One of the 32 participants was excluded from analyses because their pretest blood levels exceeded mean + 3 standard deviation over all participants, when considering the combination of THC + metabolites level (sum of THC, THC-COOH and 11-OH-THC) and CBD level.Footnote 3 This reduced the THC + CBD strain group from 17 to 16 participants (see Table 1).
As seen in Table 2, the strain groups did not significantly differ in age, first age of regular cannabis use, or time away from the van. We did observe a significant difference in cannabis consumption for the past 30 days, showing more cannabis use in the THC group compared to the THC + CBD group.
Cannabinoid content
One participant did not weigh her or his product, so dosage results are based on only 14 subjects in the THC group. As reported in Table 2, the groups did not differ significantly in the total amount (mg) of product they consumed during at-home administration. However, they did differ in the amount (mg) of CBD and THC. As expected based on product content and group assignment, and as shown in Table 2, results indicated that each group differed on THC and CBD dosages in the expected directions. The THC group had the highest THC doses and the CBD group had the highest CBD doses.
Cannabinoid plasma biomarker levels
Cannabinoid plasma biomarker levels (Table 2) were analyzed in a mixed-design ANOVA with 2 sessions (pretest, posttest) and 2 cannabinoid types (CBD, sum THC + metabolites) as within-subject factors, and strain (THC, THC + CBD) as a between-subject factor. Pre-test THC levels fell < 10 ng/mL on average across both groups, supporting that participants complied with day of abstinence procedures prior to their mobile laboratory study appointment.
Analysis of cannabinoid plasma biomarker levels revealed a main effect of session, F(1,29) = 11.44, p < .001, \( {\eta}_p^2 \) = 0.28, and a significant main effect of cannabinoid type, F(1,29) = 16.12, p < .001, \( {\eta}_p^2 \) = 0.36. Cannabinoid type interacted with strain group, F(1,29) = 5.25, p < .05, \( {\eta}_p^2 \) = 0.15, showing that sum THC + metabolite levels were higher for the THC group compared to the THC + CBD group (pbf < .05). Cannabinoid type interacted with session, F(1,29) = 7.69, p < .01, \( {\eta}_p^2 \) = 0.21, showing that the level of sum THC + metabolites was higher at posttest (i.e., after cannabis use) compared to pretest (pbf < .001). There was a significant 3-way interaction between cannabinoid type, strain group, and session, F(1,29) = 5.42, p < .05, \( {\eta}_p^2 \) = 0.16. When this interaction was decomposed with Bonferroni-corrected post hoc tests, they indicated that the strain groups did not differ on any pretest levels, but posttest sum THC + metabolites levels were higher for the THC group than the THC + CBD group (pbf < .001). When testing each measure separately (Table 2), we only observed a significant difference for THC levels at pretest. Posttest CBD levels were higher for the THC + CBD group than the THC group, whereas posttest THC levels and sum THC + metabolites were higher for the THC group than the THC + CBD group.
Cannabis dose and strain effects on memory
First, we ran a regression model (Eq. 1) to examine how cannabinoid levels (THC + metabolites and CBD) were associated with accuracy (d’). The model revealed that the level of THC + metabolites was significantly negatively correlated to accuracy (p < .05, \( {\eta}_p^2 \) = 0.28) (Fig. 2a), but neither the effect of CBD (Fig. 2b) nor the THC × CBD interaction was significant. This result was observed across the two strain groups, and neither THC nor CBD blood levels were significantly correlated with d′ within each strain group.
Second, accuracy (d′, Fig. 3) was analyzed in a mixed-design analysis of variance (ANOVA) with session (pretest, posttest) as a within-subject factor, and strain group (THC, THC + CBD) as a between-subject factor. d′ significantly decreased between pre- and post-test, F(1, 29) = 5.84, p < .05, \( {\eta}_p^2 \) = 0.17, and d′ was significantly higher for the THC + CBD group compared to the THC group, F(1, 29) = 6.05, p < .05, \( {\eta}_p^2 \) = 0.17. The significant session × strain group interaction, F(1,29) = 7.90, p < .01, \( {\eta}_p^2 \) = 0.21, showed that accuracy was lower at posttest than pretest for the THC group (pbf < .01), but not for the THC + CBD group. We also observed that the accuracy at posttest was lower for the THC group than for the THC + CBD group (pbf < 0.01). Additionally, sum THC + metabolite blood plasma levels were included as a covariate since it significantly predicted memory accuracy in the regression analysis and because the THC content of the product consumed by the THC + CBD group was lower in THC. As performed in previous analyses, we used the log transform of metabolite data. The covariate log (THC) was significant, F(1,28) = 7.79, p < .01, \( {\eta}_p^2 \) = 0.22. The significant session × strain group interaction, F(1,28) = 6.18, p < .05, \( {\eta}_p^2 \) = 0.18, showed similar results as before, with lower accuracy at posttest compared to pretest for the THC group (pbf < .01), but not for the THC + CBD group. Also, the accuracy at posttest for the THC group was lower than for the THC + CBD group (pbf < 0.01).
Consistent with our approach for d’, each of the other performance measures was separately analyzed in a mixed-design analysis of variance (ANOVA) with session (pretest, posttest) as a within-subject factor, and strain (THC, THC + CBD) as a between-subject factor. Results are presented without a covariate. When adding log (THC) as a covariate, no significant effects were observed for the 4 measures. Analysis of false alarm (FA) rate indicated a significant main effect of session, F(1, 29) = 18.45, p < .001, \( {\eta}_p^2 \) = 0.39, showing a higher rate of FA at posttest compared to pretest. Session also interacted with strain for FA, F(1, 29) = 4.86, p < .05, \( {\eta}_p^2 \) = 0.14, such that only the posttest FA rate was higher for the THC group than for the THC + CBD group (Table 3). Analysis of response bias (c) indicated a significant effect of session, F(1, 29) = 5.79, p < .05, \( {\eta}_p^2 \) = 0.17, such that subjects were somewhat conservative pretest (tended to respond “no” more than “yes”) but somewhat liberal posttest (tended to respond “yes” more than “no”). Analysis of hit rate and reaction time revealed no significant effects. The presence of significant posttest hit rate effects in the t tests (Table 3), but not in the ANOVA, suggests that ANOVA did not have sufficient power to detect the session × strain interaction for this outcome.