Causal impact of gambling-related cognitive distortions on the severity of gambling disorder: A one-year multicenter longitudinal study in treatment-seeking patients in Japan

Abstract

Background and Aims

Gambling-related cognitive distortions (GRCD) sustain gambling behaviors despite adverse consequences. While previous studies have shown an association between GRCD and gambling disorder (GD) severity, few have conducted causal analyses. We aimed to examine temporal changes in GRCD among treatment-seeking patients and to investigate whether GRCD predicts subsequent gambling severity.

Methods

This one-year follow-up study was conducted in collaboration with 20 addiction-specialized medical institutions. The participants were 100 male patients diagnosed with GD (mean age: 37.2±8.5). We also examined the differences in GRCD improvement based on gambling relapse after treatment initiation. GRCD were assessed using the Gambling-Related Cognitions Scale (GRCS), and gambling severity was measured using the Problem Gambling Severity Index (PGSI). Gambling engaement was assessed using a self-report questionnaire. Temporal changes were analyzed using repeated-measures ANOVA, and the causal relationship between the GRCD and the PGSI was examined using a cross-lagged panel model (CLPM).

Results

Overall, GRCD significantly decreased within 6 months of treatment initiation and then stabilized. The abstinence group had significantly lower GRCD scores than the non-abstinence group at all the timepoints. CLPM identified Perceived Inability to Stop Gambling (IS) as the only GRCD subscale that significantly predicted PGSI scores at 12 months.

Discussion

This study demonstrates that GRCD predict the severity of subsequent gambling disorders. In particular, IS has been identified as a critical target for interventions.

Conclusions

These findings provide valuable evidence of a causal relationship between GRCD and gambling severity and underscore the importance of targeting IS in treatment.

Introduction

Background of treatment for patients with gambling disorder in Japan

Gambling disorder (GD) negatively affects both physical and mental health, and is linked to various social issues, including family conflicts, workplace difficulties, financial losses, and involvement in illegal behaviors (Grant & Potenza, 2022; Stefanovics & Potenza, 2022). According to a 2020 nationwide general population survey, past-year prevalence of GD in Japan was estimated to be 2.2%—equivalent to approximately 2 million adults. This rate is relatively high compared to other countries (Nitta, Koga, Toyama, & Matsushita, 2023), and is perhaps due to the widespread accessibility of “pachi-slot”—devices similar to slot machines—and “pachinko,” both of which are Japan-specific Electronic Gambling Machines (EGMs) available throughout the country (Higuchi, 2020). However, specialized medical care for Japanese GD individuals has long been underdeveloped. The major shift occurred when the 2016 legalization of casinos in Japan raised public concern about gambling-related harm. In response, the Japanese government enacted the Basic Act on Countermeasures against Gambling Addiction in 2018, which expanded consultation and treatment services.

Previous Japan-based studies on treatment-seeking patients with GD have primarily involved retrospective medical record reviews (Harada, Inaba, Kumagai, Sonomoto, & Kumagai, 2010; Moriyama, 2016; Ohta, 2008; Yamada et al., 2023). These studies identified that Japanese GD patients are predominantly male (7:1 to 9:1), often begin gambling in their teens, and develop problematic gambling in their 30 s or 40 s. Additionally, they tend to have higher education levels, divorce rates, and smoking prevalence than the general population. Pachinko and pachi-slot account for 50–90% of gambling addiction cases, and alcohol dependence and mood disorders are common comorbidities.

As described above, studies targeting treatment-seeking patients with GD in Japan have primarily relied on cross-sectional designs; there exist no nationwide longitudinal surveys.

Gambling-related cognitive distortions as an important factor associated with the onset and prognosis of gambling disorder

Gambling-related cognitive distortions (GRCD) refer to gambling-specific cognitive beliefs held by both regular and pathological gamblers, and have been conceptualized by several researchers as false perceptions of gambling success and beliefs about the ability to control gambling outcomes (Delfabbro & Winefield, 2000; Gadboury & Ladouceur, 1989; Jacobsen, Knudsen, Krogh, Pallesen, & Molde, 2007). These distorted cognitions are thought to sustain gambling behavior despite continued loss, and have been associated with both the onset and maintenance of pathological gambling (Goodie & Fortune, 2013; Spurrier & Blaszczynski, 2014; Tan & Tam, 2024).

Moreover, GRCD are considered a critical factor influencing GD prognosis, with previous cross-sectional studies revealing GRCD's impact on GD severity in different cultural contexts (Estévez et al., 2021; Ledgerwood et al., 2020; Mallorquí-Bagué et al., 2019; Philander & Gainsbury, 2023; Shirk et al., 2018). Although few causal analyses have been conducted in these studies, one longitudinal study of young adults (aged 18–21) at two timepoints found that individuals with greater gambling severity tended to develop more pronounced cognitive distortions over time (Nicholson, Graves, Ellery, & Afifi, 2016). Furthermore, individuals with existing cognitive distortions are more likely to develop gambling problems (Yakovenko et al., 2016), although evidence indicates a bidirectional GRCD–gambling severity relationship (Leonard & Williams, 2016). However, these findings are based on non-clinical samples, limiting their generalizability to severe-GD individuals.

Clinical studies have explored the GRCD–treatment prognosis relationship. Ledgerwood et al. (2020) conducted a retrospective medical record review of a residential treatment facility, finding that greater reductions in the GRCD were associated with improved treatment outcomes, whereas stronger predictive control distortions increased the risk of treatment dropout. Shirk et al. (2018) examined 345 treatment-seeking patients, finding that GRCD mediate the relationship between depressive symptoms and gambling severity. To date, a Cognitive-Behavioral Therapy (CBT) has been demonstrated as an effective treatment for GD (Eriksen et al., 2023; Ribeiro, Afonso, & Morgado, 2021). Rossini-Dib, Fuentes, and Tavares (2015) assessed outpatients receiving either psychoeducation alone or CBT-based interventions, discovering that cognitive restructuring was crucial in treatment success. Similarly, CBT interventions' effectiveness in improving GRCD has been confirmed (Chrétien, Giroux, Goulet, Jacques, & Bouchard, 2017; Fortune & Goodie, 2012). Conversely, Dunsmuir, Smith, Fairweather-Schmidt, Riley, and Battersby (2018) reported that some GRCD components—gambling expectancy and interpretive bias—were not associated with compulsive gambling symptoms. These findings suggest that, while some cognitive distortions may be linked to GD severity, others may not directly contribute to GD prognosis. Furthermore, these studies have only evaluated short-term treatment effects and have not observed long-term modifications of GRCD.

Although GRCD has been recognized as key in GD, previous studies have primarily relied on cross-sectional data or short-term intervention follow-up, with little evidence regarding how GRCD evolves over time and affects long-term GD prognosis. Given Japan's distinct gambling environment, we must examine whether the existing evidence applies to this population.

To address these research gaps, we conducted the Japan Collaborative Clinical Study on GD (JaCCS-G)—a one-year multicenter follow-up study—at 20 specialized medical facilities; it was the first prospective longitudinal study of GD in Japan. This paper presents the findings of that study, with a focus on GRCD.

Objectives:

1. Investigate the temporal changes and improvement processes in GRCD after treatment initiation in GD patients.

2. Examine whether there were differences in GRCD improvement between the non-abstinence and abstinence groups during the one-year follow-up period.

3. Determine whether improvements in GRCD contribute to reductions in subsequent gambling problem severity and identify which GRCD aspects most strongly predict prognosis.

Methods

Details of the JaCCS-G are available elsewhere. The following is an overview of the study procedure and participant recruitment (Fig. 1).

Fig. 1.

Flow chart of the study participants

Recruitment and baseline procedures

A total of 211 potential participants were recruited from collaborating institutions and assessed for eligibility based on the following criteria. Inclusion criteria: a) aged 20–65, b) DSM-5 diagnosis of GD, and c) ability to provide informed consent. Exclusion criteria: a) cognitive or language difficulties and b) any condition deemed, by the attending psychiatrist, unsuitable for participation. As part of the baseline clinical assessment, psychiatrists conducted diagnostic interviews using the Structured Clinical Interview for DSM-5 Research Version (SCID-5-RV; First, Williams, Karg, & Spitzer, 2020). Trained professionals also collected sociodemographic, medical history, and gambling-related variables, including counseling experience, age of onset of problematic gambling, primary gambling forms, frequency, bet amounts, and debts. Two individuals were excluded for not meeting the inclusion criteria.

Eligible participants (n = 209) were enrolled and provided written informed consent. They were asked to complete a set of self-administered baseline questionnaires, including the Problem Gambling Severity Index (PGSI) and Gambling-Related Cognitions Scale (GRCS), within four weeks. Finally, 202 participants (mean age 36.3 ± 8.7 years, range = 20–60, 96.5% male) completed the baseline survey. All procedures were conducted between February and September 2021. Participants received a 1,000-yen gift voucher for each completed survey.

Follow-up survey

Of the 202 participants who completed the baseline survey, 183 (mean age 36.3 ± 8.7 years, 96.7% male) consented to follow-up surveys at 1, 3, 6, 9, and 12 months after the one-year period. Participants self-reported treatment retention, type of treatment received (e.g., doctor visits, CBT, psychotherapy), and gambling behavior (abstinence or non-abstinence) during the follow-up. Since there were only six females, they were excluded from the analysis to ensure the results' validity and avoid bias. One hundred male participants with complete follow-up data were included in the analysis.

Measures

Gambling-Related Cognitions Scale (GRCS)

The GRCS was developed to comprehensively measure GRCD in the general population and individuals with GD (Raylu & Oei, 2004). It was designed to assess erroneous gambling beliefs across multiple dimensions, providing a standardized measure that has been widely adopted, translated internationally, and demonstrated high reliability and validity (Del Prete et al., 2017; Donati, Ancona, Chiesi, & Primi, 2015; Iliceto et al., 2015; Leonard & Williams, 2015; Yang, Wu, Wen, Lu, & Li, 2014). Yokomitsu, Takahashi, Kanazawa, and Sakano (2015) validated the Japanese version of GRCS. The GRCS consists of 23 items rated on a 7-point scale (1 = strongly disagree, 7 = strongly agree), with higher scores indicating stronger distortions. It comprises the following five subscales; Interpretive Bias (IB) (four items): Attributing gambling wins to personal skill while blaming losses on external factors.

Illusion of Control (IC) (four items): Believing that certain actions or rituals can influence gambling outcomes. Predictive Control (PC) (six items): Thinking that a win is imminent based solely on previous gambling outcomes. Gambling Expectancies (GE) (five items): Expecting gambling to provide benefits or relieve distress. Perceived Inability to Stop Gambling (IS) (four items): Feeling incapable of quitting gambling.

The Cronbach's alpha coefficients for all subscales in this study ranged from 0.739 to 0.933, indicating high internal consistency.

Problem Gambling Severity Index (PGSI)

The PGSI assesses problem-gambling severity in both general and clinical populations (Ferris & Wynne, 2001). It comprises nine items rated from 0 (never) to 3 (almost always), with total scores ranging from 0 to 27. A score of eight or more on the Japanese version (So, Matsushita, Kishimoto, & Furukawa, 2019) is indicative of problem gambling. The PGSI has exhibited strong validity and reliability across cultures and high sensitivity and specificity against the DSM-IV criteria (Caler, Garcia, & Nower, 2016; Holtgraves, 2009; Orford, Wardle, Griffiths, Sproston, & Erens, 2010). Previous studies also found a linear relationship between PGSI scores and gambling-related harms such as psychological distress and suicidality (Ipsos & GambleAware, 2023; Wardle, Kesaite, Tipping, & McManus, 2023). Therefore, we treated the PGSI as a continuous variable to enhance analytical accuracy. The PGSI exhibited high internal reliability (α = 0.748–0.927) at all timepoints.

Gambling involvement during the follow-up period

Gambling involvement during follow-up was determined based on a single question: “Have you gambled from the date of your first visit until today?” Data were collected at five timepoints: 1, 3, 6, 9, and 12 months. Participants were classified as “non-abstinence” if they answered “yes” at any timepoint and “abstinence” if they answered “no” at all timepoints.

Ethics

This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Kurihama Medical and the Addiction Center Ethics Committee (Approval No. 427). All participants were informed about the study and provided informed consent. All the collected data, including personal information, were stored in a web-based database developed for this study.

Statistical analysis

Repeated measures ANOVA

To examine the temporal changes in GRCD over the one-year follow-up period, a repeated-measures one-way ANOVA was conducted with three timepoints (baseline, 6 months, and 12 months) as the independent variables. The dependent variable was the mean score of the five GRCS subscales. Post-hoc pairwise comparisons were conducted using the Bonferroni correction to identify when pairs of timepoints (baseline vs. 6 months, baseline vs. 12 months, and 6 months vs. 12 months) showed significant differences.

Subsequently, the participants were categorized into two groups (abstinence or non-abstinence) based on their self-reported gambling behavior during the follow-up period. To examine the differences in the mean scores for the PGSI and GRCS subscales across the three timepoints between the two groups, a repeated-measures two-way ANOVA was conducted, with time as the within-subjects factor and group (abstinence vs. non-abstinence) as the between-subjects factor. Bonferroni adjustment was also performed.

Path analysis using cross-lagged panel model

We conducted a cross-lagged panel model (CLPM) using path analysis to explore the temporal and bidirectional causal relationships between PGSI and GRCS scores. The CLPM is a statistical tool designed to examine the causal relationships between two variables using longitudinal data, accounting for the possibility of mutual influence (Finkel, 1995). A summary of our model is presented in Fig. 2.

Fig. 2.

The conceptual framework for the cross-lagged panel model in current study

Note: PGSI: Problem Gambling Severity Index, GRCS: Gambling-related cognitions scale.

“a”, “b,” “c,” “d” indicates each path coefficients between variables.

When “a” or “c” are significantly positive, it indicates that scores of the PGSI contributes to GRCS subscales. Conversely, when “b” or “d” are significantly positive, it suggests that the scores of the GRCS subscale contributes to PGSI.”

“e” indicates residual variance (error term) in the cross-lagged panel model, representing variance not explained by the predictors.

“f”, “g”, “h” indicates the correlation coefficient between GRCS and PGSI at each time point.

The GRCS subscales were used as cognitive distortion indicators, whereas the PGSI scores represented GD severity. PGSI and GRCS scores at three timepoints were examined; baseline (BL) scores predict 6-month scores, and 6-month scores predict 12-month scores. This approach is valuable for investigating the interaction between GD severity and cognitive distortion over time.

Each GRCS subscale was analyzed separately to determine its causal relationship with PGSI scores. When the path coefficients a or c are significantly positive, “GD severity influences cognitive distortions.” Conversely, when the coefficients b or d are significantly positive, “cognitive distortions contribute to GD severity” (Fig. 2).

All analyses were performed using IBM SPSS version 29.0 and Amos version 28.0. Statistical significance was p < 0.05.

Results

Participants' sociodemographic factors and clinical characteristics

Table 1 presents the participants' sociodemographic and clinical characteristics. All respondents were male (mean age 37.2 ± 8.5 years) and 70% were in their 30–40 s. Approximately 75% were employed full-time and more than half belonged to Japan's middle-income bracket (the mean annual income for men in 2020 was approximately 35,000 USD). Around 30% were relatively high-income earners, 67% were married, and roughly 30% had a college degree. When participants were asked about the types of gambling that caused problems at baseline, the most frequently reported types were pachinko (59%), pachi-slot (53%), and horse racing (35%). More than half had debt of one million yen (approximately 9,000 USD) or more. The mean age of first gambling experience was 19.1 years, with habitual gambling beginning at 20.5 years old, and problematic gambling developing in the late 20 s (onset mean age 28.2 ± 6.8). Regarding cohabitation status, 77% lived with their family, 15% lived alone, and 6% were separated from their family. Regarding smoking status, 46% were current smokers, 18% were ex-smokers, and 35% had never smoked.

Table 1.

Sociodemographic and clinical characteristics of treatment-seeking patients with gambling disorder

Variables	Categories	Analysis Sample (N = 100)		Non-responders (N = 77)		χ2 test
		n	%	n	%
Age (years)	20–29	19	19.0	19	24.7	n.s.
	30–39	49	49.0	38	49.4
	40–49	22	22.0	14	18.2
	50–59	9	9.0	6	7.8
	60 <=	1	1.0	0	0.0
Occupation	Self-employed	2	2.0	2	2.6	n.s.
	Full-time employed	76	76.0	59	76.6
	Part-time employed	9	9.0	7	9.1
	Student	2	2.0	0	0.0
	Unemployed/Retired	10	10.0	8	10.4
	No response	1	1.0	1	1.3
Marital status	Married	66	66.0	34	44.2	x2(2) = 10.309 p < 0.01
	Divorce	8	8.0	6	7.8
	Never married	24	24.0	36	46.8
	No response	2	2.0	1	1.3
Cohabitants	Living with family	77	77.0	53	68.8	n.s.
	Separation	6	6.0	5	6.5
	Living alone	15	15.0	18	23.4
	No response	2	2.0	1	1.3
Education level	Junior high school	5	5.0	6	7.8	n.s.
	High school	51	51.0	32	41.6
	College	14	14.0	10	13.0
	University graduate or higher	28	28.0	27	35.1
	No response/other	2	2.0	2	2.6
Smoking status	Smoker	46	46.0	42	54.5	n.s.
	Ex-smoker	18	18.0	10	13.0
	Non-smoker	35	35.0	25	32.5
	No response	1	1.0	0	0.0
Variables	Categories	Analysis Sample		Non-responders		χ2 test
		(N = 100)		(N = 77)
		n	%	n	%
Types of gambling caused problems*	Pachinko	59	59.0	58	75.3	x2(1) = 4.471 p < 0.05
	Pachi-slot	53	53.0	51	66.2	n.s.
	Horse Racing	35	35.0	27	35.1	n.s.
	Cycle Racing	13	13.0	10	13.0	n.s.
	Boat Racing	23	23.0	12	15.6	n.s.
	Auto Race	2	2.0	1	1.3	n.s.
	Lottery	4	4.0	2	2.6	n.s.
	Sports betting	3	3.0	0	0.0	n.s.
	Casino***	5	5.0	6	7.8	n.s.
	Mahjong	3	3.0	5	6.5	n.s.
	Baccarat	3	3.0	0	0.0	n.s.
	FX	8	8.0	6	7.8	n.s.
	Stock margin trading	4	4.0	3	3.9	n.s.
	Others	6	6.0	3	3.9	n.s.
Annual income**	Low (-$18,203)	10	10.0	8	10.4	n.s.
	Lower middle ($18,204-$45,508)	47	47.0	37	48.1
	Upper middle ($45,509-$72,813)	27	27.0	15	19.5
	High ($72,814-)	7	7.0	2	2.6
	No response	9	9.0	15	19.5
Debt amount**	<$9,101	12	12.0	5	6.5	n.s.
	$9,102-$45,508	38	38.0	39	50.6
	$45,509-$91,016	25	25.0	18	23.4
	$91,017-$273,050	18	18.0	10	13.0
	$273,051-$455,083	4	4.0	0	0.0
	No response	3	3.0	5	6.5
		Mean	(SD)	Mean	(SD)	T -test
Onset age of gambling (years)	first gambling	19.14	3.12	18.86	3.28	n.s.
	habitual gambling	20.50	3.78	20.72	4.0	n.s.
	problem gambling	28.28	6.78	26.36	5.7	t(171) = 1.984 p = 0.049
PGSI total scores at baseline		18.28	4.82	17.57	5.01	n.s.

Note: The total sample consisted of 100 male patients who sought treatment at specialized medical institutions for addiction.

*Multiple response options.

**Responses received in Japanese yen were converted to US dollars using the average exchange rate from January 2021 to January 2022 at the time the survey was conducted (1 US dollar = 109.87 yen).

*** Any casinos are currently illegal in Japan; we asked about patients' experiences playing at legal casinos overseas.

A comparison between the analysis sample (n = 100) and non-responders (n = 77) revealed significant differences in the following variables: Non-responders were more likely to have never been married (46.8% vs. 24.0%, χ²(2) = 10.309, p < 0.01) and had a higher proportion of pachinko as their primary gambling activity (75.3% vs. 59.0%, χ²(1) = 4.471, p < 0.05). Additionally, their problem-gambling onset age was slightly lower (mean age 26.4 ± 5.7 years) than that of the analyzed sample (mean age 28.3 ± 6.8 years, t(171) = 1.984, p = 0.049). However, no significant differences were observed in other demographic and gambling behavior variables (Table 1).

Temporal changes in GRCS and PGSI scores across follow-up periods of the total samples

Table 2 presents a summary of the descriptive statistics and repeated-measures ANOVAs for the PGSI total score and each GRCS subscale at each follow-up timepoint.

Table 2.

Temporal changes in GRCS and PGSI mean scores across timepoints for the total sample

	Baseline		6 months		12 months		Repeated-measures ANOVA
	Mean	(SD)	Mean	(SD)	Mean	(SD)	F(2, 198)	p	partial η2	Post-hoc
PGSI	18.28	4.83	14.36	7.25	10.44	8.09	46.36	<0.001	0.32	BL > 6 M > 12 M
GRCS: GE	13.32	5.55	9.38	4.97	9.44	5.71	32.28	<0.001	0.25	BL > 6 M, 12 M
GRCS: IC	10.05	5.42	6.87	3.81	6.17	3.53	36.97	<0.001	0.27	BL > 6 M, 12 M
GRCS: PC	19.11	7.51	13.58	7.22	12.42	7.03	45.43	<0.001	0.32	BL > 6 M, 12 M
GRCS: IS	19.73	7.68	14.16	8.54	14.15	9.18	31.86	<0.001	0.24	BL > 6 M, 12 M
GRCS: IB	16.29	6.28	11.07	6.19	10.26	6.48	45.30	<0.001	0.31	BL > 6 M, 12 M

PGSI: Problem Gambling Severity Index.

GRCS: Gambling-related Cognition Scale.

GRCS:GE; Gambling Expectancy, GRCS:IC; Illusion of Control, GRCS:PC; Predictive Control, GRCS:IS; Perceived Inability to Stop Gambling, GRCS:IB; Interpretative Bias.

The PGSI scores at the 6-month and 12-month follow-ups showed a significant positive correlation for all GRCS subscales (r = 0.22–0.57, ps < 0.05).

Subsequently, to examine the longitudinal changes in scores for each GRCS subscale at follow-up, a repeated-measures ANOVA was conducted with three timepoints (baseline, 6 months, and 12 months) as independent variables and each of the five GRCS subscales as dependent variables. The analysis revealed significant main effects of timepoints for all GRCS subscales (Fs(2, 198) = 31.86–45.43, ps < 0.001, partial η²s = 0.24–0.32). Subsequent multiple comparisons to examine the mean score differences between each timepoint showed that the scores at 6 and 12 months were significantly lower than those at baseline for all GRCS subscales (ps < 0.001). However, no significant differences were found between the scores at 6 months and 12 months (ps = 0.06–0.99). This indicates that the overall trend for treated patients was for cognitive distortions to improve significantly at 6 months and then plateau at 12 months.

For the PGSI mean scores, the repeated measures ANOVA results indicated a significant main effect of timepoint (F(2, 198) = 46.36, p < 0.001, partial η² = 0.32). Multiple comparisons revealed significant differences between every timepoint, indicating a linear decline in the PGSI scores. However, at 12 months, the mean PGSI score was 10.4, still above the cutoff score of 8, indicating that many participants still had some type of gambling problem one year later.

Temporal changes in GRCD: comparison between abstinence and non-abstinence groups

We analyzed temporal changes in the GRCS scores based on the participants' gambling behavior during follow-up. Throughout the one-year follow-up period, participants who reported no gambling were assigned to the abstinence group (n = 50), and those who gambled at least once to the non-abstinence group (n = 50).

First, we show the mean scores of PGSI for both groups at each follow-up point and the results from a repeated-measures two-way ANOVA in Fig. 3. A significant time × group interaction was observed for the PGSI scores (F(2, 196) = 12.56, p < 0.001, partial η² = 0.11). In the abstinence group, the PGSI scores significantly decreased during both follow-up periods (baseline to 6 months: p = 0.003, Cohen's d = 0.61; 6–12 months: p < 0.001, Cohen's d = 1.03). Meanwhile, the non-abstinence group showed a significant decrease from baseline to six months (p < 0.001, Cohen's d = 0.67). However, no further decrease was observed between 6 and 12 months.

Fig. 3.

Temporal changes in PGSI mean score over a one-year follow-up: Comparison between abstinence and non-abstinence groups

Note: The X-axis represents the survey time points, and the Y-axis indicates the PGSI mean scores for each group. The values in the figure represent the mean PGSI scores for each group at each time point.

Next, Fig. 4 shows the GRCS subscale scores at each time point. In both groups, all subscales decreased significantly from baseline to 6 months, but no changes were observed between 6 and 12 months (Fs(1, 236) = 0.72–2.85, n.s.). Each subscale showed a significant main effect for time (GE: F(2, 236) = 35.28, p < 0.001, partial η² = 0.25; IC: F(2, 236) = 27.26, p < 0.001, partial η² = 0.27; PC: F(2, 236) = 41.30, p < 0.001, partial η² = 0.32; IS: F(2, 236) = 35.54, p < 0.001, partial η² = 0.25; IB: F(2, 236) = 46.41, p < 0.001, partial η² = 0.32). Post hoc tests confirmed significant differences between baseline and 6 months and between baseline and 12 months, but not between 6 and 12 months (ps < 0.001). This trend was consistent across all subscales, regardless of gambling behavior during follow-up.

Fig. 4.

Changes in gambling-related cognitive distortions over a one-year follow-up: Comparison between abstinence and non-abstinence groups

Note: GRCS: Gambling-related Cognition Scale.

GRCS:GE; Gambling Expectancy, GRCS:IC; Illusion of Control, GRCS:PC; Predictive Control, GRCS: IS; Perceived Inability to Stop Gambling, GRCS:IB; Interpretative Bias.

The X-axis represents the survey time points, and the Y-axis indicates the mean scores for each subscale.

All subscales except IC exhibited a significant group effect, with the abstinence group consistently scoring lower than the non-abstinence group. No significant time- or group-related interactions were observed. However, in the non-abstinence group, GE, IS, and IB decreased from baseline to 6 months, and then showed a non-significant increasing trend at 12 months, suggesting that once-improved cognitive distortions may potentially worsen again over time. Significant group main effects were found for all subscales except IC: abstinence group scores were significantly lower than non-abstinence group scores (GE: F(1, 98) = 15.11, p < 0.001, partial η² = 0.13; IC: F(1, 98) = 1.73, p = 0.96, partial η² = 0.02; PC: F(1, 98) = 8.14, p = 0.03, partial η² = 0.08; IS: F(1, 98) = 38.40, p < 0.001, partial η² = 0.28; IB: F(1, 98) = 16.04, p < 0.001, partial η² = 0.14).

Causal relationship between GRCD and GD severity

We conducted separate three-wave CLPM analyses for each GRCS subscale (IB, IC, PC, GE, and IS) to examine their respective longitudinal relationships with PGSI scores. Only the model including the IS subscale (Perceived Inability to Stop Gambling) demonstrated statistically significant cross-lagged paths. The CLPM results for the other GRCS subscales (IB, IC, PC, and GE) are provided in supplementary Figs 1–4.

Figure 5 presents the results of the CLPM analysis of IS. The model fit was good, with GFI = 0.943 and CFI = 0.907, both exceeding 0.90. Standardized regression coefficients between PGSI and IS at each timepoint showed significant moderate covariance (BL: β = 0.354, p < 0.001; 6 m: β = 0.408, p < 0.001; 12 m: β = 0.414, p < 0.001).

Fig. 5.

Cross-lagged panel models examining the bidirectional association between “Gambling severity” as assessed by the PGSI and the GRCS subscale “Perceived inability to stop gambling” from baseline to 12-month follow-up

Note: GRCS: IS = perceived inability to stop gambling, a subscale of the Gambling-related Cognitions Scale.

PGSI: Problem Gambling Severity Index.

Bold lines represent paths with regression coefficients that are significant at the 5% level.

Dotted lines indicate non-significant path coefficients.

“e” indicates residual variance (error term) in the cross-lagged panel model, representing variance not explained by the predictors.

†p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001

Mondel fit indices; χ²[4] = 19.277, p < 0.001, GFI = 0.943, CFI = 0.907.

Regarding temporal stability, PGSI showed a significant association only between baseline and 6 months (p < 0.05), while IS demonstrated moderate to high associations across all timepoints (BL→6m: β = 0.427, p < 0.001; 6 m → 12 m: β = 0.698, p < 0.001). Finally, examining the cross-lagged paths, the path from baseline to 6 months was not significant, whereas the path from 6-month IS scores to 12-month PGSI scores was significant (β = 0.31, p < 0.01), suggesting that perceived inability to stop gambling at 6 months predicted GD severity at 12 months.

Discussion

Clinical characteristics and gambling behavior among patients treated for GD in Japan

In our sample, the onset of gambling problems was at age 28.2 ± 6.8 years, and the mean age at first diagnosis was 37.1 ± 8.5 years. This suggests that 70% developed gambling disorders by 30—slightly younger than the 40 s reported in other countries (Ledgerwood et al., 2020; Rossini-Dib et al., 2015; Schluter et al., 2019). Additionally, participants exhibited higher socioeconomic stability (e.g., education, marriage, and employment rates) relative to population-level estimates from national statistics (Statistics Bureau of Japan, 2021). However, these comparisons were not statistically tested and are presented solely to provide contextual reference in interpreting the results.

Moreover, most participants had gambling-related debt, with 20% owing over 10 million yen—nearly twice Japan's mean annual income. The most frequently reported gambling activities were pachinko (59%), pachi-slot (53%), and horse racing (35%), aligned with the national survey data (Nitta et al., 2023).

Non-response bias analysis showed that dropouts were more likely to be unmarried, pachinko gamblers, and have an earlier gambling onset than respondents. However, no significant differences were found in other key demographics or gambling behaviors, suggesting that attrition did not significantly distort the findings.

Temporal changes in GRCD

We examined longitudinal changes in GRCD among participants over a one-year follow-up period after initiating treatment, finding a significant reduction in the mean scores of all GRCS subscales from baseline to the 6-month follow-up, but no significant changes between the 6-month and 12-month follow-up points, and the scores remained stable. This suggests that cognitive impairment in GD patients tends to improve relatively early during the treatment process. In fact, at the 6-month follow-up, the mean scores of the three GRCS subscales—Gambling Expectancy (GE), Predictive Control (PC), and Illusion of Control (IC)—were comparable to those reported for the general adult male population in previous studies (Raylu & Oei, 2004). These results suggest that GRCD tend to improve during relatively early periods of treatment initiation, mirroring results of previous studies examining CBT's effectiveness (Chrétien et al., 2017; Dunsmuir et al., 2018; Fortune & Goodie, 2012). One potential explanation for this is the psychological and social changes that occur when seeking treatment at specialized medical institutions. Receiving a diagnosis, discussing accumulated gambling-related issues, and recognizing GD as a medical condition may all contribute to restructuring patients' cognition of gambling. However, this study did not explore the specific factors underlying these improvements, highlighting the need for further research to elucidate the mechanisms driving the reduction of cognitive distortion.

Comparison of the abstinence and non-abstinence groups in improving GRCD

When comparing the abstinence and non-abstinence groups, there were no significant interaction effects between time and group for any GRCS subscale. This indicates that, in both groups, cognitive distortions declined from baseline to 6 months and then plateaued after 12 months. Although no group × time interaction was found, significant main effects of group were observed for four subscales (GE, PC, IS, and IB), indicating that the abstinence group showed consistently lower cognitive distortions across timepoints. These results suggest that those who refrain from gambling for an extended period exhibit fewer cognitive distortions than those who continue to gamble. Such findings align with prior cross-sectional studies that have linked GRCD with GD prognosis (Ciccarelli, Griffiths, Nigro, & Cosenza, 2017; Nicholson et al., 2016).

Notably, IC (Illusion of Control) was the only subscale with no significant group effect. However, the subscale scores improved over time in both groups; at the 12-month follow-up, the mean scores (5.7 in the abstinence and 6.7 in the non-abstinence groups) were within the range reported for healthy adult males in the original GRCS development study (6.20 ± 3.98; Raylu & Oei, 2004). While no statistical comparisons were performed, we referred to these normative data from healthy populations to provide contextual reference in interpreting the scores. This suggests that, when patients receive treatment, they may realize that their belief in the ability to influence gambling outcomes through specific actions or rituals is unfounded, regardless of whether they continue to gamble. Furthermore, the abstinence group showed a gradual decline in GE, IS, and IB scores over time, whereas the non-abstinence group showed slight increases between 6 and 12 months. Although these increases were not statistically significant, they suggest that individuals who resume gambling after treatment may experience a partial resurgence of cognitive distortions that have improved.

Regarding IS (Perceived Inability to Stop Gambling), participants in the abstinence group had maintained complete gambling abstinence for 12 months. Nevertheless, the mean IS score in the abstinence group remained relatively high at the 12-month follow-up. Their positive responses to IS subscale items such as “I can't function without gambling,” “It is difficult to stop gambling as I am so out of control,” and “I will never be able to stop gambling” indicate that the perception of being unable to control their gambling behavior was resistant to change, even after a full year of abstinence. This suggests that the IS subscale may capture a particularly persistent aspect of distorted cognition, highlighting the need for targeted cognitive interventions as part of long-term treatment strategies.

As to longitudinal changes in gambling severity, a significant group × time interaction was found for PGSI scores. This suggests that the abstinence group steadily improved over time, whereas the non-abstinence group improved over the first 6 months but showed no further change between 6 and 12 months.

Overall, the trajectories of GRCS subscales and PGSI scores showed different longitudinal patterns between the abstinence and non-abstinence groups. While cognitive distortions tended to improve early after treatment initiation and then remained stable in both groups, gambling severity continued to decline throughout the follow-up period only in the abstinence group. These findings suggest that behavioral changes such as sustained abstinence may contribute to the reduction in gambling severity. However, although there were qualitative differences in the trajectories of cognitive distortions, such as a slight rebound in the non-abstinent group, no clear differences were observed throughout the one-year follow-up period. Longer-term follow-up studies are needed to examine the relationship between abstinence and improvements in gambling-related cognitive distortions among treatment-seeking patients.

GRCD predict future GD severity

Path analysis using a CLPM revealed that only IS (perceived inability to stop gambling) demonstrated a predictive relationship with gambling severity. IS at the 6-month follow-up significantly predicted PGSI scores at 12 months, whereas baseline IS did not predict PGSI scores at 6 months. One possible explanation for the lack of a significant association in the early phase is that this period may be strongly influenced by external factors—such as clinical interventions and social support—that were not included in our model. These unmeasured variables may have temporarily attenuated the direct influence of cognitive distortions on gambling severity during the initial stages of recovery. These results suggest that perceived inability to stop gambling, particularly when it persists beyond the initial treatment phase, may play a key role in the maintenance of gambling problems. Therefore, increasing patients' sense of control over their gambling behavior may help reduce the risk of future gambling severity.

Beyond the temporal pattern, among the five GRCD dimensions, only IS demonstrated a clear predictive relationship with gambling severity, while the other four dimensions—GE, IC, PC, and IB—did not. One possible explanation is that the IS may be qualitatively different from the other subscales in the types of cognitive distortions it captures. For example, Interpretation Bias (IB), Illusion of Control (IC), and Predictive Control (PC) focus on distorted beliefs regarding the outcomes of gambling, whereas Gambling Expectancies (GE) reflect beliefs about the anticipated effects or rewards of gambling. In contrast, IS encompasses more self-referential beliefs related to one's perceived control or self-efficacy in refraining from gambling. These distinctions may explain why IS emerged as the only significant predictor of subsequent gambling severity in this study.

Moreover, our CLPM was explicitly designed to examine the potential bidirectional relationship between gambling severity and gambling-related cognitive distortions. However, the PGSI-to-IS paths were statistically non-significant, with large p-values (baseline to 6 months: p = 0.519; 6–12 months: p = 0.787) (Fig. 5). These results suggest that gambling severity had minimal influence on subsequent perceptions of inability to stop gambling. Nevertheless, the possibility of confounding factors not measured in this study cannot be ruled out. For example, subconscious cognitive processes such as cue reactivity, spontaneous or intrusive thoughts about gambling, deficits in inhibitory control, or trait impulsivity could plausibly affect both gambling severity and the perceived (or actual) inability to stop gambling. Future research incorporating such variables may offer a more nuanced understanding of whether and how gambling severity contributes to cognitive distortions over time.

In summary, although unmeasured factors may warrant consideration in future research, improving IS early in treatment may promote subsequent improvement in severity. A previous study by Guillou-Landreat et al. (2016) investigated suicide risk and cognitive distortions in patients seeking treatment for GD and showed that major depression, anxiety, and IS influenced suicide risk, highlighting the importance of assessing IS during consultations. Another study showed that, among the pre-treatment GRCS scores of patients who received residential treatment, IS and IB were associated with the number of DSM-5 diagnostic criteria (Ledgerwood et al., 2020). Our findings are consistent with previous evidence showing that IS is a key predictor of GRCD, supporting the importance of focusing on IS in the care of GD patients. Furthermore, considering that the IS scores of the abstinence group at the 12-month follow-up in this study were not sufficiently improved, it is desirable to develop treatments that act on cognitive distortions, such as treatments that focus on improving IS.

Although our findings are based on Japanese patients in a gambling environment dominated by EGMs and horse racing, these forms of gambling are widespread worldwide and may, therefore, be applicable to individuals from other cultures and ethnicities with similar gambling habits. EGMs are widely recognized as significant risk factors for problem gambling (Allami et al., 2021; Moreira, Azeredo, & Dias, 2023; Orlowski et al., 2020), and offer important implications for researchers and clinicians working to prevent and treat GD.

Limitations

This study has some limitations. The CLPM used captures temporal associations between variables; however, it has notable limitations, including a failure to separate within-person changes from stable between-person differences and an inability to control for potential unmeasured confounders influencing both predictor and outcome variables. These methodological issues have also been observed in addiction research (Littlefield et al., 2022). Thus, causal interpretations should be made cautiously, as prospective effects may have been overestimated. Future research would benefit from applying models that better account for within-person variability and potential confounding factors.

Another limitation of this study was lack of data due to participant dropout. However, we confirmed, through analysis, that the non-respondents’ characteristics were not significantly different from those of the respondents. However, all potential biases cannot be avoided, and the impact of GRCD may have been underestimated. Future studies should aim to strengthen the efforts to retain participants. Additionally, only men were included in the analysis; therefore, it is difficult to directly apply the results to women. In future studies, measures such as oversampling should be considered because the prevalence of GD in women is nine times lower than in men. Finally, the participants' gambling involvement during the follow-up period was based on self-reporting; therefore, the existence of a self-reporting bias could not be denied, and the number of people in the non-abstinence group may have been underestimated.

Conclusions

Despite these limitations, this study is the first longitudinal investigation in Japan to examine GRCD in treatment-seeking GD patients. Our findings show that cognitive distortions significantly improved within six months, with IS emerging as a key predictor of future gambling severity. Notably, IS remained resistant to change even in abstinent individuals, highlighting its role in GD persistence. These findings offer new insights into GRCD and emphasize the need for targeted therapeutic interventions.

Supplementary data

jba-14-1267-s001.pdf ^{(355.5KB, pdf)}

Funding Statement

Funding sources: This research is funded by the Ministry of Health, Labour and Welfare Scientific Research Grant, project number 19GC1016.