Population

The UK Biobank (https://www.ukbiobank.ac.uk/) is a cohort study involving over 500 000 participants aged 40–69 who were recruited between 2006 and 2010. At recruitment, all participants were invited to 22 assessment centres in the UK, where they provided written informed consent and completed a touchscreen questionnaire, verbal interview, physical examination and biological sample collection. During 2012–2013, approximately 20 000 of these participants underwent the first repeated assessment. During 2014–2019, a subset of participants participated in the ongoing second and third repeat assessments, which included brain magnetic resonance imaging (MRI) examination. Image data capture and processing have been described in detail elsewhere (https://biobank.ndph.ox.ac.uk/).

The present study was a cohort study based on the UK Biobank, including all participants with available information on hearing ability and tinnitus at recruitment (N = 143 804) (Fig. 1). As we were primarily interested in new-onset depression and anxiety, we excluded participants with a previous diagnosis of depression or anxiety at recruitment, leaving 129 610 in the final cohort. We then followed these participants from the date of recruitment until the first diagnosis of depression (or anxiety), loss to follow-up, death or 30 September 2021, whichever came first. Participants who withdrew their informed consent were excluded (opt-out list dated 22 May 2022).

Fig. 1

Flow chart of study population selection. DTT, digit triplet test.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human subjects were approved by the NHS National Research Ethics Service (reference number: 16/NW/0274). We applied and obtained access to the UK Biobank data through application number 76 517. This study was also approved by the Swedish Ethical Review Authority (DNR: 2022-01516-01).

Definitions of hearing loss and tinnitus

Hearing ability was evaluated by the digit triplet test (DTT) (Supplementary Methods available at https://doi.org/10.1192/bjo.2023.634), which is used to estimate hearing ability in noise.^{Reference Van den Borre, Denys, van Wieringen and Wouters13} Based on the speech reception threshold (SRT), hearing loss was categorised as ‘yes’ (SRT ≥ −5.5 dB) or ‘no’ (SRT < −5.5 dB).^{Reference Dawes, Fortnum, Moore, Emsley, Norman and Cruickshanks14} Tinnitus was determined according to the question ‘Do you get or have you had noises (such as ringing or buzzing) in your head or in one or both ears that lasts for more than five minutes at a time?’ in the touchscreen questionnaire at recruitment and classified as ‘yes’ or ‘no’. According to hearing loss and tinnitus status at recruitment, we divided the cohort participants into the following categories: ‘tinnitus’, ‘hearing loss’, ‘both’ and ‘none’. In the secondary analyses, we further divided hearing loss as ‘insufficient hearing’ (SRT −5.5 to −3.5 dB) or ‘poor hearing’ (SRT > −3.5 dB) based on the result for the better-performing ear. We also divided tinnitus as ‘not bothersome’ or ‘bothersome’ based on the answer to the question ‘How much do these noises worry, annoy or upset you when they are at their worst?’ We included 134 456 participants who responded to the tinnitus questionnaire in the secondary analysis of severity of tinnitus, and 148 584 participants who completed the DTT test in the secondary analysis of severity of hearing loss (Fig. 1).

Diagnoses of depression and anxiety

We ascertained diagnoses of depression and anxiety through health-related outcome data (Category 1712, https://biobank.ctsu.ox.ac.uk/ukb/label.cgi?id=1712) from the UK Biobank, according to ICD codes (Supplementary Table 1). Category 1712 includes complete in-patient hospital data since 1997 obtained through the Hospital Episode Statistics database in England, the Patient Database for Wales and the Scottish Mortality Record, as well as primary care data since 1990, covering about 45% of the UK Biobank participants, obtained through primary care computer system suppliers in England, Wales and Scotland.^15,16 It also includes information from death registers and conditions self-reported at the assessment centres. We used ICD-10 codes F32 and F33 to ascertain depression and ICD-10 codes F40 and F41 to ascertain anxiety. The diagnoses of depression and anxiety in these data sources have been partly validated against detailed clinical evaluation, demonstrating a positive predictive value of over 80%.^{Reference Burns, Rigby, Mamidanna, Bottle, Aylin and Ziprin17}

Brain MRI

Using the available brain MRI data in the second repeat assessment from three dedicated centres, we selected eight brain regions of interest (ROIs) according to the literature.^{Reference Landgrebe, Langguth, Rosengarth, Braun, Koch and Kleinjung9–Reference Schmidt, Zimmerman, Bido Medina, Carpenter-Thompson and Husain12,Reference Svennerholm, Bostrom and Jungbjer18–Reference Cheng, Xu, Zhou, Qu, Li and He21} Subsequently, we studied three subcortical volumes (accumbens, amygdala and hippocampus), as well as five regional grey matter volumes (central opercular cortex, cingulate gyrus, hippocampus, inferior temporal gyrus and parahippocampal gyrus). To assess the specificity of the results, we also studied the whole intracranial volume and whole grey matter volume, as well as the volume of four brain regions with no relevance to hearing – the IX cerebellum (vermis), sensory cortex (postcentral gyrus), primary motor cortex (precentral gyrus) and vision cortex (occipital pole). We analysed the brain volumes in the left and right hemispheres separately as well as by subregion when relevant.

All data on brain volumes were obtained from T1-weighted MRI images by voxel-based morphometry, including FSL FAST for cortical structures (Category 1101), FSL FIRST for subcortical structures (Category 1102) and SIENAX analysis for total brain volume as previously shown.^{Reference Alfaro-Almagro, Jenkinson, Bangerter, Andersson, Griffanti and Douaud22} To reduce the possible impact of misclassification, in the analysis of brain volumes, we excluded participants who reported a different status of hearing loss and tinnitus between recruitment and second repeat assessment, leaving 4980 participants in the final analysis of brain volumes (43.2%).

Covariates

Information on demographic and lifestyle factors including birth year, sex, education, annual household income and insomnia status was collected through a touchscreen questionnaire. Townsend deprivation score was calculated based on residential postcode at recruitment and used as a proxy of socioeconomic status. Information on psychiatric disorders (other than depression and anxiety) before recruitment was collected from health-related outcome data (Category 1712), according to ICD codes (Supplementary Table 1).

Statistical analysis

We first described the baseline characteristics of the study participants, by hearing loss or tinnitus status. We used analysis of variance and chi-squared tests to determine the statistical significance of the differences among all four groups. We then used a univariable Cox model to assess the associations of these baseline characteristics, and of hearing loss and tinnitus, with the risk of depression or anxiety.

In the primary analysis, we used a multivariable Cox model to investigate the associations of hearing loss or tinnitus with the risk of depression and anxiety after adjusting for covariables shown to be statistically significantly associated with the risk of depression and anxiety in the univariable models; these included sex, age in years, educational attainment (college, tertiary, secondary or below secondary), income (>£100 000, £52 000–100 000, £31 000–51,999, £18 000–30 999 or <£18 000), insomnia status (usually, sometimes or never), Townsend index (by quartile distribution), history of other psychiatric disorders (yes or no) and assessment centre. To reduce the possibility of reverse causality, we used a 90 day lag time in all analyses. The proportional hazards assumption was examined using a test based on Schoenfeld residuals and found to hold. The results are presented as hazard ratios (HRs) with 95% confidence intervals. In addition to the Cox model, we used a flexible parametric model to explore changes in HR during the follow-up period, with the same adjustment and use of lag time as in the Cox model. To assess potential effect modification, we also performed stratified analyses by sex (male or female), previous psychiatric disorder (yes or no) and insomnia status (usually, sometimes or never). In the secondary analyses, we investigated the severity of hearing loss and tinnitus on the risk of depression and anxiety using the same models.

In the analysis of brain volumes, we first used a multivariable linear regression model to assess the associations of hearing loss, tinnitus or both with the different brain volumes, after adjustment of the same set of covariates as in the analyses of depression and anxiety. Second, to examine whether the associations for the selected ROIs were independent of whole brain volume, we further adjusted for the whole intracranial volume in the multivariable analyses of different ROIs.

Statistical analyses were conducted using R (version 4.0.2). Statistical significance was set at P < 0.05. All statistical tests were two-sided. Owing to multiple testing in the analyses of brain volume, the P-values in the multivariable linear regression were corrected by the Benjamini–Hochberg method. We considered a false discovery rate (FDR) of <0.05 to indicate statistical significance.