The app includes 50 negative symptoms along with “wellness” that the user can select as the target of their cannabis treatment, with the user capable of treating more than one symptom simultaneously in a session. Out of these 51 options, we selected the three distress-related symptoms available for selection in the app: agitation/irritability, anxiety, and stress. The app also includes 47 side effects (called “feelings” in the user interface), which the user can report at any time during a session. The available symptoms and side effects were generated through focus groups, by the app developers, and by beta user suggestion. Sessions where patients treated a distress-related symptom were included. Only sessions with baseline symptom intensity levels exceeding zero were included in order to allow for the existence of a treatment effect. We further restricted our sample to symptom levels reported within 4 h post-cannabis consumption, similar to previous investigations (Cuttler et al. 2018; Vigil et al. 2018). In other words, we included only sessions with at least one post-cannabis symptom level reported within 4 h. A total of 23,055 cannabis administration sessions, recorded by 4127 individuals, reported a baseline symptom intensity of one or greater for at least one cannabis administration session used to treat anxiety, agitation/irritability, or stress. We further restricted the sample to include only sessions that reported inhaling dried, natural flower, the most common and homogenous type of cannabis product recorded in the Releaf App data (Stith et al. 2019), leaving 14,693 sessions recorded by 3061 users. Because THC and CBD levels are not mandatory recording, these variables are less commonly reported, and our sample is, therefore, further reduced when we restrict the sample to cannabis administration sessions with a full set of product characteristics (subspecies, inhalation method, and THC and CBD levels) reported. We also did not include sessions with THC or CBD levels exceeding 30%/dry wt. because levels exceeding 30% are unlikely to occur naturally in the Cannabis plant. Our THC and CBD measures are not mutually exclusive product categories, but rather track potencies, from 0 to 100%, as voluntarily reported by users, presumably based on product labeling. (Including only sessions with THC and CBD reported potentially biases our sample towards sessions using products purchased from dispensaries. All recreational and medical retail markets in the U.S. require labeled independent potency testing by certified laboratories, but individuals, who may, for example, be home cultivating, are unlikely to have access to the necessary equipment or be willing to pay prices designed for commercial retailers testing large product batches.) The final analysis sample includes 2306 cannabis administration sessions by 670 individuals who recorded at least one user session between June 06, 2016, and February 23, 2019. Among these sessions, 18.3% reported agitation/irritability, 43.3% reported anxiety, and 38.4% reported stress. Side effect reporting is optional, so our side effect analysis is restricted to a sample of 1519 sessions recorded by 559 users.
The study outcomes are the change in symptom severity level (symptom relief) and the prevalence of side effects following cannabis consumption. Symptom relief is measured as the minimum symptom severity level within 4 h minus the baseline symptom intensity. All cannabis sessions in our final sample include at least one symptom update within 4 h following cannabis consumption with 2.6 (SD = 1.8) symptom updates in the average session. The resulting symptom relief outcome ranges between − 10 (maximum relief) and 9 (maximum exacerbation). In addition to our primary outcome, maximum symptom relief, we also report results for symptom relief within the specified time periods of 1, 2, 3, and 4 h, i.e., the last symptom level reported within that time period minus the baseline symptom intensity. To measure the prevalence of side effects, we used dummy variables to indicate if the user reported any of the side effects in the category as well as variables measuring the proportion of total side effects selected by the user within each category.
A multivariable panel regression approach was used to analyze the association between symptom intensity level and cannabis use and the association between product characteristics and symptom relief, controlling for baseline symptom intensity and session length (minutes). To address the concern that symptom intensity changes in response to cannabis reported by the same user are systematically correlated due to individual-specific characteristics, user-specific fixed effects models were used to account for time-invariant user-specific attributes. As such, the effect of cannabis use on symptom intensity level was estimated from a comparison of symptom intensity levels reported by the same user before and after cannabis use. Similarly, the effect of product characteristics on symptom relief was estimated from a comparison across different products by the same user, rather than a comparison across users.
To examine the average effect of cannabis on symptom intensity by symptom type, we regressed symptom intensity levels on a dummy variable equal to one if symptom intensity was reported after cannabis use and equal to zero if reported before cannabis use, controlling for individual fixed effects and running the regressions separately by symptom type.
To explore the effect of product characteristics on symptom relief, we regressed symptom relief on the product characteristics, including THC and CBD content, labeled subtype (hybrid, C. indica, or C. sativa), and inhalation method (joint, pipe, and vaporizer). Our primary THC and CBD measures are the potency from 0 to 30%/dry wt. Our plant subspecies variables distinguish between C. indica, C. sativa, and hybrid Cannabis strains. While the colloquial distinction between C. indica and C. sativa has been widely discounted by the scientific community (Piomelli and Russo 2016), we included these labels because they are still commonly incorporated into Cannabis consumer purchasing decisions. For example, Ontario’s government-run online cannabis store differentiates between sativa- and indica-dominant strains as does Leafly, the largest aggregator of consumer-friendly cannabis information in the world with more than 100 million visitors each year. We include inhalation method (joint, pipe, or vaporizer) because joints typically are thought to contain lower quality cannabis than loose flower and vaporizing can occur at lower temperatures than combustion via joint or pipe, making controlling for these characteristics potentially important. Our regressions are run for the overall sample and for the three subsamples defined by symptom type. In addition to including product characteristics, we also controlled for session-level pre-cannabis use symptom intensity and session length (minutes up to 4 h—symptom updates beyond 4 h are not included in our analysis). Baseline symptom intensity is included in all regressions because higher starting symptom levels are associated with greater symptom relief (Vigil et al. 2018). Session length (time from start until the last symptom was reported within 4 h) is included because the effects of inhaled cannabis may vary systematically with session length. Throughout our regression analyses, standard errors were clustered at the user level and to control for heteroskedasticity and arbitrary correlation among sessions entered by the same user.
In addition to our continuous THC and CBD potency measures, we further explore the relationship between THC, CBD, and symptom relief using categorical THC and CBD measures to capture nonlinearities in the effect of THC and CBD on symptom relief. We divided our sample fairly evenly into low THC = < 9%, medium THC = 10–19%, and high THC = 20–30%; and low CBD = 0%, medium THC = 1–9%, and high CBD = 10–30%.
Because we find THC to be a primary driver of symptom relief in the results and it might vary with the other product characteristics, we also test for whether plant subspecies or inhalation method influences the effect of THC on symptom relief, by interacting our continuous measure of THC with those product characteristics. A statistically significant interaction term could arise if, for example, vaporization of cannabis occurs at lower temperatures than combustion of flower in a pipe or joint and this affects THC bioavailability or if joints systematically contain lower grade flower, in which, for example, a greater amount of THC may have already degraded into CBN (cannabinol). We also interact THC with session length to test for variation in the effect of THC over time within 4 h.
We conduct two robustness checks on our symptom relief regression approach. First, because our regression design is inherently based on repeated sessions entered by the same user, we test the robustness of our main results to including only users who entered at least three, four, or five sessions respectively. Second, we extend our time-to-effect analysis by exchanging the maximum symptom relief reported within 4 h for the difference between baseline symptom intensity and the last symptom level reported within 1, 2, 3, and 4 h.
For the side effect outcomes, we use the same regression approach, including the three categories of product characteristics (subtype, inhalation method, and cannabinoid content), baseline symptom intensity level, and session length, along with user fixed effects.
All statistical analyses are conducted using Stata 15.1 (Stata corporation, U.S.).