Background

Frailty is a common clinical syndrome in older adults that is associated with adverse health outcomes and increased mortality over time [1]. There is ongoing debate as to how best to define frailty, but most approaches are based on declining body mass, strength, endurance, balance, gait speed, and daily activity [2, 3]. Several well-validated frailty models have been established and many screening tools are available for use in clinical practice and research [1, 4]. One popular approach is based on the model that arose from the Cardiovascular Health Study (Fried criteria). It identifies an individual as frail when three or more of the following criteria are present: unintentional weight loss, self-reported exhaustion, muscle weakness, slow walking speed, and low physical activity [2]. Nonetheless, the lack of consensus on how best to define frailty may have contributed to the rather discrepant prevalence estimates of frailty in later life, ranging from 4 to 59 % [5, 6] in the community and to nearly 70 % among institutionalized older people [7].

Recent research findings have shown that frailty is not a static syndrome, and many of its signs and symptoms can be reversed, at least partly, with appropriate interventions, such as physical activity, diet and review of medications [8,9,10]. These promising results have led to renewed efforts to uncover other potentially modifiable risk factors associated with frailty, and hearing loss is one of them. Hearing loss has a high prevalence among aged people. Age related hearing loss (ARHL) is the most common cause of hearing loss, affecting 40 % of the population aged 65 years and over [11]. Data from the 2005–2006 National Health and Nutritional Examination Survey in the USA suggested a 63 % prevalence of hearing loss in adults aged 70 years and above [12]. Notwithstanding its high prevalence, the proportion of people with ARHL who receive appropriate treatment remains low [13]. A UK study reported that only one-third of hearing impaired adults older than 70 years who could have benefit from hearing aids actually owned them, and among those with hearing aids, 10 % never used them [14].

There are a number of potential pathways linking hearing loss to frailty [15]. Hearing loss is reported to increase the risk of incident falls in older adults [13, 15, 16], and a significantly great odds of depression has also been observed [17]. Hearing loss has also been associated with poor physical function [18, 19] and reduced activities of daily living [19, 20] among older adults. Panza et al. [21] suggested that age-related hearing loss and frailty, as well as cognition, share pathophysiological pathways, and that the use of hearing devices could potentially alleviate frailty. However, opposing evidence suggesting there is no association between these two also exist [22, 23] and the available evidence is far from conclusive.

Given the negative health outcomes associated with frailty and the high prevalence of hearing loss in this age-group, it seems important to determine if this association is likely to be causal. An initial step in this process would be to determine if there is an association between hearing loss and frailty by reviewing currently available studies investigating this. Thus, this systematic review and meta-analysis aimed to examine the association between hearing loss and frailty in later life. Specifically, we wanted to determine if in older adults aged 50 years or older, does hearing loss increase the risk of frailty compared with normal hearing? We hypothesised that hearing loss would increase the risk of frailty in older adults aged 50 years or over.

Methods

Study Eligibility

This review focused on observational studies and was limited to papers published in English. We searched cross-sectional, cohort and case-control studies investigating the association between hearing loss and frailty. Studies were included if: (1) the subjects of the study were aged 50 years or above; (2) hearing loss was self-reported or measured by audiology tests; (3) the presence of frailty was ascertained through the use of a valid measure. Studies were excluded if: (1) the type or cause of individuals’ hearing loss was reported and the hearing loss may not be permanent, for example, it was conductive or caused by other health issues such as tumour or injury; (2) the study did not offer a clear description of hearing and frailty assessment.

Search Strategy

We conducted a comprehensive literature search of Medline, PsychINFO, Embase, and the Cochrane Collaboration databases from 2000 to February 2021. This time limit was set because studies on frailty emerged prominently after 2000 according to previous review articles [4, 5, 24]. The following search terms and Boolean operators were used to source articles: (“Hearing impairment” OR “Hearing loss” OR deafness OR “hearing deficit”) AND (frailty OR frail OR morbidity OR mortality OR fall OR “physical function”). Additional studies were sought from article reference lists, review articles, conference abstracts, Google Scholar and Open Grey.

Study Selection

Identified citations from the electronic and manual searches were screened for eligibility by RT and full text documents of potentially eligible articles were retrieved. These articles were further assessed for final eligibility by all authors.

Data Extraction

Data extracted from the final articles included: authors, sample characteristics, study characteristics, measurement and criteria of hearing loss and frailty, and quantitative data for the purpose of meta-analysis. Authors were contacted when necessary if study information or data were not reported in the published articles.

Quality Assessment and Risk of Bias

The quality of the articles were assessed using the Newcastle-Ottawa Scale (NOS)[25]. This scale was developed for use in cohort and case-control studies. The scale is adapted for cross-sectional studies, according to the work of Modesti et al. [26]. Articles were assessed according to eight items that are categorized into three domains, including selection, comparability and outcome. Articles were classified as good, fair or poor quality according to their score in each section. Good quality was defined as having three or four stars in the selection domain (maximum five), one or two stars in the comparability domain (maximum two), and two or three stars in the outcome domain (maximum three); Fair quality was defined as having two stars in the selection domain, one or two stars in the comparability domain, and two or three stars in the outcome domain; Poor quality was defined as having one or zero stars in the selection domain, or zero stars in the comparability domain, or one or zero stars in the outcome domain. Articles rated as having poor quality were excluded from the review. Quality assessment was conducted by RT and agreed by all authors.

Statistical Analyses

Stata 16.1 software was used for the meta-analyses (StataCorp LLC, 2019) using the ‘metan’ command. We used the measure of risk and respective 95 % confidence interval of each individual study to calculate the overall summary estimate of risk for all studies. Results that were adjusted for confounders were used when available. In this review, we use the generic term ‘risk ratio’ to describe the odds ratio, relative risk or hazard ratio reported by the individual studies. Separate analyses were undertaken for different study designs (i.e., cross-sectional and cohort). A random effects model was chosen because of study heterogeneity. Most studies reported hearing and frailty status in a dichotomous manner. Some studies reported these in several categories according to their severity, such as normal, mild, moderate or severe for hearing loss, and normal, pre-frail or frail for frailty. For these studies, data of the most severely impaired category was used for the purposes of the meta-analyses.

Heterogeneity and Sensitivity Analyses

Q and I2 tests were conducted to examine heterogeneity. Significant Q test results (P < 0.10) provides evidence of heterogeneity. I2 test was used to quantify heterogeneity. Heterogeneity with I2 values lower than 40 % is generally considered low, 41-60 % is considered medium, and I2 values over 60 % is considered high. Heterogeneity was then explored by conducting subgroup analyses and sensitivity analyses. Sensitivity analysis was conducted by repeating the meta-analysis while sequentially removing individual studies.

Results

Study Selection

Figure 1 summarises the results of the systematic search of the literature. Five thousand and thirty (5030) articles were screened for eligibility, but only 16 fulfilled criteria for inclusion and these resulted in 13 cross-sectional and 5 longitudinal datasets (two studies reported both cross-sectional and longitudinal data [27, 28]). Among these, two studies [29, 30], and the longitudinal component of one study [27] did not include sufficient data for meta-analysis. Fourteen studies were included in the quantitative analysis with 12 sets of cross-sectional data (including 12,313 participants) and three sets of longitudinal data (including 3042 participants).No case-control studies met our inclusion criteria.

Fig. 1
figure 1

Flow chart showing results of the literature search

Full size image

Study Characteristics

Table 1 summarises the 16 included studies. Among them, nine studies identified hearing loss through simple self-reported questions [28, 31,32,33] (e.g. “Is your hearing excellent, very good, good, fair or poor?”), questionnaires [23, 34, 35], or subjective judgement of examiners [22, 27]. The other seven studies used validated audiology methods, including pure-tone audiometry [13, 36], whisper test [29, 30, 37, 38] and finger friction test [39], to measure hearing.

Table 1 Summary of included studies
Full size table

The studies used various methods to define frailty. Most (n = 13) used the Fried criteria [2] to define frailty status, but the criteria for each component were slightly modified, as described in Table 2. Kmail et al. (2016) [13] used two simple assessments to define physical frailty, Sable-Morita et al. [39] used the Kihon Checklist (KCL) score, and Doba et al. [35] used the Canadian Study for Health and Aging Clinical Frailty Scale (CSHA-CFS). The odds ratio of four studies [27, 33, 36, 38] used in the meta-analysis were calculated by us using demographic data reported by the studies. Another four studies [29, 30, 35, 37] reported only unadjusted statistics (i.e., not adjusted for covariates). All other eight studies [13, 22, 23, 28, 31, 32, 34, 39] were adjusted for age, gender, and other additional factors according to their study design.

Table 2 Summary of criteria used for Fried Phenotype components
Full size table

Follow-up periods for the five longitudinal studies varied, with 10 years of follow up reported by Kamil et al. (2016) [13], five years by Doba et al. [35] and Cheung et al. [27], four years by Liljas et al. [28] and one year by Lorenzo-López et al. [30].

Three studies were conducted in the USA [13, 31, 32], six studies based in Asia [22, 23, 27, 35, 37, 39], four in Europe [28, 30, 34, 36], one each in Turkey [29] and Brazil [38], and one study [33] was a multinational study involving five countries. Most of the participants were community dwelling aged people, while Sable-Morita et al. [39] recruited participants from a diabetes mellitus outpatient clinic, and Cheung et al. [27] recruited participants through aged-care services. Herr et al. [33] did not specify the recruitment setting.

Fourteen studies were included in the meta-analysis. Among them, seven [22, 23, 28, 31, 32, 35, 37, 39] reported odds ratios, while Kamil et al. (2016) [13] reported hazard ratio and Buttery et al. [34] reported relative risk ratio. We calculated the odds ratios of four studies [27, 33, 36, 38]. Two studies were not suitable for meta-analysis. Cakmur [29] completed a cross-sectional study of 168 older participants in rural Turkey, aiming to investigating the prevalence of frailty in this area and its correlates. Hearing loss was reported to have a statistically significant association with frailty, but the specific data were not described. Lorenzo-López et al. [30] reported the results of a longitudinal study involving 749 community-dwelling older adults in Spain and followed their frailty status for one year. The frailty status transitions (progressed, regressed, no change or death) and their associated factors were reported, but these data could not be included in the meta-analyses because it did not report the risk of incident frailty. In this study, hearing impairment at baseline was associated with higher risk of experiencing worsening of frailty over time. Cheung et al. [27]reported that hearing impairment is associated with worse frailty status transition over a 5-year period. These longitudinal data are not included in the meta-analysis, but we calculated the odds ratio for the cross-sectional data and included this in the meta-analysis.

Risk of Bias Within Studies

Study quality is summarised in eTable 2. According to the Newcastle-Ottawa Scale (NOS) [25] the quality of the studies included in the review were good or fair. The quality of two studies were considered as fair. No cross-sectional studies reported characteristics of the non-responders, but two [23, 33] reported the response rate. One possible reason for this could be that the data used in most cross-sectional studies were from previous studies and the samples were restricted to participants with information available to address the aims of the studies. Thus, non-response rate was not available. Participants of Sable-Morita et al. [39] were outpatients with diabetes recruited from a single clinic, which limits the generalisability of the findings. The sample size of this study, and of Naharci et al. [32], Cheung et al. [27], Closs et al. [38] and Mohd Hamidin et al. [23] was relatively modest, with 283, 484, 165, 255 and 279 participants, respectively. Among the longitudinal studies, Lorenzo-López et al. [30] reported that 28.3 % of participants were lost to follow-up. Liljas et al. [28] used prospective data from a national study and only participants with complete data were included in the study. In Doba et al. [35], assessment measures were self-rating. The forms were later reviewed by trained nurses, but the validity of the assessment of hearing loss seemed less certain than that of a structured interview, test or health record. Kamil et al. (2016) [13] reported 35 participants were frail or severely frail, but these people were not excluded from follow-up, so that prevalent cases may have contaminated the sample.

Meta-analysis of the primary outcome - hearing loss and frailty

The overall association between hearing loss and frailty for cross-sectional studies included in this meta-analysis was associated with a RR of 1.87 (95 %CI 1.63–2.13). For longitudinal studies the RR was 1.56 (95 %CI 1.29–1.88) – see Fig. 2.

Fig. 2
figure 2

Forrest plot showing overall risk ratio of cross-sectional studies and longitudinal studies

Full size image

Risk of bias across studies

There was high (I2 = 75.2 %) and low (I2 = 15.9 %) heterogeneity present in the meta-analysis of cross-sectional and longitudinal studies, respectively. The funnel plots indicated that publication bias may have been present and that positive studies may have been more likely to appear in print (figure not shown).

Sensitivity Analyses

Sensitivity analyses were conducted for cross-sectional studies to determine the impact of individual reports on the outcome of the meta-analysis. The meta-analysis was repeated by sequentially removing individual studies – the results are depicted in eFigure 1. These were not performed for longitudinal studies due to the small number of publications available. No single study had a significant effect on the overall effect, although removal of Liljas et al. [28] did result in a larger summary effect estimate (RR 2.21, 95 %CI 1.84–2.64). The heterogeneity remained high, whichever study was removed.

Subgroup analyses

Subgroup analyses were conducted for cross-sectional studies according to the methods used to assess hearing, the sample size and the quality of studies. Longitudinal studies were not included in these additional analyses because of the limited number of studies.

Hearing assessment method

Among cross-sectional studies, Castellana et al. [36], Closs et al. [38], Ng et al. [37] and Sable-Morita et al. [39] used validated audiology assessment method to identify hearing loss, while in the rest eight studies hearing ability were either self-reported [23, 28, 31,32,33,34] or subjectively judged by examiners [22, 27]. When separately analysed (Fig. 3), both groups had similar pooled effect size, but the group using an audiology assessment method had a much smaller heterogeneity when compared with the other group (RR 1.83, 95 %CI 1.46–2.31, I2 = 47.1 % vs. RR 1.88, 95 %CI 1.60–2.22, I2 = 81.9 %).

Fig. 3
figure 3

Subgroup analyses of cross-sectional studies according to the methods used to assess hearing ability

Full size image

Sample size

The association between hearing loss and frailty remained significant regardless of study size although heterogeneity did vary across the three sample size categories (eFigure 2). The group of study with more than 2000 participants had the smallest heterogeneity. The effect size of this association seemed more pronounced for the studies with participants less than 1000. However, only a small number of studies were available for each category.

Quality of studies

Estimates of study quality did not seem to affect the results of the meta-analysis (eFigure 3), although heterogeneity was higher in those studies rated as ‘good’ quality compared with ‘fair’ quality (I2 = 78.9 % vs. I2 = 0 %). The summary effect was also higher in the ‘fair’ quality studies, but it was associated with a less precise effect estimate (RR 2.38, 95 %CI 1.47–3.86 vs RR 1.83, 95 %CI 1.59–2.10).

Discussion

The aim of this systematic review and meta-analysis was to determine if there is an association between hearing loss and frailty by reviewing currently available studies. We found that the risk of frailty is greater among older people with than without hearing loss, regardless of study design (cross-sectional or cohort), method of assessment of hearing and frailty, sample size or study quality.

Apart from the positive association between hearing loss and increased risk of frailty, other findings merit mention. Kamil et al. (2014) [31] showed that self-reported hearing loss was associated with frailty in women but not men. However, their subsequent longitudinal study in 2016 [13] using objective hearing assessment found increased risk for both men and women with moderate or greater hearing loss. They also reported that the use of hearing aids was not significantly associated with decreased frailty risk [31]. Sable-Morita et al. [39] reported higher risk of frailty in hearing impaired older adults with diabetes. They also indicated that previous studies suggested that people with diabetes may have higher prevalence of hearing loss than those without, although the physiological mechanisms supporting the purported association remain unclear. Naharci et al. [32] evaluated the association between self-reported hearing loss and frailty among four ethnic groups and observed an increase in risk only among Afro-Caribbeans, but not in African Americans, Hispanics or European Americans. Their sampling and assessment strategies (e.g., possible unbalanced distribution of severe hearing loss in the samples) could explain their results.

There are a number of strengths and limitations worth considering in this review. The possible publication bias suggests that the effect of the association between hearing loss and frailty may have been overestimated. Limiting our literature search to English language may also raise questions regarding the generalisability of the findings. We were only able to include 14 publications with 15 separate datasets (total number of participants = 13,959) in the meta-analyses, thereby limiting the power and generalizability of the study. Nonetheless, most studies included in this review had good methodological quality and our sensitivity and subgroup analyses suggest that the observed associations are most likely robust.

Heterogeneity, especially clinical heterogeneity, needs to be considered. Hearing ability in most studies were self-reported or measured by simple screening test such as whisper test or finger friction test. The self-reported degree of hearing impairment tends to be underestimated by middle-aged to older adults [40], so that false negatives may have contaminated the samples. Only two studies used pure-tone audiometry test, which provides an accurate result of individual’s hearing thresholds. Also, most of the studies we reviewed did not report information about the degree of hearing loss. The results would have been more informative if participants within the same degree of hearing were analysed separately, as the effect of hearing loss on the risk of frailty may be ‘dose-dependent’. Kamil et al. (2016) [13] is one of the two studies that used pure-tone audiometry, reporting a 11 % increased risk of frailty with per 10 dB increase in hearing thresholds. Future studies should consider excluding participants with hearing loss caused by reasons other than presbycusis and distinguishing participants whose hearing loss has/has not been treated.

Likewise, the heterogeneity associated with the assessment of frailty raises doubts about how best to interpret these results. Most of the studies (n = 15)[13, 35] that we included in these analyses focused on physical frailty only, while Sable-Morita et al. [39] used the KCL, and considered frailty included not only physical, but also social and psychological aspects. This checklist assesses seven areas: exercise/fall, instrumental activities of daily living, cognition, mood, malnutrition, oral function, and social activities of daily living. It is similar to the Fried phenotype, albeit more inclusive. Thirteen studies used slightly modified Fried criteria. In addition, Naharci et al. [32] noted a higher proportion of exhaustion among people with than without self-reported hearing loss. There may be merit in exploring the association between hearing loss and each frailty domain in greater detail.

Eleven out of 16 articles reviewed used available data from existing population studies. These large studies collect data from multi-disciplines and were not specifically designed for hearing and frailty analysis. This could account for the limitations in study design in most of the articles we reviewed. Moreover, most studies had a cross-sectional design, so that the direction of the observed associations was not clear. Five articles reported longitudinal results, but follow-up periods varied from 1 year to 10 years.

Currently there is insufficient evidence to conclude that a causal relationship between hearing loss and frailty exists. More research is needed to investigate the plausibility of such hypotheses. It is important to further confirm and investigate this relationship. If hearing loss is found to be a marker of frailty, then health care providers should consider medical review for frail or pre-frail status when hearing loss is present. If hearing loss contributes significantly to frailty, then appropriate management may delay frailty and death in later life. Either way, considering the high prevalence of hearing loss among aged adults and the relatively straightforward management of this, regular assessments of older adults at risk of hearing loss and the proper management of hearing impairment could improve the quality of life of older adults and, potentially, lower costs of healthcare and social support systems by delaying the onset of frailty.

Conclusions

Our review and meta-analysis of observational studies suggest that hearing loss is associated with higher prevalence and risk of frailty in later life. Whether this relationship is causal remains to be determined.