Intelligence quotient decline following frequent or dependent cannabis use in youth: a systematic review and meta-analysis of longitudinal studies | Psychological Medicine | Cambridge Core

Introduction Cannabis is the most frequently used illicit substance worldwide, with the prevalence of lifetime cannabis use highest in young people (Degenhardt et al., Reference Degenhardt, Ferrari, Calabria, Hall, Norman, McGrath and Whiteford2013). Cannabis use in adolescence is consistently associated with poorer mental health outcomes including increased risk of mood disorders, self-harm and suicidality (Gobbi et al., Reference Gobbi, Atkin, Zytynski, Wang, Askari, Boruff and Dendukuri2019; Twomey, Reference Twomey2017). Cannabis use is also associated with markedly poorer psychosocial outcomes across the lifespan in diverse indices such as educational attainment, employment, relationships, welfare dependency, risk of motor accidents, social mobility and income (Fergusson, Horwood, & Beautrais, Reference Fergusson, Horwood and Beautrais2003; Fergusson, Lynskey, & Horwood, Reference Fergusson, Lynskey and Horwood1996; Hall, Reference Hall2015). There is strong evidence demonstrating an association between cannabis and psychotic disorders, particularly frequent use of high tetrahydrocannabinol potency cannabis (Di Forti et al., Reference Di Forti, Quattrone, Freeman, Tripoli, Gayer-Anderson, Quigley and van der Ven2019). Cannabis use has been estimated to be associated with approximately 12 and 15 excess life-years lost in women and men, respectively, in Danish register data (Weye et al., Reference Weye, Momen, Christensen, Iburg, Dalsgaard, Laursen and Plana-Ripoll2020). Earlier initiation of cannabis use and frequent cannabis use in adolescence are risk factors for later cannabis dependency (Leung, Chan, Hides, & Hall, Reference Leung, Chan, Hides and Hall2020). Only a minority of those who have used cannabis more than five times in adolescence remit from use in mid-life, indicating the persistence of cannabis use (Perkonigg et al., Reference Perkonigg, Goodwin, Fiedler, Behrendt, Beesdo, Lieb and Wittchen2008). One in three youth who use cannabis weekly or more frequently is cannabis-dependent (Leung et al.,

Reference Leung, Chan, Hides and Hall2020). The legalisation of cannabis and a decreasing perception of harm in adolescent and young adult populations is likely to lead to increased use, particularly in vulnerable populations, resulting in negative public mental health consequences. (Mauro et al., Reference Mauro, Newswanger, Santaella-Tenorio, Mauro, Carliner and Martins2019). Cannabis use during youth is of particular concern, as the developing brain may be particularly susceptible to harm during this period (Lubman, Cheetham, & Yücel, Reference Lubman, Cheetham and Yücel2015). A New Zealand cohort study has shown that persistent cannabis dependency from adolescence to midlife has previously been associated with a clinically significant eight-point decline in Intelligence Quotient (IQ) (Meier et al., Reference Meier, Caspi, Ambler, Harrington, Houts, Keefe and Moffitt2012). The long-term effect of cannabis on intelligence is under-research. A recent study has found that even minimal incidental use of cannabis in adolescence is associated with morphological brain volume changes (Orr et al., Reference Orr, Spechler, Cao, Albaugh, Chaarani, Mackey and Garavan2019). A meta-analysis of cross-sectional MRI studies found replicated evidence of reduced grey matter in the CB1R rich areas of the hippocampus and the amygdala associated with cannabis use (Rocchetti et al., Reference Rocchetti, Crescini, Borgwardt, Caverzasi, Politi, Atakan and Fusar-Poli2013). Previous meta-analyses show inconsistent and heterogeneous findings for both global and specific cognitive domains relating to cannabis use. Two reviews found some evidence for deficits in attention, executive functioning, memory and learning, motor function deficit and verbal cognition (Ganzer, Broning, Kraft, Sack, & Thomasius, Reference Ganzer, Broning, Kraft, Sack and Thomasius2016; Grant, Gonzalez, Carey, Natarajan, & Wolfson, Reference Grant, Gonzalez, Carey, Natarajan and Wolfson2003).

Two further meta-analytic studies found multidomain and overall cognitive deficits associated with cannabis use, however, reported that the results could be attributed to residual (i.e. may be related to recent use) rather than chronic effects (Schreiner & Dunn, Reference Schreiner and Dunn2012; Scott et al., Reference Scott, Slomiak, Jones, Rosen, Moore and Gur2018). The majority of studies included in these reviews have been case-control or cross-sectional studies generally containing small samples that may not be representative of the general population. Representative longitudinal cohort studies accounting for pre-cannabis exposure IQ may better inform whether frequent or dependent cannabis use in youth has a deleterious effect on IQ over time at a population level. This is to our knowledge the first meta-analysis of longitudinal IQ change in relation to cannabis use in adolescence. The primary aim of this study is to quantitatively synthesize the available literature examining the longitudinal association between frequent/dependent cannabis use and IQ change from pre-exposure baseline in young people. We had a number of exploratory analyses. We explored whether we could disentangle the effects of chronic v. residual effects from the available longitudinal literature. Chronic effects are defined as effects lasting beyond a period of 28 days from last use and residual effects are effects lasting up to 28 days from last use (Pope et al., Reference Pope, Gruber, Hudson, Cohane, Huestis and Yurgelun-Todd2003). We also explored whether frequent/dependent cannabis use was associated with verbal and performance IQ decline, and lower baseline full scale, verbal and performance IQ. Methods We preregistered our review with PROSPERO (ID no. CRD42019125624). We searched Embase, PubMed and PsychInfo from inception to 24 January 2019. We developed our search strategy through an iterative process with an information specialist to maximise the number of potential articles available for screening (see supplementary details for full search summary). Two authors independently screened articles by title and abstract to identify articles suitable for full-text review, following this, two authors screened articles by full text for inclusion in systematic review and meta-analysis. We included prospective cohort studies of non-treatment seeking youth from samples recruited from the community with a baseline measurement of IQ prior to participants initiating cannabis use.

We specified that the onset of cannabis use should have occurred at or before age 26. We specified that participants should have both a baseline and follow-up measure of IQ. We specified that studies should have at least a verbal and performance subtest of IQ allowing construction of a short form full-scale IQ composite measure. We considered articles or conference abstracts published in English. We defined our cannabis exposure as at minimum weekly use for 6 months and/or >25 reported lifetime uses and/ or diagnosis of cannabis dependency. The rationale for these thresholds was that approximately 1/3rd of weekly or greater adolescent cannabis users are cannabis-dependent and that studies would vary in how they measured cannabis use (i.e. some would measure lifetime use, some would define frequency, some would use diagnostic assessments) (Leung et al., Reference Leung, Chan, Hides and Hall2020). We defined the control group as having used no or minimal cannabis (i.e. <5 lifetime uses). Where studies presented multiple groups i.e. frequent/dependent former and current users corresponding to chronic effects and residual effects respectively, we decided a priori to include them as one group in the main analysis, and attempt to separate them in exploratory analyses. Two authors (EP, SS) using a pre-specified template extracted data independently. Disagreements were resolved with consensus through discussion. Where estimation of effect size was not possible with the available data or whereby the analytic strategy of the source data did not meet our inclusion criteria, we contacted authors to provide additional data/clarification. Two authors calculated effect sizes (EP, CM) agreement was 100%. We used WebPlotDigitizer to extract information from figures (Rohatgi, Reference Rohatgi2020). We collected information from individual studies, where available, on a number of different potential confounding factors in extracted adjusted estimates. This varied by study (see online Supplementary eTable 1) and included current depression diagnosis or symptoms, alcohol use, tobacco use, use of other drugs, educational attainment, psychotic symptoms, socio-economic status, gender, maternal educational level, attention deficit hyperactivity disorder symptoms or diagnosis, maternal substance use during pregnancy, age at initial and follow-up testing, and recency of cannabis use. We extracted final adjusted standardized mean differences that authors reported. Comprehensive information regarding individual study level data is available in the online supplement.

We used the Newcastle-Ottawa Scale to assess the risk of bias in individual studies and present the findings in our results and supplementary materials (Wells et al., Reference Wells, Shea, Connell, Peterson, Welch, Losos and Tugwell2014). The Newcastle Ottawa Scale is a ten-point rating tool that assesses the quality of selection, comparability and outcome in an individual study. Two authors (EP, AON) calculated the Newcastle Ottawa Scale and agreement was initially 96% (cohen's kappa = 0.9). Following consensus discussion and provision of additional information, the agreement was 100%. We used the Campbell Collaboration effect size calculator to calculate effect sizes except in linear mixed models where they were calculated in Stata according to Feingold's description (Feingold, Reference Feingold2015; Reference WilsonWilson). We chose a priori a random-effects model to estimate the pooled Cohen's d statistic. We chose this model due to the expected heterogeneity in study-level characteristics. We calculated the I 2 statistic to measure heterogeneity between studies. We present funnel plots to inspect publication bias and results of the Vevea and Hedges weight-function model for publication bias (Vevea & Hedges, Reference Vevea and Hedges1995). We used metan command function in Stata version 15 for our analysis (Harris et al., Reference Harris, Bradburn, Deeks, Harbord, Altman and Sterne2008). Results We identified 2875 papers and conference abstracts for screening after removal of duplicates. We identified 33 papers for full-text screening. We included seven studies that met our criteria (Fried, Watkinson, & Gray, Reference Fried, Watkinson and Gray2005; Jackson et al., Reference Jackson, Isen, Khoddam, Irons, Tuvblad, Iacono and Baker2016; Meier et al., Reference Meier, Caspi, Ambler, Harrington, Houts, Keefe and Moffitt2012, Reference Meier, Caspi, Danese, Fisher, Houts, Arseneault and Moffitt2018; Mokrysz et al., Reference Mokrysz, Landy, Gage, Munafò, Roiser and Curran2016; Ross et al.,