Gambling has become ubiquitous, with almost half (46.2%; 95% CI 41.7–50.8) of adults globally reporting gambling in the past year. ^{Reference Tran, Wardle, College-Frisby, Taylor, Lynch and Rehm1} Gambling harm caused by problems related to gambling is a major public health challenge. ^{Reference Wardle, Degenhardt, Marionneau, Reith, Livingstone and Sparrow2} Problematic patterns of gambling, which are those that cause multiple problems for the individual in personal, family, financial and employment domains, are estimated to have a global prevalence of 1.4% (95% CI 1.06–1.84). ^{Reference Tran, Wardle, College-Frisby, Taylor, Lynch and Rehm1} A further 8.7% (95% CI 6.6–11.3) of adults engage in ‘any risk’ gambling, which includes individuals who experience at least one adverse consequence from gambling. ^{Reference Tran, Wardle, College-Frisby, Taylor, Lynch and Rehm1} Persistent or recurrent patterns of gambling form part of the diagnostic criteria for gambling disorder within both the DSM-5 and the ICD-11. Globally, up to 2.2% (0.9–3.9) of adults experience problematic gambling or gambling disorder; this includes up to 54.7 (23.403–99.583) million men and 25.3 (11.483–44.607) million women. As these estimates illustrate, the scale and extent of gambling harm show no signs of abating, and neither do the increasing costs of clinical treatment and support services.

Gambling disorder is associated with increased risks of suicide, mortality and comorbidity. ^{Reference Karlsson and Håkansson3–Reference Wardle and McManus6} In a meta-analytic literature review, Kristensen et al ^{Reference Kristensen, Pallesen, Bauer, Leino, Griffiths and Erevik4} identified pooled prevalence rates of 31.6% (95% CI 29.1–34.3) for lifetime suicidal ideation and 13.2% (95% CI 11.3–15.5) for lifetime suicide attempts, respectively, among individuals with gambling problems. To date, however, few studies have sought to investigate the association between gambling disorder and suicide mortality. ^{Reference Boering, Jones, Patel, Leightley and Dymond7} Using a whole-population cohort design with routinely collected data from Norway, Kristensen et al ^{Reference Kristensen, Baravelli, Leino, Pallesen, Griffiths and Erevik8} found that suicide was the leading cause of death among individuals with gambling disorder between 2008 and 2021. That is, among 6899 people with gambling disorder, 37 of 148 deaths (25%) were by suicide. Moreover, suicide risk did not differ from that seen with people with comorbid conditions such as anxiety disorders, depression and personality disorders. These findings indicate that gambling disorder is associated with increased suicide risk in a potentially separable manner from that involving comorbid disorders. ^{Reference Karlsson and Håkansson3} Other studies linked patient registries in Sweden and Canada, respectively, with national cause of death data sources and found increased risk of suicide among individuals recorded as ‘gambling’ cases. ^{Reference Karlsson and Håkansson3,Reference Kristensen, Baravelli, Leino, Pallesen, Griffiths and Erevik8,Reference Reccord, Power, Hatfield, Karaivanov, Mulay and Wilson9} To infer gambling caseness, two studies analysed the records of individuals with a diagnosis of ‘gambling disorder’ defined as ‘pathological gambling’ (ICD-10: F63.0) using Read codes from ICD-10, whereas another defined gambling as ‘a history of gambling’. ^{Reference Reccord, Power, Hatfield, Karaivanov, Mulay and Wilson9} Previous linked registry-based studies ^{Reference Karlsson and Håkansson3,Reference Kristensen, Baravelli, Leino, Pallesen, Griffiths and Erevik8} have found suicide to be the leading cause of mortality among people with a gambling disorder diagnosis and have demonstrated increased suicide mortality rates (SMRs) for people with a diagnosis compared with the general population in Sweden (SMR 5.1; 95% CI 3.71–7.06) and Norway (SMR 15.1; 95% CI 8.7–21.6), respectively. Reccord et al ^{Reference Reccord, Power, Hatfield, Karaivanov, Mulay and Wilson9} assessed differences in suicide rates between rural and urban areas in Canada and found a significant association between gambling and suicide (P = 0.007). To date, these are the only existing data linkage studies of gambling-related suicide using routinely collected data. ^{Reference Boering, Jones, Patel, Leightley and Dymond7}

Healthcare contacts recorded through routinely collected data present opportunities for early intervention and treatment, as well as potential data sources for further research. ^{Reference Chadd, Caute, Pettican and Enderby10} Routinely collected data – that is, data gathered from administrative and clinical records – may provide both increased statistical power owing to large sample sizes and insight into the pragmatic nature of healthcare contact and treatment. ^{Reference Harron11} Linkage of a range of data-sets from services such as general practice, out-patient appointments, hospital admissions and annual deaths can provide a comprehensive understanding of how individuals experiencing gambling harm access healthcare before their death. Individuals who die by suicide are more likely to access health services, including general practice and emergency departments, in the year before their death. ^{Reference Pearson, Saini, Da Cruz, Miles, While and Swinson12,Reference Da Cruz, Pearson, Saini, Miles, While and Swinson13} Evidence from Wales (in the UK) has shown that 98.3% of those who died by suicide had been in contact with healthcare services in the year before their death, and healthcare contacts were higher in the week before their death. Individuals who died by suicide were also more likely to attend emergency departments rather than primary care or out-patient services. ^{Reference DelPozo-Banos, Rodway, Lee, Rouquette, Ibrahim and Lloyd14}

Globally, help-seeking rates for gambling are low, and many people experiencing harm from gambling never seek treatment. ^{Reference Bijker, Booth, Merkouris, Dowling and Rodda15} In the UK, individuals experiencing problematic gambling may be more likely to consult their general practitioner for mental health concerns and to be admitted to hospital and receive psychological counselling compared with those who do not gamble. ^{Reference Cowlishaw and Kessler16} A Canadian study found that individuals with a history of problem gambling who subsequently died by suicide had lower rates of accessing front-line health and social services in the year leading up to their death and throughout their lives. ^{Reference Séguin, Boyer, Lesage, McGirr, Suissa and Tousignant17} High rates of comorbidity with mental disorders including depression, schizophrenia, generalised anxiety disorder, obsessive–compulsive disorder, panic disorder and personality disorders, as well as addictive behaviours such as alcohol use disorder, increase the range of potential mental health treatment-seeking opportunities experienced by people with gambling disorder.

To date, no study has examined the association between gambling diagnoses and mental health treatment utilisation in the months preceding death by suicide. Here, we conducted a study of healthcare contacts, gambling, suicide and mental health disorders using population-based linked data at the person level, with a case–control design, across a study period of 30 years. Our primary aim was to assess for the first time the degree to which gambling predicts suicidal death using routinely collected healthcare data. Our secondary aim was to measure the degree to which gambling diagnoses predict mental health treatment utilisation.

Method

Study design

We conducted a case–control study to compare individuals who died by suicide and had a past diagnosis of gambling in Wales between 1 January 1993 and 31 December 2023 with those who did not die by suicide (Fig. 1). A random sample of 3000 all-cause decedents were selected as comparators (using the SQL ‘rand’ function to identify a random sample). This comparator sample size was deemed optimal for clinical prediction models. ^{Reference Christodoulou, van Smeden, Edlinger, Timmerman, Wanitschek and Steyerberg18–Reference Lyons, Jones, John, Brooks, Verplancke and Ford20} Gambling was defined with either pathological/compulsive gambling or betting (primary care data Read codes E2C31, Xa2ke, Eu63000-1, ZV4k500) and gambling and betting or pathological gambling (hospital admission ICD-10 codes Z72.6 and F63.0; Table 1). Suicidal death was defined as fatal intentional self-harm in accordance with the Office for National Statistics (ICD-10 codes X60-X84; see Supplementary Materials available at https://doi.org/10.1192/bjo.2025.10867). ^{Reference John21–Reference Thomas, Davies, Metcalfe, Windmeijer, Martin and Gunnell23}

Table 1 Inclusion criteria for cases

Φ is the count suppressed due to low cell size following statistical reporting requirements.

Data sources

Gambling diagnoses, suicide deaths and health contacts were obtained from the Welsh Longitudinal General Practice Dataset, the Annual District Death Extract), the Patient Episode Database for Wales, and the Outpatient Dataset for Wales (Table 2), all housed within the Secure Anonymised Information Linkage (SAIL) Databank. ^{Reference Lyons, Jones, John, Brooks, Verplancke and Ford20,Reference Jones, Ford, Jones, Dsilva, Thompson and Brooks24} SAIL is a privacy-protecting Trusted Research Environment holding anonymised data from the whole population of Wales. We undertook probabilistic linkage of health and mortality records using a unique identifier (anonymised linkage field) generated by SAIL.

Table 2 Included data-sets linked in SAIL, variables and coding framework

Case and comparator definitions

We defined gambling cases as those with either primary care data Read codes or hospital admission ICD-10 codes for gambling, with death recorded as suicide and an index date of death between 1993 and 2023. Within SAIL, we age- and sex-matched each gambling case to five all-cause decedents who had also had contact with primary care, emergency departments, hospitals and out-patient services in Wales since 1993.

Measures and variables

Demographics

We extracted sex and age at death from the Annual District Death Extract data source (Table 2).

Diagnostic variables

We extracted mental health diagnostic variables assigned for cases and comparators based on disorders known to be highly comorbid with gambling, such as alcohol use disorder and depression, ^{Reference Vaughan and Flack25,Reference Grant, Kushner and Kim26} and other disorders including generalised anxiety disorder, obsessive–compulsive disorder, personality disorders, schizophrenia, schizotypal and delusional disorders, post-traumatic stress disorder and panic disorder. Further details of the coding frameworks and data sources can be found in the Supplementary Material.

Healthcare contacts

A healthcare ‘contact’ was a recorded entry in healthcare records. Contacts included those recorded in primary care (e.g. general practice settings), hospital admissions with an episode with a psychiatric specialty or treatment (Read coded thesaurus of clinical terms versions 2 and 3), mental health primary diagnosis (ICD-10 codes ‘F’), and out-patient attendances with a psychiatric specialty or treatment.

Statistical methods

Individual record linkage and generation of data tables were conducted using Eclipse software version 2024-12 R for Windows (Eclipse Foundation, Belgium; https://eclipseide.org/) and SQL within SAIL. Descriptive statistics were produced before matching of cases to comparators. ^{Reference Rassen, Shelat, Myers, Glynn, Rothman and Schneeweiss27} Binary logistic regressions with predictor variables of any treatment contact related to gambling diagnosis (e.g. caseness), mental health diagnoses in in-patient (including day-cases), out-patient and primary care (i.e. general practitioner) settings were conducted. We conducted nested case–control analysis of regression and tests of association for cases (n = 92) and controls (n = 460) (Fig. 1). The backward elimination method of stepped regression was used to assess predictors of suicide, and all predictors which did not contribute to the model and exhibited high P-values were removed. The Hosmer–Lemeshow test was used to determine goodness of fit. Chi-squared tests of association were conducted, with mental health treatment contacts as the dependent variable (n = 552) and gambling, alcohol and mental health diagnoses as independent variables. The eta coefficient was used as an effect size estimate.

Fig. 1 Flow diagram of cases and controls.

Ethics statement

This study makes use of patient data held in the SAIL Databank; individuals’ informed consent was not necessary, as all data were anonymised. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees, and with the Helsinki Declaration of 1975, as revised in 2013. Ethical approval was obtained from SAIL’s Information Governance Review Panel (project number 1564).

Results

Study population

We identified a cohort of n = 92 gambling cases who died by suicide. After ten duplicate records had been removed from the all-cause decedents (n = 897 976), we identified a final group of n = 2990 comparators.

Individuals with a history of gambling disorder who died by suicide were 61.5 years old on average (s.d. 13.1), and the majority were male (71%). Almost half of cases had a depression diagnosis (45.65%) and 25% had an alcohol-related diagnosis. Most gambling diagnoses were made in the general practice setting (51.09%), and 26.09% were made in an in-patient setting (Table 3). Compared with the control group, individuals in the case cohort were younger (61.5 years (s.d. 13.1) for cases v. 79.56 years (s.d. 12.35) for controls) and had a higher prevalence of depression and alcohol use disorder the year before death (Table 3).

Table 3 Demographic characteristics, mental health (depression), alcohol use and gambling diagnoses in primary and secondary care settings and suicidal deaths among cases and comparators in the year before death

Most gambling-related diagnostic codes (47%) were found in primary care data rather than in-hospital data (24%). Crucially, cases and comparators differed significantly with respect to the incidence of suicidal death only.

Mental health predictors of suicidal death

Having a history of gambling disorder was a predictor of suicidal death (odds ratio 30.94 (95% CI 3.57–268.28), P = 0.002). In this model, alcohol use disorders, depression, and schizophrenia, schizotypal and delusional disorders were not significant predictors of suicide (Table 4). The Hosmer–Lemeshow test returned a greater P-value, indicating improved goodness of fit (χ ² = 1.48, d.f. = 8, P = 0.83), and the omnibus test was significant at χ ² = 22.16, d.f. = 4 and P ≤ 0.001.

Table 4 Mental health predictors of suicidal death (n = 552)

Mental health treatment contact

Mental health treatment utilisation is summarised in Table 5. We found that cases had more frequent contact with mental health treatment services overall (P < 0.001) than comparators, particularly in-patient contacts (P < 0.001); there was no significant difference for out-patient contacts.Footnote ^a As shown in Table 6, only individuals with a gambling diagnosis were significantly likely to have had mental health treatment contacts before their death (P = 0.005).

Table 5 Mental health treatment utilisation (contact) characteristics

Table 6 Chi-squared tests of association (n = 552) between mental health treatment contacts