxiety Inventory (STAI-S-SF; Marteau and Bekker 1992). Participants rated six statements about their current state (e.g., “I am tense”) and rate them on a 4-point scale (1- “Not at all,” 4- “Very much”). The STAI-S-SF has been shown to possess good reliability (α = .82; Marteau and Bekker 1992). It also produces acceptable validity, generating similar scores to those obtained using the full 20-item STAI-S (Spielberger 1983) (.91 total score correlation; Marteau and Bekker 1992), which is sensitive to rapid state-dependent fluctuations in anxiety (Rossi and Pourtois 2012). To calculate total anxiety scores, the three positive STAI-S-SF items were first reverse scored. Next, all scores were summed then multiplied by 20/6 to yield total scores between 20 and 80.
Subjective mood was assessed with the ten-item International Positive and Negative Affect Schedule, Short Form (I-PANAS-SF; Thompson 2007). Participants were asked to rate the extent to which they presently feel a list of positive affect (e.g., “Alert,” “Inspired”) and negative affect (e.g., “Upset”, “Hostile”)-related items on a 5-point scale (1- “Very slightly or not at all,” 5- “Extremely”). Both positive and negative affect subscales of the I-PANAS-SF have been shown to possess adequate reliability (α = .78 and .76, respectively), as well as acceptable convergent validity with measures of subjective well-being (Thompson 2007). To calculate total scores, the five items from each subscale within the I-PANAS-SF were summed to create a positive affect and negative affect score (subscale scores range between 5 and 25).
Subjective drug effects were assessed using the six-item Brief Biphasic Alcohol Effects Scale (B-BAES; Rueger et al. 2009). Participants rated how well three sedation items (e.g., “Sedated”) and three stimulation items (e.g., “Energized”) described their current feelings on a 10-point scale (1- “Not at all,” 10- “Extremely”). The subscales within the B-BAES correlated highly (.92–.97) with the full form of the BAES (Martin et al. 1993), demonstrating adequate criterion validity, and showed excellent internal consistency reliability (α = .89–.91; Rueger and King 2013). Though the B-BAES was initially developed to evaluate the biphasic stimulation and sedation effects associated with alcohol use, the questions are not specific to alcohol and thus were used to assess subjective sedation- and stimulation-related drug effects in this study. To calculate total scores, the three items from each of the two subscales within the B-BAES were summed to create a sedation and stimulation score (subscale score range between 10 and 30). Additionally, two researcher-compiled NRS items (“intoxicated,” “relaxed”) rated on a 10-point scale (1- “Not at all,” 10- “Extremely”) were included in the assessment of potential drug-related effects.
Once eligibility was confirmed via telephone screening, participants were scheduled for an initial orientation session (~ 30 min). After providing consent, participants had their weight measured, which they were informed would determine the dose of oil that they would be given during their experimental sessions. Demographic and CBD belief rating information was collected. Two experimental sessions were then scheduled between 10:30 and 18:00 h. The laboratory setting, testing procedure, and time of day was kept constant for each participant across all sessions to minimize circadian fluctuations in the stress response (Nicolson and van Diest 2000). Female participants not using birth control were tested during the luteal phase of their menstrual cycle to minimize menstrual cycle-related fluctuations in the stress response (Barel et al. 2018). Experimental sessions were separated by a minimum of 1 week and a maximum of 1 month.
All participants received CBD-free hemp seed oil across both experimental sessions but received different instructions during each session about the CBD content of the oil (told CBD-containing vs. told CBD-free), in randomized order. This produced two conditions: (a) told CBD, administered CBD-free; (b) told CBD-free, administered CBD-free, allowing for an assessment of the effects of CBD-related expectancy, independent from pharmacology (Sutton 1991). Experimenters were blind to the expectancy condition, as oil was administered by an independent blinder who otherwise did not interact with the participant, and participants were blind to the actual pharmacology of the oil.
Experimental sessions (~ 3 h) took place following a minimum of 72 h of abstinence from cannabis, given the ~ 31 h half-life of CBD and THC in infrequent users (Smith-Kielland et al. 1999; Millar et al. 2018). Additionally, 12 h of abstinence from alcohol, tobacco smoking, and other drug consumption was required (Holford 1987; Benowitz and Jacob 1994). Participants were also required to abstain from caffeine for a minimum of 2 h (Benowitz et al. 1995) as well as to fast for 1 h prior to their session. Abstinence from substances was verified via self-report since all participants were pre-screened to be healthy, infrequent cannabis users with no current substance dependencies. To increase compliance to the study procedures, participants were sent multiple email reminders about their upcoming experimental sessions and the respective abstinence requirements.
Following the collection of baseline subjective and HRV data, participants were administered hemp seed oil sublingually by an independent blinder. To enhance the believability of the drug content instructions, participants were presented with their assigned oil in packaging consistent with the instructions provided. Participants were informed during the consent process, and by the independent blinder, there would be a 90-min “absorption period” following oil administration (to mimic the absorption period of CBD; Zuardi et al. 2017). During this period, participants were provided with neutral word puzzles and reading material to pass the time. For the second time (post-absorption), participants completed the same battery of assessments as used at baseline. To induce stress and state-anxiety, the MAST protocol was administered by the experimenter. Immediately following completion of the MAST, subjective measures were re-administered for the third time (post-stress), and for a final time 10 min later (recovery).
At the end of each experimental session, participants were asked about the CBD content of the oil that they had self-administered, with the following response options “CBD oil,” “CBD-free/hemp seed oil,” or “Unsure.” This served as a manipulation check to determine whether participant beliefs regarding drug assignment were consistent with the CBD content stated in their instructions. It was decided a priori that sessions where participant did not believe the CBD content information provided would be excluded from the analyses to avoid confounding the interpretation of results. Lastly, to ensure that the deceptive nature of the study was not revealed to potential future participants, full debriefing of the nature and aims of the study was delayed until study data collection concluded.
Data acquisition and ECG pre-processing
Raw ECG data were extracted from Lead II with LabChart Pro software (HRV 2.0 module; ADI). HRV was calculated by extracting beat-to-beat RR intervals. Ectopic beats were excluded from analyses, using the Lomb-Scargle Periodogram, to enable exclusion of ectopic beats without interpolation. To reduce baseline wandering, all ECG signals were passed through a high pass filter (.5 Hz). All segments used in analyses were visually inspected for ectopic beats and noise. If noise or ectopic beats exceeded 5% of total beats in an ECG segment, they were excluded. An artifact-free 5-min segment during the first 70 min of the session was selected as a baseline. A 5-min segment was selected during the anticipation phase of the MAST (anticipation), and the two 5-min segments during the arithmetic and cold pressor components of the MAST were averaged to compute HRV during acute stress (stress). The final 5-min segment was selected 10 min after the MAST (recovery). Additionally, an ECG-derived respiration rate (EDR) was manually calculated from raw Lead II ECG as per recommendations (Brugnera et al. 2018).
Heart rate (HR) and the root mean square successive difference (RMSSD), a time-domain index of HRV, were extracted from the RR data. RMSSD is a widely used index of HRV that is thought to reflect parasympathetic output and successful emotional regulation (Laborde et al. 2017).
Statistical analyses were conducted in SPSS, version 25.0. Generalized estimating equations (GEE) were used for all primary analyses because they have robust estimators and can accommodate missing data as well as non-normal distributions (Hubbard et al. 2010). Multiple models per outcome were conducted to determine the optimal fit for the data based on the lowest number of parameters and the lowest Quasi Likelihood under Independence Model Criterion (QIC). First, the dependent variable was visually screened to identify plausible distributions, which were then compared with the covariate structure specified as Unstructured. Once an optimal distribution was chosen, plausible covariate structures were tested. The Exchangeable correlation matrix tended to be most parsimonious among all models.
Subjective outcomes included stress, state anxiety (STAI-S-SF), mood (I-PANAS-SF positive affect and negative affect), and drug effects (B-BAES sedation and stimulation; intoxication; relaxation). For subjective outcomes, time (baseline, post-absorption, post-stress, recovery), and expectancy condition (CBD, CBD-free) were entered as repeated factors. Physiological outcomes included HR and HRV (i.e., RMSSD). The physiological outcomes were analyzed in a similar fashion to the subjective outcomes. Specifically, Time (baseline, anticipation, stress, recovery) and Expectancy condition (CBD, CBD-free) were entered as repeated factors, and EDR was entered as a covariate to control for respiratory influences on HR and HRV (Brugnera et al. 2018). Effects of interest for subjective and physiological outcomes included main effects of time and interactions between Time and Expectancy condition. Planned post-hoc pairwise comparisons using tests of simple effects were used to probe main effects of time and Time by Expectancy condition interactions. We also examined whether a priori beliefs about CBD influenced corresponding stress, anxiety, and mood responses according to expectancy condition. The corresponding CBD belief rating and baseline outcome values were entered as covariates. Time (all time points following oil administration: post-absorption, post-stress, recovery) and expectancy condition (CBD, CBD-free) were entered as repeated factors. Effects of interest included expectancy condition by belief interactions on overall stress, anxiety, and mood ratings. Given that GEE in SPSS does not have the capability of probing interactions involving continuous predictors using tests of simple effects, we used “geepack” in R (version 4.0) to probe significant interactions involving CBD belief rating. Post-hoc tests of simple effects involved contrasts between expectancy condition across three levels of CBD belief ratings (i.e., terciles). All p-values less than .05 were considered significant. Additionally, the Benjamini-Hochberg method (Benjamini and Hochberg 1995) was used to control for the false discovery rate (FDR) within each model (i.e., family) tested. The FDR threshold was set at .05 such that there was a 5% chance that any finding within each model was a false discovery. All p-values were reported in their original format unless the FDR threshold was exceeded, in which case both adjusted and unadjusted p-values were reported.