1.1 Background

Cholesteatoma is an invasive, non-cancerous cyst of the middle ear whose expansion and destruction of surrounding bone results in progressive hearing loss, discharge and potentially serious intracranial complications including meningitis and abscess [1]. The only known treatment is surgical excision, which itself can contribute to hearing loss. Cholesteatoma incidence varies between populations but is reported up to 9.81 per 100 000 adults per year in Finland [2]. Not much is known about why cholesteatoma forms but several risk factors have been identified, including male sex [1], family history [3, 4] and certain developmental disorders affecting cranial morphology, including Turner syndrome and cleft palate [5]. While these risk factors are well known, associations with deprivation [2, 6, 7] and smoking [8] have occasionally been reported but require further investigation. Meanwhile, increased incidence in white over non-white populations is often referenced [1, 9-11], but original epidemiological data supporting this is scant. Furthermore, cholesteatoma incidence has been shown to differ between populations and over time.

The UK BioBank (UKBB) is a database and large research resource that includes in depth health and genetic data from a cohort of ~500 000 participants recruited from the UK between 2006 and 2010. Hospital inpatient data is available for the whole cohort, with primary care data available for less than half of the participants [12]. Data about health and lifestyle were collected by questionnaire at recruitment and UKBB use algorithms to define health outcomes from self-reported health information, hospital inpatient data and death data.

Greater understanding of cholesteatoma, its risk factors, and relationships to other middle ear disease are needed to inform study design for the discovery of preventative measures or non-surgical treatments.

1.2 Aims and Objectives

This study aims to use retrospective data about UKBB participants to characterise lifetime prevalence and demographics related to cholesteatoma amongst a cohort aged > 52 years. We also compare cholesteatoma directly to non-cholesteatoma middle ear disease (NC-MED) to determine whether risk factors and associations are shared or unique. In addition, we aimed to compare these results to publicly available statistics from an equivalent Finnish Biobank, FinnGen (https://www.finngen.fi/en). This will be useful for other researchers considering the use of this biobank for the study of cholesteatoma and related conditions.

2.1 Study Design and Setting

This is a retrospective case–control study of data from UKBB participants (application ID 61632) who were drawn from the UK population between 2006 and 2010 aged 40–69 at recruitment. All participants provided informed consent through electronic signature at their baseline assessment.

UKBB has obtained Research Tissue Bank (RTB) approval from its ethics committee that covers our use of downloaded data. The Research Ethics Committee (REC) approval number for UKBB is 16/NW/0274. Researchers can acquire UKBB data by registering for access: (https://www.ukbiobank.ac.uk/enable-your-research/register). Publicly available demographic data and statistical results from FinnGen release 9 were accessed via Risteys (https://r9.risteys.finngen.fi/endpoints/H8_CHOLEASTOMA) in June 2023.

2.2 Participants

From the UKBB cohort, we identified individuals as having (or very likely having had) cholesteatoma. In this study, we use UKBB demographic data, including ethnicity, sex, birth year and deprivation and clinical classification codes from health records, to characterise cholesteatoma prevalence and associated diseases amongst this cohort.

2.3 Variables

2.4 Study Size

The study size was determined by the availability of eligible data within the Biobank and FinnGen databases.

2.5 Statistical Analyses

2.5.3 Cox Hazard Regression for Comparison to FinnGen

The FinnGen comparison data include Cox hazard ratios for ICD-10 codes before and after cholesteatoma. To compare the UKBB data to FinnGen, we performed Cox hazard regression on the set of codes with significant hazard ratios in FinnGen. UKBB cholesteatoma cases with time to event information (n = 650) were compared to ear disease-free controls (n = 496 667) as in Risteys. Age and sex were used as covariates.

Figure 1 shows handling of missing data.

3.1 Descriptive Data

3.2 Main Results

Significant associations with cholesteatoma incidence were found for sex (male AOR = 1.33, p < 0.001, 95%CI = [1.179, 1.491]), deprivation (AOR = 1.08, p < 0.001, 95%CI = [1.059–1.097]), age (AOR = 1.02, p < 0.001, 95%CI = [1.011, 1.026]), Black ethnicity (AOR = 0.35, p = 0.0035, 95%CI = [0.175, 0.71]) and other/unknown ethnicity (AOR 0.48, p = 0.042, 95%CI = [0.241, 0.973]) (Table 1, Figure 2d). The ORs obtained in sensitivity analysis for demographic factors generally agreed with the AORs obtained from the unmatched data except for the other/unknown ethnicity, showing that imbalance did not greatly affect the results (Table S3).

Comparing cholesteatoma to NC-MED shows a similar male bias with an AOR of 1.30 (p < 0.001, 95%CI = [1.142, 1.486]), meaning the cholesteatoma group differs from NC-MED about as much as it differs from the controls (AOR = 1.33; Cases vs. control; Figure 2b). The cholesteatoma and NC-MED ear disease cohorts do not differ significantly in age or smoking status (p > 0.05) (Figure 2a,c). Deprivation was significantly associated with cholesteatoma but to a lesser extent than when compared to healthy ears (AOR 1.02, p = 0.040, 95%CI = [1.001, 1.042]) and the NC-MED group has a more similar distribution of deprivation to the cholesteatoma group than the controls (Figure 2e).

3.3 ICD-10 Associations With Cholesteatoma

Briefly, 56 ICD-10 codes were significantly associated with cholesteatoma after Bonferroni multiple testing correction (p < 0.05; Table 2). Hierarchical clustering of Jaccard distance between these codes reveals two main groups: common conditions, including chronic obstructive pulmonary disease (OR = 2.03, 95%CI = [1.653, 2.495]), disorders of lipoprotein metabolism and other lipidaemia (OR = 1.48, 95%CI = [1.28, 1.703]) and gastro-oesophageal reflux (OR = 1.51, 95%CI = [1.288, 1.771]); and diseases of the sinuses and middle ear and their complications (Figure 3).

The strongest associations (all p values < 0.001) were with other disorders of middle ear and mastoid (OR = 242.12, 95%CI = [203.651, 287.867]), other disorders of tympanic membrane (OR = 163.63, 95%CI = [133.92 9, 199.912]), suppurative and unspecified otitis media (OR = 144.54, 95%CI = [124.259, 168.125]), otalgia and effusion of the ear (OR = 78.22, 95%CI = [64.143, 95.383]) and perforation of the tympanic membrane (OR 76.07, 95%CI = [64.549, 89.646]). The overlap with cholesteatoma for each of these conditions was 100–263 persons. Otitis externa (OR = 34.97, p < 0.001, 95%CI = [27.429, 44.577]) and other diseases of inner ear (OR = 6.04, p < 0.001, 95%CI = [3.946, 9.235]) were also associated with cholesteatoma. Known complications of cholesteatoma were also strongly associated, including sensorineural and conductive hearing loss (OR = 25.81, 95%CI = [21.65, 30.774]), other hearing loss (OR = 9.94, 95%CI = [8.514, 11.608]), facial nerve disorders (OR = 10.10, 95%CI = [7.391, 13.793]) and bacterial meningitis (OR = 41.78, 95%CI = [24.919, 70.056]).

The most strongly associated non-ear code not known to be a cholesteatoma complication was F17, mental and behavioural disorders due to tobacco use with an OR of 2.34 (p < 0.001, 95%CI = [1.942, 2.813]), followed by chronic sinusitis, with an OR of 4.09 (p < 0.001). Several other respiratory conditions are represented including chronic obstructive pulmonary disease, asthma and chronic rhinitis with and without nasal polyps. Some congenital anomalies affecting the ear and head (Q17, Q16, Q75, Q96) were also strongly associated with cholesteatoma (OR = 33.58–83.24, p ≤ 0.011), although the number of overlapping cases was small (n = 2–4).

3.4 Comparison to FinnGen Significant Associations

Diseases occurring before cholesteatoma with significantly increased hazard ratios in both biobanks were otosclerosis and sleep apnoea (Table 3). Hazards of otosclerosis, epilepsy and chronic kidney disease were significantly increased after cholesteatoma in both biobanks.

4.1 Key Results

Here we present an epidemiological study of cholesteatoma in the UKBB to establish demographic factors associated with cholesteatoma in this cohort and to identify associated diseases using ICD-10 codes. We also compared these results with summary analyses performed on data from the Finnish biobank FinnGen. Our study replicated some demographic risk factors for cholesteatoma identified in previous studies. We also found a large overlap between cholesteatoma and other middle ear diseases, and the demographics of these groups are similar except for sex and ethnicity.

4.2 Associations With Sex, Deprivation and Smoking

Male sex was an important risk factor for cholesteatoma compared to disease-free controls (AOR = 1.33, 95%CI = [1.179–1.491]). A male predominance is well established in the literature and the sex ratios of 1:13 females to males in UKBB and 1:1.67 in FinnGen are similar to the reported range [2, 17]. Both cholesteatoma and NC-MED were associated with higher deprivation, with cholesteatoma associated with the highest deprivation (median − 1.40, −1.70 and −2.14, respectively). Deprivation is associated with risk behaviours such as smoking [18, 19], and the most disadvantaged are both more likely to require healthcare and less likely to access it [20]. These factors may contribute to greater rates of ear disease and cholesteatoma. However, it is also possible that increased deprivation results from cholesteatoma due to the economic impact of hearing loss. Increased odds of several common diseases found in this study may also reflect generally poorer health in the cholesteatoma group. Meanwhile, the odds ratio associated with age for both cholesteatoma and NC-MED is likely due to older participants having had longer to develop any disease.

Rates of cholesteatoma have also been reported to vary with ethnicity [1], although original epidemiological data is rarely presented. Both rates of cholesteatoma and NC-MED varied between ethnicities in this analysis, although the only consistent significant difference was for the Black ethnicity, who had reduced odds of both cholesteatoma and NC-MED. The Chinese and Asian groups also had increased odds of NC-MED, but the sample sizes may have been too small to properly quantify their risk of cholesteatoma.

Although smoking was not significantly associated with cholesteatoma according to the survey results, the disease code F17 mental and behavioural disorders due to use of tobacco was (OR 2.34, 95%CI = [1.942, 2.813]). This suggests that infrequent or discontinued smoking affects risk of cholesteatoma less than heavy smoking. This confirms the results of Kaylie et al. [8], where they found a higher rate of cholesteatoma and more severe disease in smokers with chronic ear problems, but that former smokers had similar outcomes to non-smokers > 5 years after smoking cessation. Smoking may increase susceptibility to ear disease by impairing mucociliary function [8], which may also impact cholesteatoma risk.

4.3 Sex Distinguishes Cholesteatoma From Other Middle Ear Disease

A key finding was that the cholesteatoma and NC-MED cohorts were more similar to each other than the controls on several characteristics. Sex ratio was a key difference distinguishing cholesteatoma from NC-MED, with the AOR for cholesteatoma being similar when compared to NC-MED or controls (AOR = 1.3 vs. NC-MED; AOR = 1.33 vs. controls). Furthermore, sex ratio did not differ significantly between middle ear disease and the control group (p = 0.49; Table S4). This means that either males are not at increased risk of general ear disease, only of cholesteatoma; or that they are at increased risk of all forms of ear disease and at increased risk of sequelae, including cholesteatoma.

4.4 Cholesteatoma Is Significantly Associated With Other Inflammatory Ear and Respiratory Disease

In this study, we found a large overlap between the incidence of cholesteatoma and other middle ear disease. Many of these are likely consequences of cholesteatoma or related ear conditions, such as hearing loss, otorrhoea, tympanic perforation, mastoiditis, Eustachian tube disorders and several forms of otitis media. A history of ear disease, including chronic otitis media is well-established risk factors in cholesteatoma: Kemppainen et al. [2] report history of otitis in 72.4% of cases, and Castle [21] reports common concurrent diseases including otic polyp and tympanosclerosis. Whether inflammation precedes cholesteatoma or is a symptom of it cannot be determined from this study design. However, children with repeated ear infection requiring ventilation tube insertion are at increased risk of cholesteatoma [22], implying that increased susceptibility to middle ear infection also increases cholesteatoma susceptibility, either directly or because both result from poor ear ventilation. Tympanic retraction is also a risk factor for cholesteatoma, though most retractions do not progress to cholesteatoma [23].

This study also found an association between cholesteatoma and chronic respiratory inflammation including rhinosinusitis and asthma. Chronic sinusitis may impact Eustachian tube function and ventilation of the middle ear, possibly contributing to cholesteatoma risk, or this may represent a general susceptibility to chronic inflammation of the airways.

Increased hazards of otosclerosis both before cholesteatoma in UKBB and FinnGen is a surprising finding as the co-occurrence of these conditions is extremely rarely reported [24]. Otosclerosis is a disease of the labyrinth resulting from remodelling and overgrowth of the bone at the base of the stapes [25]. Though it is possible that a background of inflammation in the middle ear could increase risk of otosclerosis, these conditions rarely occur together and can easily be misdiagnosed. Because of the small number of overlapping cases (n = 5 in UKBB, 48 in FinnGen), this may be the cause of this association.

4.5 Known Complications of Cholesteatoma and a New Potential Association

We detected known rare complications of cholesteatoma such as facial nerve disorders (including Bell's palsy) and bacterial meningitis [26]. Similarly, other diseases of the nervous system such as encephalitis, myelitis and encephalomyelitis, and sequelae of inflammatory diseases of the central nervous system may also represent rare complications of cholesteatoma, which occur when the cyst erodes through the temporal bone. The risk of epilepsy following cholesteatoma was also significantly increased in both UKBB and FinnGen. Some of these cases could be a complication of cholesteatoma where there is intracranial involvement, and epilepsy has occasionally been described in this context [27, 28]. Alternatively, epilepsy itself may be a risk factor, as it is often associated with a high burden of comorbid disease [29].

We also detected increased odds of psoriasis and other local infections of skin and subcutaneous tissue with cholesteatoma. Cholesteatoma expresses several proteins, which are markers of inflammatory skin conditions, including members of the S100 protein family [30].

4.6 Study Limitations

The limitations of this study reflect the limitations of BioBank-based observational studies. UKBB is biassed towards females, participants are wealthier and have fewer health conditions than the general population and is majority of white European background [31]. It is also likely that not all cholesteatoma cases could be identified due to the lack of ICD-10 data prior to its introduction in 1995. We mitigated this by including likely cholesteatoma cases from related ICD-10 and operative codes, but there are probably still missing cases, including childhood disease. Furthermore, no distinction has been made between congenital and acquired cholesteatoma in this study as they are impossible to distinguish from ICD codes.

Demographics such as deprivation were collected at the time of recruitment and may differ from the time of disease onset. Additionally, deprivation is based on postcode and may not reflect the status of the individual. Different demographic groups may also differ in their willingness to engage with a doctor, which will influence the apparent effect of these factors.

Numbers of cases and controls were highly imbalanced. To retain as much information as possible, unmatched controls were used for each analysis which can inflate p values. We performed sensitivity analysis for the demographic features using 1:5 matched controls to check that the effect of imbalance was not too severe. However, some ICD-10 codes have even more extreme imbalance so should be interpreted with caution.

Conclusions about causality cannot be drawn from observational associations, for example, deprivation could be a sequela or a risk factor for the morbidity associated with cholesteatoma. Time to event information also reflects time of diagnosis, not necessarily the order of disease onset.