Survey analyses find abstainers can inflate apparent links between substance…

AIMS: This study addressed the implications of including substance abstainers in investigations of the epidemiology, structure and nature of substance use disorder (SUD). DESIGN: We first conducted a Monte Carlo simulation to quantify bias introduced when 'structural zeros' (all-zero response patterns consistent with non-exposure) are added to symptom data used to model SUD. We simulated two correlated latent SUD factors and varied the proportion of all-zero rows to estimate inflation in (1) the recovered correlation between disorders and (2) the apparent strength (variance explained in the indicators) and unidimensionality of latent SUD factors. We then illustrated the implications of the simulation results in secondary cross-sectional analyses of the National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) wave 1 (2001-2002), directly comparing estimates including abstainers (i.e. symptom indicators structurally coded as 0 via interview skip patterns/nonuse) and excluding abstainers (restricting to exposed individuals reporting past-year use). SETTING: United States. PARTICIPANTS: NESARC included 43 093 participants (57% female, 57% White, 19% Black, 3% Asian, 2% American Indian/Native American, 19% Hispanic or Latino) aged 18 years or older. MEASUREMENTS: Past-year Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) SUD criteria/diagnoses [alcohol use disorder (AUD), cannabis use disorder (CUD), tobacco use disorder (TUD), opioid use disorder (OUD)] and consumption frequency were assessed using the Alcohol Use Disorder and Associated Disabilities Interview Schedule-IV (AUDADIS-IV). We examined: (1) associations between consumption frequency and SUD diagnosis; (2) dissociable validity profiles of consumption vs. SUD across 30 external validators spanning other substance use, mental disorders and psychosocial functioning; and (3) cross-SUD overlap and evidence for general addiction liability using latent variable models. RESULTS: In simulations, adding all-zero response patterns produced substantial upward bias in recovered cross-disorder associations and increased apparent unidimensionality, even when true associations were modest. In NESARC, abstention was common (e.g. 96.3% for cannabis; 98.4% for opioids). Consumption and SUD showed dissociable relationships with external validators, with consumption generally less indicative than SUD of dysfunction broadly construed. Including abstainers substantially strengthened the observed association between consumption and SUD, correlations between different SUDs (e.g. AUD-CUD correlation was higher when abstainers were included than when restricting to users of both substances) and often strengthened general SUD factors (i.e. loadings were higher). SUMMARY: Including abstainers in epidemiologic studies can conflate substance exposure/consumption with substance use disorder liability, inflate cross-substance use disorder associations and strengthen apparent evidence for general addiction liability. We recommend adopting designs and analyses that explicitly separate exposure from disorder conditional on exposure.