Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Aug 1.
Published in final edited form as: Int J Audiol. 2020 Dec 7;60(8):598–606. doi: 10.1080/14992027.2020.1853260

Predicting Hearing Aid Use in Adults: The Beaver Dam Offspring Study

Lauren K Dillard 1,2, Amy L Cochran 1,3, Alex Pinto 4, Cynthia G Fowler 2, Mary E Fischer 4, Ted S Tweed 2,4, Karen J Cruickshanks 1,4
PMCID: PMC8180532  NIHMSID: NIHMS1653132  PMID: 33287599

Abstract

Objective:

The purpose of this study was to (i) develop a model that predicts hearing aid (HA) use and (ii) determine if model fit is improved by adding factors not typically collected in audiologic evaluations.

Design:

Two models were created and evaluated. The ‘clinical’ model used factors typically collected during audiologic clinical evaluations. The ‘expanded’ model considered additional clinical, health and lifestyle factors to determine if the model fit could be improved (compared to clinical model). Models were created with Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression with 10-fold cross validation. Predictive ability was evaluated via receiver operating characteristic curves and concordance statistics (c-statistics).

Study Sample:

This study included 275 participants from the Beaver Dam Offspring Study, a prospective longitudinal cohort study of aging, with a treatable level of hearing loss and no HA use at baseline.

Results:

The clinical and expanded models report predictors important for HA use. The c-statistics of the clinical (0.80) and expanded (0.79) models were not significantly different (p=0.41).

Conclusions:

Similar predictive abilities of models suggest audiological evaluations perform well in predicting HA use.

INTRODUCTION

Despite negative consequences of hearing loss (HL), few people who may benefit from hearing aids (HA) use them. Population-based studies estimate that only 9% to 17% of eligible individuals acquire HAs (Hartley et al., 2010; Fischer et al., 2011; Gopinath et al., 2011; Moon et al., 2015) and that among the few who acquire HA, the subsequent use of HA is even lower (Lupsakko et al., 2005; Moon et al., 2015). Long-term HA use is also low, as it is estimated approximately 40% of individuals with HAs continue use after 10 years (Davis et al., 2007). This is particularly surprising given increased severity of HL with age (Cruickshanks et al., 1998, 2003, 2010). The low level of HA use across populations is a concern, particularly given the potential of HAs to reduce psychological, social and emotional effects of HL and thus improve health related quality of life (Chisolm et al., 2007).

In attempts to better understand low rates of HA use, population-based epidemiological research has evaluated audiologic, demographic, health and lifestyle factors associated with HA use. Consistently, population-based studies have reported associations of increased HA use with worse pure-tone hearing sensitivity, worse self-perception of HL and higher reported hearing handicap (Popelka et al., 1998; Hartley et al., 2010; Fischer et al., 2011; Gopinath et al., 2011; Bainbridge & Ramachardran 2014; Fisher et al., 2015; Moon et al., 2015). A longitudinal population-based study in Australia reported that HA use is associated with older age (Gopinath et al., 2011). Socioeconomic factors (e.g., higher education) have also been associated with HA use, particularly in US populations, which likely reflects the high costs of HAs in the US (Popelka et al., 1998; Fischer et al., 2011; Bainbridge & Ramachandran 2014). However, even in countries where the cost of HAs is free or government subsidized, HA use is low (Hartley et al., 2010; Gopinath et al., 2011; Fisher et al., 2015), which suggests that the price of HAs is not the only reason for non-use. Certain health factors have also been associated with HA non-use, including smoking (Hartley et al., 2010), history of stroke (Gopinath et al., 2011), cognitive impairment (Fisher et al., 2015), use of a greater number of medications (Fisher et al., 2015), balance problems (Moon et al., 2015) and myopia (Moon et al., 2015). The association of HA use and lifestyle factors is less clear. Although Fisher et al. (2015) reported that increased HA use was associated with the number of leisure activities, other lifestyle factors (e.g., number of people in home, marital status) have not been associated with HA use in population-based studies (Fischer et al., 2011; Gopinath et al., 2011).

As compared to population-based epidemiological studies, audiological basic science studies have reported additional or different predictors of HA use. These differences are likely a result of differences in study populations (general population versus clinical samples) and measures collected. For example, some evidence from audiologic studies suggests that hearing sensitivity did not predict HA use or satisfaction (e.g., Hickson et al., 1999; Cox et al., 2007), whereas population-based epidemiological studies showed an association between hearing sensitivity and HA use (e.g., Fischer et al., 2011; Gopinath et al., 2011; Fisher et al., 2015). A study conducted with cross-sectional survey data of Swiss HA owners reported women had a higher prevalence of HA use than men (Staehelin et al., 2011) although gender has not been associated with HA use in population-based studies (e.g., Fischer et al., 2011; Gopinath et al., 2011). Extensive audiologic research has described factors contributing to HA use, success and satisfaction, including the importance of social support (Singh et al., 2015; Hickson et al., 2014), attitude/expectations (Hickson et al., 1999; Humes et al., 2003; Hickson et al., 2014) and self-efficacy (Kricos et al., 2006), factors which are often not collected in population-based studies. Recent reviews have summarized the range and complexity of factors associated with HA-related outcomes in the literature (Knudsen et al., 2010; Meyer & Hickson, 2012; McCormack & Fortnum, 2013; Ho-Yee Ng & Yuen Loke, 2015). The focus of this paper is on hearing aid use because the study did not include information on satisfaction or success.

Many population-based and audiologic studies on HA use utilize samples of older adults (e.g., Fischer et al., 2011; Bainbridge & Ramachandaran, 2014; Hickson et al., 2014) although HL often develops in middle age (Cruickshanks et al., 1998, 2003, 2010; Nash et al., 2011). In middle-aged populations, HL has been associated with reduced work-related productivity and limitations (Nachtegaal et al., 2012) as well as increased risk of earlier retirement (Helvik et al., 2012). Earlier intervention with HA in middle-aged adults may contribute to prevention of hearing-related limitations in middle-age and later in life, given that those with prior HA experience are more likely to continue HA use over time (Saunders & Jutai, 2004). Considering general differences in health and/or lifestyle in middle-aged versus older adults, factors related to HA use may differ between these age groups. For example, when compared to older adults, middle-aged adults often have less severe hearing impairment (e.g., Cruickshanks et al., 1998; Nash et al., 2011) and may have different communication needs given higher rates of employment, better general health and higher levels of social engagement.

Population-based research has indicated HA use may be associated with factors captured in audiological evaluations (e.g., hearing sensitivity, hearing handicap) as well as factors often not captured in audiological evaluations (e.g., smoking, stroke). Although audiological evaluations are relatively comprehensive, clinicians may be able to better predict who will use HA by incorporating additional diagnostic tests or by obtaining a more comprehensive picture of patients’ health and lifestyles.

Predictive models are a novel approach to comprehensively examine factors potentially related to HA use. Such models are often used in medical research to predict the development of a disease (Pencina et al., 2009; Wilson et al., 1998) or treatment outcomes (Guldvog et al., 1994; Simmen et al., 2008). Results from predictive models can be converted into clinical risk scores, which indicate the likelihood of an event given certain patient characteristics (Wilson et al., 1998). In audiology, the use of a risk score to predict HA use may serve as a helpful screening tool for clinicians to identify individuals likely to use HA. It may be possible to implement a risk score in population-based screening programs, primary care, or audiology clinics, and use of such a tool may aid in earlier identification and intervention for HL in middle-aged adults.

The purpose of this study was to (i) develop a model that predicts HA use and (ii) determine if model fit was improved by adding factors not typically collected in audiologic evaluations. To accomplish these purposes, two models were created and evaluated in a sample of primarily middle-aged individuals with hearing impairment from a population-based cohort. The ‘clinical model’ considered factors collected during a typical audiologic evaluation and the ‘expanded model’ considered additional factors, including tests of auditory processing and more in-depth factors related to demographics, health motivation, lifestyle and social practices, medical history and health and cognitive function and mental health.

METHODS

Study population

Participants were members of the Beaver Dam Offspring Study (BOSS), a prospective longitudinal study of aging and sensory disorders (2005-2015). Participants of BOSS are the middle-aged adult offspring of the participants of the population-based Epidemiology of Hearing Loss Study (EHLS; 1993-2020). The BOSS and EHLS cohorts have been described in other publications (e. g., Cruickshanks et al., 1998, 2003; Nash et al., 2011). BOSS examinations, conducted every 5 years since baseline, include measures of hearing, other sensory disorders and extensive information on health and lifestyle. Data from the baseline examination (2005-2008) were used to determine inclusion and for covariates.

There were 3298 participants with data at the baseline examination of BOSS. This study included BOSS participants with no HA use at baseline and with a treatable HL as defined by the World Health Organization (WHO; 2020) (pure tone average [PTA] at 0.5, 1.0, 2.0 and 4.0 kHz > 25 dB HL in the better ear) or the U.S. Department of Veteran’s Affairs (VA; 2014) (threshold ≥40 dB HL at any frequency 0.5-4.0 kHz or threshold ≥ 26 dB HL at any three frequencies 0.5-4.0 kHz). Of the 418 participants with a treatable HL, we excluded 39 HA users at baseline, 97 who were lost to follow up over 10 years and 2 who attended follow-up examinations but did not answer the question regarding HA use. Therefore, 280 participants were eligible for inclusion in the study.

Measurements

Participants underwent a series of standardized examinations and interviews conducted by examiners trained and certified in all study protocols. Similar methodology was used for all data collection cycles. The outcome variable was self-reported HA use, collected by the question: “Do you use your hearing aid now?” at 5- or 10-year follow-up examinations (2010-2013, 2015-2017).

The measurements section of the methods is organized by describing the covariates used in the clinical model, then by describing the covariates used in the expanded model. A more detailed description of methods is included in supplementary materials (http://tandfonline.com/doi/suppl).

Covariates included in the clinical model

Diagnostic Tests, Pure-tone Audiometry & Tympanometry

Audiometric testing was performed in accordance with American Speech-Language-Hearing Association guidelines and in compliance with standards defined by the American National Standards Institute (American Speech-Language-Hearing Association 2005; American National Standards Institute 1999, 2010). Otoscopy and tympanometry were performed prior to audiometric testing. Testing was performed in sound-treated booths with clinical audiometers (calibrated every 6 months). TDH-50P earphones and ER-3A insert earphones (in cases of probable ear canal collapse) were used. Pure-tone air-conduction thresholds were obtained in both ears at 0.5, 1, 2, 3, 4, 6, and 8 kHz and bone-conduction thresholds were obtained at 0.5, 2, and 4 kHz. Masking was used as necessary (Cruickshanks et al., 1998, 2003). PTAs were calculated in each ear from frequencies 0.5, 1, 2, 4 kHz and 6 and 8 kHz as measures of hearing sensitivity in speech and high frequencies, respectively. Air bone gaps were defined as a difference between air- and bone-conduction thresholds of ≥15 dB HL in either ear at 0.5, 2.0 or 4.0 kHz. Tympanometry was performed using a GSI-37 tympanometer. Abnormal tympanometric results were defined as having peak admittance at 226 Hz (Ytm) ≤0.1 or ≥3.0 mmho, or an ear canal volume ≥3 ml in either ear.

Otologic or Medical History

Self-reported data were collected on presence and severity of HL, history of middle ear problems (otologic aching, pressure, fullness, or discharge), history of noise exposure, use of assistive technology and presence of significant tinnitus (Nondahl et al., 2002, 2010, 2011, 2013). A slightly modified version of the hearing handicap inventory screener for the elderly and for adults (HHIE/A-S) was administered (Ventry & Weinstein 1982; Newman et al., 1990). For analysis purposes only, the categories ‘sometimes’ and ‘yes’ were combined to form a binary variable in order to ensure adequate variability in response distributions. Therefore, HHIE/A-S questions indicate any level (‘sometimes’ or ‘yes’) of reported hearing handicap. One HHIE/A-S question, ‘Does a hearing problem cause you to attend religious services less often than you would like?’ was not included in analysis due to lack of variability in response (n=6 answered yes or sometimes, n=274 answered no).

Additional self-reported data include history of head injury, episodic or chronic dizziness not related to sudden changes in position or issues keeping balance while walking in the past year (Schubert et al., 2019).

Demographics

Demographic information included age and self-reported health insurance coverage.

Lifestyle & Social Factors

Participants reported the number of people living in their home as well as their current employment status (classified into working or not working).

Covariates included in the expanded model

Auditory Processing Tests

Word recognition in competing message (WRCM) was measured in a sound-treated booth using the Northwestern University Auditory Test Number 6 (NU-6; 25-word list presented to the better ear based on the threshold at 2 kHz), and results are presented as percent correct. The dichotic digits test (DDT) was administered in the right ear directed recall condition with 25 sets of triple-digit pairs, and results are presented as the score correct (possible range 0-75 digits).

Demographics

Demographic factors were sex and education. Education was categorized as less than a high school graduate (< 12 years), a high school graduate (12 years), some college (13-15 years) and college graduate or beyond (16+ years).

Health Motivation

Participants self-reported frequent consumption of a multi-vitamin (≥4 days/week), exercise (sufficient to work up a sweat ≥1 day/week), number of weekly servings of fruits and vegetables and receipt of a flu shot in the past year (yes/no). Participants also self-reported use of glasses (for distance, computer or a book). A variable indicating history of parental HA use was created by linking participant data to parental data (participants of EHLS). Participants had a positive history of parental HA use if at least one of their parents self-reported HA use at any EHLS phase.

Lifestyle & Social Factors

Participants self-reported smoking history and alcohol consumption. A variable indicating time not spent communicating with others was derived by summing responses to questions reporting number of weekday hours spent watching TV, on the computer or reading a book.

Medical History & Overall Health

Measures of height and weight were used to calculate body mass index (BMI). BMI was calculated as weight in kilograms (kg) divided by height in meters (m) squared. Obesity was defined as a BMI ≥ 30 kg/m2. Laboratory data were used to measure hypertension and serum total cholesterol. Participants self-reported history of cardiovascular disease, diabetes and cancer treatment with chemotherapy or radiation. A proxy of overall health was created from a self-report of whether or not participants spent a week in bed from illness or injury (previous 3 months) were hospitalized at least overnight (previous year). Participants also reported history of severe headache or migraine (previous 3 months).

Cognitive Function & Mental Health

The Trail Making Test Part B (TMT-B) and GPB were administered separately as tests of cognitive function. The TMT is a test of attention, processing speed and executive function (Arbuthnott & Frank 2000; Reitan 1992; Strauss et al., 2006). The GPB (Lafayette Instruments, Lafayette, IN, USA) is a test of attention, executive and psychomotor function (Ashendorf et al., 2009; Strauss et al., 2006). For both the TMT and GPB, the score was recorded as the number of seconds the participant took to successfully complete each test, with a longer time indicating poorer performance. Participants completed the Centers for Epidemiological Studies Depression Scale (CES-D; Radloff 1977). A total score of ≥16 out of 60 classified the participant as having depressive symptoms.

Statistical Methods

Two models, the clinical model and the expanded model, were built and compared using the process described below.

First, models were created using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression with 10-fold cross validation. Final LASSO model selection was guided by predictive power ascertained via cross-validation procedures which selected the model with the lowest sum of squared residuals (Tibshirani 1996; Efron et al., 2004). Results obtained from LASSO models show the level of ‘importance’ (ranked in order from high to low importance) of individual predictors and highlight the relevance of the combination of predictors. LASSO regression requires complete data for all covariates.

Second, logistic regression was performed on the list of important predictors determined by LASSO. Logistic regression models returned odds ratios (OR) and corresponding 95% confidence intervals (CI). Logistic regression was used to evaluate model performance via receiver operating characteristic (ROC) curves and concordance statistics (c-statistics). The c-statistic measures the model’s ability to discriminate between those who do and do not experience the outcome (HA use) (Kuhn & Johnson, 2013). C-statistics were calculated as the area under ROC curves and were presented as an estimated value with corresponding 95% CIs. According to Hosmer and Lemeshow (2000), the c-statistic evaluates model discrimination as follows: 0.5 suggests no discrimination, 0.7 to <0.8 is considered acceptable discrimination, ≥0.8 to 0.9 is considered excellent discrimination and ≥0.9 is considered outstanding discrimination.

The clinical model used covariates commonly collected during clinical audiologic evaluations (standard diagnostic tests, otologic and medical history, hearing handicap, some lifestyle, social and demographic factors) and was developed using the number of individuals with complete data for all covariates (n=270; Figure 1). The expanded model used important predictors from the clinical model in addition to covariates hypothesized to improve model fit (auditory processing tests, demographic, health motivation, lifestyle & social, medical history & overall health and cognitive function & mental health factors). The expanded model was developed using the number of individuals with complete data for all covariates considered in the expanded model (n=251; Figure 1). ROC curves were generated independently for both models on the minimum common dataset containing complete data for all important predictors (n=275; Figure 1). ROC curves were superimposed, and statistical significance of difference between the c-statistics was evaluated using a chi-square test (DeLong et al., 1988).

Figure 1:

Figure 1:

Open in a new tab

Pathway of factor selection, model construction and model comparison for clinical and expanded models.

RESULTS

Of the 275 participants in the minimum common dataset, 77.8% (n=214) were men and 22.2% (n=61) were women. The average age of participants was 56.8 (standard deviation [SD] 9.5, range 28-84) years and the majority (74%) of participants were between ages 45 and 65 years of age at baseline. The number of participants who fit both the VA and WHO definitions of HL was 102; 167 participants fit only the VA definition and 6 participants fit only the WHO definition. Baseline characteristics of covariates in the clinical and expanded models are in Tables 1 and 2, respectively. The 10-year cumulative incidence of HA use was 24.3% (n=68).

Table 1:

Baseline characteristics of all covariates considered in the clinical model.

n, mean (SD, range) or n (%)
Yes No
Diagnostic Tests
PTA (0.5-4.0 kHz) 275, 24.4 (8.3, 6.3-51.3)
PTA (6.0-8.0 kHz) 275, 50.6 (19.9, 5.0-100.0)
Air bone gap (≥15 dB HL, either ear) 67 (24.4%) 208 (75.6%)
Abnormal tympanometry 38 (13.8%) 233 (84.7%)
Otologic or Medical History
Self-reported HL (yes/no) 233 (85.4%) 40 (14.7%)
Self-reported HL severity (poor or fair vs excellent, very good or good) 138 (50.2%) 137 (49.8%)
Middle ear problems (aching, pressure, fullness, or discharge) 98 (35.6%) 177 (64.4%)
Noise exposure 192 (69.8%) 83 (30.2%)
Use of assistive listening device or closed captions 29 (10.6%) 246 (89.5%)
Significant tinnitus 62 (22.6%) 212 (77.4%)
History of head injury 85 (30.9%) 190 (69.1%)
Dizziness problems 35 (12.7%) 240 (87.3%)
Balance problems 29 (89.5%) 246 (10.6%)
(HHIE/A-S)
Does a hearing problem cause you to feel embarrassed when you meet new people? 99 (36.0%) 176 (64.0%)
Does a hearing problem cause you to feel frustrated when talking to members of your family? 112 (40.7%) 163 (59.3%)
Do you have difficulty hearing when someone speaks in a whisper? 217 (78.9%) 58 (21.1%)
Do you feel handicapped by a hearing problem? 89 (32.4%) 186 (67.6%)
Does a hearing problem cause you difficulty when visiting friends, relatives or neighbors? 117 (42.6%) 158 (57.5%)
Does a hearing problem cause you to have arguments with family members 61 (22.2%) 214 (77.8%)
Does a hearing problem cause you difficulty when listening to TV or radio? 120 (43.6%) 155 (56.4%)
Do you feel that any difficulty with your hearing limits or hampers your personal or social life? 79 (28.7%) 196 (71.3%)
Does a hearing problem cause you difficulty when in a restaurant with relatives or friends? 140 (50.9%) 135 (49.1%)
Does a hearing problem cause you difficulty hearing/understanding coworkers, clients, or customers? 144 (52.4%) 131 (47.6%)
Does a hearing problem cause you difficulties in the movies or in the theater? 55 (20.0%) 220 (80.0%)
Demographics
Age (years) 275, 56.8 (9.5, 28.0-84.0)
Has health insurance 261 (94.9%) 14 (5.1%)
Lifestyle & Social Factors
Number of people in home 275, 1.4 (1.1, 0-6)
Currently has a job 202 (73.5%) 73 (26.6%)
Open in a new tab

Table 2:

Baseline characteristics of all factors considered in the expanded model.

n, mean (SD, range) or n (%)
Yes No
Auditory Processing Tests
Dichotic Digits Right Ear Directed Recall (total score) 271, 74.0 (3.3, 39.0-75.0)
Word Recognition in Noise (%) 272, 48.9 (17.0, 12.0-88.0)
Demographics
Sex (Female) 61 (22.2%) 214 (77.8%)
Education 0-12 yr 110 (40.0%) --
      13-15 yr 93 (33.8%) --
      16+ yr 72 (26.2%) --
Health Motivation
Current use of multi-vitamin 137 (49.8%) 138 (50.2%)
Regular exercise (at least once/week) 151 (54.9%) 124 (45.1%)
Weekly servings of fruits and vegetables 275, 14.3 (8.6, n/a†)
Flu vaccination, past year 120 (43.6%) 155 (56.4%)
Current use of glasses (distance, computer, book) 255 (92.7%) 20 (7.3%)
Parental hearing aid experience 115 (41.8%) 160 (58.2%)
Lifestyle & Social
Past/current smoker 159 (57.8%) 115 (41.8%)
Alcohol intake  0 gm/week 32 (11.6%) --
        >0-14 gm/week 85 (30.9%) --
        >14 to 74 gm/week 71 (25.8%) --
        >74 to 140 gm/week 41 (14.9%) --
        >140 gm/week 46 (16.7%) --
Time not spent communicating (hours) 275, 4.4 (0.2, 0.3-12.0)
Medical History & Overall Health
Obesity (BMI ≥30) 167 (61.0%) 107 (39.1%)
Hypertension 155 (56.4%) 120 (43.6%)
Serum total cholesterol (mg/dl) 272, 198 (36.5, 104-339)
History of cardiovascular disease 35 (12.7%) 240 (87.3%)
Diabetes 37 (13.6%) 236 (86.5%)
History of chemotherapy/radiation for cancer treatment 7 (2.6%) 268 (97.5%)
Spent wk in bed from illness/injury, hospitalized ≥ overnight 26 (9.5%) 248 (90.5%)
Experience severe headaches/migraines, past 3 months 31 (11.3%) 244 (88.7%)
Cognitive Function & Mental Health
Trail Making Test (B) time (sec) 274, 79.5 (32.4, 34.0-301.0)
Grooved Pegboard Test (GPB) time (sec) 275, 81.0 (19.4, 51.0-241.0)
Depressive symptoms 38 (14.3%) 227 (85.7%)
Open in a new tab
†

Not applicable (n/a) because responses were recorded categorically

Clinical Model

LASSO regression identified the following covariates as important (ranked in order from high to low importance) in the clinical model: (1) higher PTA (0.5-4.0 kHz), (2) self-reported hearing difficulties listening to the TV/radio and (3) in restaurants, (4) balance and (5) middle ear problems. That is, individuals with a higher PTA (per 1 dB HL increase: OR 1.11, CI 1.07-1.15), individuals with self-reported difficulties hearing the TV/radio (OR 2.88, CI 1.49-5.55) and/or individuals with self-reported difficulties hearing in restaurants (OR 1.83, CI 0.95-3.54) had higher odds of using HA. Reduced odds of HA use were associated with balance problems (OR 0.43, CI 0.14-1.29) and/or middle ear problems (OR 0.59, CI 0.30-1.15). Table 3 displays predictors and the corresponding OR and CIs. The c-statistic for the clinical model is 0.80 (CI 0.73-0.86) (Figure 2).

Table 3:

Important predictors from the clinical model with corresponding odds ratios (OR) and 95% confidence interval (CI). Predictors are presented in order of importance.

Characteristic OR (95% CI)
PTAper +1 dB HL increase (0.5-4.0 kHz), better ear 1.11 (1.07-1.15)*
Self-reported difficulties hearing TV/radio (HHIE/A-S) 2.88 (1.49-5.55)*
Self-reported difficulties hearing in restaurant (HHIE/A-S) 1.83 (0.95-3.54)
Balance Problems 0.43 (0.14-1.29)
Middle Ear Problems 0.59 (0.30-1.15)
Open in a new tab
*

significant at p<0.05

Figure 2:

Figure 2:

Open in a new tab

Superimposed ROC curves for clinical and expanded models. The difference between the c-statistics of models is not statistically significant (p=0.41).

Expanded Model

The expanded model considered important covariates from the clinical model and additional factors related to auditory function, demographics, health motivation, lifestyle and social practices, medical history and health and cognitive function and mental health (Table 3). LASSO regression identified the following covariates as important (ranked in order from high to low importance): (1) higher PTA (0.5-4.0 kHz), (2) self-reported hearing difficulties listening to the TV/radio and (3) in restaurants, (4) balance problems, and (5) history of cardiovascular disease. In other words, individuals with a higher PTA (per 1 dB HL increase: OR 1.12, CI 1.08-1.16), individuals with self-reported hearing difficulties hearing the TV/radio (OR 2.89, CI 1.49-5.52) and/or individuals with self-reported hearing difficulties in restaurants (OR 1.52, CI 0.78-2.94) had higher odds of HA use. Reduced odds of HA use were associated with balance problems (OR 0.46, CI 0.15-1.39), and history of cardiovascular disease (OR 0.67, CI 0.25-1.80). Table 4 displays the predictors ranked by level of importance and the corresponding OR and CIs. The c-statistic for the expanded model is 0.79 (CI 0.73-0.85) (Figure 2).

Table 4:

Important predictors from the expanded model with corresponding odds ratios (OR) and 95% confidence interval (CI). Predictors are presented in order of importance.

Characteristic OR (95% CI)
PTAper +1 dB HL increase (0.5-4.0 kHz), better ear 1.12 (1.08-1.16)*
Self-reported difficulties hearing TV/radio (HHIE/A-S) 2.89 (1.49-5.52)*
Self-reported difficulties hearing in restaurants (HHIE/A-S) 1.52 (0.78-2.94)
Balance Problems 0.46 (0.15-1.39)
History of cardiovascular disease 0.67 (0.25-1.80)
Open in a new tab
*

significant at p<0.05

Comparison of Clinical and Expanded Models

Both models reported the most important predictors of HA use were higher PTA, self-reported hearing difficulties listening to the TV/radio and/or in restaurants and not having balance problems. The clinical model also reported middle ear problems as an important predictor associated with reduced odds of HA use. However, in the expanded model, middle ear problems was not an important predictor. Instead, the expanded model reported history of cardiovascular disease as an important predictor associated with reduced odds of HA use. The c-statistics for the clinical (0.80, CI 0.73-0.86) and expanded (0.79, CI 0.73-0.85) models are not significantly different (p=0.41). Figure 2 displays the superimposed ROC curves and c-statistics.

DISCUSSION

This study developed models to predict HA use in a cohort of primarily middle-aged adults. Both the clinical and expanded models reported factors important for predicting HA use and had high predictive abilities, but the predictive abilities of the models (c-statistics) were not significantly different.

The 10-year cumulative incidence of HA use (24.3%) is low but is consistent with population-based research on prevalent HA use and incident HA acquisition over 5 or 10 years. In EHLS, Fischer et al. (2011) reported 5- and 10-year cumulative incidence of HA acquisition as 14.3% and 35.7%, respectively, in participants with HL. In the Blue Mountains Hearing Study, Gopinath et al. (2011) reported the 5-year incidence of HA ownership and use at 24.3% and 23.4%, respectively, in hearing impaired individuals.

When interpreting LASSO regression models, one must focus on the overall predictive ability of the model (c-statistics), which encompasses information regarding importance of individual predictors as well as the relationships of the predictors. Therefore, lack of statistical significance of ORs for individual predictors does not diminish their importance in the models. Important predictors of HA use in the clinical model were higher PTA (0.5-4.0 kHz), self-reported difficulties hearing the TV/radio and in restaurants, and lack of balance and middle ear problems. In the expanded model, important predictors for HA use were higher PTA (0.5-4.0 kHz), self-reported difficulties hearing the TV/radio and in restaurants, lack of balance problems, and no history of cardiovascular disease. Although the main focus of this work is on the predictive abilities of the models, the individual predictors and their relationships provide a basis for understanding factors important in predicting HA use.

Broadly, factors deemed important in the models indicate the relevance of hearing severity, self-reported hearing difficulties and health on HA use, which is consistent with existing research. For example, population-based studies have associated HA use and/or acquisition with worse hearing sensitivity and higher hearing handicap (Popelka et al., 1998; Hartley et al., 2010; Fischer et al., 2011; Gopinath et al., 2011; Bainbridge & Ramachardran 2014; Fisher et al., 2015; Moon et al., 2015). The present study used each question of the HHIE/A-S separately to measure hearing handicap and found that self-reported difficulties listening to the TV/radio and in restaurants were important predictors. These two questions, in particular, may most accurately capture the daily life of study participants and thus the situations in which there is the most hearing handicap. The importance of these two questions in particular does not necessarily indicate them as best for predicting HA use, but rather suggests that in general, self-reported hearing handicap is an important predictor of HA use. In the South Korean NHANES, Moon et al. (2015) reported balance problems were associated with reduced HA use, as was reported in the present study. Few participants in the present study had known otologic conditions and therefore, it is likely that balance problems were idiopathic or neurologic and not vestibular or otologic. Self-reported balance problems likely indicate restricted mobility, which may limit one’s ability to seek healthcare (Pearson et al., 2004). Additionally, limited mobility may diminish the need for HA use, if for example, individuals spend most of their time at home. Other important predictors associated with reduced odds of HA use were middle ear problems (clinical model) and history of cardiovascular disease (expanded model). These conditions are also associated with HL (Cruickshanks et al., 2003; Nash et al., 2016) and may indicate worse overall health. Population-based studies have also demonstrated the association of HA use with other factors such as age (e.g., Gopinath et al., 2011). The non-importance of certain factors in the models may be a result of the LASSO regression modeling technique. LASSO generates the most efficient model by selecting the fewest number of important predictors that results in the highest predictive ability of the model whereas reporting all factors (i.e., a ‘less efficient’ model) associated with an outcome is often done via other modeling techniques (e.g., logistic regression).

When considering factors predictive of HA use in this study, it is valuable to consider potential reasons for differences observed in population-based epidemiological studies versus audiologic basic science studies (see Kukull & Ganguli, 2012 for an overview of study generalizability). Consistent with methodology of other population-based epidemiological studies of HA use (e.g., Fischer et al., 2011; Gopinath et al., 2011), this study considered participants with a treatable level of HL (i.e., at risk for HA use). Hearing sensitivity is often better in studies of the general population than in clinical cohorts which may increase the ability to detect an effect of PTA on HA use. This may partially explain why population-based studies (e.g., Fischer et al., 2011; Gopinath et al., 2011) have found an effect of PTA on HA use whereas some audiologic studies have not (e.g., Hickson et al., 2014). It is notable that in this sample, the mean PTA was relatively low (24.4 dB HL) as a result of the definitions of treatable HL. The VA definition requires any three frequencies from 0.5-4.0 kHz to be ≥26 dB HL and does not average across the frequency range. Audiologic studies have contributed vast amounts of knowledge on factors associated with HA use, as well as other HA-related outcomes (e.g., satisfaction, success), that are not always measured in population-based studies (for a review, see Knudsen et al. [2010], Meyer & Hickson [2012], McCormack & Fortnum [2013] and Ho-Yee Ng & Yuen Loke [2015]). Future population-based epidemiological studies may benefit from measuring additional potentially associated factors such as social support, attitude/expectations, self-efficacy, HA type and HA fitting data (e.g., Bertoli et al., 2009; Hickson et al., 2014; Singh et al., 2015; Singh & Launer, 2016; Bisgaard & Ruf, 2017).

Both the clinical and expanded models had excellent predictive abilities (c-statistics 0.79-0.80) (Hosmer & Lemeshow, 2000). Given their similar predictive abilities (p=0.41), it is not possible to assert which model is statistically better. This indicates that information commonly collected during audiologic evaluations (i.e., factors in the clinical model) likely results in excellent prediction of HA use. Information collected in audiologic evaluations (e.g., case history) often varies by patient presentation, clinician and other factors, and therefore the clinical model is not representative of all audiologic evaluations.

LASSO regression is an optimal method for creating predictive models that can later be developed into clinical risk scores. Ideally, clinical risk scores use parsimonious models, which include the minimum amount of information that results in high predictive ability. Use of risk scores may improve clinical efficiency by saving time and resources needed to collect information from patients (Pavlou et al., 2015). Data from the population-based Framingham Heart Study were used to create a risk score to predict coronary heart disease (Wilson et al., 1998). Although Wilson et al. (1998) used a different regression model (Cox proportional hazards), they evaluated their model with methods similar to those included in this paper (i.e., ROC curves and c-statistic). Other cohort studies have successfully used LASSO regression models for risk prediction and development of risk scores (Airakninsen et al., 2018). The present study, which used LASSO regression to build and evaluate models predicting HA use, is a first step toward building a risk score to predict HA use in adults. However, LASSO regression models should first be validated using an external dataset (Pavlou et al., 2015). Future research should incorporate additional potential predictors associated with HA use in audiologic studies (described above), detailed information on HL severity and configuration, and should use a nationally representative sample of individuals to maximize generalizability of the model. Future research should also consider whether models can be applied across age groups (i.e., middle-aged versus older adults).

Upon validation of this model, it would be possible to use a risk score as a screening tool to predict HA use in a variety of settings. For example, risk scores could be incorporated in population-based audiologic screening programs to identify potential HA candidates and encourage them to seek clinical care. In primary care clinics, risk scores could be used to identify potential HA users and correctly refer them to audiology clinics. In middle-aged adults, a risk score that successfully predicts HA use may improve early intervention for middle-aged adults and may lead to continued HA use with age. In audiological practice, clinicians may use risk scores to identify patients most likely to use (and therefore benefit) from HAs. Additionally, clinicians able to identify important predictors of HA use or non-use can provide targeted interventions to their patients. For example, if patients with treatable levels of HL do not realize their hearing handicap, clinicians may wish to provided targeted counseling to help patients realize the situations in which HL affects them most.

The strengths of this study include a large, well-characterized population with follow-up examinations with a high retention rate. All data were measured with standard protocols at multiple time points by trained and certified examiners. The richness of available data allowed for a holistic consideration of the factors that may influence HA use. However, there are some study limitations. Because the sample was restricted to those with treatable HL, the sample size was relatively small. Therefore, it was not possible to consider all factors hypothesized to be related to HA use due to lack of variability in distributions. LASSO requires complete data for each variable so factors with a high number of missing values were not examined. There is the potential for residual confounding if there are unmeasured factors that influence HA use (e.g., social support, attitude/expectations, self-efficacy, HA fitting details), some of which are described in other studies (e.g., Knudsen et al., 2010; Hickson et al., 2014; Singh et al., 2015; Singh & Launer, 2016). As in most observational studies of HA use, it is not possible to disentangle factors contributing to HA acquisition, HA use, and hearing health care seeking. Care seeking and HA acquisition may be influenced by factors including, but not limited to, race/ethnicity, socioeconomic factors social support and auditory factors such as PTA (Meyer et al., 2014; Nieman et al., 2016). Fine-grained data on HA use (e.g., hours per day) were not available and it was not possible to account for impacts of changing technology (e.g., rechargeable HA) over the course of this study (up to 15 years). Lastly, BOSS participants are ethnically homogenous, non-Hispanic white individuals. Therefore, findings may not be generalizable to all populations and it was not possible to investigate racial or ethnic differences in factors predictive of HA use.

CONCLUSIONS

Two models, a clinical and expanded model, highlighted factors that predict HA use. There was no difference in the predictive ability of the models, which suggests audiological evaluations gather much of the necessary information to predict HA use. Additional research in population- or clinical-based cohorts is necessary to validate model findings, which may eventually be transformed into a risk score to predict HA use in adults.

Supplementary Material

1

SOURCES OF FUNDING & ACKNOWLEDGEMENTS

This work was supported by R01AG021917 (KJC) from the National Institute on Aging and an unrestricted grant from Research to Prevent Blindness.

Footnotes

DECLARATION OF INTEREST STATEMENT

The authors have no other funding, financial relationships, or conflicts of interest to disclose.

REFERENCES

  1. Airaksinen J, Jokela M, Virtanen M, et al. (2018). Prediction of long-term absence due to sickness in employees: Development and validation of a multifactorial risk score in two cohort studies. Scandinavian Journal of Work, Environment and Health, 44(3), 274–282. 10.5271/sjweh.3713 [DOI] [PubMed] [Google Scholar]
  2. American Speech-Language-Hearing Association. (2005). Guidelines for manual pure-tone threshold audiometry [Guidelines]. Available from www.asha.org/policy.
  3. American National Standards Institute (ANSI). (1999). Maximum permissible ambient noise levels for audiometric test rooms. ANSI S3.1e1999. New York, NY: American National Standards Institute. [Google Scholar]
  4. American National Standards Institute (ANSI). (2010). Specification for audiometers. ANSI S3.6e2010. New York, NY: American National Standards Institute. [Google Scholar]
  5. Arbuthnott K & Frank J (2000) Trail Making Test, Part B as a measure of executive control: Validation using a set-switching paradigm. Journal of Clinical and Experimental Neuropsychology, 22, 518–528. [DOI] [PubMed] [Google Scholar]
  6. Ashendorf L, Vanderslice-Barr JL , McCaffrey RJ (2009). Motor tests and cognition in healthy older adults. Applied Neuropsychology. 16, 171–176. [DOI] [PubMed] [Google Scholar]
  7. Bainbridge KE, & Ramachandran V (2014). Hearing aid use among older U.S. adults: The National Health and Nutrition Examination Survey, 2005–2006 and 2009–2010. Ear & Hearing, 35(3), 289–294. 10.1097/01.aud.0000441036.40169.29 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bertoli S, Staehelin K, Zemp E, Schindler C, Bodmer D, & Probst R (2009). Survey on hearing aid use and satisfaction in Switzerland and their determinants. International Journal of Audiology, 48(4), 183–195. [DOI] [PubMed] [Google Scholar]
  9. Bisgaard N, & Ruf S (2017). Findings from EuroTrak surveys from 2009 to 2015: Hearing loss prevalence, hearing aid adoption, and benefits of hearing aid use. American Journal of Audiology, 26(3S), 451–461. [DOI] [PubMed] [Google Scholar]
  10. Chisolm TH, Johnson CE, Danhauer JL, et al. (2007). A systematic review of health-related quality of life and hearing aids: Final report of the American Academy of Audiology task force on the health-related quality of life benefits of amplification in adults. Journal of the American Academy of Audiology, 18(2), 151–183. [DOI] [PubMed] [Google Scholar]
  11. Cox RM, Alexander GC, & Gray GA (2007). Personality, hearing problems, and amplification characteristics: Contributions to self-report hearing aid outcomes. Ear and Hearing, 28, 141–162. [DOI] [PubMed] [Google Scholar]
  12. Cruickshanks K, Wiley T, Tweed T, Klein BEK, Klein R, Mares-Perlman J, & Nondahl D (1998). Prevalence of Hearing Loss in Older Adults in Beaver Dam , Wisconsin. American Journal of Epidemiology, 148(9), 879–886. [DOI] [PubMed] [Google Scholar]
  13. Cruickshanks K, Tweed TS, Wiley T, Klein BEK, Klein R, Chappell R, … Dalton D (2003). The 5-Year incidence and progression of hearing loss. Archives Otolaryngology Head & Neck Surgery, 129, 1041–1046. Retrieved from www.archoto.com [DOI] [PubMed] [Google Scholar]
  14. Cruickshanks KJ, Nondahl DM, Tweed TS, Wiley TL, Klein BE, Klein R, ... & Nash SD (2010). Education, occupation, noise exposure history and the 10-yr cumulative incidence of hearing impairment in older adults. Hearing research, 264(1–2), 3–9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Davis A, Smith P, Ferguson M, & Stephens D (2007). Acceptability, benefit and costs of early screening for hearing disability. Health Technology Assessment, 11(42), 1–293. 10.3310/hta11420 [DOI] [PubMed] [Google Scholar]
  16. DeLong ER, DeLong DM, & Clarke-Pearson DL (1988). Comparing the areas under two or more correlated receiver operating characteristic curves: A nonparametric approach. Biometrics, 44(3), 837. 10.2307/2531595 [DOI] [PubMed] [Google Scholar]
  17. Department of Veterans Affairs. (2014). Prescribing and providing eyeglasses, contact lenses, and hearing aids. VHA Directive 1034(1). [Google Scholar]
  18. Efron B, Hastie T, Johnstone I, & Tibshirani R (2004). Least Angle Regression. Annals of Statistics, 32, 407–499. [Google Scholar]
  19. Fischer ME, Cruickshanks KJ, Wiley TL, Klein BEK, Klein R, & Tweed TS (2011). Determinants of hearing aid acquisition in older adults. American Journal of Public Health, 101(8), 1449–1455. doi: 10.2105/ajph.2010.300078 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Fisher DE, Li CM, Hoffman HJ, Chiu MS, Themann CL, Petersen H, … Cotch MF (2015). Sex-specific predictors of hearing-aid use in older persons: The age, gene/environment susceptibility-Reykjavik study. International Journal of Audiology, 54(9), 634–641. 10.3109/14992027.2015.1024889 [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Gopinath B, Schneider J, Hartley D, Teber E, McMahon CM, Leeder SR, & Mitchell P (2011). Incidence and predictors of hearing aid use and ownership among older adults with hearing loss. Annals of Epidemiology, 21(7), 497–506. 10.1016/j.annepidem.2011.03.005 [DOI] [PubMed] [Google Scholar]
  22. Guldvog B, Løyning Y Hauglie-Hanssen E, & Bjørnæs H (1994). Predictive Factors for Success in Surgical Treatment for Partial Epilepsy: A Multivariate Analysis. Epilepsia, 35(3), 566–578. [DOI] [PubMed] [Google Scholar]
  23. Hartley D, Rochtchina E, Newall P, Golding M, & Mitchell P (2010). Use of hearing aids and assistive listening devices in an older Australian population. Journal of the American Academy of Audiology, 21(10), 642–653. 10.3766/jaaa.21.10.4 [DOI] [PubMed] [Google Scholar]
  24. Helvik AS, Krokstad S, & Tambs K (2013). Hearing loss and risk of early retirement. The HUNT study. The European Journal of Public Health, 23(4), 617–622. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Hickson L, Hamilton L, & Orange SP (1986). Factors associated with hearing aid use. Australian Journal of Audiology, 8, 37–41. [Google Scholar]
  26. Hickson L, Timm M, & Worrall L (1999). Hearing aid fitting: Outcomes for older adults. Australian Journal of Audiology, 21, 9–21. [Google Scholar]
  27. Hickson L, Meyer C, Lovelock K, Lampert M, & Khan A (2014). Factors associated with success with hearing aids in older adults. International Journal of Audiology, 53(sup1), S18–S27. [DOI] [PubMed] [Google Scholar]
  28. Hosmer DW & Lemeshow S (2000). Applied logistic regression, 2nd edition. John Wiley & sons, New York, NY, pp 160–164. [Google Scholar]
  29. Humes LE, Wilson DL, & Humes AC (2003). Examinations of differences between successful and unsuccessful elderly hearing aid candidates matched for age, hearing loss and gender. International Journal of Audiology, 42 , 432–441. [DOI] [PubMed] [Google Scholar]
  30. Kricos PB (2006). Processing Capabilities Audiologic Manage- ment of Older Adults With Hearing Loss and Compromised Cognitive/Psychoacoustic Auditory. Trends in Amplification, 10(1). [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Kuhn M & Johnson K (2013). Applied Predictive Modeling. Springer Publishing. [Google Scholar]
  32. Kukull WA, & Ganguli M (2012). Generalizability: The trees, the forest, and the low-hanging fruit. Neurology, 78(23), 1886–1891. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Lupsakko TA, Kautiainen HJ, & Sulkava R (2005). The non-use of hearing aids in people aged 75 years and over in the city of Kuopio in Finland. European Archives of Oto-Rhino-Laryngology, 262(3), 165–169. 10.1007/s00405-004-0789-x [DOI] [PubMed] [Google Scholar]
  34. McCormack A, & Fortnum H (2013). Why do people fitted with hearing aids not wear them? International Journal of Audiology, 52(5), 360–368. 10.3109/14992027.2013.769066 [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Meyer C, & Hickson L (2012). What factors influence help-seeking for hearing impairment and hearing aid adoption in older adults?. International Journal of Audiology, 51(2), 66–74. [DOI] [PubMed] [Google Scholar]
  36. Meyer C, Hickson L, Lovelock K, Lampert M, & Khan A (2014). An investigation of factors that influence help-seeking for hearing impairment in older adults. International Journal of Audiology, 53(sup1), S3–S17. [DOI] [PubMed] [Google Scholar]
  37. Moon J, Baek SY, & Cho YS (2015). Hearing aid use and associated factors in South Korea. Medicine, 94(42), e1580. 10.1097/MD.0000000000001580 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Nachtegaal J, Festen JM, & Kramer SE (2012). Hearing ability in working life and its relationship with sick leave and self-reported work productivity. Ear and Hearing, 33(1), 94–103. [DOI] [PubMed] [Google Scholar]
  39. Nash SD, Cruickshanks KJ, Klein R, Klein BEK, Nieto JF, Huang GH, Pankow JS, & Tweed TS (2011). The prevalence of hearing impairment and associated risk factors: The Beaver Dam Offspring Study. Archives of Otolaryngology - Head and Neck Surgery, 137(5), 432–439. 10.1001/archoto.2011.15 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Newman CW, Weinstein BE, Jacobson GP, Hug GA (1990). The Hearing Handicap Inventory for Adults: Psychometric adequacy and audiometric correlates. Ear & Hearing, 11(6):430–433. [DOI] [PubMed] [Google Scholar]
  41. Ng JHY, & Loke AY (2015). Determinants of hearing-aid adoption and use among the elderly: A systematic review. International Journal of Audiology. 10.3109/14992027.2014.966922 [DOI] [PubMed] [Google Scholar]
  42. Nieman CL, Marrone N, Szanton SL, Thorpe RJ Jr, & Lin FR (2016). Racial/ethnic and socioeconomic disparities in hearing health care among older Americans. Journal of Aging and Health, 28(1), 68–94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Nondahl DM, Cruickshanks KJ, Wiley T L, Klein BEK, Tweed TS (2002). Prevalence and 5-year incidence of tinnitus among older adults: The Epidemiology of Hearing Loss Study. Journal of the American Academy of Audiology, 13(6), 323–31. [PubMed] [Google Scholar]
  44. Nondahl DM, Cruickshanks KJ, Wiley TL, Klein BEK, Klein R, Chappell R, & Tweed TS (2010). The ten-year incidence of tinnitus among older adults. International Journal of Audiology, 49(8), 580–585. 10.3109/14992021003753508 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Nondahl DM, Cruickshanks KJ, Huang GH, Klein BEK, Klein R, Nieto JF, & Tweed TS (2011). Tinnitus and its risk factors in the Beaver Dam Offspring Study. International Journal of Audiology, 50(5), 313–320. 10.3109/14992027.2010.551220 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Nondahl DM, Cruickshanks KJ, Huang GH, Klein BEK, Klein R, Tweed TS, & Zhan W (2012). Generational differences in the reporting of tinnitus. Ear & Hearing, 33(5), 640–644. 10.1097/AUD.0b013e31825069e8 [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Pavlou M, Ambler G, Seaman SR, Guttmann O, Elliott P, King M, & Omar RZ (2015). How to develop a more accurate risk prediction model when there are few events. British Medical Journal, 351, 1–5. 10.1136/bmj.h3868 [DOI] [PMC free article] [PubMed] [Google Scholar]
  48. Pearson OR, Busse MR, van Deursen RWM, Wiles C M (2004). Quantification of walking mobility in neurological disorders. QJM: An International Journal of Medicine, 97(8), 463–475. [DOI] [PubMed] [Google Scholar]
  49. Pencina MJ, D’Agostino RB, Larson MG, Massaro JM, & Vasan R S (2009). Predicting the 30-year risk of cardiovascular disease: The Framingham Heart Study. Circulation, 119(24), 3078–3084. 10.1161/CIRCULATIONAHA.108.816694 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Popelka MM, Cruickshanks KJ, Wiley TL, Tweed TS, Klein BEK, & Klein R (1998). Low prevalence of hearing aid use among older adults with hearing loss: The Epidemiology of Hearing Loss Study. Journal of the American Geriatrics Society, 46(9), 1075–1078. 10.1111/j.1532-5415.1998.tb06643.x [DOI] [PubMed] [Google Scholar]
  51. Radloff LS (1997). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measures, 1, 385–401. [Google Scholar]
  52. Reitan RM (1992). Trail Making Test Manual for Administration and Scoring. Reitan Neuropsychology Laboratory; Tucson, AZ. [Google Scholar]
  53. Saunders GH, Jutai JW (2004). Hearing specific and generic measures of the psychosocial impact of hearing aids. Journal of the American Academy of Audiology, 15(3), 238–248. doi: 10.3766/jaaa.15.3.6 [DOI] [PubMed] [Google Scholar]
  54. Schubert CR, Cruickshanks KJ, Fischer ME, Chen Y, Klein B E K, Klein R, & Pinto AA (2017). Sensory Impairments and Cognitive Function in Middle-Aged Adults. Journals of Gerontology - Series A Biological Sciences and Medical Sciences, 72(8), 1087–1090. 10.1093/gerona/glx067 [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Simmen BR, Bachmann LM, Drerup S, Schwyzer HK, Burkhart A, & Goldhahn J (2008). Development of a predictive model for estimating the probability of treatment success one year after total shoulder replacement - cohort study. Osteoarthritis and Cartilage, 16(5), 631–634. 10.1016/j.joca.2007.10.010 [DOI] [PubMed] [Google Scholar]
  56. Singh G, Lau ST, & Pichora-Fuller MK (2015). Social support predicts hearing aid satisfaction. Ear and Hearing, 36(6), 664–676. [DOI] [PubMed] [Google Scholar]
  57. Singh G, & Launer S (2016). Social context and hearing aid adoption. Trends in Hearing, 20, 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  58. Strauss E, Sherman EMS, Spreen O (2006). A Compendium of Neuropsychological Tests: Administration, Norms, and Commentary. 3. Oxford University Press; New York, NY: 2006. [Google Scholar]
  59. Staehelin K, Bertoli S, Probst R, Schindler C, Dratva J, & Stutz EZ (2011). Gender and hearing aids: Patterns of use and determinants of nonregular use. Ear & Hearing, 32(6), e26–e37. 10.1097/AUD.0b013e3182291f94 [DOI] [PubMed] [Google Scholar]
  60. Tibshirani R (1996). Regression Shrinkage and Selection Via the Lasso. Journal of the Royal Statistical Society: Series B (Methodological), 58(1), 267–288. 10.1111/j.2517-6161.1996.tb02080.x [DOI] [Google Scholar]
  61. Ventry IM, & Weinstein BE (1982). The Hearing Handicap Inventory for the Elderly: A new tool. Ear & Hearing, 3, 128–134. [DOI] [PubMed] [Google Scholar]
  62. Vestergaard Knudsen L, Öberg M, Nielsen C, Naylor G, & Kramer SE (2010). Factors Influencing Help Seeking, Hearing Aid Uptake, Hearing Aid Use and Satisfaction With Hearing Aids: A Review of the Literature. Trends in Amplification, 14(3), 127–154. 10.1177/1084713810385712 [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. World Health Organization (2020). Deafness and hearing loss factsheet. Retrieved from https://www.who.int/news-room/fact-sheets/detail/deafness-and-hearing-loss
  64. Wilson PWF, D’Agostino RB, Levy D, Belanger AM, Silbershat H, & Kannel WB (1998). Prediction of coronary heart disease using risk factor categories. Circulation, 97(18), 1837–1847. 10.1161/01.CIR.97.18.1837 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1653132-supplement-1.pdf (66.6KB, pdf)

RESOURCES