Study Sample and Population
Currently there is no single database available that could provide information on the variables of hearing impairment and its severity in China. Therefore, we combined two data sets: (1) the CHARLS 2011, 2013, 2015 pooled data; (2) the Hearing Survey 2019 data. The data of CHARLS 2011, 2013, 2015 are based on the national level without the severity of hearing impairment, and the Hearing Survey 2019 is only from a city but with the severity of hearing impairment, which is evaluated by objective audiometric measurement.
The first dataset CHARLS is a long-term tracking project, with its baseline survey conducted in 2011 and follow-up surveys in 2013 and 2015, eying on the community-dwelling Chinese adults aged 45 years old or above (25). CHARLS adopted the method of PPS (Probability Proportional to Size) and sampling from maps in 450 villages/resident committees, 150 counties/districts and 28 provinces (26), and combined detailed socioeconomic data with high-quality health data to address challenges of China’s aging problem. After data pooling, collating and cleaning, we got the nationally representative 20,099 samples from CHARLS pooled data, with 14,930 samples in baseline 2011.
The second data set Hearing Survey 2019 is the baseline survey of a randomized controlled trial in Linyi City, Shandong Province of China to investigate the association between the severity of hearing impairment and health outcomes. We included middle-aged and older adults with moderate hearing impairment and above (4 frequencies (0.5, 1, 2, 4 kHz)>40 dB), and with no hearing aids use at baseline. Those in the treatment group were prescribed with hearing aids, while those in the control group received no intervention. The primary outcome is changes in their health conditions over a 12-month period. According to the list provided by the hearing center of Linyi Disabled Persons’ Federation, 500 people were randomly selected, of which 36 patients were approached and were unwilling to give informed consent. In addition, 12 patients were excluded from the study because they were unable to understand the questionnaire, and 22 patients did not meet the inclusion criteria for the Hearing Survey 2019. We finally got the sample size of 430, of which 376 people were 45 years old and above.
Independent variables: hearing impairment and its severity
In CHARLS 2011, 2013 and 2015, hearing status was derived by asking whether respondents have hearing problems or not. The result is a subjective answer “yes” or “no”. Prior studies have proved that a single-item question about an individual’s hearing ability is moderately useful and valid to assess hearing and can be used for a population-based study (27, 28).
Hearing Survey 2019 measures hearing status by pure‐tone average (PTA) at the thresholds of 0.5, 1, 2, and 4k Hz. According to the classification by World Health Organization, those are defined as moderate hearing impairment if PTA > 40 dB and PTA ≤ 60 dB, severe hearing impairment if PTA > 60 dB and PTA ≤ 80 dB, and profound hearing impairment if PTA > 80 dB (29).
Outcome variables: physical and mental health status
In both CHARLS and Hearing Survey, we incorporated several variables to measure health status. Physical health was measured by whether they had chronic diseases (30), impaired ADLs (31) and impaired IADLs (32); Mental health was measured by whether they had depressive symptoms (33).
(1) Chronic diseases: Individuals were asked whether they had been diagnosed with chronic diseases, including hypertension, dyslipidemia, diabetes, cancer, chronic lung disease, liver disease, heart attack, stroke, kidney disease, digestive disease, arthritis or asthma. If individuals reported having at least one of these chronic diseases, we defined them as having chronic diseases.
(2) ADLs: Individuals were asked whether they had any difficulty with activities of daily living, including dressing, bathing or showering, eating, getting into or out of bed, using the toilet, or controlling urination and defecation. Impaired ADLs was defined as difficulty or inability with any of the activity (34).
(3) IADLs: Individuals were asked whether they had any difficulty with instrumental activities of daily living, including doing household chores, preparing hot meals, shopping for groceries, making phone calls, taking medications, or managing money. Impaired IADLs was defined as difficulty or inability with any of the activity (34).
(4) Depressive symptoms: Both surveys applied the screening tool Center for Epidemiologic Studies Depression Scale-10 items (CES-D-10) Chinese version to detect depressive symptoms (35). The CES-D-10 Chinese version has been shown to have good sensitivity, specificity, and predictive value (36). Subjects responded to the CES-D by rating the frequency of each mood occurred during the past week on a four-point scale, ranging from 0 (“none of the time”) to three (“most of the time”) (35). A cut-off score ≥ 10 on the total 0-30 CES-D-10 was optimal to identify individuals at risk of depressive symptoms (37).
Covariates
Prior research has identified several confounders that are associated with hearing impairment and health, and should be included in the analysis as covariates (5, 7, 18). These include age (continuous variable), gender (male or female), residency (rural or urban), educational attainment (illiterate, finishing primary school, finishing middle school and above), marital status (married or partnered; alone) and annual income. However, income is a less useful measure in late life due to exiting the labor force (38), and its missing value occupies nearly half in CHARLS, so we decided not to include income as the covariate.
Statistical Analyses
Descriptive analyses were used to present sample characteristics and hearing status. We used analysis of Variance (ANOVA) and the Chi-square test to compare characteristics between different hearing impairment categories. Logistic regression models and marginal effects were employed to predict the probabilities of having chronic diseases, impaired ADLs, impaired IADLs, and depressive symptoms with changes in hearing status while potential confounders were controlled (39). Before conducting the logistic regression, we have tested the dual collinearity of independent variables by the correlation matrix and the variance inflation factor (VIF) (40), and found no strong dual collinearity of independent variables in the Logistic regression. The software Stata version 14.0 for Mac was utilized for statistical analyses. All hypothesis tests were two-sided, with a P-value less than 0.05 considered statistically significant.