120,691 Cannabis Symptom Records Were Not 120,691 Independent Patients
A cannabis-app analysis linked later recorded sessions with slightly less symptom relief. Its large record count should not be confused with the number of independent patients or treated as proof that every patient develops tolerance at the same rate.
People, sessions and symptoms were different units
The main analysis included 120,691 symptom-specific observations from 42,005 cannabis sessions recorded by 16,395 app users. A person could report several symptoms during one session and could contribute multiple sessions. The investigators restricted analysis to each user’s first ten recorded sessions.
The outcome was the lowest reported symptom intensity during a session minus its starting value. It therefore represented the greatest reported relief within that session, not an average over the entire session, a clinician’s assessment or a direct measure of daily functioning.
The models addressed some repeated-measure problems
The investigators included user fixed effects and clustered standard errors by user. These methods address stable individual differences and correlation among the same person’s records. They do not turn repeated reports into independent people or eliminate changing health, medication, product and reporting circumstances.
Later sessions were associated with a small decline in relief. Reporting of negative, positive and context-dependent side effects also decreased. The pattern is consistent with tolerance, but less reporting can have several explanations, including changing engagement with the app.
The most adjusted analysis used a smaller sample
Adding product THC, CBD and dose information reduced the symptom-analysis sample to 25,639 observations from 4,787 users. Product potency information was optional and unavailable in a directly comparable percentage format for some products. The largest sample therefore does not describe every adjusted result.
THC and reported dose were associated with symptom relief, while CBD did not show a significant association in the main fully adjusted model. These observational coefficients are not dosing instructions. Product choice and use were not randomly assigned.
What this evidence cannot establish
The study did not quantify blood, serum or other biological cannabinoid concentrations. It cannot distinguish pharmacokinetic from pharmacodynamic tolerance in an individual, establish impairment, or show that product rotation reverses tolerance. Its first-ten-session findings also should not be projected uncritically over months or years.
For evaluating a cannabis-treatment claim, retain the actual denominator, outcome definition and model-specific sample. The study offers a useful real-world association with substantial uncertainty about mechanism and long-term clinical importance.
Related analysis: evaluating cannabis tolerance claims.
Primary research
Stith SS; Li X; Brockelman F; Keeling K; Hall B; Vigil JM. Cannabis tolerance reduces symptom relief. Frontiers in pharmacology (2025). DOI: 10.3389/fphar.2025.1496232.
