Exposure variables
Hearing loss
The HUNT hearing study included pure-tone audiometry, otoscopy and a comprehensive questionnaire. Pure-tone air conduction (HTL) hearing threshold levels were measured in accordance with ISO 8253–1 [9], with fixed frequencies at the eight test frequencies 0.25, 0.5, 1, 2, 3, 4, 6 and 8 kHz, according to an automatic procedure (“press the button when you hear a sound”). Masking was not applied and bone conduction thresholds were not assessed. The elderly or unable to follow the automatic procedure benefited from manual audiometry. The audiometry procedure has been described in detail previously [4]. We used the average of the hearing thresholds measured at frequencies 0.5, 1, 2 and 4 kHz in the better hearing ear with the definition of the global burden of disease (GBD) of hearing loss to construct a categorical variable with hearing normal (hearing threshold 20 dB) as exposed.
Tinnitus
We constructed a dichotomous variable to compare participants with hearing loss and tinnitus (exposed) to normal hearing and no tinnitus as the reference category. For the construction of this variable, we used the following question from the HUNT2 questionnaire: “Have you experienced ringing in the ears/tinnitus during the last 12 months?” “. Participants with missing values were excluded from this particular analysis.
Result variables
Sick leave and invalidity pension
We obtained annual data on doctor-certified sick leave (episodes longer than 16 days) and disability pension from 1996 to 2016 from Statistics Norway. Based on the personal identification number assigned to all Norwegian citizens, data from Statistics Norway was linked at the individual level with data from the HUNT survey. ID numbers were removed before researchers had access to matched data. We coupled the HUNT data with individual records covering episodes of sick leave between 1998 and 2016.
Modifiers or confounders of potential effects
age and sex
In all analyzes we adjusted for age and sex.
Education
Data from Statistics Norway on education was used to construct this variable. The level of education was divided into 4 groups: primary education, secondary education, university
A dummy variable was created by combining primary school and secondary school to give the “lower education” group and university NOT= 13). We adjusted for education in all analyses.
White collar/blue collar
Occupational codes from Statistics Norway were available from 1990 and 1980. We used NYK codes (Nordisk yrkesklassifisering; “Nordic Occupational Classification”), based on the International Standard Classification of Occupations, ISCO-58. [10]. At the single-digit level, occupations are divided into 12 major groups: 0 = Technical work, physical sciences, humanities and arts; 1 = Administrative, executive and managerial work; 2 = Office work; 3 = Sales work; 4 = Agriculture, forestry and fishing work; 5 = Mining works and quarries, etc. ; 6 = Transport and communication works; 7 and 8 = Manufacturing and construction work; 9 = maintenance work; A = military work; X = Occupation not declared. We categorized occupational codes 0 through 3 as “mostly white collar” and codes 4 through 9 and A as “mostly blue collar.” People who were not registered with a professional code (not working or lack of registrations/missing data) and people with professional code X (undeclared profession), were excluded from this specific analysis (NOT= 6495).
statistical analyzes
We used STATA version 17.0. Statistical tests were calculated at a 95% confidence interval.
Hearing loss and sick leave
We assessed the association between hearing loss in 1996-1998 and time to first episode of medically certified absence from work (1996-2011) using Cox proportional hazards regression. We chose to end the monitoring period in 2011 because a Norwegian pension reform was introduced that year. It allowed people over the age of 62 to combine work and retirement, which would have made the period before and after 2011 non-comparable in terms of censorship at the time of retirement. We analyzed the time in calendar days until the incident of sick leave during the follow-up period, starting from the date of the hearing test in 1996-1998 (baseline), ending at the end of 2011.
We used Statistics Norway (SSB) data on employment status to construct censorship categories. SSB data on employment status was categorized into 5 groups: employed, self-employed, unemployed, out of the labor market (retired, disabled, student, homemaker) or in labor market programs. Labor market programs are part of the Norwegian social protection system and aim to improve the chances of finding a job, by offering job search measures, work experience and vocational training.
Participants were censored at the year they left (unemployed, out of the labor force, or in labor market programs), at the date of death, or at the end of the follow-up period with an average duration of follow-up of 6.7 years and a maximum of 15.0 years. This equates to a total follow-up time of 145,723 person-years.
Hearing loss and invalidity pension
We assessed the association between hearing loss in 1996-1998 and time to first disability pension (1996-2011) using the same method as for sick leave. Participants were censored at the year of retirement from work, the year of death, or the end of the follow-up period with an average follow-up time of 14.1 years and a maximum of 15 years and a time total follow-up of 304,000 person-years.
Finally, we performed subgroup analyzes investigating the risk of certified medical absence from work and the risk of receiving a disability pension when the exposed group was defined as people with both hearing loss and tinnitus.
For the analyzes above, we adjusted for age (using a continuous age variable), gender, and education. We chose not to adjust for cardiovascular risk factors or smoking because although many cardiovascular risk factors were associated with hearing loss, the effects were small. [11]. We also assessed whether the associations between hearing loss and sick leave or disability pension were modified by age, gender, education or occupational category by performing stratified analyzes and testing interaction terms with hearing loss variable as a continuous variable (hearing loss *age, hearing loss *gender, hearing loss *education, hearing loss *occupational category). We only performed an interaction analysis between hearing loss and the mentioned variables, not for the group with hearing loss and tinnitus, due to the way this variable was constructed. The post-estimate proportional hazards test was used to test proportional hazards. The proportional hazards assumption was met in all models.
