Effects of hearing impairment and hearing aid use on the incidence of cognitive impairment among community-dwelling older adults: evidence from the Taiwan Longitudinal Study on Aging (TLSA)

This study’s population included participants of the TLSA, which was a longitudinal, population-based survey initiated by the Health Promotion Administration, Ministry of Health and Welfare, Taiwan [15]. A three-stage systematic random sampling design was used to select an equal probability adult samples aged 60 or above. Data were collected through face-to-face personal interviews conducted by trained interviewers. Respondent follow-ups were then conducted every 3 to 4 years. A total of seven surveys were conducted in 1989, 1993, 1996, 1999, 2003, 2007, and 2011 (i.e., Waves I–VII). The details and design of the TLSA have been described elsewhere [16]. This study analyzed datasets from Waves IV–VII (i.e., 1999–2011).

First, of the 6091 eligible older adults, we excluded 227 individuals who had only completed one wave of investigation (Fig. 1). Second, we excluded 384 individuals due to missing data for either the Short Portable Mental Status Questionnaire (SPMSQ) or sensory impairment items. Third, we excluded 156 individuals with a history of cancer. Finally, we excluded 1672 individuals with cognitive impairments as defined by receiving SPMSQ scores ≤4 during their initial assessments [17]. Exclusion of older adults with cognitive impairments could ensure the accuracy of the responses in the initial assessments which may be influenced by cognitive status. Ultimately, data from 3658 older adults without cognitive impairments were included for analysis. HI was defined by self-reported hearing loss or hearing aid use. The final sample was then divided into HI (n = 775) and non-HI (n = 2883) groups.

We gathered subject data on age, sex, height, weight, level of education, marital status, living arrangement, self-rated health condition, health-related behaviors, self-reported economic pressure, concurrent comorbidities, and the use of assistive device (Table 1).

This study evaluated respondent physical and psychosocial status using multidimensional scores from the social support scale, social participation scale, mobility scores, instrumental activities of daily living (IADL) scale, Center for Epidemiologic Studies Depression (CES-D) scale, SPMSQ, and quality of life (QoL) scale [18‐22]. Higher total scores for the social support scale (0–20), social participation scale (0–6), SPMSQ (0–9), and QoL scale (0–12) represented better condition. On the other hand, lower total scores for mobility scale (0–8), IADL scale (0–5), and CES-D (0–24) represented better condition. The details of each scales are given below.

The social support scale consisted of four questions designed to measure the four corresponding items of emotional support (What degree of care is received from family members, relatives, and/or friends?), instrumental support (Is someone available if you need help?), and information support (Do you feel that your family provides useful help? and Does your family consult your opinion when they make decisions?) [22]. Each item was scored from 1 to 5, with total scores ranging from 0 to 20.

Respondents answered the social participation scale to indicate whether they participated in group activities through one or more of six types of social organizations (i.e., hobby related clubs, religious or church groups, political groups, retired associations, elderly-related associations, and volunteer groups) [19]. Each type was coded as yes (score = 1) or no (score = 0), with total scores ranging from 0 to 6.

For physical function assessments, the mobility scale included the difficulty level of standing for 15 min, stooping, reaching overhead, grasping with fingers, lifting or carrying 25 pounds, running for 20–30 m, walking 200–300 m, and climbing stairs [20]. The IADL scale was used to assess five activities as proposed by Lawton and Brody [21]. Respondents who reported difficulty, were unable to carry out tasks, or received help and/or used equipment when performing tasks were coded as having difficulty (i.e., 1 = yes, 0 = no).

Depressive symptoms were measured using the shorter form of the CES-D scale [23]. This included the eight items of “my appetite was poor,” “everything I did took effort,” “my sleep was restless,” “felt depressed,” “felt lonely,” “people were unfriendly,” “felt sad,” and “could not get going.” Responses ranged from “rarely or none of the time” to “most or all of the time” to produce scores between 0 and 3, respectively, with total scores ranging from 0 to 24 [20].

The QoL scale consisted of 12 items proposed by Neugarten and colleagues [24], including “is your life better than that of most others?,” “are you satisfied with your life?,” “are you interested in the things in which you are engaged?,” “have the most recent years contained the best days of your life?,” “would you like to change your past life?,” “do you expect something good to happen in the future?,” “should your life be better than it is now?,” “are you bored with most things you do?,” “do you feel old and tired?,” “does most of your life meet your expectations?,” “do you feel that you are living in a safe environment?,” and “are you satisfied with your living environment?” Each item was rated as yes (1) or no (0), with total scores ranging from 0 to 12.

The dependent variable of this study was the incident occurrence of a > 4 point deficit on the SPMSQ, while the independent variable was self-reported hearing deficit. This study applied the modified version of SPMSQ to the exclusion criteria, propensity-score matching covariates, and primary outcome to constitute criteria for cognitive impairment. This study included responses to the nine following SPMSQ items: “where are you located now?,” “what is your home address?,” “what day is it?,” “what month is it?,” “what year is it?,” “how old are you?,” “who are the current and last presidents?,” and “subtract 3 from 20 four consecutive times” [17]. Participants with four or more errors were described as having cognitive impairment, which was our primary outcome, and several other cohort studies supported this cut-off point [25].

We assessed the robustness of our findings by conducting a sensitivity analysis designed to access unmeasured confounding factors related to the exposures and outcomes by calculating E-values [26]. The E-value is the minimum strength of association, on the risk ratio scale, that unmeasured confounders would need to have with both the exposure and outcome, conditional on the measured covariates, to fully explain away a specific exposure-outcome association.

Propensity-score matching was conducted through the PSMATCH procedure provided by SAS (Statistics Analysis System Institute Inc., Cary, NC, USA). This study used descriptive statistics to assess patient demographics. The adjusted HRs of incidences of cognitive impairment were calculated using the Cox proportional hazard model while adjusting for possible confounders. All analyses were conducted using SAS version 9.4.

We used the formula to calculate the E-value for the effect estimate [26, 27]. For unmeasured confounders associated with HI and cognitive impairment during the 12-year follow-up, the E-value formula produced E = 1.45 for the estimate. The result could be interpreted that an unmeasured confounder that was associated with both the HI and cognitive impairment by a risk ratio of 1.45-fold each, above and beyond the measured confounders, but weaker confounding would not do so. As such, the E-value provided evidence for this study’s robustness.

Although this study generated important findings, they should be interpreted with some cautions. First, although we used multiple strategies to minimize the effects of confounders through propensity-score matching, this observational study may entail residual confounding factors and thus cannot prove causality. Second, we did not evaluate HI severity and possible mechanisms of HI among respondents, which was related to cognitive impairment. In addition, the severity of HI may make progress through the follow-ups, which was likely to affect the SPMSQ score. However, it is more important to detect HI during community screening than it is to determine severity. Third, although two groups were balanced at the baseline, some time-varying changes including later onset of hearing loss in the non-HI group could influence the result. However, these effects could only make the HRs of our results underestimated. Therefore, the HI group may have higher risk of incident cognitive impairment. Fourth, compliance with hearing aids is often much less than with wearing glasses in the older adults [35]. Therefore, the effect of hearing aid on the prevention of cognitive decline might be underestimated. Finally, the sample size was relatively limited in regard to hearing aid users. Future clinical trials are therefore needed to confirm how hearing aid use may prevent cognitive decline.