Study population
The study utilised data from the English Longitudinal Study of Ageing (ELSA). The ELSA is a longitudinal prospective cohort study that collects multidisciplinary data from a nationally representative sample of community-dwelling middle-aged and older (aged 50 and above) adults in England [12]. The study started in 2002 and is collecting responses every 2 years on participants’ health, social, wellbeing and economic circumstances. The current sample contains data from eight Waves, covering the period 2002–2017 [13]. As the ELSA follows a longitudinal design, the sample is comprised of a sequence of observations on the same individuals across Waves and the refreshment samples (Cohorts 3, 4, 6 and 7) [13]. Proxy interviews were carried out in case an ELSA panel member refused to further participate [14]. In our analyses, we used the full dataset (74,699 person-years) of self-reported hearing data from all eight Waves of the ELSA.
The ELSA follows the sampling strategy of the Health Survey for England (HSE), which ensures that every address on the small users’ Postcode Address File (PAF) in England has an equal chance of inclusion. Field household contact rates of over 96% were achieved. The study excluded cases not belonging to the target population through ‘terminating events’, such as deaths, institutional moves and moves out of England since taking part in the HSE [15].
Outcomes
Hearing acuity
Self-rated hearing data was collected from participants across all Waves. According to the study’s documentation, self-reported HL was defined as declarations of fair or poor hearing on a five-point Likert scale (excellent, very good, good, fair or poor) or ‘Yes’ responses to the question concerning whether or not the participants find it difficult to follow a conversation if there is background noise (e.g. noise from a TV, a radio or children playing) [13, 16].
Geographical variables
The geographically related information of the ELSA dataset was in the form of identifiers such as the Government Office Region (GOR) [17], and indices that are used as measure of poverty of different geographical areas, such as the Index of Multiple Deprivation (IMD). The geographical variables were provided to the first author under a Special License and Secure Access agreement (UK Data Service Project Number: 121175).
Each respondent’s geography is determined by their residence postcode at the time of the survey interview date. Different versions of the IMD were provided for the eight Waves of the ELSA: IMD 2004 [18] for Waves 1–3, IMD 2007 [19] for Wave 4, IMD 2010 [20] for Waves 5–7 and IMD 2015 [21] for Wave 8. The IMD was provided in quintiles (the first quintile being the least deprived, the fifth being the most deprived).
The nine GORs represent the highest tier of sub-national division in England (North East, North West, Yorkshire and the Humber, East Midlands, West Midlands, East of England, London, South East, South West).
Covariates
For covariates, we examined non-modifiable factors (age, sex), partly modifiable indicators of socio-economic position (SEP) (education, occupation, income, wealth) and alcohol consumption as a fully modifiable lifestyle risk factor for HL. Age was assessed both as a discrete (as only certain values could be taken) and categorical variable in three groups (50–64, 65–74, 75–89). We used this categorisation to allow for a comparison with Benova et al. [22], who examined the association of SEP with self-reported hearing difficulties in Wave 2 of the ELSA.
We considered five categories regarding highest educational attainment: no qualifications, foreign or other, O level Certificate of Secondary Education, A level (Level 3 Qualification of the National Qualifications Framework) and a degree or higher education.
Tertiles of self-reported occupation were based on the National Statistics Socio-economic Classification (NS-SEC): routine and manual occupations; intermediate; managerial and professional. The relative financial position of the participants was captured by quintiles of net household income (the first quintile being the lowest, the fifth being the highest). Wealth was examined in quintiles of the net total non-pension wealth reported at the household unit level (the first quintile being the highest, the fifth being the lowest).
Alcohol consumption was selected as the only lifestyle factor that was consistently recorded in all Waves. We constructed a continuous variable to represent the sum of units of alcohol that each participant consumed during the last 7 days. This variable was dichotomised into those that consumed more than 14 units of alcohol in the last 7 days and those that did not, using the Chief Medical Officer’s Drinking Guidelines [23].
Data analysis
Categorical variables are presented as absolute (n) and relative (%) frequencies, while continuous variables are presented through their mean and standard deviation. We used the full dataset from the eight Waves (74,699 person-years) to strengthen the argument that there is a correlation between spatial variables and HL over time. A small number of cases (one in Wave 0 and eight in Wave 2) in the geographical identifiers had missing values because the address was located within Wales (which uses its own deprivation index). Due to the low proportion of missingness in the variables, records with missing data were excluded from analyses (3.2% of all records in listwise deletion). We used Bartlett’s test for homogeneity of variances to test that age variances were equal for all samples. Following this, we applied one-way analysis of variance (ANOVA) to compare the means of age among GOR samples in all Waves. We also computed adjusted predictions at the means (APMs) and the marginal effects at the means (MEMs) [24] for the HL prevalence in each Wave of the ELSA, with age, sex, education, occupation, income, wealth, IMD and alcohol consumption as the factor variables.
We used local spatial analysis statistical tools for analysing spatial distributions, patterns, processes and relationships in the geographical data. We used the Spatial Join tool to aggregate the number of cases of self-reported HL to total responses of hearing acuity in each polygon (GOR) in order to visualise the prevalence rates of HL per GOR. We used the Natural Breaks (Jenks) classification to optimise the arrangement of the sets of HL values into ‘natural’ classes, a method also known as the goodness of variance fit (GVF). Furthermore, we used the Hot Spot Analysis (Getis-Ord Gi*) as a mapping cluster tool to identify the locations of statistically significant Hot Spots and Cold Spots. The Getis-Ord Gi* is an inferential statistic for the conceptualisation of spatial relationships, used when one is looking for unexpected spatial spikes of high values. In essence, this tool works by looking at each feature within the context of neighbouring features and assessing whether high or low values cluster spatially. Due to the small scale of the analysis, we chose this local spatial statistic tool so that the value of each feature could be included in its own analysis, along with the neighbouring features.
The Getis-Ord local statistic is given as:
$$ {G}_i^{\ast }=\frac{\sum_{j=1}^n{w}_{i,j}{x}_j-\overline{X}{\sum}_{j=1}^n{w}_{i,j}}{\sqrt[S]{\frac{\left[n{\sum}_{j=1}^n{w}_{i,j}^2-{\left({\sum}_{j=1}^n{w}_{i,j}\right)}^2\right]}{n-1}}} $$
(1)
Here, xj is the attribute value for feature j, wi, j is the spatial weight between feature i and j, n is equal to the total number of features and:
$$ \overline{X}=\frac{\sum_{j=1}^n{x}_j}{n} $$
(2)
$$ S=\sqrt{\frac{\sum_{j=1}^n{x}_j^2}{n}-{\left(\overline{X}\right)}^2} $$
(3)
The \( {G}_i^{\ast } \) statistic is a z-score, so no further calculations are required.
The spatial relationship was defined according to the ‘Contiguity Edges Corners’, a method that was selected in order to allow all neighbouring polygon features that share a boundary or node to influence the target polygon feature’s computations. Confidence levels of 90, 95 and 99% were considered in the calculations of Getis-Ord Gi*. Data were analysed using Stata version 14 [25] and ESRI ArcGIS Desktop 10.7.1 [26].