Gambling onset and progression in a sample of at-risk gamblersfrom the general population$

Elizabeth Carneiro a,b,n, Hermano Tavares c, Marcos Sanches d, Ilana Pinsky a,e,Raul Caetano f, Marcos Zaleski a,g, Ronaldo Laranjeira a,e

a Instituto Nacional de Ciência e Tecnologia para Políticas do Álcool e Outras Drogas (INPAD, National Science and Technology Institute for Public Policies onAlcohol and Other Drugs) of the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, National Council for Scientific and TechnologicalDevelopment), São Paulo, Brazilb Addictions and Other Impulse Control Disorders Unit, Santa Casa da Misericórdia, Rio de Janeiro, Brazilc Gambling Outpatient Unit, Institute and Department of Psychiatry, University of São Paulo, Brazild Ipsos-Reid, Toronto, Canadae The Federal University of São Paulo, São Paulo, Brazilf University of Texas School of Public Health, Dallas, TX, USAg The Federal University of Santa Catarina, Florianópolis, Brazil

a r t i c l e i n f o

Article history:Received 22 October 2012Received in revised form11 September 2013Accepted 22 January 2014Available online 1 February 2014

Keywords:Pathological gamblingGambling trajectoryCommunity sampleGambling sub-population

a b s t r a c t

The goal of this study was to investigate gambling-related behavior, onset and progression in a sample ofat-risk gamblers from the community. A national household survey was conducted in Brazil, coveringindividuals 14 years old or older. Subjects were screened for at-risk gambling, those testing positiveanswered a questionnaire about gambling progression, preferred games and DSM-IV pathologicalgambling criteria. Out of 3007 respondents, 118 were considered at-risk gamblers according to the Lie/Bet Questionnaire. According to the DSM-IV, 32.7% and 24.9% of those were considered problem andpathological gamblers, respectively. Early at-risk gamblers (onset prior to 20 years of age), were morelikely to be male, to prefer non-commercially structured games, and to chase losses while gambling.Young pathological gamblers (under 35 years of age) progressed faster from regular to problem gambling(roughly 2 years) than mature pathological gamblers (12 years). Such findings had not been describedbefore because previous reports focused mostly on clinical samples that lack young, male, early-onsetgamblers. Gambling programs have not satisfactorily covered this segment of gamblers. Outreachstrategies and early interventions should be provided to prevent these individuals from rapidly evolvinginto pathological gambling.

& 2014 Elsevier Ireland Ltd. All rights reserved.

1. Introduction

Pathological gambling is an increasing concern. Studies fromvarious countries report pathological gambling prevalence ratesranging from 0.4% to 2.1% (Weinstock et al., 2008). There is evidencethat the prevalence of pathological gambling increases as gambling ismade more available (Cox et al., 2005).

Gambling is distributed in the population across a continuumthat can be divided in stages ranging from non-gambling to socialgambling, problem gambling and severe gambling addiction (i.e.,

pathological gambling). Scientific reports on gambling havefocused on the pathological extreme of that continuum becausemost studies focus mainly on clinical samples. Therefore,community-based studies are important because they can circum-vent treatment-seeking bias and provide balanced samples thatrepresent the various degrees of gambling. In fact, a community-based survey conducted in New Zealand found that the proportionof individuals experiencing gambling harms was greater than thatof those strictly meeting the criteria for problem or pathologicalgambling (Walker et al., 2012), underscoring the importance ofinvestigating at-risk, subclinical gamblers, who most likely willnot be found in clinical settings.

The paucity of community-based data on gambling is noteworthy,particularly in low- and middle-income countries (LAMICs). In ahousehold survey in Brazil, problematic gambling was associatedwith male gender (3:1 male-to-female ratio), low level of education,unemployment and living in a metropolis (Tavares et al., 2010).

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/psychres

Psychiatry Research

0165-1781/$ - see front matter & 2014 Elsevier Ireland Ltd. All rights reserved.http://dx.doi.org/10.1016/j.psychres.2014.01.035

☆This study was funded, in part, by the Brazilian Secretaria Nacional Antidrogas(SENAD, National Agency on Drug Dependence; Grant no. 017/2003).

n Correspondence to: Instituto Nacional de Ciência e Tecnologia para Políticas doÁlcool e Outras Drogas (INPAD), Rua Ministro Arthur Ribeiro 27, Jardim Botânico,Rio de Janeiro, RJ CEP 22461-230, Brasil. Tel./fax: þ55 21 22660295.

E-mail address: Elizabeth.buba@uol.com.br (E. Carneiro).

Psychiatry Research 216 (2014) 404–411

Spritzer et al. (2011) showed that adolescent gambling was asso-ciated with school dropout and low religiosity. Another LAMICcommunity-based study, conducted in South Africa, confirmed theassociation between at-risk/problem gambling and high urbaniza-tion, with a predominance of engagement in unlicensed non-commercial gambling (Dellis et al., 2013). In developed countries,disordered gambling has been associated with lower age, low income(Ladouceur et al., 2005) and minority status (Black et al., 2012).Regardless of the level of income or social development, problemgambling seems to be strongly associated with a host of demographicvariables that are indicative of loneliness and poor social integration(Afifi et al., 2010; Tavares et al., 2010).

In clinically based studies (Tavares et al., 2001, 2003), it hasbeen reported that women progress rapidly from social to patho-logical gambling and express a preference for electronic gamblingmachines. Nelson et al. (2006) studied gambling trajectories in2256 pathological gamblers entering treatment. The authors foundthat women tended to initiate gambling later than did men,although late onset of gambling behavior seemed to be a betterdeterminant of rapid progression to pathological gambling thanwas gender. In addition to its impact on gambling progression, ageof onset is thought to be a marker of pathological gamblingsubgroups. It has been demonstrated that, in comparison withlate-onset gamblers, early-onset gamblers are more often male,have more comorbidities (psychiatric and medical) and showgreater gambling severity (Burge et al., 2004). Accordingly, datafrom community-based surveys have also indicated that earlyadolescent onset of gambling behavior is associated with conductdisorder (Welte et al., 2009) and greater gambling severity(Rahman et al., 2012).

Using data from the Brazilian household survey on gambling(Tavares et al., 2010), the present study had the following goals:(I) to investigate the associations between at-risk gambling andgender, socio-demographic characteristics, preferred game,gambling onset, recognition and persistence of gambling pro-blems, and gambling progression; (II) to investigate differencesbetween early- and late-onsets at-risk gamblers; (III) to exploreage-based subgroups within the category of pathologicalgamblers.

2. Methods

2.1. Sampling

The Brazilian National Survey on Alcohol Consumption Patterns, which wasaimed at assessing drinking and harmful alcohol-related behaviors, was carried outbetween November 2005 and April 2006. It included a gambling behaviorassessment section (Tavares et al., 2010). Further details of the assessment packageand related methodology are available in Castro-Costa et al. (2008). It was ahousehold survey that included 143 cities, amounting to 325 census sectors (IBGE,2008). The sample was designed to represent the Brazilian population aging 14years or older, of both genders, encompassing rural and urban areas nationwide.The survey did not include non-Portuguese speakers living in Brazil; individualswith mental retardation or any other condition that would preclude consent andcompliance with the investigation; indigenous individuals living on reservations orin institutions such as hospitals and shelters, or individuals engaged in internshipsor military service.

For the sample construction, a three-stage probabilistic procedure was adopted(Hansen et al., 1993). First, cities were stratified by size and region. In each of the 25strata established, cities were ordered by mean income, then submitted tosystematic selection proportional to the estimated size of the local population(IBGE, 2008). Finally, census sectors were selected from the chosen cities.

In the second stage, the jurisdictions were ordered according to the meanincome of the population and then selected according to population and sectorsize. A household counting was performed in each sector. The households wereselected according to a random numerical table, the target being eight interviewsby sector. The number of selected households was set considering non-responserates previously reported (DataUFF, 2009).

The interviewer was required to approach any given selected household at leastthree times at different times of the day on two different days, at least one of them

being a weekend day. No substitution of household was allowed in case of contactfailure. Once contact had been successfully established, the interviewer surveyedall members of the household aging 14 years old or older. The resident whose dateof birth was the closest to the current date was selected to be interviewed. Weightswere attributed according to procedures applied in recent major North-Americansurveys: first a base weight inversely corresponding to the probability of householdselection, then another related to the total number of inhabitants in the household.A non-response weight adjusted for gender and education was used. Finally, a post-stratification weight was applied to adjust the demographic variables to populationnumbers previously reported (IBGE, 2008).

The sampling error was established at 2% for a confidence interval of 95%. Thefinal sample was composed of two parts: a normal sample, representative of theBrazilian population, and an oversample of subjects between 14 and 17 years of age,which allowed for some targeted analysis of this subpopulation (Spritzer et al.,2011). Those two parts were selected in the same way and merged into a singlerepresentative sample of the Brazilian population after proper weighting. Withinthe normal sample, the response rate was 66.4%; of the 3800 potential subjectsidentified in the initial random selection, 1278 could not be contacted. The totalsample therefore comprised 3007 subjects: 2522 from the regular sample and 485from the adolescent oversample. The response rate for the oversample wasapproximately the same, but cannot be calculated exactly, because for some ofthe non-response households the list of residents did not include resident ages toindicate if there were eligible adolescents for the survey.

The interviews were conducted face-to-face, by trained interviewers, and lastedan average of 50 min. A pilot study was conducted to detect difficulties and specificneeds for additional interview training. One hundred and fifty interviewers weretrained; a post-hoc analysis was performed comparing the interview data betweeninterviewers – no interviewer bias was identified. In addition, roughly 5% of theinterviewees (n¼150) were re-assessed to determine whether the interviews hadbeen conducted appropriately.

The gambling section of the survey began with the Lie/Bet Questionnaire (LBQ),a two-item screening tool for pathological gambling (Johnson et al., 1998; Götestamet al., 2004), based on DSM-IV criteria 3 and 7 (American Psychiatric Association,1994): “Have you ever felt the need to bet more and more money?”; and “Have youever had to lie to people important to you about how much you gambled?” Anaffirmative response to either of those questions indicates probable problemgambling, with a reported sensitivity and specificity of 99% and 91%, respectively.Of the 3007 subjects in the final sample, 118 tested positive on the LBQ and weretherefore considered at-risk gamblers. Only those 118 individuals were submittedto parts 2 and 3 of the gambling interview, which investigate gambling behaviorand DSM-IV lifetime diagnostic criteria for pathological gambling. Given that otheraddiction related behaviors were investigated simultaneously, we opted not tosubmit all 3007 subjects to parts 2 and 3 of the gambling interview and to focusrather on the subset of at-risk gamblers, i.e., those who had been substantiallyinvolved with gambling over their lifetimes, which could have, at some point, led topathological gambling.

The study was approved by the Research Ethics Committee of the FederalUniversity of São Paulo, located in the city of São Paulo, Brazil.

2.2. Scales and variables

2.2.1. Demographic variablesThe variables investigated were gender, age, education level, “currently study-

ing”, ethnicity, marital status, employment status, household monthly income,region of birth, region of residence, and religious affiliation.

2.2.2. Gambling variablesPart 2 of the gambling section comprised questions adapted from a questionnaire

previously devised to investigate gambling progression, addressing the age at onsetof regular gambling, age at first experiencing a gambling-related problem, preferredgame and types of gambling-related problems (Tavares et al., 2003). Regulargambling was defined as gambling once a month for at least 2 months in a row.This timeframe was established to avoid overestimation of early-onset status byexcluding reports of occasional gambling. The current preferred game was deter-mined by asking subjects on which game they had been spending most of theirmoney and was classified as follows: non-commercially structured games, dividedinto card games and clandestine lotteries (casinos are outlawed in Brazil; at the timeof the survey, gambling via card games most often occurred informally at bars andclandestine venues); state lotteries (including sports lotteries); new games (electro-nic gambling), other games; and none (stopped gambling). Previous studies,evaluating clinical samples, have indicated that recollection of past-preferred gamesbased on financial expenditure was not reliable because monetary inflation requiredseveral currency changes in the past. Hence, past-preferred game was identified asthe game most frequently played at gambling onset. The first gambling-relatedproblem was defined as the first serious difficulty directly caused by gambling,classified as a family problem, a financial problem or other. The time to gamblingprogression (calculated only for pathological gamblers) was estimated by subtractingthe age at onset of regular gambling from the age at which the first gambling-relatedproblem occurred.

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411 405

Part 3 of the gambling section included the National Opinion Research CenterDSM-IV Screen for Gambling Problems (NODS – Gerstein et al., 1999), which wasused in order to investigate lifetime diagnostic criteria for pathological gambling insubjects Z18 years of age, and the DSM-IV-Juvenile (DSM-IV-J) criteria (Fisher,2004), which were applied in subjects 14–17 years of age. Subjects meeting at leastfive of the 10 NODS criteria or at least four of the nine DSM-IV-J criteria werecategorized as being pathological gamblers. A parallel analysis was conducted inorder to determine whether the use of separate instruments for different agegroups in the survey could have biased the diagnosis of pathological gambling. TheNODS and DSM-IV-J share six criteria (preoccupation with gambling, tolerance,withdrawal, gambling to escape problems, chasing losses and lying about gam-bling). We devised an alternative gambling classification based on the six-sharedcriteria, which yielded prevalence estimates that were slightly higher for thesample as a whole but maintained the ratio between the adolescents and theadults. Hence, age-adapted instruments did not skew the diagnosis towardsadolescents or adults but rather made it more stringent (Tavares et al., 2010).One final question in part 3 addressed the issue of whether the subject thought thatgambling was still a problem. The aim was to have a sense of the persistence ofgambling-related problems. Gambling severity was estimated by summing thenumber of NODS and DSM-IV-J pathological gambling criteria met (proportionallytransformed from nine-criteria to a 10-criteria base). The six shared criteria werecompared separately.

The following dependent variables were defined:

– “Gambling category” (N¼118), divided into three possibilities: social gambler(the individual was positive for the LBQ, but fulfilled no DSM-IV criteria for PG),problem gambler (adults: positive for one up to four DSM-IV criteria for PG;adolescents: positive for one up to three DSM-IV-J criteria for PG), andpathological gambler (adults: positive for five or more DSM-IV criteria;adolescents: positive for four or more DSM-IV-J criteria);

– “Gambling onset” (N¼118), data analysis showed a 20-year-old median forregular gambling onset, hence at-risk gamblers were divided into two cate-gories: early gamblers (initiated regular gambling prior to 20 years of age), andlate gamblers (initiated regular gambling at 20 years or later);

– “Gambling generation” (specified only for pathological gamblers, N¼33), dataanalysis showed a 35-year-old median for age, thus pathological gamblers weredivided into two categories: young gamblers (o35 years of age), and maturegamblers (Z35 years of age);

2.3. Statistical analysis

Data were adjusted for sampling probability selection and non-response rate.The sample was adjusted to sex, age and country region distribution according topost-stratification weights. The Statistical Package for Social Sciences (SPSS)Complex Sample module for weighed comparisons (SPSS, 2004), version 13.0,was used for data analysis.

Analysis of Variance (ANOVA) and Pearson's test of independence or Likelihoodratio test were used respectively for continuous and categorical variables on apreliminary univariate analysis. Demographics and gambling variables werecompared for the dependent variables “Gambling category” and “Gambling onset”(Tabachnick and Fidell, 2001).

A multivariate analysis was conducted selecting variables that reached sig-nificance at 0.10 or lower in the univariate analysis. Backward multinomial logisticregression models were built for “Gambling category” and “Gambling onset” asfollows: (I) independent variables entered as a block; (II) non-significant variableswere withdrawn step-by-step, until all remaining variables in the model weresignificant at.05 or less; and (III) excluded variables were introduced back into themodel one at a time, to check whether they may fit in the final model. Relative riskratios (RRR) and odds ratios (OR) were calculated respectively for multinomial andbinary regressions. The small number of pathological gamblers did not yield theconstruction of a regression model, hence in this case only the outcomes of theunivariate analysis were presented in the results section.

3. Results

3.1. Gambling categories: demographics and gambling behavior

Table 1 displays a summary of the demographic data bygambling category: 42.3% did not meet any of the DSM-IV criteriafor pathological gambling and were therefore classified as socialgamblers; 32.7% were classified as problem gamblers; and 24.9%were classified as pathological gamblers. The positive predictivevalue (PPV) of our screening tool was estimated at 57.7%, which isin keeping with previous reports of LBQ screening (Götestamet al., 2004). Two variables reached significance on Pearson's test

of independence: educational status and region of birth, and weresubsequently introduced into the multivariate model as controlvariables.

Table 2 shows the distribution of the gambling variables bygambling categories. The medians for the onset of regular gam-bling and for the occurrence of the first gambling-related problemcoincided at 20 years. Two variables reached significance onPearson's test of independence: “First problem caused by gam-bling” and “Current preferred game”. Less than 10% of socialgamblers reported having had a problem caused by gambling,whereas nearly 29% of problem gamblers and 63% of pathologicalgamblers reported a gambling problem. These gambling variableswere introduced along with the previously selected demographicvariables, educational status and region of birth, as a block in abackward multinomial logistic regression model. Some categoriesof region of birth had to be merged according to cultural andhistorical similarities in order to fit the model (South/Southeastand North/Center-West). Independent variables that did not pre-sent a significant contribution to the model were excluded step-by-step in the following order: “Current preferred game” and“Currently studying”. Table 3 displays the main results obtainedfrom the final model.

3.2. Gambling onset: early and late gamblers

Table 4 shows all demographic and gambling variables thatreached significance at 0.10 or lower on the univariate comparisonbetween early and late gamblers. Gambling progression was notcalculated for the whole sample, because only 38 subjects reportedhaving ever experienced a gambling problem, thus time elapsedbetween onset of regular gambling and the experience of the firstgambling problem could not be established.

All of the variables described in Table 4 were entered into abackward logistic regression model in which age at gamblingonset was the dependent variable. Table 5 displays the mainfindings. Early-onset gamblers were more likely to be male andto chase losses while gambling. Late-onset gamblers were morelikely to prefer commercially licensed gambling such as statelotteries and new games (electronic gambling).

3.3. Pathological gamblers: young and mature gamblers

Investigation into the pathological gamblers subgroup pointed to aclear division in the age distribution with a break at a 35 years oldmedian (see Fig. 1). In comparisonwith mature pathological gamblers,young pathological gamblers experienced the first gambling-relatedproblem at an earlier age (20.3 years, [S.D.¼10.7] vs. 41.9 years,[S.D.¼11.8], p¼0.001), progressed more rapidly from regular toproblem gambling (1.84 years, [S.D.¼1.24] vs. 12.75 years,[S.D.¼8.54], p¼0.008) and presented greater gambling severity (meannumber of DSM-IV pathological gambling criteria met: 7.74,[S.D.¼2.41] vs. 6.31, [S.D.¼3.85], p¼0.029).

4. Discussion

In the present study, the performance of the LBQ was compar-able to that reported in previous studies, in which it was found tohave high sensitivity and specificity, but low PPV for the diagnosisof pathological gambling (Götestam et al., 2004). In fact, we foundthat a considerable proportion (40%) of the individuals who testedpositive on the LBQ were nevertheless classified as social gamblers(because they did not meet any of the NODS or DSM-IV-J criteriafor a diagnosis of pathological gambling). It is noteworthy thatthe two items on the LBQ are derived from the DSM-IV, as well asthe items on the NODS and DSM-IV-J. This apparent conflict can be

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411406

explained by differences in wording among the instruments. TheLBQ employs short questions addressing the largest possibletimeframe (e.g., “Have you ever felt the need to bet more andmore money?”). The corresponding items in the NODS and DSM-IV-J are more stringent because they address discrete periods oflife and present a more detailed description of the behavior infocus. Therefore, it is more likely that a subject will respondnegatively to items on the NODS or DSM-IV-J.

The findings of the present study, in which the focus was on asub-group of gamblers (those identified as at-risk gamblers,n¼118), are in keeping with those of the previous study in whichour group evaluated the Brazilian National Survey on AlcoholConsumption Patterns study sample as a whole (N¼3007; Tavareset al., 2010). On both studies, we found that being a student(enrolled in any formal education program) seemed to protectagainst problem and pathological gambling, although in thepresent study that variable did not remain in the regression model

when other variables were analyzed conjointly. On the other hand,the region of birth variable was significant in the univariate andmultivariate analyses. Apparently, being born in the southernregion, the most highly developed area of Brazil, is a protectionagainst disordered gambling, regardless of the region of residence.Conversely, being born in a low socioeconomic status regionincreases the risk for developing a gambling disorder, whichsuggests that not only current but also past social vulnerabilityfavors the involvement with habit-forming stimuli. In fact, there issome evidence that childhood socioeconomic status is associatedwith later drug use (Daniel et al., 2009).

On the other hand, male gender, which was strongly associatedwith gambling in ours and other previous reports, did not showsuch an association when only at-risk gamblers were evaluated.This could be due to the loss of power resulting from the smallersample size. However, it is also possible that gender is morerelevant to the likelihood of gambling involvement than to the risk

Table 1Univariate analysis of demographic variables, by gambling category, in a sample of at-risk gamblers (N¼118) a.

Demographic variable Gambling category F p

Social Problem Pathological(n¼49) (n¼36) (n¼33)

Gender, %Male 45.3 [48.8] 28.6 [44.4] 26.1 [43.1] 1.17 0.313Female 34.0 [47.7] 44.3 [50.0] 21.6 [41.4]

Age in years 41.2 [15.0] 41.2 [17.1] 40.5 [18.8] 0.20 0.984

Education, %Elementary school 42.0 [51.4] 24.1 [44.5] 33.9 [49.3] 0.92 0.442Middle school 41.2 [48.3] 43.2 [48.6] 15.6 [35.6]High school or beyond 43.8 [46.2] 32.5 [43.6] 23.7 [39.6]

Educational status, %Student 29.8 [52.0] 63.4 [54.7] 6.8 [28.6] 3.61 0.032Non-student 43.8 [48.1] 29.2 [44.1] 27.1 [43.0]

Ethnicity, %European 56.7 [48.8] 25.2 [42.8] 18.1 [37.9] 1.56 0.177African-European 27.9 [41.5] 50.7 [46.2] 21.4 [37.9]African 34.6 [52.3] 29.2 [50.0] 36.2 [52.9]Other 27.6 [33.9] 33.5 [35.7] 38.9 [36.9]

Marital status, %Never married 31.0 [46.5] 47.1 [50.2] 21.9 [41.6] 1.40 0.237Married or cohabitating 49.7 [47.2] 22.2 [39.2] 28.1 [42.4]Unmarried 38.5 [53.0] 41.3 [53.6] 20.2 [43.7]

Employment status, %Employed/self-employed 44.0 [47.6] 32.0 [44.7] 24.0 [41.0] 0.88 0.519Unemployed – 54.4 [50.7] 45.6 [20.7]Homemaker 60.1 [44.8] 31.6 [42.6] 8.3 [25.3]Retired 45.8 [49.6] 27.8 [44.6] 26.4 [43.9]Student 37.7 [72.3] 38.8 [72.7] 23.5 [63.3]

Monthly incomeb, Mean 835.0 [914.4] 924.8 [650.8] 1040.6 [822.3] 0.69 0.504

Region of birth, %South/southeast 58.3 [48.6] 21.4 [40.4] 20.4 [39.7] 3.19 0.020North/central-west 16.2 [35.5] 22.5 [48.0] 28.7 [43.6]Northeast 36.1 [47.9] 35.9 [47.8] 28.0 [44.8]

Region of residence, %Southeast 48.9 [50.2] 27.4 [44.8] 23.7 [42.7] 1.88 0.119Northeast 31.1 [47.9] 31.5 [48.1] 37.3 [50.1]South 67.3 [44.6] 21.7 [39.2] 11.0 [29.8]Central-west 28.4 [54.5] 48.5 [60.4] 23.1 [50.9]North 24.5 [33.6] 57.7 [38.6] 17.7 [29.9]

Religious affiliation, %Catholic 45.3 [47.9] 31.5 [44.7] 23.3 [40.7] 1.30 0.264Protestant 22.3 [45.2] 38.2 [52.8] 39.5 [53.1]Other 25.0 [41.8] 42.1 [47.6] 32.8 [45.3]None 64.0 [47.7] 26.9 [44.1] 9.1 [28.6]

a All results expressed as mean [S.D.].b In Brazilian reals (R$; exchange rate at the time of the survey, R$ 1.00¼US$ 0.45).

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411 407

of developing at-risk gambling. We found few differences amongthe gambling categories regarding gambling behaviors andgambling-related problems. As expected, financial problems weresignificantly associated with being a pathological gambler. How-ever, that association is somewhat tautological, given that thediagnosis of pathological gambling partly relies upon money-related criteria and financial damage.

As previously mentioned, we found that 20 years of age was acritical point for the onset of regular and problem gambling. Amongthe at-risk gamblers, we identified significant differences between theearly-onset and the late-onset gamblers. Male gender, chasing lossesand preferred game at gambling onset remained significantly asso-ciated with the age at onset of gambling, even after the multivariateanalysis had been adjusted for demographic characteristics (age,

educational status, marital status and employment status). Our findingthat male gender was strongly related with early onset of gamblingbehavior is in agreement with those of population-based surveyscarried out in high-income countries (Blanco et al., 2006). The fact thatthe same remains true in a LAMIC setting such as Brazil is furtherevidence that gender plays a role in shaping gambling behaviorbeyond the socioeconomic scenario. Our finding that early onsetincreases the likelihood of chasing losses suggests that early-onsetgamblers show unhealthier patterns of gambling behavior. Early andlate onset gamblers also differed regarding gaming preference.Although the preferred game at gambling onset and current preferredgame variables were both selected for inclusion in the multivariateanalysis, only the former remained in the final model. Among late-onset gamblers, the most common preferred games at gambling onsetwere state lotteries and the so-called new games, which encompassmainly electronic gambling options. In Brazil, the status of electronicgambling has vacillated between legal and illegal over the years, butremained legal from 1995 to 2004. Therefore, the preferred game atgambling onset among the late-onset gamblers reflects a choice toengage in only lawful and socially tolerated gambling. In contrast, theearly-onset gamblers seem to give preference to non-commercialunlicensed gambling, represented by card playing (poker playing hasonly recently managed to achieve a legal status, all other forms of cardbetting remain unregulated), as well as by the popular illegal lotteriesthat are run in the country.

The results from the analysis of the subgroup of pathologicalgamblers align with findings from the comparisons between earlyand late at-risk gamblers. The result about the earlier age of firstgambling problem of young pathological gamblers compared tomature pathological gamblers is not just an artifact resulting fromthe earlier age of gambling onset, as it can be seen from the rathercompressed period of time elapsed between onset of regulargambling and emergence of gambling related problems. In addition,the severity of pathological gambling, as determined by the meannumber of DSM-IV criteria met, appeared to be greater in the younggamblers than in the mature gamblers. Taken together, these findings

Table 2Univariate analysis of gambling variables, by gambling category, in a sample of at-risk gamblers (N¼118).a

Variable Gambling category F p

Social Problem Pathological(N¼49) (N¼36) (N¼33)

Age at gambling onset, years 23.45 [9.1] 22.7 [11.5] 21.19 [11.3] 0.53 0.594

Preferred game at gambling onset, %Card games 42.8 [46.4] 25.1 [40.6] 32.2 [43.8] 0.92 0.484Illegal lotteries 37.1 [48.1] 37.6 [48.3] 25.3 [43.3]State lotteries 56.1 [45.1] 32.0 [42.4] 11.9 [29.4]New games (electronic gambling) 25.9 [48.8] 26.7 [49.3] 47.4 [55.6]Other games 34.4 [58.0] 45.9 [60.8] 19.6 [48.5]

First gambling-related problem (type), %None 54.1 [48.9] 32.8 [46.1] 13.1 [33.1] 5.70 o0.001Family problem 15.5 [37.4] 47.2 [51.6] 37.3 [50.0]Financial problem 3.2 [18.4] 24.8 [45.4] 72.0 [47.1]Other 33.1 [53.2] 19.4 [44.7] 47.5 [56.5]Age at first gambling-related problem, years 23.99 [11.1] 26.07 [16.3] 29.10 [14.6] 0.56 0.585

Current preferred game, %Card games 40.3 [48.6] 32.6 [46.5] 27.1 [44.1] 2.34 0.019Illegal lotteries 37.6 [46.4] 47.4 [47.8] 15.0 [34.3]State lotteries 61.3 [45.2] 24.2 [39.7] 14.5 [32.7]New games (electronic gambling) 22.7 [47.7] 41.5 [56.2] 35.8 [54.6]Other games 43.4 [54.2] 34.7 [52.1] 21.9 [45.3]None (stopped gambling) – 8.3 [29.7] 91.7 [29.7]

Gambling still a problem, %Yes 19.6 [36.2] 37.6 [44.2] 42.8 [45.2] 1.58 0.212No 46.1 [50.8] 31.9 [47.5] 21.9 [42.2]

a All results expressed as mean [S.D.].

Table 3Final model of the multinomial backward logistic regression of demographic andgambling variables, by gambling category, in a sample of at-risk gamblers(N¼118).a

Independent variables Relative risk ratio*

Problem gamblers Pathological gamblers

First gambling-related problem [type]None 1.15 [0.13–10.15] 0.19 [0.02–1.56]Family problem 3.56 [0.26–47.97] 1.54 [0.14–16.53]Financial problem 14.39 [0.67–307.92]† 18.03 [1.15–283.58]‡

Other – –†p¼0.087; ‡p¼0.040

Region of birthSouth/southeast 0.36 [0.12–1.05]§ 0.43 [0.13–1.42]North/central-west 2.51 [0.71–8.96] 1.38 [0.39–4.91]Northeast – –§p¼0.061

Model data: F[10;45]¼2.595, p¼0.014, Nagelkerke's R2¼0.334.a All results expressed as relative risk ratio and 95% CI [minimum–maximum].n Social Gamblers is the reference category, for the independent variables the

reference category is the last one.

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411408

strengthen the impression that we may be dealing with two types ofat-risk gamblers, one of which is particularly concerning, comprisingindividuals who are predominantly male, have an early gambling

onset, have an initial preference for unregulated gambling and tendto progress rapidly to severe disordered gambling. The window forearly intervention is narrower in such individuals, who therefore

Table 4Univariate analysis of demographic and gambling variables, by age at gambling onset,a in a sample of at-risk gamblers (N¼118).b

Demographic variable Early-onset gamblers Late-onset gamblers F p

(n¼57) (n¼61)

Gender, %Male 61.1 [47.8] 38.9 [47.8] 23.6 o0.001Female 11.9 [32.6] 88.1 [32.6]

Age, years 33.06 [15.3] 48.42 [14.8] o0.001

Educational statusStudent 89.9 [34.3] 10.1 [34.3] 8.3 0.005Non-student 43.5 [48.1] 56.5 [48.1]

Marital status, %Never married 85.7 [35.3] 14.3 [35.3] 11.0 o0.001Married or cohabitating 34.1 [44.7] 65.9 [44.7]Unmarried 27.8 [48.8] 72.2 [48.8]

Employment status, %Employed/self-employed 50.6 [47.9] 49.4 [47.9] 4.7 0.050Unemployed 92.7 [26.4] 7.3 [26.4]Homemaker – 100Retired 33.6 [47.0] 66.4 [47.0]Student 100.0 [0.0] –

Preferred game at gambling onset, %Card games 62.7 [45.3] 37.3 [45.3] 2.6 0.068Illegal lotteries 48.5 [49.8] 51.5 [49.8]State lotteries 29.0 [41.2] 71.0 [41.2]New games (electronic gambling) 46.6 [55.6] 53.4 [55.6]Other games 75.7 [52.4] 24.3 [52.4]

Current preferred game, %Card games 70.3 [45.3] 29.7 [45.3] 3.3 0.015Illegal lotteries 54.9 [47.7] 45.1 [47.7]State lotteries 23.1 [39.1] 76.9 [39.1]New games (electronic gambling) 50.1 [57.0] 49.9 [57.0]Other games 58.3 [54.0] 41.7 [54.0]None (stopped gambling) 73.9 [47.2] 26.1 [47.2]

Chasing losses, %Yes 33.6 [48.4] 66.4 [48.4] 10.3 0.002No 65.8 [46.4] 34.2 [46.4]

a Early onset (o20 years of age) vs. late onset (Z20 years of age).b All results expressed as mean [S.D.].

Table 5Demographics and gambling variables for early and late onset gamblersa, backward logistic regression (N¼118).b

Variable Coefficient [SE] OR [95% CI] DF1c DF2c Wald F p

Gender 1 54 12.903 0.001Male 2.7 [0.75] 15.024 [3.311–68.173]Femalea – –

Age �0.67 [0.03] 0.935 [0.889–0.983] 1 54 7.238 0.009

Educational status 1 54 4.958 0.030Svtudent 2.02 [0.91] 7.567 [1.223–46.808]Non-studenta – –

Chasing losses 1 54 4.238 0.044Yes 1.72 [0.84] 5.582 [1.046–29.795]Noa – –

Preferred game at gambling onset 4 51 2.658 0.043Card games �0.04 [0.98] 0.965 [0.136–6.841]Illegal lotteries �0.17 [0.89] 0.842 [0.141–5.036]State lotteries �1.70 [0.82] 0.183 [0.035–0.952]New games �2.16 [0.99] 0.116 [0.016–0.849]Other gamesa – –

a Reference category.b Model data: F[8;47]¼3.797, p¼0.002, Nagelkerke's R2¼0.594.c Degrees of freedom.

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411 409

merit even greater attention. This last finding contradicts our ownprevious reports, in which gambling progression was found to bemost rapid among middle-aged female pathological gamblers. Itunderscores the relevance of investigating gambling in communitysamples, thereby circumventing the biases inherent to samplescomposed of treatment-seeking individuals. Indeed, the absence ofthis particular segment of young and severe male gamblers fromclinical reports is a further suggestion that the current treatmentprograms have failed to target them. Specific recruiting methodsshould be developed in order to bring them to the attention ofclinicians.

The establishment of a division between early- and late-onsetgamblers mirrors attempts at classifying alcoholics. Although noconsensus has yet been reached, various studies have investigatedthe typology of early- and late-onsets alcohol dependence. Baborand Caetano (2006) found that individuals with the early-onsetsubtype of alcohol dependence show signs of greater impulsivityand severity. It is noteworthy that our community-based findingsare in agreement with those previously reported for addictionsother than gambling, even in diverse sociocultural scenarios. Thebroader age bracket adopted in the present study (Z14 years ofage) might have contributed to the identification of youngergamblers, underscoring the importance of including adolescentsin population surveys. Indeed, despite the fact that adolescentshave been shown to have a high prevalence of mental disordersand to frequently be involved in gambling (Spritzer et al., 2011),this population segment is often underrepresented in gamblingstudies.

Our study has some limitations. For a brief period at the time ofthe survey, there were riots, supposedly carried out by organizedgroups, in some jurisdictions. The consequent heightened securityconcerns were likely responsible, in part, for the modest responserate. However, that probably represents only a slight bias, becauseit seems unlikely that only gambling households would be lessprone to participate when security risks are high. Our surveytapped into various sorts of addiction related behaviors andfactors. To keep the interview to a reasonable duration, somesections were preceded by screening items, such as the LBQ, aprocedure that have skewed the selection toward at-risk gamblers.In addition, we did not assess inter-rater reliability and wecollected no further data on the subjects who tested negative onthe LBQ. Therefore, the sensitivity and specificity of the screeningtool could not be determined, although the PPV value suggeststhat the screen was inclusive and that few, if any, problemgamblers went undetected. Furthermore, milestone ages wereself-reported with no collateral checking.

Nevertheless, the careful training and vetting of the inter-viewers, together with the rigorous methodology employed inthe stratified household survey, resulted in a community samplethat we believe was representative of at-risk gamblers in thecommunity, including adolescents. Thus, we were able to identify

a group of young male gamblers who are at risk of rapidlyprogressing to more severe forms of gambling, with a preferencefor unregulated gambling. Age restrictions or maybe the type ofgambling preferred by this segment of the population mightexplain its underrepresentation in clinical settings. Outreachstrategies and early interventions should be provided in order toprevent these youth from evolving to pathological gambling.

References

Afifi, T.O., Cox, B.J., Martens, P.J., Sareen, J., Enns, MW, 2010. Demographic and socialvariables associated with problem gambling among men and women inCanada. Psychiatry Research 178, 395–400.

American Psychiatric Association, 1994. Diagnostic and Statistical Manual of MentalDisorders, 4th ed. American Psychiatric Association, Washington, DC

Babor, T.F, Caetano, R., 2006. Subtypes of substance dependence and abuse:implications for diagnostic classification and empirical research. Addiction101 (Suppl. 1), S104–S110.

Black, D.W., McCormick, B., Losch, M.E., Shaw, M., Lutz, G., Allen, J, 2012. Prevalenceof problem gambling in Iowa: revisiting Shaffer's adaptation hypothesis. Annalsof Clinical Psychiatry 24, 279–284.

Blanco, C., Hasin, D.S., Petry, N., Stinson, F.S., Grant, B.F., 2006. Sex differences insubclinical and DSM-IV pathological gambling: results from the NationalEpidemiologic Survey on Alcohol and Related Conditions. Psychological Med-icine 36, 943–953.

Burge, A.N., Pietrzak, R.H., Molina, C.A., Petry, N.M., 2004. Age of gambling initiationand severity of gambling and health problems among older adult problemgamblers. Psychiatric Services 55, 1437–1439.

Castro-Costa, E., Ferri, C.P., Lima-Costa, M.F., Zaleski, M., Pinsky, I., Caetano, R.,Laranjeira, R., 2008. Alcohol consumption in late-life—the first BrazilianNational Alcohol Survey (BNAS). Addictive Behaviors 33, 1598–1601.

Cox, B.J., Yu, N., Afifi, T.O., Ladouceur, R., 2005. A national survey of gamblingproblems in Canada. Canadian Journal of Psychiatry 50, 213–217.

Daniel, J.Z., Hickman, M., Macleod, J., Wiles, N., Lingford-Hughes, A., Farrell, M.,Araya, R., Skapinakis, P., Haynes, J., Lewis, G., 2009. Is socioeconomic status inearly life associated with drug use? A systematic review of the evidence. Drugand Alcohol Review 28, 142–153.

DataUFF, 2009. Núcleo de Pesquisas da Universidade Federal Fluminense (Internet).Niterói, Brazil: Universidade Federal Fluminense; (updated 2013 September 5).Pesquisa Social Brasileira (updated 2009 November 03; cited 2013 September 8).Available from: ⟨http://www.uff.br/datauff/PESB.htm⟩.

Dellis, A., Spurrett, D., Hofmeyr, A., Sharp, C., Ross, D., 2013. Gambling participationand problem gambling severity among rural and peri-urban poor South Africanadults in KwaZulu-Natal. Journal of Gambling Studies 29, 417–433.

Fisher, S., 2004. Measuring pathological gambling in children: the case of fruitmachines in the U.K. Journal of Gambling Studies 8, 263–285.

Gerstein, D., Murphy, S., Toce, M., Hoffmann, J., Palmer, A., Johnson, R., 1999.Gambling Impact and Behavior study: Report to the National Gambling ImpactStudy Commission. National Opinion Research Center, Chicago, IL

Götestam, K.G., Johansson, A., Wenzel, H.G., Simonsen, I.E., 2004. Validation of thelie/bet screen for pathological gambling on two normal population data sets.Psychological Reports 95, 1009–1013.

Hansen, H.M., Hurwitz, W.N., Madow, W.G., 1993. Sample Survey Methods andTheory, vol. I and II. John Wiley & Sons, Inc., New York.

IBGE, 2008. Instituto Brasileiro de Geografia e Estatística (Internet). Rio de Janeiro, Brazil:Instituto Brasileiro de Geografia e Estatística; (updated 2013 September 5).Estimativas das populações residentes, em 11 de julho de 2008, segundo osmunicípios (Estimated resident populations on July 1, 2008, by municipality)(updated 2008 December 12; cited 2013 September 8). Available from: ⟨http://www.ibge.gov.br/home/estatistica/populacao/estimativa2008/POP_2008_TCU.pdf⟩.

Johnson, E.E., Hamer, R.M., Nora, R.M., 1998. The Lie/Bet Questionnaire for screen-ing pathological gamblers: a follow-up study. Psychological Reports 83,1219–1224.

Ladouceur, R., Jacques, C., Chevalier, S., Sévigny, S., Hamel, D., 2005. Prevalence ofpathological gambling in Quebec in 2002. Canadian Journal of Psychiatry 50,451–456.

Nelson, S.E., Laplante, D.A., Labrie, R.A, Shaffer, H.J., 2006. The proxy effect: genderand gambling problem trajectories of Iowa Gambling Treatment Programparticipants. Journal of Gambling Studies 22, 221–240.

Rahman, A.S., Pilver, C.E., Desai, R.A., Steinberg, M.A., Rugle, L., Krishnan-Sarin, S.,Potenza, M.N., 2012. The relationship between age of gambling onset andadolescent problematic gambling severity. Journal of Psychiatric Research 46,675–683.

Spritzer, D.T., Rohde, L.A., Benzano, D.B., Laranjeira, R.R., Pinsky, I., Zaleski, M.,Tavares, H., 2011. Prevalence and correlates of gambling problems among anationally representative sample of Brazilian adolescents. Journal of GamblingStudies 27, 649–661.

Tabachnick, B.G., Fidell, L.S., 2001. Using Multivariate Statistics, 4th Edition Allyn &Bacon, Needham Heights, MA

Tavares, H., Zilberman, M.L., Beites, F.J., Gentil, V., 2001. Gender differences ingambling progression. Journal of Gambling Studies 17, 151–159.

Fig. 1. Age distribution in a sample of at-risk gamblers (N¼118).

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411410

Tavares, H., Carneiro, E., Sanches, M., Pinsky, I., Caetano, R., Zaleski, M., Laranjeira,R., 2010. Gambling in Brazil: lifetime prevalences and sociodemographiccorrelates. Psychiatry Research 180, 35–41.

Tavares, H., Martins, S.S., Lobo, D.S., Silveira, C.M., Gentil, V., Hodgins, D.C., 2003.Factors at play in faster progression for female pathological gamblers: anexploratory analysis. Journal of Clinical Psychiatry 64, 433–438.

Walker, S.E., Abbott, M.W., Gray, R.J., 2012. Knowledge, views and experiences ofgambling and gambling-related harms in different ethnic and socio-economicgroups in New Zealand. Australian and New Zealand Journal of Public Health36, 153–159.

Welte, J.W., Barnes, G.M., Tidwell, M.C., Hoffman, J.H., 2009. Association betweenproblem gambling and conduct disorder in a national survey of adolescents andyoung adults in the United States. Journal of Adolescent Health 45, 396–401.

Weinstock, J., Ledgerwood, D.M., Modesto-Lowe, V., Petry, N.M., 2008. Ludomania:cross-cultural examinations of gambling and its treatment. Revista Brasileira dePsiquiatria 30, S3–S10.

E. Carneiro et al. / Psychiatry Research 216 (2014) 404–411 411