Alcohol related disorders among elite male football players in Sweden: nationwide cohort study

Data sources

Detailed information about the data sources and the methods used to construct the study cohort are provided in the supplementary material. Information about football players was collected from data sources of all current and former football players in the Swedish top tier of competition, Allsvenskan, and was compiled by the Swedish Association of Football Historians and Statisticians. These data comprised the full name, birth date, seasons and number of games played, number of goals scored, teams represented and player position (outfield or goalkeeper). In databases run by the Swedish Tax Agency and the Federation of Swedish Genealogical Societies, the name and birth date were used to search for the player’s personal identity number, a number given to all inhabitants in Sweden which enables linkage of individual level data across data sources (supplementary material and supplementary table 1). We used several nationwide health and administrative registers, including the Total Population Register (vital status, region of residence, region of birth), Patient Register (outcomes), Cause of Death Register (outcomes), Prescribed Drug Register (outcomes), Statistics Sweden (income, education), and Swedish Military Conscription Register (results from psychological and physical tests at military conscription).

Study population

A flowchart for the study population is shown in supplementary figure 1. All football players who had played at least one game in Allsvenskan from its first season in 1924 to 2019 were included. We excluded players whose personal identity number were not available in the databases (eg, because players had removed themselves from public databases, had died before the introduction of identity numbers, or were foreign players without an identity number). We also excluded players for whom a personal identity number could not be identified because of multiple individuals having the same name and birth date, and it was not possible to identify the football player among them. We excluded players who were not registered in the Total Population Register, which started in 1969.

Each football player was matched in a 1:10 ratio with men from the general population, based on year of birth, region of residence, and vital status, to form a base cohort. The matching variables had no missing data. The matching date was 1 January in the year when the football player played his first game in Allsvenskan or was first registered as a resident in Sweden (if this occurred after the first year of play in Allsvenskan), and matching was performed using the Total Population Register. For players who played their first season before 1969 (ie, the start of the Total Population Register), we selected men from the general population matched by vital status and region of residence on 1 January 1969. The date when players and men from the general population were matched constituted cohort entry.

From the base cohort, we excluded players (and matched men from the general population) and men from the general population who were not born in Sweden and who had their first date of residence in the country after age 15 years. We used this exclusion criterion because foreign players are likely to emigrate during or shortly after their football career, which would preclude long term follow-up.

Outcomes

The primary outcome was alcohol related disorders recorded in the Cause of Death Register (primary or contributing causes of death), National Patient Register (primary or secondary diagnosis during hospital or outpatient visits), or Prescribed Drug Register (filled prescriptions for drugs used to treat alcoholism). The secondary outcome was disorders related to misuse of other drugs. The ICD-8, ICD-9, ICD-10 (international classification of diseases, 8-10th revisions) codes and Anatomical Therapeutic Chemical codes used for the definition of the outcomes are shown in supplementary table 2.

Statistical analysis

Football players and men from the general population were followed from cohort entry to emigration, outcome event, death, or end of follow-up (31 December 2020). Emigration, death that did not include an outcome event, and end of follow-up constituted censoring events. We used Cox regression analysis with age as the time scale and adjusted for region of residence, place of birth, and calendar time as a time varying covariate (supplementary table 3) to estimate the hazard ratio for the risk of the primary outcome among football players versus men from the general population.

We performed additional analyses. Rules and norms regarding alcohol consumption among football players have differed substantially over time in Sweden.1819 We therefore modelled the interaction between the year of the player’s first season in the top tier of competition using restricted cubic splines with 3 knots and a flexible parametric survival model20 to calculate the hazard ratio for football players versus matched men from the general population by year of the football player’s first season.

Some of the most high profile cases of alcohol addiction have occurred among players famous for their goal scoring abilities. Previous studies have shown associations between personality traits and alcohol consumption.21 Therefore, we hypothesised that factors potentially associated with scoring goals (eg, sensation seeking),21 might also be associated with the risk of alcohol related disorders. Further, football players who have played many games and seasons in the top tier could have been more exposed to drinking culture associated with elite football, fame, and public scrutiny. Another possibility is that football players with drinking problems could have terminated their careers early. We therefore hypothesised that the number of games and seasons in the top tier might be associated with alcohol related disorders.

In analyses restricted to football players, we used the Cox regression model additionally adjusted for birth cohort (categorical variable; supplementary table 3) to separately assess the association with alcohol related disorders for number of goals scored per 10 games (among outfield players who had played at least 10 games), the number of top tier games played, and the number of seasons played, modelled using cubic splines with 3 knots. We also assessed the risk of alcohol related disorders among outfield players versus goalkeepers.

Studies have indicated that elite athletes could be at increased risk of alcohol related disorders after retirement from their elite careers141522 owing to changes in life circumstances associated with retirement. We therefore assessed whether the risk of alcohol related disorders among football players versus men from the general population varied across age. As outcome data from death certificates and hospital admissions might poorly reflect the timing of the first diagnosis, we set the earliest cohort entry to 1 January 2002, one year after the start of nationwide coverage for outpatient visits in the National Patient Register. We excluded football players (and matched men from the general population) and men from the general population who had died, emigrated, or had any history of alcohol related disorders before cohort entry. We calculated the hazard ratio of the risk of alcohol related disorders for football players versus men from the general population using the total follow-up time in this analysis, and in another analysis modelling age using restricted cubic splines with 3 knots.

We performed a sensitivity analysis. The earliest possible cohort entry was in 1969. If players and men from the general population had different risks of death with alcohol related disorders, the exclusion of those who died before 1969 could introduce left truncation bias owing to a depletion of individuals susceptible to alcohol related disorders.23 We therefore conducted an analysis including only those players and those men from the general population who were 49 years or younger at cohort entry.

We performed exploratory analyses assessing the impact of accounting for some characteristics that might differ between elite football players and men from the general population. Some men from the general population could have had severe health or social problems that precluded participation in the workforce and increased the risk of alcohol related disorders. Therefore, we performed two analyses that excluded men from the general population with zero registered income at cohort entry, and at age 40-44 years (supplementary table 4). For these analyses, we excluded men from the general population without available income data and football players with no matched men from the general population for whom income data was available. We then performed the analysis with and without exclusion of men from the general population with zero registered income.

Next, we used data from the Military Conscription Register. Firstly, we performed the analysis including all football players and men from the general population born in 1951-87, because military conscription was mandatory at around age 18 years for men born during this period.24 Individuals living abroad, those with certain health conditions, and those with functional impairments were exempt from conscription or tests (supplementary material (data sources) and supplementary table 4), so we performed the analyses excluding football players and men from the general population born 1951-87 without complete data on test results, including stress reliance, cognitive ability, body mass index, muscle strength, and cardiorespiratory exercise capacity (supplementary table 4). We then performed analyses adjusting for stress resilience and cognitive ability scores.252627 Finally, we further adjusted the model for body mass index, muscle strength, and cardiorespiratory exercise capacity.

The secondary outcome of disorders related to other drug misuse was also assessed. ICD-8 codes, which were used before 1987, did not contain diagnoses suitable for capturing the secondary outcome in death certificates, so we set the earliest date of cohort entry to 1 January 1987, when nationwide coverage in the National Patient Register for inpatient diagnoses began. In these analyses, we excluded players who died or emigrated before 1987 and their matched men from the general population, and men from the general population who died or emigrated before 1987. We also excluded players and men from the general population aged 65 years or older in 1987 to avoid potential left truncation bias.23

Analyses were conducted using SAS software (version 9.4) and Stata (version 16.1). Hazard ratios whose confidence intervals did not overlap 1.0 were considered as significant.

Patient and public involvement

No patients were involved in setting the research question, nor in the design, conduct, or interpretation of the study. The study is based on anonymised nationwide register data and there is no planned dissemination of results directly to study participants. Active elite football players were not involved in the study design and planning because it was not deemed crucial for conducting the study and because of resource (time and funding) constraints.