Evaluation of THC-Related Neuropsychiatric Symptoms Among Adults Aged 50 Years and Older

This systematic review and metaregression analysis estimates the association between the delta-9-tetrahydrocannabinol (THC) dose of cannabinoid-based medicines and neuropsychiatric adverse events among adults aged 50 years and older.

Rationale for age cut-off: For this meta-regression analysis, we chose mean age  50 years as the cut-off as the clinical conditions (diabetes, cancer, neurodegenerative disorders, cancer etc) for which CBMs are often considered afflict people more commonly from around this age. This period of life onwards is also characterised by multi-morbidities, polypharmacy and age-related bodily changes that may affect pharmacokinetics 49 and tolerability of medications. In addition, the age range as well as median and interquartile range of the mean ages of study participants included in the studies that constitute our meta-analysis clearly indicate that people over 65 and 75 years are currently being recruited into studies of CBMs for various indications. However, while there is a larger evidence base of studies with mean age of participants  50 years this is very modest for studies where all participants are  65 years (n=3 studies for THC studies and n=1 for THC:CBD study), the typical cut-off age for defining 'elderly'. In light limited power of individual RCTs to unravel patterns of side-effects, the growing use of CBMs in the elderly and the general perception that they are safe to use, it may be particularly important to examine this by synthesizing currently available evidence to help inform about the safety and tolerability profile of CBMs in those aged 50 years and over rather than wait for the evidence base to mature. Therefore, we have taken the approach to focus on studies with a mean participant age  50 years which also capture a substantial number of people who are  65 years. Future attempts at evidence synthesis may be able to focus only on studies of people  65 years when a sufficient number of studies have accumulated.

Data extraction
All relevant available data for examination of the safety and tolerability of different CBMs (THC:CBD combination or THC or CBD alone) was collected from eligible studies. This was complemented with information from ClinicalTrials.gov and author responses. Data was extracted for study design, participant characteristics, indication, dosage and duration of intervention, all cause and treatment-related AEs and SAEs, AE-related withdrawals and deaths. AEs and SAEs were coded according to the Medical Dictionary for Regulatory Activities (MedDRA) 'system organ classes' (SOC). Data was also extracted for the top 5 (as reported by each study) AEs for each SOC, where available. Data extraction and coding was verified by a medically qualified researcher and discrepancies resolved following discussions with senior researcher. In the present report, we focus on the AEs categorized under the nervous system or psychiatric disorder of MedDRA SOC to investigate their association with dose/s of CBM/s used.

Quality assessment
We used the GRADE (Grading of Recommendations Assessment, Development and Evaluation) criteria to assess the overall quality of evidence and rate risk of bias, publication bias, imprecision, inconsistency, indirectness, and magnitude of effect 50 . We have summarised the GRADE ratings of very low-, low-, moderate-, or high-quality evidence to reflect the extent to which we have confidence in the effect estimates are correct 51 . This was done by one reviewer (KM) and checked by a second reviewer (LV), and disagreements were resolved via discussion with a third reviewer (SB).
A broad range of disease conditions/ clinical indications were investigated in these RCTs and included: Alzheimer's disease, Parkinson's disease, Huntington's disease. Amyotrophic lateral sclerosis, Multiple sclerosis, motor neuron disease, neuropathic pain, cancer (cancer or chemotherapy related anorexia, pain or nausea/vomiting), type 2 diabetes mellitus, chronic obstructive pulmonary disease, fibromyalgia, raised intraocular pressure, cervical dystonia, healthy, pancreatitis, obstructive sleep apnoea and Levodopa induced dyskinesia in Parkinson's disease.

Data synthesis and analysis:
We estimated total exposure to active intervention in person-years by first calculating this for each individual study by multiplying the number of subjects in the active intervention arm with the duration of treatment for that arm for each study and then adding up these study-specific values for all studies under each broad category (THC, THC:CBD) of intervention investigated here.
We estimated pooled effect-sizes if there were 2 or more RCTs for each individual neuropsychiatric AE within each broad category of intervention (THC, THC:CBD) under the random-effects model using the restricted maximum-likelihood estimator because of anticipated heterogeneity. For each broad category of intervention, analyses combined both parallel-arm and crossover RCTs, with the latter treated as parallel-arm design 52 for pooled analyses. We estimated incident rate ratio (IRR) for individual AEs. Studies with more than one active treatment arm were treated as independent studies. We combined the data for all conditions for the analysis of AEs.
To test our primary hypothesis, we carried out meta-regression analyses under the random-effects model using the restricted maximum-likelihood estimator to examine the association of individual neuropsychiatric AEs with the dose of THC used in THC studies and separately with the dose of THC and CBD used in THC:CBD studies.
For our primary analysis we focused on all studies with available data where the mean age of study participants was ≥50 years. We also explored the possibility to carry out sensitivity analysis by restricting the analyses to studies where all participants were  65 years of age. Sensitivity analyses at a different age cut-off of  65 years was not possible as there was data from fewer studies (3 THC studies and 1 THC:CBD study) than recommended for meta-regression analyses. We investigated heterogeneity using forest plots and the I 2 statistic and report these in Table 1 and forest-plots. We also carried out formal outlier and influence detection diagnostics 53 for the AE of self-reported 'thinking/ perception disorder', which identified two studies as being influential 18,28 . The association between THC dose and 'thinking/ perception disorder' no longer remained significant, though the direction of effect did not change. While outlier/ influence diagnostics are popular in meta-analysis, others have also recommended against their routine use 54 , particularly because of challenges in distinguishing true outliers (where the data is erroneous) from large errors in sampling 55 . Instead, they may be seen as a method of sensitivity analysis and inform confidence in findings 53 . In the present case, this is especially important because one of the influential studies 28 had the largest sample size (329 participants in the active intervention arm and 164 in the control intervention arm) as well as the longest duration of treatment (~3 years; with only one other study involving 1 year of treatment) in the meta-analytic dataset. Further, we observed that heterogeneity (as indexed by the I2 statistic) decreased from I2=35.73% when the analysis was conducted without THC dose as a moderator (i.e. simple meta-analysis investigating the pooled effect of THC treatment compared to control treatment in the RCTs included on the AE of self-reported 'thinking/ perception disorder') to I2=1.46% when THC dose was included as a moderator, pointing towards the appropriateness of the meta-regression results presented herein. Therefore, these studies have not been excluded but the results presented and discussed with a caveat.
Statistical analyses were performed using the metafor package in R (version 3.6.3) 56 .