Problem Gambling in Early Adulthood: a Population-Based Study

The Avon Longitudinal Study of Parents and Children (www.​bristol.​ac.​uk/​alspac) is a birth cohort study which enrolled mothers in early pregnancy in the Bristol and surrounding areas in England in 1991–1992. It has detailed information on parents (Fraser et al. 2012) and children (Boyd et al. 2013), collected prospectively at multiple times during pregnancy and throughout childhood. Sources of data include self-report surveys; clinical assessments; birth, medical, and educational records; and biological samples. The initial recruitment resulted in a core cohort of 14,541 pregnancies and 13,988 children alive at 12 months. A total of 913 additional participants have been enrolled in the study since the age of 7 years with 195 of these joining since the age of 18 years. This additional enrolment provides a baseline sample of 14,901 participants who were alive at 1 year of age.

The study website contains details of all the data that are available through a fully searchable data dictionary (http://​www.​bris.​ac.​uk/​alspac/​researchers/​data-access/​data-dictionary/​). Ethical approval for the ALSPAC was obtained from local research ethics committees. Informed consent for the use of data collected via questionnaires and clinics was obtained from participants following the recommendations of the ALSPAC Ethics and Law Committee at the time.

A total of 9000 young participants in the ALSPAC were invited to complete gambling surveys either on paper or online as part of a bigger survey on other aspects of the participants’ life, when the young people were aged 20 years (2012–2013) and 24 years (2016–2017). If a participant completed a paper survey, the data were entered by ALSPAC staff into online databases.

Young people’s participation in specific gambling activities during the past year was assessed at both time points using items derived from the British Gambling Prevalence Survey 2007 (Wardle et al. 2008) (see Supplementary Table 1). Problems with gambling were assessed using the Problem Gambling Severity Index (PGSI), a nine-item scale derived from a 31-item larger screen, the Canadian Problem Gambling Inventory (Ferris and Wynne 2001). Only participants reporting gambling in the past year were asked to complete the PGSI, and respondents were categorized as non-problem gamblers (score of 0), low-risk gamblers with few or no identified negative consequences (score 1 or 2), moderate-risk gamblers leading to some negative consequences (score 3 to 7), and problem gamblers with negative consequences and possible loss of control (score 8 or more). Due to low numbers, moderate-risk gamblers were pooled with problem gamblers for analyses, as has been done in many previous studies (Canale et al. 2016; Potenza et al. 2011). The term ‘at-risk’ gambling includes both the low-risk and the moderate-risk/problem gambling groups. The gambling activity of the mothers was captured separately using the PGSI reported in a self-completion questionnaire when the children were aged 18 years in 2010–2011.

The choice of antecedents was informed by previous research by the study team and reviews of the gambling literature (Griffiths 2011), and were clustered into child, parental, and socio-economic (SES) factors. Child variables included gender, IQ measured in a research clinic at age 8 years using a short version of the WISC III (lowest quartile, IQ < 90), frequency of video gaming reported by questionnaire at age 13–14 years, smoking and alcohol usage reported by questionnaire at 16.5 years, hyperactivity and conduct problems reported at age 16.5 years using the Strengths and Difficulties Questionnaire, locus of control assessed at age 16.5 years, sensation seeking at age 17 years, education/employment status at age 17, diagnosed depression at age 17 years, and mental well-being at 17.5 using the Warwick-Edinburgh Mental Well-being Scale (WEMWBS). Parental and SES variables included maternal highest education level in pregnancy, financial difficulties in pregnancy, maternal problem gambling when child was aged 18 years, and parental monitoring (reported by young people aged 17 years—see Supplementary Table 2). More detailed information about the variables used is provided in Supplementary Table 3.

Outcomes used for investigating effects of problem gambling in young adulthood were depression, anxiety, self-harm, criminal activity, use of illicit drugs, smoking cigarettes, use of alcohol, and employment. All outcomes were collected by self-report using online and paper questionnaires when participants were aged 24 years. More detailed information about these outcome variables is provided in Supplementary Table 3.

All available data were used to conduct univariate analyses on the stability of problem gambling and the antecedents to problem gambling. Categorical data were analysed using Chi-square tests and numerical data using ANOVA. For categorical variables, we also looked at adjusted residuals after the Chi-square tests to see if the observed cell counts for the reference category differed to that expected. The larger the residual, the more different from the average values and that expected. A general rule of a residual of 2 was used as signifying a large enough difference (equating to 95% confidence).

Variables significant at p < 0.10 in the univariate analyses were then included as covariates in multivariate analyses. Due to the number of antecedents analysed, a large loss to follow-up and missing data on antecedents, outcomes, and the PGSI, multivariate analyses adjusting for other significant factors would not be possible without multiple imputation. Multiple imputations using 50 imputations by the chained equations method were performed utilizing the mi impute command in STATA v.15. Variables included in the imputation model were all of those included in the final regression models (many of which predicted missingness in the other variables). Additionally, a few auxiliary variables also associated with missingness (stressful live events score, maternal age at birth, gambling regularity at age 17 and 24 years; see Supplementary Material 1 for more details). We included non-gamblers in the analyses, so imputed up to the number of participants where gambling status (yes/no) at age 20 years could be established (N = 4263). PGSI was imputed conditional on gambling status (i.e. only those who did gamble were imputed for this variable). A comparison of the unadjusted associations using all available data and the imputed dataset is contained in Supplementary Table 4. Adjustment for other variables was carried out in a stepwise procedure using binomial (for binary variables) or multinomial (for variables with > 2 categories) logistic regressions: (i) adjusted for all child variables; (ii) adjusted for child variables and parental variables; and (iii) adjusted for child variables, parental variables, and SES variables. A final minimal model was created adjusting only for variables where confidence intervals did not cross 1 to avoid problems from over-adjusting. Unadjusted and minimal adjusted models (fully adjusted for significant covariates) are shown in main results, and models adjusted for child variables and family variables separately can be provided on request. Odds ratios and 95% confidence intervals (95% CI) are reported.

Figure 1 shows participant flow at age 20 years. Out of 9061 questionnaires sent out, 4263 (48%) were returned of which 2624 participants answered all PGSI items. More females (61%) than males completed the PGSI. The distribution of gambling activities among non-problem gamblers, low-risk gamblers, and moderate-risk/problem gamblers for those that gambled regularly is illustrated in Fig. 2. This shows that moderate-risk/problem gamblers gambled on a wide range of activities, particularly on the internet, whereas non-problem gamblers limited their gambling to the lottery, scratch cards and football pools, and playing bingo. Figure 3 shows the change in distribution of activities between age 20 and 24 years for moderate-risk/problem regular gamblers only. Similar to age 20 years, at 24 years moderate-risk/problem gamblers gambled on a wide variety of activities, with no activity increasing in frequency, but some activities like sport betting, football pools, and private betting decreased in frequency. The amount spent/month on gambling at age 24 years showed a clear trend, with 92% of non-problem gamblers spending < £10/month, whereas in the group of moderate-risk gamblers 29% spent < £10 and 56% spent £11–£100/month. Problem gamblers spent the most, with 24% reporting spending £11–£100, and 62% spending > £100/month.

The proportion of participants in the combined moderate-risk/problem gambling group remained stable across time from 6.5% at 20 years to 6.9% at 24 years, whereas the proportions of low-risk gamblers decreased, and non-problem gamblers increased from age 20 to 24 years (Table 1). There was a strong male gender bias: males comprised 74.1% of the moderate-risk/problem group at 20 years, and 67% at 24 years. Moderate-risk/problem gambling was strongly associated with regular (at least weekly) gambling at both 20 and 24 years. Analysis of the 995 participants (38%) who provided PGSI data at 20 and at 24 years (Table 2) clearly shows that being a moderate-risk/problem gambler at age 20 years was highly predictive of being one at age 24 years (49% compared with 8% of low-risk gamblers and 1% of non-problem gamblers at age 20 years).

Univariate analyses of demographic factors associated with moderate-risk/problem gambling at age 20 years using all available data are presented in Table 3. Important individual factors associated with subsequent moderate-risk/problem gambling were male gender, preference for video gaming at 13 years, higher scores on hyperactivity and conduct difficulties measured at 16 years, an external locus of control measured at 16 years, high sensation seeking scores at 17 years, and weekly smoking and alcohol use at 16 years. However, locus of control proportions and means of sensation seeking scores did not differ greatly between low-risk and moderate-risk/problem gambling groups. For male participation and preference for video gaming, adjusted residuals showed that observed frequencies were higher than expected for both low-risk and moderate-risk/problem gambling. The proportion of regular smokers and alcohol users also increased with increased severity of problem gambling, but smoking proportions did not differ between low-risk and moderate-risk/problem gamblers. Family characteristics which were risk factors for moderate-risk/problem gambling were financial difficulties, lower maternal education level, a history of maternal problem gambling, and lower levels of parental monitoring reported by young people at 17 years.

The frequency of participants in the different categories of PGSI remained very similar in the imputed dataset compared with all available data (Table 4). Unadjusted and adjusted multivariate analysis of moderate-risk/problem gambling at 20 years against outcomes collected at 24 years are summarized in Table 5 for all six pairwise PGSI category comparisons using the imputed dataset. Depression showed an association with any at-risk gambling, in that both low-risk and moderate-risk/problem gamblers were more likely to be depressed at 24 years compared with non-problem gamblers. Adjustment had very little effect on depression outcomes. However, those that did not gamble at all were more anxious and depressed than non-problem gamblers, suggesting a less straightforward relationship between gambling and anxiety/depression. Regular cigarette smoking and intake of illicit drugs were associated with any at-risk gambling (no difference between low-risk and moderate-risk/problem gambling) but attenuated after adjustment and only remained strong when comparing any at-risk gambling with no gambling at all.

Regular alcohol use was associated with any at-risk gambling, but a stronger association was seen with moderate-risk/problem gambling which remained after adjustment. More problematic use of alcohol was only associated with moderate-risk/problem gambling and remained very strongly associated even after adjustment. Involvement in crime was also higher in low-risk and moderate-risk/problem gamblers but associations attenuated after adjustment. Although any gambling at age 20 years was associated with high social media use at 24 years, odds ratios were highest in the moderate-risk/problem gambling group. Although no financial data were available on the economic consequences of problem gambling, non-problem and low-risk gamblers were more likely to be employed compared with non-gamblers.

Supplementary Table 5 summarizes the effect of covariates on all outcomes. The two covariates which influenced most of the outcomes were gender and regular smoking aged 16 years. Males were more likely to consume alcohol regularly and be involved in crime but less likely than females to be depressed, anxious, self-harming, and frequent users of social media. Apart from affecting smoking and alcohol use at 24 years, regular smokers at 16 years were also more likely to be anxious, self-harming, and involved with drugs and crime. Despite some strong effects of covariates, however, many of the individual associations with gambling on outcomes remained after adjustment. Adjustment for family factors had minimal effect on outcomes.