Study design and participants
A cross-sectional, nationally representative survey of persons 15 years and older in 2017 living in South Africa was analysed. This multistage stratified random cluster population-based household sample is described elsewhere [21]. In brief, the 2015 national population sampling frame [22] was utilized to draw 1000 small area layers (SALs) that were stratified by South Africas nine provinces, and locality types. In each of the 1000 SALs, 15 households were randomly selected to participate and all individuals living in the selected household that slept there the night before were invited to participate. It is important to point out that this paper utilized racial categorization where “Coloured” or mixed race is defined children born to parents of Black African and either White and/or Indian/Asian race groups as per South Africa’s Apartheid government’s Act 30 of 1950. This is done to correct the inequalities of the previous apartheid regime.
Study procedures
Participants were handed an informed consent form to read together with a trained interviewer. If the participant agreed to participate, they signed the consent form. For those younger than 18 years old, parental consent was sought together with youth assent. If either the youth or the parent did not sign the assent/consent form, no interview was conducted. All interviews were done in private and kept confidential. The questionnaire was electronic, the interviewer administered and completed on a tablet using CSPro software. Data collection started in December 2016 and ended in February 2018. The household response rate was 82.2% and the individual response rate to be interviewed was 93.6% [21]. The participants did not receive any payment or gifts for the interview. For this paper, we restricted the sample to those with complete cannabis use measurement.
Measures
Non-daily and daily cannabis use was assessed using the question: “In the past three months, how often have you used cannabis (dagga, marijuana, pot, grass, hash, etc.)?” from the “Alcohol, Smoking and Substance Involvement Screening Test (ASSIST)” [23]. Response options were “Never, once or twice, monthly, weekly, or almost daily.” “Non-daily” was defined as “once or twice, monthly, or weekly” and “almost daily” as “daily” cannabis use.
Past three months, other drug use was assessed with six items (cocaine, amphetamine, inhalants, sedatives, hallucinogens, and opiates) from the (ASSIST)” [23]. The six items were summed to define any other drug use in the past 3 months. Cronbach’s alpha for the 6-item other drug use measure was 0.97 in this sample.
Alcohol use disorder was assessed using the Alcohol Use Disorders Identification Test (AUDIT) [24] and was scored as in a previous survey in South Africa [25]. Among adults (20 years and above) a cut-off score of 8 or more [24] and among adolescents (15-19 years), 5 or more [26] for classifying alcohol use disorder. (Cronbach alpha 0.87 in this sample).
Sociodemographic factors included age, sex, highest educational level, living status, population group, employment status, and residence status [21].
Psychological distress was assessed with the Kessler Psychological Distress Scale (K10), with scores 20 or more indicating psychological distress [27]. Cronbach’s alpha for the K10 was 0.92 in this sample.
Multimorbidity was assessed with self-reported health care provider diagnosed hypertension, diabetes, HIV positive, cancer and heart disease.
Health care utilization was sourced from the question: When was the last time you went to see a health professional (doctor, nurse, traditional healer, etc.)? Response options ranged from 1=within the last 6 months to 4=never.
Data analysis
All statistical analyses were conducted using STATA software version 14.0 (Stata Corporation, College Station, TX, USA). The data were weighted to make the sample representative of the target population in South Africa. Descriptive statistics were used to summarize the sample and cannabis use prevalence characteristics. Unadjusted and adjusted (including variables significant at p<0.05 in univariate analysis) multinomial logistic regression was used to predict nondaily and daily cannabis use, with no past 3-month cannabis use as the reference category. In addition, unadjusted and adjusted (including variables significant at p<0.05 in univariate analysis) logistic regression was used among active (past 3 months) cannabis users to predict daily versus nondaily cannabis use. Taylor linearization methods were applied to account for the complex study design and the sampling weight. Results from (multinomial) logistic regression analyses are reported as (relative risk ratios) odds ratios (ORs), and 95% confidence intervals (CIs). Missing values (<1.8% for any study variable) were excluded and p<0.05 considered significant.