The Longitudinal Impact of Psychosocial Factors on Cognition and Hearing in Younger and Older Adults During the COVID-19 Pandemic

Abstract

Purpose:

In March 2020, a unique situation unfolded wherein the U.K. government announced social restriction measures to reduce the spread of the virus that causes COVID-19. Various measures remained in place until April 2021, with older adults, who were considered clinically vulnerable, being placed under stricter restrictions. This study aimed to determine the effect of psychosocial factors, including loneliness, depression, and engagement in various recreational lifestyle activities, on hearing and cognitive function in younger and older adults during the COVID-19 pandemic.

Method:

One hundred twelve older adults aged 60–82 (M = 70.08, SD = 5.89) years and 121 younger adults aged 18–29 (M = 20.52, SD = 2.63) years participated online between June 2020 and February 2022. Participants completed questionnaires assessing loneliness, depression, auditory and lifestyle engagement, and hearing ability, as well as behavioral tasks assessing auditory function and global cognition. All measures were completed 12 times at 4-week intervals.

Results:

Linear mixed-effects analyses found that, of the variables examined, increased loneliness was significantly associated with poorer auditory function. There were no main effects of time during the pandemic on auditory or cognitive outcomes. However, the interaction between time and age group significantly affected global cognition; in younger adults, global cognition decreased over time, whereas older adults displayed an unexpected positive change.

Conclusions:

These data show that there are associations between loneliness and auditory function but provide a lack of support for the impact of time experiencing auditory deprivation, or other psychosocial factors, on hearing and cognitive function. Such observations may be underpinned by motivational differences, learning effects, or sample biases. Future research may wish to investigate these factors further, to determine how psychological factors such as loneliness affect hearing and cognitive processes across diverse participant groups.

Supplemental Material:

https://doi.org/10.23641/asha.30936590

As the population ages, health issues grow in prevalence, placing increasing pressure on health care systems. Hearing loss (HL) is one of the most common conditions in older age, affecting over 70% of people aged 70+ years in the United Kingdom (Royal National Institute for Deaf People, 2020). Many age-related health conditions are associated with, or have been shown to exacerbate, one another. For example, HL is associated with increased levels of loneliness and depression (Lawrence et al., 2020; Mick et al., 2014; Shukla et al., 2020). It has also been recognized internationally that both social isolation and HL are potentially modifiable risk factors for dementia. In fact, if the risk factors of social isolation and HL are indeed causal for dementia and were removed, then it is hypothesized that dementia cases could be reduced by as much as 4% and 8%, respectively (Livingston et al., 2024). However, further high-quality longitudinal data are required to elucidate the true nature of the HL–dementia relationship.

Several hypotheses have been proposed to explain the association between age-related HL (ARHL) and dementia (Lindenberger & Baltes, 1994; Powell et al., 2021). A number of these hypotheses suggest that HL has a causal effect on cognitive function. In brief, HL has been suggested to affect cognitive function directly via (a) increasing listening effort depleting cognitive resources (Pichora-Fuller et al., 2016) and (b) increasing auditory deprivation, which occurs when the brain is deprived of sound, leading to neuroanatomical changes (e.g., Lin et al., 2014). These direct pathways are suggested to affect global brain function and structure in a way that compromises cognitive functioning (Fitzhugh et al., 2019; Panouillères & Möttönen, 2018).

Researchers have also suggested that HL may causally affect cognition via an indirect psychosocial pathway (Shukla et al., 2020). HL significantly impacts psychosocial factors, including feelings of loneliness, isolation, and depression, due to a reduction in the quantity and quality of social interactions (Jayakody et al., 2018). Difficulty listening, particularly in noisy environments, may lead older adults with HL to withdraw from social interactions due to communication challenges or embarrassment and stigma (David et al., 2018). This social withdrawal may exacerbate auditory deprivation, due to reduced engagement with auditory-rich and cognitively stimulating environments. This, in turn, may modulate the association between HL and cognitive decline. Importantly, older people may be particularly vulnerable to loneliness and depression due to the higher prevalence of living alone (Age UK, 2019), and this risk may be further increased by HL (Bott & Saunders, 2021; Maharani et al., 2019). Understanding the effect of psychosocial factors on both HL and cognition in older age is essential to shed light on the factors that might contribute to increased well-being and healthy brain aging.

During the height of the COVID-19 pandemic, the U.K. public experienced social distancing, enforced isolation, and restricted means of communication in various forms, from March 2020 until January 2022. This overwhelming period of unprecedented change enabled researchers to investigate how loneliness and isolation might affect sensory and cognitive function across age ranges. Considering that older adults may be more likely to experience loneliness and isolation as well as HL (Age UK, 2019), compared to younger adults, it is conceivable that older adults may have been disproportionately affected by pandemic-related restrictions. Associations between sensory impairments and psychosocial factors including social participation, social network size, and loneliness have been widely reported (Mick et al., 2018; Ray et al., 2018), and theoretical frameworks have been proposed detailing anchor stages (from listening disengagement to social withdrawal and loneliness) to describe the relation between HL and social isolation (Motala et al., 2024). Importantly, during the height of the pandemic, older adults and other clinically vulnerable populations were provided with stricter social distancing guidance. As such, older adults may have been at a greater risk of social withdrawal and reduced social communication, leading to auditory deprivation, particularly in terms of reduced in-person social contact. This could have led to long-term consequences for hearing and cognitive function. Understanding how social factors relate to both cognitive and hearing function is imperative for identifying intervention pathways targeting HL and cognitive decline.

Hearing could be affected if environmental auditory deprivation, due to social distancing and isolation, leads to tangible changes in the auditory cortex and associated brain areas used for processing speech in noise. Deprivation of auditory input, due to HL, is associated with atrophy of the brain regions associated with hearing (Slade et al., 2022), which could negatively affect speech perception ability. This atrophy may occur because HL-related damage to the auditory periphery leads to distorted auditory representations, reduces access to verbal and emotional information in speech, and decreases the amount of auditory information sent to the brain, leading to atrophy of auditory and association areas (Griffiths et al., 2020). Similarly, during the height of the pandemic, a deprived auditory environment was created due to social restrictions and poor listening environments (i.e., use of face coverings, online calls, Perspex screens), which could have negatively affected the capacity for speech understanding. Indeed, deprivation of auditory input, due to prolonged wearing of earplugs, has been shown to alter neural responses to speech (Munro & Blount, 2009). Furthermore, social distancing has been shown to negatively impact the quality of communication and connection with others (Wood et al., 2024).

Cognition may also be affected in a similar way. Increased social interactions give rise to mentally stimulating situations that benefit cognitive function (Sommerlad et al., 2019). According to the cognitive reserve hypothesis, engaging in social activities is a key aspect of building cognitive reserve that may help to protect against age- and disease-related declines in cognitive function (Oosterhuis et al., 2023; Stern et al., 2020). Similarly, social contact has been internationally recognized as a protective factor against dementia (Livingston et al., 2024), and a recent scoping review indicates that social isolation and loneliness relate to poor cognitive function in older adults (Cardona & Andrés, 2023). A variety of assessments may be employed to measure cognition such as the Mini-Mental State Examination (MMSE; Folstein et al., 1975) and the Montreal Cognitive Assessment (MoCA; Hobson, 2015). These standardized assessments are generally employed to test the presence of cognitive impairment. They comprise several domains of cognition including short-term and working memory, executive functioning, processing speed, or reaction time. Importantly, these domains are considered to be sensitive to age-related declines in cognition (Deary et al., 2009; Murman, 2015). Furthermore, studies indicate these cognitive domains may be affected by psychosocial factors. For example, memory recall and executive function abilities have been found to be related to loneliness (Lara et al., 2019; Luchetti et al., 2020; Sin et al., 2021), and processing speed has been found to be related to social isolation (Hajek et al., 2020).

During the COVID-19 pandemic, we explored the indirect psychosocial pathway hypothesis, also known as the “cascade hypothesis” (Dawes et al., 2015; Dhanda et al., 2024). According to this hypothesis, social withdrawal, isolation, and possibly resulting loneliness, further exacerbate auditory deprivation, due to reduced engagement with auditory-rich and cognitively stimulating environments. This deprivation then negatively affects hearing and/or cognitive function. In a previous study, subjective hearing ability (measured by the Speech and Spatial Qualities of Hearing Scale; Noble et al., 2013) exacerbated the impact of social distancing on depression, loneliness, and memory in older adults (Littlejohn et al., 2022). The present study builds on these findings, taking a lifespan approach by comparing younger and older adults and measuring longitudinal outcomes of auditory function (comprising both subjective hearing ability and speech-in-noise perception [SPiN]) and global cognitive function across a period of 12 months during the pandemic.

Consistent with our preregistration protocol, data were collected between June 2020 and January 2022. All participants joined the study between June 2020 and February 2021, and data were collected over the subsequent 48 weeks for each participant. For context, the first U.K. national lockdown, the government-ordered mandate to stay at home, was announced in March 2020; a second national lockdown was then announced in November 2020; and a third was announced in January 2021. Between these dates, the United Kingdom experienced numerous changes to social contact, including various local lockdowns and a tiered system of restrictions. Restrictions to social contact remained in place until the end of 2021, with the last measures of compulsory face mask wearing and mandatory NHS COVID passes finally being removed in January 2022 (Institute for Government, 2022).

The aim of this preregistered study was to determine the effect of a period of enforced social isolation and restriction on both hearing and cognitive function in younger and older adults. As such, the primary predictors were (a) loneliness, determined via self-report scales, and (b) time, which ranged from Time Point 1 to Time Point 12, with each time point separated by 4 weeks. We also included secondary predictors, which we hypothesized to interact with the primary variables, including (a) age group (older vs. younger), (b) hearing status (ARHL vs. no HL), (c) depression, (d) engagement in auditory activities, and (e) engagement in lifestyle activities. Table 1 outlines the hypotheses.

Table 1.

The hypothesized impact of the predictors on cognitive and auditory outcomes.

Hypotheses 1 and 2

Main effects of time or loneliness on global cognition (H1a and H1b) or on auditory function (H2a and H2b): Both global cognition and auditory function are expected to worsen over time and with greater loneliness.

Secondary hypotheses

Two-way interaction effects between time and age, loneliness and age, time and hearing status, or loneliness and hearing status on global cognition (H3a, H3b, H3c, and H3d, respectively) or on auditory function (H4a, H4b, H4c, and H4d, respectively): Any negative changes in cognitive or auditory function with time or loneliness will be greater for older adults or for those with HL. Two-way interaction effects between time and depression or between loneliness and depression on global cognition (H5a and H5b, respectively) or on auditory function (H5c and H5d, respectively): Any negative changes in cognitive or auditory function with time or loneliness will be greater for participants with increased depressive symptoms. Two-way interaction effects between time and auditory engagement or between loneliness and auditory engagement on global cognition (H6a and H6b, respectively) or on auditory function (H6c and H6d, respectively): Any negative changes in cognitive or auditory function with time or loneliness are expected to be greater for participants with lower engagement in auditory activities. Two-way interaction effects between time and lifestyle engagement or between loneliness and auditory engagement on global cognition (H7a and H7b, respectively) or on auditory function (H7c and H7d, respectively): Any negative changes in cognitive or auditory function with time or loneliness are expected to be greater for participants with lower engagement in lifestyle activities.

Exploratory hypotheses

We report on exploratory analyses of the main effects of age, hearing status, depressive symptoms, auditory engagement, and lifestyle engagement on both global auditory and cognitive function. We also report on the impact of study start date, operationalized as the months passed since the first government mandated lockdown in the U.K. when the participant began taking part.

Materials and Methods

Ethical approval was obtained from Lancaster University's Faculty of Science and Technology Ethics Committee (FST19175).

Transparency and Openness Statement

The study was preregistered on the Open Science Framework (https://osf.io/67rwh/). Any deviations from the protocol are described below.

Deviations From Preregistration

Statistical inference. In our preregistered analysis plan, we reported that the statistical inference criteria would be p < .05 for determining significant results. However, on reflection, in identifying the need to test multiple hypotheses, we decided to apply a correction factor to this criterion to reduce the likelihood of Type I error. We corrected the p value for determining statistical significance, over the number of hypotheses tested (n = 24), to the more conservative threshold where p value < .002 would be classed as significant.

Analysis. As detailed in our preregistered analysis plan, the start date (i.e., the number of months since the first U.K. lockdown) was included as a fixed-effects covariate. However, despite indicating a plan to model how this variable interacted with other variables of interest, we chose not to do this to simplify the amount of statistical analysis in the absence of clear hypotheses concerning this parameter.

Participants

The sample initially consisted of 112 older adults (62 women, 50 men) aged 60–82 (M = 70.08, SD = 5.89) years both with (n = 55) and without (n = 57) self-reported HL and 121 younger adults (85 women, 36 men) aged 18–29 (M = 20.52, SD = 2.63) years with self-reported normal hearing. The required sample size was determined by an a priori power analysis to detect a moderate effect size of Cohen's f = .25 at 90% power and an α of .05, using GLIMMPSE software (Version 3) for calculating power and sample size for linear mixed models (Kreidler et al., 2013), as detailed in the associated preregistration (https://osf.io/67rwh/).

The sample was self-selected with participants recruited through advertisements on Lancaster University's Research Participation (SONA) System and Centre for Ageing Research Participant Panel, as well as the University of the Third Age, social media, and local print media. Inclusion criteria required that participants be right-handed and monolingual speakers of English, have normal or corrected-to-normal vision, and present no history of neurological, language, or speech disorders. Participants completed a cognitive screening questionnaire, the self-report version of the Informant Questionnaire on Cognitive Decline, and participants scoring 3.65 or higher were excluded before participation, as this has been suggested as an appropriate cutoff (Jansen et al., 2008), where scores > 3.65 indicate potential cognitive decline. The study was approved by Lancaster University Faculty of Science and Technology Research Ethics Committee (Reference No. FST20091).

Participant Attrition

At month 12, the sample consisted of 165 participants, an attrition rate of 29.18%. At this time point, there were 58 younger adults (36 women, M_age = 20.90 years) and 107 older adults (59 women, M_age = 70.11 years). Due to attrition, there were missing data points across the months, which are detailed in Supplemental Material S1.

Materials

Self-Report Predictor Measures

Hearing status. Participants were asked to self-report any clinical or perceived HL, using a single-item question: “Do you have any hearing disorders or hearing loss?” Response options included “lifelong hearing loss,” “age-related hearing loss,” “other hearing disorder,” or “no hearing loss.” All younger adults reported no HL, as required for study participation. Only older participants who either had no HL or experienced acquired HL in later life were able to participate, as we were primarily interested in ARHL rather than lifelong HL or deafness. As such, any older adults who had experienced lifelong deafness or HL were ineligible, and any who selected the “other” category were asked further questions about their hearing to check for eligibility. These data were used to group older adults into two hearing status groups: ARHL or no HL. For older adults, 57 reported no HL (age: M = 68.70 years, SD = 5.58 years), and 55 reported having ARHL (age: M = 71.51 years, SD = 5.92 years). Of those who reported having ARHL, 31 reported being bilateral hearing aid users, and five reported being unilateral users.

Loneliness. Loneliness was measured using two questionnaires: the Lubben Social Network Scale-6 (LSNS-6; Lubben et al., 2006) and the UCLA Loneliness Scale Version 3 (UCLA-LS3; Russell, 1996). The LSNS-6 is a six-item questionnaire used to assess an individual's perception of social support available to them and frequency of contact with their social networks. An example question is “How many relatives did you see or hear from at least once a month?” Participants responded using a 6-point scale containing the following choices: “none,” “one,” “two,” “three or four,” “five to eight,” or “nine or more.” The questionnaire is reported to have good reliability (Cronbach's α = .83) in older adult populations (Lubben et al., 2006). The UCLA-LS3 is a 20-item questionnaire used to assess feelings of loneliness and disconnection from others. An example question is “How often do you feel alone?” and “How often did you feel that you lacked companionship?” Participants respond using a 4-point rating scale containing the following choices: “never,” “rarely,” “sometimes,” or “always.” The questionnaire has been shown to have high reliability, in terms of internal consistency (Cronbach's α ranging from .89 to .94) and test–retest reliability (r = .73), across age ranges (Russell, 1996). Two questionnaires were employed here to ensure that the index captured both social network size (social loneliness) and feelings of loneliness (emotional loneliness). By assessing both constructs, we ensure that we capture multiple constructs of loneliness that may have been affected during the pandemic. A composite measure of loneliness was created by standardizing the total scores within each questionnaire and then calculating the mean of the total scores on each measure, per person, with higher scores indicating greater loneliness.

The test–retest reliability for the loneliness composite across the 12 time points of data collection was estimated with intraclass correlation coefficients (ICCs) using the “psych” package in R (Revelle, 2024). ICCs were used on the data after influential outliers were removed for both linear mixed-effects models (see details of this procedure in the Results section), in which either global cognition or auditory function was the outcome of interest, because different data points may have been excluded as influential data points across the two models. We report the results of two-way mixed-effects models for absolute agreement, ICC(2,1), and consistency, ICC(3,1). For the data included in the global cognition model and in the auditory function model, the estimated agreement was .90, 95% confidence interval (CI) [.88, .92], and the estimated consistency was .90, 95% CI [.88, .92]. The loneliness composite was found to have good internal consistency across the 12 time points of data collection (Koo & Li, 2016).

Depression. Depression was measured using the Beck Depression Inventory-I (BDI-I; Beck et al., 1961). The BDI-I is a 21-item questionnaire used to evaluate the severity of depressive symptoms experienced by a participant over the previous week. For each item, the participant selected one of four statements that range in intensity, each scored on a scale from 0 to 3, for example, “I do not feel sad” (0), “I feel sad” (1), “I am sad all the time and I can't snap out of it” (2), or “I am so sad or unhappy that I can't stand it” (3). The questionnaire has been shown to have high reliability (Cronbach's α < .75) and validity (Beck et al., 1988; Richter et al., 1998). The measure of depression was created by calculating the total score, with higher scores indicating greater depressive symptoms.

We estimated test–retest reliability for the depression scores across the 12 time points of data collection with ICCs in R using “psych” (Revelle, 2024). For the data included in the global cognition model, the estimated agreement was .77, 95% CI [.73, .80], and the estimated consistency was .77, 95% CI [.73, .80]. For the data included in the auditory function model, the estimated agreement was .76, 95% CI [.73, .80], and the estimated consistency was .77, 95% CI [.73, .80]. The depression measure was found to have good internal consistency (Koo & Li, 2016).

Auditory and lifestyle engagement. A 10-item self-report questionnaire measured engagement in auditory and lifestyle activities (Slade et al., 2023). Participants estimated how many hours they spent doing certain activities in an average week in the previous month on a scale of 0–50 hr.

Auditory engagement was measured using the first seven items, which measured how much time participants estimated they spent doing auditory activities across active (e.g., in-person or online socializing) and passive (e.g., listening to audiobooks) listening domains. The questionnaire assessed three factors: Items 1–3 assessed in-person communicative auditory engagement, Items 4–5 assessed online communicative auditory engagement, and Items 6–7 assessed online noncommunicative auditory engagement. The questionnaire items were weighted based on the level of auditory engagement they were designed to assess. The score obtained from Items 1–3 for in-person communication was multiplied by 0.3. The score from Items 4–5 for remote communication was multiplied by 0.2. The score from Items 6–7 for noncommunication activities was multiplied by 0.1. The decision to employ these weightings was made a priori and preregistered and was designed to ensure that activities that involved greater in-person communication were given greater importance in the total score derived for this measure. The measure intended to tap into the auditory and social exposures of the participants in the study during the pandemic, comprising both passive listening and socially active listening. Greater weighting is placed on more active, and thus more cognitively involved, auditory activities. The resulting scores were totaled to provide an auditory engagement score, with higher scores indicating greater auditory engagement.

We estimated test–retest reliability for the auditory engagement scores across the 12 time points of data collection with ICCs in R using “psych” (Revelle, 2024). For the data included in the global cognition model and in the auditory function model, the estimated agreement was .61, 95% CI [.56, .65], and the estimated consistency was .61, 95% CI [.56, .65]. The auditory engagement measure was found to have moderate internal consistency (Koo & Li, 2016).

Lifestyle engagement was measured using the final three items of the engagement questionnaire, which measured the time participants estimated that they spent engaged in various lifestyle activities such as hobbies or sports. The total score obtained from the summed responses to the three items provided a total lifestyle engagement score, with higher scores indicating greater lifestyle engagement or participation.

We estimated test–retest reliability for the lifestyle engagement scores across the 12 time points of data collection with ICCs in R using “psych” (Revelle, 2024). For the data included in the global cognition model, the estimated agreement was .66, 95% CI [.62, .71], and the estimated consistency was .67, 95% CI [.62, .71]. For the data included in the auditory function model, the estimated agreement was .67, 95% CI [.62, .71], and the estimated consistency was .67, 95% CI [.63, .71]. The lifestyle engagement measure was found to have moderate internal consistency (Koo & Li, 2016).

Outcome Measures

Global cognition. Global cognition was measured using a battery of four cognitive assessments: (a) the forward digit span (e.g., Wechsler Adult Intelligence Scale [WAIS]; Wechsler, 1997), (b) the backward digit span (e.g., WAIS; Wechsler, 1997), (c) the Deary–Liewald choice reaction time task (Deary et al., 2011), and (d) the Stroop Color–Word Test (Scarpina & Tagini, 2017; Stroop, 1935). These measures were employed to assess aspects of cognitive functioning (short-term and working memory, executive functioning, and processing speed) that may not necessarily be relevant to auditory cognitive performance during speech understanding but are typically assessed in standard assessments of cognitive decline; these aspects have shown age-related declines in previous research (Bopp & Verhaeghen, 2005; Folstein et al., 1975; Hobson, 2015). The scores calculated within each task were standardized (z scored) and then totaled to provide a composite score, following the preregistered protocol. Higher scores indicate better global cognitive performance.

We estimated test–retest reliability for the composite global cognition measure across the 12 time points of data collection with ICCs in R using “psych” (Revelle, 2024). The estimated agreement was .56, 95% CI [.51, .61], and the estimated consistency was .56, 95% CI [.51, .61]. The global cognition measure was found to have moderate internal consistency (Koo & Li, 2016).

Forward digit span: This task was used to assess short-term memory (e.g., Wechsler, 1997). Participants were presented with eight sets of number sequences containing two sequences per set, in order of difficulty. The sequence length ranged from two digits in Set 1 to nine digits in Set 8. In a trial, participants saw a fixation cross (1 s), followed by each number in the sequence (1 s for each number), and then a response screen, where they were asked to type the number sequence. After the response, participants saw a blank screen for 1 s before the next trial began. The task ended if two sequences in a set were recalled incorrectly. The number of correctly recalled sequences was totaled, with higher scores indicating better short-term memory performance; scores ranged from 0 to 16.

Backward digit span: This task was used to assess working memory (e.g., Wechsler, 1997). Participants were presented with seven sets of number sequences containing two sequences per set, in order of difficulty. The sequence length ranged from two digits in Set 1 to eight digits in Set 7. In a trial, participants were presented with a fixation cross (for 1 s), followed by each number in the sequence (1 s for each number), and then a response screen, where they were asked to type the number sequence in the reverse order. After the response, participants saw a blank screen for 1 s before the next trial began. The task ended if two sequences in a set were recalled incorrectly. The number of correctly recalled sequences was totaled, with higher scores indicating better working memory performance; scores ranged from 0 to 14.

Deary–Liewald choice reaction time: This task was used to assess processing speed (Deary et al., 2011). Participants were presented with four on-screen squares in a horizontal line in a randomized order. In a trial, a target “x” appeared in one of the four squares, and the participant used their number keys to indicate which box the target appeared in, where 1 indicated the box furthest left and 4 indicated the box furthest right. The intertrial interval varied between 1 and 3 s, and there were 40 trials in total. The response time for when the target position was identified was recorded to provide a mean reaction time. The mean was reversed (i.e., raw score × −1) prior to calculating the global cognition composite so that better reaction time performance was indicated by higher numbers to be consistent with the other cognitive measures.

Stroop color–word test: This task was used to assess executive function (Cohen et al., 1990; Stroop, 1935). The task consisted of three conditions, each containing 48 trials: words only (W), colors only (C), or color-words (CW), resulting in 144 trials in total, with trials presented in condition blocks. In the words-only condition, participants were presented with a fixation cross (1 s) followed by a word (either “RED,” “GREEN,” “YELLOW,” or “BLUE”) in white text on a gray background. The participant was instructed to recall the word they saw by pressing one of the “R,” “G,” “Y,” or “B” keys. The keys corresponded to colors sharing the same initial: R = red, G = green, Y = yellow, and B = blue. In the colors-only condition, participants were presented with the repeated letter “X” in either red, green, yellow, or blue text. Participants were instructed to recall the color of the Xs by pressing one of the “R,” “G,” “Y,” or “B” keys. In the color-word condition, participants were presented with the color word (either “RED,” “GREEN,” “YELLOW,” or “BLUE”) printed in incongruent colored text (e.g., the word “BLUE” printed in red color). Participants were instructed to recall the color of the text, not the word itself, by pressing one of the “R,” “G,” “Y,” or “B” keys. An interference score was calculated using a method adapted from Golden (1978). First, the number of correct responses out of a possible 48 in each condition was calculated (i.e., W, C, CW), and then the predicted color–word (PCW) score was calculated as

$$\mathrm{PCW}=\frac{48}{\left(\frac{\left(\left(48\times \mathrm{W}\right)+\left(48\times \mathrm{C}\right)\right)}{\left(\mathrm{W}\times \mathrm{C}\right)}\right)}$$ (1)

The PCW value is then subtracted from the participant's score in the incongruent color–word condition to provide an interference score, with higher scores indicating better ability to inhibit interference: Interference score = CW–PCW.

Auditory function: Auditory function was measured using two assessments: (a) Speech, Spatial and Qualities of Hearing scale short version (SSQ-12; Noble et al., 2013) and (b) an online SPiN test, based on the Bamford–Kowal–Bench Speech-in-Noise test (BKB-SIN, Etymotic Research). The scores calculated within each task were standardized (z scored) and then totaled to provide a composite score, following the preregistered protocol. Higher scores indicate better auditory function.

We estimated test–retest reliability for the composite auditory function scores across the 12 time points of data collection with ICCs in R using “psych” (Revelle, 2024). The estimated agreement was .83, 95% CI [.80, .86], and the estimated consistency was .83, 95% CI [.80, .86]. The auditory function measure was found to have good internal consistency (Koo & Li, 2016).

Subjective hearing ability: Subjective hearing ability was measured using the SSQ-12 (Noble et al., 2013). This 12-item questionnaire assessed subjective hearing ability. Participants responded on a 10-point Likert scale, with 0 = very poor hearing ability and 10 = perfect hearing ability. The scores were averaged over all items, with better hearing ability indicated by higher scores.

SPiN: SPiN was assessed using an online behavioral test (based on the BKB-SIN, Etymotic Research). Before the task, participants were asked to adjust their volume to a level that was audible but comfortable. To do this, sample sentences were presented at the highest overall level that would be presented during the test (fixed at 70 dB HL), and participants could then manually adjust their volume in response to these sentences. Once participants were happy that the volume was at a loud but comfortable level, this volume was fixed for the entire test. The speech-in-noise stimuli consisted of target sentences from the IEEE (or Harvard) corpus spoken by a British English man, in the presence of four-talker babble. The babble was created from the IEEE sentences, all voiced by a British English man, in MATLAB (The MathWorks Inc., 2024). The Praat software application (Boersma & Weenink, 2022) was used to combine the speech with different levels of babble noise to create 10 signal-to-noise ratios (SNRs) ranging from −6 to +21 dB SNR, in 3-dB steps, with four trials at each SNR. Therefore, the task consisted of 10 blocks, each containing four trials. The trials were ordered from most easy (e.g., +21 dB SNR) to most difficult (e.g., −6 dB SNR) to represent an equivalent process as employed in the clinical standard speech-in-noise assessment (BKB-SIN, Etymotic Research), on which this online task was based. The scripts used to create the stimuli can be accessed from the associated OSF repository (https://osf.io/67rwh/).

Participants were instructed to wear headphones or earphones during the task. In a trial, participants saw a fixation cross (1 s) and then heard a sentence, after which they were asked to type the sentence in a response window. In each sentence, there were five predetermined target words, each worth a point if correctly recalled. The points awarded in each SNR block were averaged across trials to create a mean score per SNR block. The test scoring method was based on the formula employed in the BKB-SIN (Etymotic Research). This scoring formula is derived from the Tillman–Olsen method (Tillman & Olsen, 1973) and was adapted for this online task to estimate the SNR required for a person to identify 50% of target words correctly (SNR-50). This calculation is based on that used for calculating spondee thresholds in a speech-in-noise task in which the SNR increases in 2-dB steps and two key words need to be identified per trial (BKB-SIN Manual, Etymotic Research). The calculation was adapted to account for the five key words per 3-dB step in this task:

$$\mathrm{SNR}-50=21+1.5+\left(2\times Y\right)–A$$ (2)

where 21 refers to the starting SNR level, 1.5 is half the step size, 2 is the number of additional predetermined target words in each trial above the step size (i.e., 5 key words – 3-dB steps = 2), Y is the number of SNR blocks where the participant's mean score was greater than 2, and A is the sum of the participant's mean scores across all SNR blocks. The score was reversed prior to calculating the auditory function score, so that a lower SNR-50 would indicate poorer performance.

Procedure

Each participant was contacted through e-mail, where they were also asked to confirm their eligibility to participate. Data were collected remotely from the participant using online platforms that controlled the presentation of experimental stimuli and collected participants' responses: Qualtrics (Qualtrics) was used to collect self-report data, and PsychoPy3 (Peirce et al., 2019) in combination with the hosting platform Pavlovia (Bridges et al., 2020) was used to collect behavioral responses. Participants were provided with URL links to the self-report measures, as well as individual links to each of the behavioral tasks. They completed the measures and tasks in the following order: (a) self-report measures, (b) forward digit span, (c) backward digit span, (d) choice reaction time, (e) Stroop Color–Word Test, and (f) speech-in-noise test. In the case of a technical issue, participants were asked to move onto the next task while the researcher resolved the potential issue. The participant was informed that they could take breaks between but not during tasks and were asked to complete all questionnaires and tasks on the same day, where possible. The date of participation was recorded. After completing all measures, the participant was provided with follow-up dates for completing the measures again. Participants were then contacted after 4 weeks to repeat the questionnaires and tasks.

Statistical Analysis

Data preprocessing and analyses were conducted in R (R Core Team, 2022). To determine the effect of the predictors on hearing and cognitive outcomes, analyses using linear mixed-effects models were conducted in R using “lme4” (Bates et al., 2015), and p values were derived using “lmerTest” (Kuznetsova et al., 2017). To test the hypotheses, two linear mixed-effects models were conducted. Linear mixed-effects models are appropriate for the analysis of data over time. They are sometimes considered preferrable over alternatives, such as cross-lagged panel or latent change score models, due to their ability to handle missing data at random across time points (Ghisletta et al., 2015; McNeish & Matta, 2018), and reliance on fewer unknown assumptions (Lucas, 2023; Rohrer & Murayama, 2023). We report the nominal p values, but we use p < .002 as the statistical inference criteria, which reduces likelihood of Type I error by correcting the original alpha level (p < .05) over the number of hypotheses tested (n = 24).

Linear Mixed-Effects Models

Two linear mixed-effects models were conducted to investigate the effects of time and loneliness, as well as the interactions between additional variables with time and loneliness, separately on the two key outcome variables: global cognition and overall auditory function. The predictors in each of the two models were time (from Time Points 1 to 12), loneliness (a composite measure from scores on the UCLA-LS3 and the LSNS-6), and the interactions between each additional variable (age, hearing status, depressive symptoms, auditory engagement, and lifestyle engagement) with time and loneliness. The start date (i.e., the number of months since the first U.K. lockdown) was included as a covariate. The outcomes in each of the models were (a) global cognition, a composite score calculated from standardized scores on a forward digit span, a backward digit span, a choice reaction time task, and a Stroop task, and (b) auditory function, a composite score calculated from standardized scores on a measure of self-reported hearing ability (SSQ-12) and a measure of SPiN. Following best practice guidelines for linear mixed-effects analyses (Jaeger, 2008; Meteyard & Davies, 2020), the categorical predictor variables age group and hearing status were sum coded using the “memisc” R package (Elff, 2024), and all other variables, measured on a continuous scale, were standardized (sample grand mean centered and divided by sample standard deviation) to ensure they were all on the same scale. Furthermore, both models were random intercept–only models, incorporating estimation of the variance associated with random between-participants variation in intercepts. A random slope model was inappropriate because between-participants variation in the slopes of the effects of by-participant individual differences are not identifiable, given the study design (Barr et al., 2013).

Influential observations and model assumptions. Influential data points were investigated using Cook's distance to detect any data points with a Cook's distance greater than 3 times the mean Cook's distance. For the global cognition model, 109 data points (of 2,796 data points; 4.22% of the data) were flagged as influential. For the auditory function model, 145 data points (of 2,796; 5.19% of the data) were flagged as influential. We investigated the effect of the removal of influential data points by fitting models without these data. For both the global cognition and auditory function models, removal of these data had no effect on statistical interpretation of the model results. We then removed influential data points for analyses. This is because the models without influential observations are likely to be less biased, as model outcomes are not as bound to specific (influential) sample data points. Across both models, the data met assumptions for linearity, homoscedasticity, and normality of residuals, and there was no multicollinearity among the variables (variance inflation factors ≤ 1.57 for the global cognition model and ≤ 1.54 for the auditory function model).

Model fitting and comparison. To determine best fit and justify the inclusion of random effects and interaction effects across our models, we compared models by obtaining the Akaike information criterion (AIC; Akaike, 1998) for various model specifications. The AIC was used as the comparison measure, because the criterion does not rely on the assumption that the true model is among the candidate models, which some researchers argue can never be the case (Burnham & Anderson, 2004). Across all models, the outcome variable (indicated by “Y”) was either global cognition or auditory function. For the global cognition models, the lower AIC value indicated that Model 1 was a better fit (see Table 2). Therefore, the data for the full global cognition model are reported here.

Table 2.

Model specifications and comparisons.

Model	Type	Specification	AIC for outcomes (Y)
			Global cognition	Auditory function
1	Full	Y ~ (Age Group + Hearing Status + Depression + Auditory Engagement + Lifestyle Engagement) × (Time + Loneliness) + Start Date + (1 | Participant)	4593.78	2803.75
2	Main effects only	Y ~ Time + Loneliness + Age + Hearing Status + Depression + Loneliness + Auditory Engagement + Lifestyle Engagement + Start Date + (1 | Participant)	4653.39	2800.24
3	Without random effect of participant	Y ~ (Age Group + Hearing Status + Depression + Auditory Engagement + Lifestyle Engagement) × (Time + Loneliness) + Start Date	5670.51	4908.54
4	Null	Y ~ (1 | Participant)	4708.23	2899.68

Note. AIC = Akaike information criterion.

For the auditory function models, the lower AIC value indicated that Model 1 was a better fit compared to Models 3 and 4 (see Table 2). However, Model 2 offered a lower AIC than the full model. Despite this, a comparison of these two models (Model 2: main effects vs. Model 1: full model) indicated that the likelihood ratio test statistic was not significant (χ² = 16.487, df = 10, p = .087), suggesting that neither model was better able to explain more variance. Therefore, the data for the model driven by our hypotheses, the full auditory function model, are reported here.

Results

Descriptive Statistics

Tables 3 and 4 provide the means and standard deviations observed in older and younger adults for each variable of interest across the two linear mixed-effects models. These statistics are represented across Time Points 3, 6, 9, and 12.

Table 3.

Descriptive statistics for the linear mixed-effects model for the outcome of global cognition: means and standard deviations of scores by age group at quarterly time points.

Variable	Time 3		Time 6		Time 9		Time 12
	YA	OA	YA	OA	YA	OA	YA	OA
Depression	8.48 (7.57)	7.04 (6.12)	7.76 (7.63)	5.94 (5.53)	6.18 (6.11)	6.34 (5.43)	6.98 (7.38)	5.83 (4.93)
Auditory engagement	12.87 (7.58)	12.78 (7.60)	12.56 (9.57)	11.34 (8.19)	12.11 (8.52)	11.33 (7.67)	12.22 (6.93)	12.51 (8.93)
Lifestyle engagement	10.55 (8.54)	22.33 (15.76)	10.77 (10.28)	18.90 (14.47)	9.87 (9.41)	17.75 (12.30)	10.33 (8.54)	18.03 (13.90)
Loneliness (composite)	0.05 (0.87)	−0.07 (0.93)	0.08 (0.85)	−0.05 (0.93)	−0.09 (0.83)	0.00 (0.94)	0.05 (0.90)	0.00 (0.91)
Global cognition (composite)	0.22 (0.45)	−0.22 (0.62)	0.14 (0.45)	−0.10 (0.63)	0.02 (0.53)	−0.02 (0.53)	−0.02 (0.60)	0.04 (0.50)

Note. Means and standard deviations after the removal of influential outliers. YA = younger adults; OA = older adults.

Table 4.

Descriptive statistics for the linear mixed-effects model for the outcome of auditory function: means and standard deviations of scores by age group at quarterly time points.

Variable	Time 3		Time 6		Time 9		Time 12
	YA	OA	YA	OA	YA	OA	YA	OA
Depression	8.38 (7.39)	7.24 (6.24)	7.93 (7.70)	5.81 (5.42)	6.85 (7.17)	6.39 (5.57)	6.90 (6.92)	5.97 (5.00)
Auditory engagement	12.46 (7.35)	12.08 (7.42)	11.86 (8.78)	11.58 (8.24)	11.75 (8.34)	11.33 (7.79)	12.82 (8.32)	12.56 (8.91)
Lifestyle engagement	10.80 (9.61)	21.12 (13.88)	10.09 (9.73)	19.56 (14.57)	9.92 (9.15)	17.65 (12.51)	11.02 (8.84)	18.11 (13.91)
Loneliness (composite)	0.06 (0.84)	0.02 (0.92)	0.08 (0.86)	−0.03 (0.92)	−0.03 (0.86)	0.03 (0.93)	0.04 (0.90)	−0.02 (0.91)
Auditory function (composite)	0.25 (0.57)	−0.30 (0.95)	0.18 (0.63)	−0.14 (0.83)	0.21 (0.51)	−0.12 (0.93)	0.27 (0.48)	−0.15 (0.95)

Note. Means and standard deviations after the removal of influential outliers. YA = younger adults; OA = older adults.

Linear Mixed-Effects Models

Model Results

Results for the global cognition model are reported in Table 5. We calculated marginal and conditional R² according to the approach set out by Nakagawa et al. (2017), using the “performance” package (Lüdecke et al., 2021). The fixed effects explained 6.5% of the variance in the data, and 57.4% was explained by both fixed and random effects. Furthermore, semipartial R² statistics were calculated for each fixed predictor using the approach set out by Nakagawa and Schielzeth (2013), using the “r2glmm” package (B. Jaeger, 2017). Of the predictors of interest for the primary and secondary hypotheses, the interaction between age and time explained 1.1% of the variance in the data, loneliness explained 0.1% of the variance, the interaction between loneliness and depression explained 0.2% of the variance, and the interactions between loneliness and age, between loneliness and auditory engagement, and between loneliness and lifestyle engagement each explained 0.1% of the variance.

Table 5.

Linear mixed-model output detailing the added variance of each fixed-effects predictor and interaction to the outcome of global cognition.

Fixed effects	β	SE	df	t	p	d
Main effects
Age group	−.089	0.062	244.44	−1.45	.149	−0.18
Hearing status	−.112	0.070	220.15	−1.61	.109	−0.22
Depression	−.015	0.032	1867.86	−0.46	.643	−0.02
Auditory engagement	.009	0.024	2051.60	0.39	.694	0.02
Lifestyle engagement	.008	0.025	2051.85	0.31	.755	0.02
Hypothesized effects
Time	−.009	0.018	1892.72	−0.49	.624	−0.02
Loneliness	−.031	0.046	949.22	−0.69	.490	−0.05
Time × Age	.133	0.020	1918.58	6.76	< .001	0.31
Time × Hearing Status	−.026	0.019	1869.31	−1.34	.179	−0.06
Time × Depression	.005	0.016	1936.24	0.30	.763	0.01
Time × Auditory Engagement	−.018	0.017	1897.02	−1.11	.268	−0.05
Time × Lifestyle Engagement	.022	0.018	1908.34	1.25	.212	0.06
Loneliness × Age	.036	0.046	983.90	0.78	.434	0.05
Loneliness × Hearing Status	.016	0.050	840.49	0.31	.755	0.02
Loneliness × Depression	.043	0.025	1973.51	1.75	.080	0.08
Loneliness × Auditory Engagement	.032	0.024	2047.79	1.35	.178	0.06
Loneliness × Lifestyle Engagement	.029	0.023	2029.21	1.26	.207	0.06

Note. This table shows the output for the linear mixed-effects model: Global Cognition ~ (Age Group + Hearing Status + Depression + Auditory Engagement + Lifestyle Engagement) × (Time + Loneliness) + Start Date + (1 | Participant). The above summary of the coefficient estimates, as well as standard errors, were estimated using “lme4” (Bates et al., 2015), with p values calculated utilizing “lmerTest” (Kuznetsova et al., 2017). Cohen's d was estimated for each of the fixed effects using “EMAtools” (Kleiman, 2021).

Results for the auditory function model are reported in Table 6. Marginal and conditional R² values indicated that the fixed effects explained 30.5% of the variance in the data, and 83.5% was explained by both fixed and random effects. Semipartial R² statistics indicated that, of the predictors of interest for the primary and secondary hypotheses, loneliness explained 1.5% of the variance in the data, the interaction between loneliness and hearing status explained 0.7% of the variance, time explained 0.1% of the variance, and the interaction between loneliness and age explained a further 0.1% of the variance.

Table 6.

Linear mixed-model output detailing the statistical contribution of each fixed-effects predictor and interaction to the outcome of auditory function.

Fixed effects	β	SE	df	t	p	d
Main effects
Age group	.006	0.060	236.56	0.10	.922	0.01
Hearing status	−.586	0.069	225.29	−8.53	< .001	−1.13
Depression	−.032	0.021	2026.17	−1.50	.134	−0.07
Auditory engagement	.013	0.015	1957.29	0.87	.386	0.04
Lifestyle engagement	−.013	0.016	1958.33	−0.78	.438	−0.04
Hypothesized effects
Time	.028	0.012	1835.43	2.42	.016	0.11
Loneliness	−.135	0.033	1675.49	−4.04	< .001	−0.20
Time × Age	.011	0.012	1864.00	0.87	.387	0.04
Time × Hearing Status	−.005	0.012	1837.49	−0.40	.687	−0.02
Time × Depression	−.018	0.010	1863.32	−1.68	.093	−0.08
Time × Auditory Engagement	−.009	0.010	1834.96	−0.88	.381	−0.04
Time × Lifestyle Engagement	.007	0.011	1844.76	0.68	.494	0.03
Loneliness × Age	.025	0.033	1739.28	0.76	.451	0.04
Loneliness × Hearing Status	−.093	0.037	1630.03	−2.52	.012	−0.12
Loneliness × Depression	.008	0.016	2013.71	0.47	.639	0.02
Loneliness × Auditory Engagement	−.022	0.016	1948.92	−1.44	.151	−0.07
Loneliness × Lifestyle Engagement	.011	0.015	1910.05	0.74	.460	0.03

Note. This table shows the output for the linear mixed-effects model: Auditory Function ~ (Age Group + Hearing Status + Depression + Auditory Engagement + Lifestyle Engagement) × (Time + Loneliness) + Start Date + (1 | Participant). The above summary of the coefficient estimates, as well as standard errors, were estimated using “lme4” (Bates et al., 2015), with p values calculated utilizing “lmerTest” (Kuznetsova et al., 2017). Cohen's d was estimated for each of the fixed effects using “EMAtools” (Kleiman, 2021).

Primary Hypotheses

There was no significant main effect of time, β = −.009, t(1892.72) = −0.49, p = .624, nor a main effect of loneliness, β = −.031, t(949.22) = −0.69, p = .490, on cognitive function. These data do not support H1a or H1b, which predicted that global cognition would worsen with time and with increased loneliness. There was also no significant main effect of time, β = .028, t(1835.43) = 2.42, p = .016, on auditory function at the p < .002 criterion level, providing no support for H2a, which predicted that auditory function would decrease with time. There was, however, a significant main effect of loneliness, β = −.135, t(1675.49) = −4.04, p < .001, on auditory function, providing support for H2b, which predicted that auditory function would decrease with increased loneliness.

Secondary Hypotheses

There was a significant interaction effect between time and age group on global cognition, β = .133, t(1918.58) = 6.76, p < .001. The shape of the interaction is inconsistent with hypothesis H3a, which predicted that any negative change in cognition with time would be greater for older adults. Instead, we find that the negative change in cognition over time only occurs in younger adults, whereas an unexpected positive change in cognition over time is observed in older adults (see Figure 1).

Figure 1.

Marginal effects plot generated using “sjPlot” (Lüdecke, 2024) showing the predicted values (95% confidence intervals) for global cognition across time points (from 1 to 12) in younger (left-hand plot) and older (right-hand plot) adults.

Despite the differing association between time and global cognition in different age groups, the effect of the interaction between loneliness and age group on global cognition was not significant, β = .036, t(983.90) = 0.78, p = .434. These data do not support H3b, which predicted that older adults would show more negative changes in cognition (than younger adults) with increased loneliness. There were also no significant interaction effects between time and hearing status, β = −.026, t(1869.31) = −1.34, p = .179, or between loneliness and hearing status, β = .016, t(840.49) = 0.31, p = .755, on global cognition, providing no support for hypothesis H3c or H3d, which predicted that older adults with HL would show increased negative changes in cognition with increased time and increased loneliness.

Similarly, in the model of auditory function outcomes, we observed no significant interaction effects between time and age, β = .011, t(1864.00) = 0.87, p = .387, or between loneliness and age, β = .025, t(1739.28) = 0.76, p = .451, providing no support for hypothesis H4a or H4b, which predicted that older adults would show poorer auditory function with increased time and increased loneliness. There was also no significant interaction effect between time and hearing status, β = −.005, t(1837.49) = −0.40, p = .687, on auditory function, providing no support for hypothesis H4c, which predicted that older adults with HL would show increased negative changes in auditory function with increased time. Furthermore, using p < .002 as the inferential statistical criterion, there was no significant interaction effect between loneliness and hearing status, β = −.093, t(1630.03) = −2.52, p = .012, on auditory function, providing no support for hypothesis H4b, which predicted that older adults with HL would show increased negative changes in auditory function with increased loneliness.

For depressive symptoms, we found no significant interaction effects between depression and time as well as between depression and loneliness on cognitive function (ps > .002). Similarly, we found no significant interaction effects between depression and time as well as between depression and loneliness on auditory function (ps > .002). These data do not support hypotheses H5a–H5d, which predicted that participants with increased depressive symptoms would show increased negative changes in cognitive and auditory function with increased time and increased loneliness.

For auditory engagement, we found no significant interaction effects between engagement in auditory activities and time as well as between engagement in auditory activities and loneliness on global cognition (ps > .002). Similarly, we found no significant interaction effects between engagement in auditory activities and time as well as between engagement in auditory activities and loneliness on auditory function (ps > .002). These data provide no support for hypotheses H6a–H6d, which predicted that participants with lower engagement in auditory activities would show increased negative changes in cognitive and auditory function with increased time and increased loneliness.

For lifestyle engagement, we found no significant interaction effects between engagement in lifestyle activities and time as well as between lifestyle engagement and loneliness on global cognition (ps > .002). Similarly, we found no significant interaction effects between engagement in lifestyle activities and time as well as between lifestyle engagement and loneliness on auditory function (ps > .002). These data do not support hypotheses H7a–H7d, which predicted that participants with lower engagement in lifestyle activities would show increased negative changes in cognitive and auditory function with increased time and increased loneliness.

Exploratory Analyses

We also report whether any of the predictor variables or covariates included in the linear mixed-effects models showed a significant main effect on either global cognition or auditory function. Despite initially not hypothesizing any main effects of these predictors (age group, hearing status, depression, auditory engagement, and lifestyle engagement), they may affect hearing or cognitive outcomes. We also included how many months had passed since the first lockdown when each person participated as a covariate, which we will explore as a main effect.

For the linear mixed-effects model predicting global cognition, none of these main effects were statistically significant (see Table 5). For the linear mixed-effects model predicting auditory function (see Table 6), there was a significant main effect of hearing status, β = −.586, t(225.69) = −8.53, p < .001, Cohen's d = −1.13, whereby older adults who reported having HL showed significantly poorer auditory function than those who did not report having HL (both older and younger adults).

Discussion

Primary Hypotheses: The Effect of Time and Loneliness on Cognitive and Auditory Function

We observed no significant effect of time or loneliness on global cognitive function. This finding was unexpected because this research took place during a time of reduced social contact, which was predicted to affect both the time and loneliness variables and, thus, cognitive performance. Previous research indicates that maintaining social contact is preventative against dementia through maintaining and strengthening cognitive reserve (Livingston et al., 2024). For example, increased contact with friends is associated with better cognitive outcomes on a global cognitive function measure (Sommerlad et al., 2019). The contradictory findings may be in part due to differences between measurements of social contact employed in previous research and our measure of self-reported loneliness. The loneliness composite we employed comprised both social and emotional loneliness, considering both perceptions of social networks and emotional support. A previous meta-analysis investigating the associations between loneliness and risk of dementia found that risk of dementia was increased with poor social engagement and poor social networks, but not with increased loneliness (Penninkilampi et al., 2018). Considering this, our use of a composite self-report measure that comprised both these components (social and emotional loneliness) may have diluted our findings, obscuring any trends or contributions of the individual subcomponents.

Furthermore, it is possible that the timeframe employed in this study time (i.e., our 48-week testing period) was not long enough to capture the effect of social distancing or loneliness on cognitive outcomes. In another study, relationships between loneliness and all-cause dementia were observed in a 20-year follow-up (Sundström et al., 2020). Additionally, in previous studies, a clinical measure of dementia or Alzheimer's disease was employed (Livingston et al., 2024; Penninkilampi et al., 2018; Sundström et al., 2020). It is possible that associations between loneliness and cognition only occur in populations with clinically significant memory declines, which were not captured within our research. For example, in a meta-analysis, loneliness was found to be associated with increased risk of Alzheimer's disease and dementia but was not associated with mild cognitive impairment (Qiao et al., 2022).

Interestingly, our findings are in line with a similar previous study conducted during the COVID-19 pandemic. The researchers found no significant associations between loneliness (as measured similarly with both the UCLA-LS3 and the LSNS-6) and behavioral tests of cognitive performance (Nogueira et al., 2022). However, they did observe significant associations between loneliness and self-reported cognitive function, which may indicate that participants perceived more subtle changes in their memory during the pandemic that were not sensitive to behavioral testing. The relationship between psychosocial factors, including feelings of loneliness, and cognition is clearly complex. Previous researchers have suggested that the association may be bidirectional (Yin et al., 2019) or that cognition may affect loneliness outcomes but not the other way around (McHugh Power et al., 2020) or may only occur significantly in specific populations (Zhou et al., 2019).

We observed a significant main effect of loneliness on auditory function; increased loneliness was associated with poorer auditory function. Associations between social factors, loneliness, and hearing difficulties are commonly reported (Bott & Saunders, 2021; Shukla et al., 2020). HL is thought to increase perceptions of loneliness through reduced social contact due to the demands of coping in challenging auditory environments. However, the effect of restricted social contact or enforced isolation on hearing outcomes is less well known; the pandemic could have theoretically exacerbated this relationship. The pandemic listening environment may have been incredibly challenging, due to increases in distance, use of face coverings (Tofanelli et al., 2022), and reliance on online communication. These factors may have increased listening difficulty and social withdrawal leading to increased auditory deprivation. Also, poorer auditory quality reduces the emotional information conveyed through the speech to the listener. Indeed, social distancing has been found to impact quality of communication and connection with others (Wood et al., 2024). In line with the “use it or lose it” view, a lack of auditory stimulation may affect auditory functioning.

However, we did not observe associations between time and auditory function, indicating that the time course of the pandemic, captured in this study, did not exacerbate hearing difficulties. It is possible that the pandemic created a unique situation in which some individuals felt speech understanding was easier or not vastly affected, which may have affected the self-reported part of our auditory composite measure. In another study, participants with cochlear implants felt less lonely and less isolated at home in a more manageable auditory environment, and they reported better speech understanding with little effort during the pandemic (Dunn et al., 2021).

The presence of an effect of loneliness on auditory functioning, but not cognitive functioning, is interesting, given that some previous research indicates a relationship between feeling lonely and poorer cognition (Cardona & Andrés, 2023). However, it is possible that if previous research employs cognitive assessments in the auditory modality (as is traditional for standardized cognitive assessments, e.g., MoCA and MMSE), then outcomes may be affected by hearing acuity, leading to overestimation of cognitive decline or poorer cognitive performance due to misheard stimuli or instructions rather than cognitive factors (Füllgrabe, 2020a, 2020b; Goodwin et al., 2021). However, this study employed cognitive assessments in the visual modality only, enabling the isolation of cognitive ability from hearing acuity or speech perception.

Secondary Hypotheses: The Interaction Effects Between Time or Loneliness and Additional Variables of Age, Hearing Status, Depression, and Engagement in Auditory and Lifestyle Activities on Cognitive or Hearing Function

In this study, we observed no significant interaction effects between time or loneliness and additional variables of age, hearing status, depression, and engagement in auditory and lifestyle activities on auditory function. It is notable that while the interaction between hearing status and loneliness was not significant at the p < .002 level, it would have reached significance at the p < .05 level. The trend indicates that individuals with higher self-reported loneliness showed poorer auditory function. In exploratory correlations, the trend was strongest among older participants with HL (r = −.33) but still present among the remaining sample with self-reported normal hearing (r = −.16). However, the effect size for this interaction was very small (Cohen's d = 0.12), and the interaction explained only 0.7% of the variance data.

We also found no significant interaction effects between time or loneliness and additional variables of hearing status, depression, and engagement in auditory and lifestyle activities on global cognition. Global cognition was also not affected by any interaction effects between loneliness and age. However, there was a significant interaction between time and age, indicating interesting differences in the effect of time on cognitive performance across the different age groups. The effect size for this interaction effect on global cognition was small–moderate (Cohen's d = 0.31). Whereas older adults showed improved performance over time, younger adult performance worsened. A possible explanation for this is motivational differences in younger and older listeners, which may affect how they engage in cognitive tasks.

There is evidence from previous research that age-related differences in motivation affect effort investment in cognitive tasks (Ennis et al., 2013). The authors found that older adults were more influenced by the importance of performing well on cognitive tasks, relative to younger adults. Several reasons may underpin such age-related differences in task motivation or in motivation to participate in research more generally. In one study, older adults were found to be motivated by the desire to understand more about their health and gain cognitive benefit (Carr et al., 2022); such motivators may arise due to increased concerns about health and memory as we age. Researchers also perceive motivational differences among participants; a surveyed group of researchers (n = 88) believed older, versus younger, adults to be more motivated participants who take part to learn about their cognitive health, to further science, and out of curiosity, rather than for course credits or monetary compensation favored by younger adults (Ryan & Campbell, 2021). Of course, such generalizations do not apply across all older and younger adults, with many factors influencing motivation. Indeed, age, employment status, and previous participation have been found to underpin the motivations to take part in research (Carr et al., 2022). Importantly, psychological factors also affect motivation; depressive symptoms are found to negatively impact reward-seeking and motivational behavior (Franzen & Brinkmann, 2016). This is important as previous research suggests that younger adults consistently reported increased psychological distress and reduced well-being during the pandemic, compared to older adults (Best et al., 2023). These age-related differences may explain the differences in cognitive performance as well as increased attrition rate observed in the younger cohort involved in this research study. Compared to the older adult sample, of which only 10 did not participate at Time Point 12, 65 younger adults dropped out by Time Point 12.

Exploratory Analyses: The Main Effects of Age, Hearing Status, Depression, Engagement in Auditory and Lifestyle Activities, or Months Since Lockdown on Cognitive or Auditory Function

Of these variables, there was only a significant main effect of hearing status on the outcome of auditory function, wherein older adults who self-reported having ARHL displayed poorer auditory function than their peers, and younger adults, who did not report having HL. The effect size for this hearing status effect on auditory function was large (Cohen's d = 1.13). This indicates that the online measures of auditory function may be sensitive to detecting hearing difficulty.

Limitations and Future Directions

Understanding the associations between psychosocial factors such as loneliness and age-related changes in hearing and cognitive function is important for identifying individuals at risk of loneliness and health declines as well as to design appropriate interventions. This study investigated the effect of time exposed to the COVID-19 pandemic–related social restrictions on cognitive and auditory outcomes.

The United Kingdom experienced vast changes across the pandemic period, including local lockdowns, tiered restrictions, and incentives such as the “Eat Out to Help Out Scheme” (HM Revenue & Customs, 2020), as well as individuals engaging in differing levels of compliance. Additionally, participants will have likely been affected differently depending on whether they were experiencing COVID-19 symptoms, as well as variances in their living, work, and study situations across the period. As such, there is variation across the study, which may have affected the linearity of the time variable and the outcomes. Additionally, the study may be limited by reliance on a self-reported measure of social and emotional loneliness. Admitting to feeling lonely can be incredibly stigmatizing (Department for Culture, Media & Sport, 2023), thus leading to biases in the measure.

Furthermore, it is possible that results were biased through the recruitment of a self-selected participant sample consisting of active and socially engaged older adults who potentially feel less impacted by pandemic-related restrictions or guidance. Factors such as computer literacy, social contacts, or socioeconomic position (SEP) may play a role in mitigating feelings of loneliness, isolation, or even cognitive decline in our sample (Cotten et al., 2013; Fakoya et al., 2020). The online nature of this research study required that participants had access to e-mail, internet connection, and a level of technical skill and digital literacy. It is probable that the participants were comfortable technology users and relatedly experienced higher levels of online social connection and auditory stimulation. It is important to note that the findings we observed may not generalize to a population of older adults with poorer digital literacy or reduced access to technology; such individuals were likely more significantly affected by pandemic-related restrictions that may have resulted in changes to their hearing or cognitive function, which we were not able to capture in this study. This highlights a potential issue for online research, in that sample recruitment may be biased to include participants who are online regularly, excluding those from different social or economic backgrounds. Importantly, research suggests that SEP and health inequalities play a critical role in hearing health, with lower SEP significantly related to increased HL (Tsimpida et al., 2019).

A further limitation that resulted from the self-selected sample is that most of the participants were female, and thus, assigned sex was not balanced across the sample. This factor was also not included in analyses. Evidence suggests that the prevalence of HL is higher in males, and importantly, both engagement with hearing health care or assistive devices and the effects of HL on other health outcomes may vary by sex (Mick et al., 2014; Reavis et al., 2023). To understand both sex and gender differences in hearing and hearing health outcomes, future researchers may wish to account for these factors. In future studies, researchers may also wish to account for SEP and additional biases within the participant sample. Furthermore, research that includes an objective measure of social connection through quantifying social interactions in the real world would provide the next step in understanding the effect of socialization on hearing and brain health. Additionally, researchers may want to consider the effect of positive social interventions in diverse populations on both cognitive and auditory outcomes to best understand future pathways for intervention for loneliness and associated health conditions in older age.

Conclusions

This study sought to understand the effect of loneliness and isolation experienced during a global pandemic on sensory and cognitive function across age ranges. During the COVID-19 pandemic, the public experienced social distancing, enforced isolation, and restricted means of communication, creating a changed auditory environment. Previous research suggests that reduced levels of auditory stimulation may affect both cognitive and auditory processing; however, in this sample, we did not find consistent significant effects of such psychosocial factors on hearing and cognitive outcomes.

Instead, cognitive performance was found to be affected only by interactions between participants' age and time (improving over time in older adults and decreasing over time in younger adults). Auditory function, however, was associated with loneliness; across all time points, poorer auditory function was related to increased self-reported loneliness. Auditory function was also affected by participants' hearing status (poorer auditory function was observed in older adults who self-reported having HL, compared to participants without HL).

Aside from the association between loneliness and auditory function, these data appear to show a lack of support for our preregistered hypotheses that auditory deprivation and reduced socialization impact hearing and cognitive function. Nevertheless, the patterns observed in the data may be underpinned by motivational differences, learning effects, sample biases, or a lack of statistical power. Interesting trends indicate an effect of the relationship between loneliness and hearing status on auditory function, wherein the correlation between increased loneliness and poorer auditory function is greater for older adults with HL. Future research may wish to investigate these effects further, over a greater period, to understand how this relationship manifests. This would provide insight into how social and psychological factors relate to both cognitive and hearing function, to identify intervention pathways targeting HL and cognitive decline.

Author Contributions

Kate Slade: Conceptualization (Lead), Data curation (Lead), Formal analysis (Lead), Investigation (Lead), Methodology (Lead), Visualization (Lead), Writing – original draft (Lead). Robert Davies: Methodology (Supporting), Formal analysis (Supporting), Writing – review & editing (Supporting). Charlotte R. Pennington: Methodology (Supporting), Writing – review & editing (Supporting). Christopher J. Plack: Conceptualization (Equal), Methodology (Supporting), Supervision (Supporting), Writing – review & editing (Supporting). Helen E. Nuttall: Conceptualization (Equal), Funding acquisition (Lead), Methodology (Equal), Project administration (Lead), Supervision (Lead), Writing – review & editing (Equal).

Data Availability Statement

All experimental scripts, stimuli, the study preregistration, and research data are openly available on the Open Science Framework at https://osf.io/67rwh/.

Funding Statement

This study was funded by the Biotechnology and Biological Sciences Research Council (Grant BB/S008527/1, awarded to Helen E. Nuttall). Christopher J. Plack was supported by the National Institute for Health and Care Research Manchester Biomedical Research Centre (Grant NIHR203308).