The data for this study comes from the National Health Interview Survey (NHIS) (2004–2013) in the United States, which is a representative, stratified, multistage probability survey. The US Centers for Disease Control and Prevention and the NCHS conducted annual household interview surveys through the Census Bureau and collected the basic health status of family members. Participants are members of the US population living in different communities. The specific design and method are all in NCHS (www.cdc.gov/nchs/nhis/about_nhis.htm; accessed 7 January 2018). The data is public and in compliance with the ethical board review of the corresponding author’s institution.

This study included representative US adult data collected in every household from 2004 to 2013. The complete reported hearing loss data was available for a total of 493,036 participants over 18 years of age who met the criteria for this study. Among them, 37,048 were excluded because of missing data on diabetes (n = 200), missing data on smoking and drinking (n = 8,524), missing data on physical activity (n = 11,119) and missing data on potential covariates (personal variables, lifestyle factors or chronic health conditions; n = 17,205), resulting in the final analytical sample of 455,988 adults.

The mortality rate is recorded by the National Death Index (NDI). It has showed that the accuracy of all-cause mortality and cause-specific mortality is consistent with NDI records [30, 31]. Cause of death is coded using the International Classification of Diseases (ICD-10) In addition to all-cause mortality, there are mainly outcomes of cause-specific mortality (CVD; heart disease; and stroke; CLRD; cancer; Diabetes; Alzheimer’s disease; Influenza and pneumonia; and accidents). See Supplementary Table SA for the ICD-10 codes.

Hearing loss status was ascertained at the initial interview via two binary coded (i.e., yes/no) items: 1) Can you hear what other people are saying when you talk to them. 2) Do family members and friends think you have difficulty hearing. Those participants responding “Yes” to the first item and “No” the second item were classified as good. Those participants responding “No” to both questions were classified as a little hearing difficulty. Those participants responding “No” to the first item and “Yes” the second item were classified as a lot of hearing difficulty. In the present study, the participants were divided according to their general hearing status, ICD-10 codes were used by the NCHS to classify hearing status as follows: as follows: 1) good hearing, 2)a little hearing difficulty, 3)a lot of hearing difficulty, 4)profoundly deaf.

Information on covariates about sociodemographic factors and lifestyle factors at baseline was obtained via examination and questionnaire-based interviews: (1) Sociodemographic factors: age (1∼18–45, 2∼45–65, 3∼65+), sex (1∼male, 2∼female), educational level (1∼Less than high school degree, 2∼High school degree, 3∼More than high school degree) (2) Lifestyle factors as follows: Alcohol intake 1∼Lifetime abstainer, 2∼Former drinker 3∼Current drinker), Smoking status (1∼Never cigarette, 2∼Former cigarette, 3∼Current cigarette), BMI (1∼<25, 2∼25–30, 3∼≥30). In addition, we included physical activity, hypertension, diabetes, stroke, coronary heart disease (CHD), or cancer. Age, sex, race, education level, income, alcohol intake, smoking status, and BMI all were treated as categorical variables.

Descriptive statistics were used to describe the distribution of parameters and baseline characteristics of US participants’ HL. We used the χ ² test to assess the baseline characteristics of participants in the four levels (good, a little trouble, a lot of trouble, deaf) of hearing loss. Chi-square independence test is used to check the classification difference of the weighted percentage of hearing loss. Among this, HL mortality is divided into three age groups according to the NHIS population: 18–45 years old, 45–65 years old, and >65 years and older. We used a multivariate Cox proportional hazards regression model with proportional assumptions to examine the hazard ratios (HR) with 95% CIs for all cause and cause-specific mortality for a little trouble hearing, a lot of trouble hearing, and deaf. To assess the different potential confounding effects on the relationship between levels of HL and mortality, we developed four models by incorporating three sets of covariates into the model in turn. We also conducted a subgroup analysis for all special cause mortality rates at the same time. In order to test the stability of the results, we conducted sensitivity analyses. Including participants with chronic diseases had little effect on risk estimates of all-cause mortality were excluded and all deaths within 2 years of follow-up were excluded.

All data analyses were weighted and conducted using STATA version 12.0 (Stata Corp, College Station, TX, United States). A two tailed p < 0.05 was considered to be statistically significant.

Among 455,988 US adults, there are 380,233 in good status, 62,491 are regarded as a little trouble hearing, 12,044 are a lot of trouble hearing, and 1,220 are in the deaf category. Of these participants, those with hearing loss were more likely to be older, men, Non-Hispanic White, more than a high school degree, coming from a middle income status, physically inactive, current drinkers. Table 1 shows the baseline characteristics of hearing loss among participants aged 18 years and older. Each baseline feature has statistically significant differences in the four categories of HL (p < 0.001).

Table 2 displays the HR of all cause and specific-cause mortality caused by HL. After several years of 479, 856 follow-up, 59,791 participants died from all causes, 13,488 from CVD deaths (10,418 deaths related to heart disease and 3,070 deaths related to stroke disease), 14,317 from cancer deaths, 3,175 from chronic lower respiratory diseases, 2,490 from accidents, 1,478 from Alzheimer’s disease, 1,804 from diabetes, 1,133 from influenza and pneumonia. In unadjusted Model 1, HL had a greater risk of all cause, CVD disease, Alzheimer’s disease, and influenza and pneumonia disease. In comparison to those good hearing, the HR of all-cause mortality were higher in little trouble hearing 2.70 (95% CI, 2.63–2.77) a lot of trouble hearing 5.46 (95% CI, 5.24–5.69) and deaf 5.29 (95% CI, 4.72–5.94). After adjusting the covariates for all-cause mortality in Model 4, the risk of death for people with a little trouble hearing decreased from 2.70 to 1.17. A large number of HL adults have a significant reduction from 5.46 to 1.45, for deaf adults, from 5.29 to 1.54. In addition, after adjusting from model 1 to model 2 (age), the risk of mortality from all nine specific-causes of disease has been greatly reduced. Similar associations were observed between HL and all cause and specific-cause mortality (Figure 1).

In terms of mortality results, a significant interaction was found between all covariates with HL. Compare people with HL diagnosed before the age of 45 vs. after the age of 45, the HL is related to the risk of all-cause mortality to those at the age of >45 years. For all-cause mortality, hearing loss risk at the age of >65 in little trouble, a lot of trouble and deaf were increased by 19% (HR1.19, 95% CI1.16-1.23), 52% (HR1.52, 95% CI 1.46–1.58), and 62% (HR1.62, 95% CI 1.42–1.84), (p < 0.001). Adults older than 65 years were associated with the highest all-cause mortality risk in those participants who were deaf. In the subgroup analysis according to these covariates (Table 3), for the same hearing condition, among the older adults (versus young adults), women (versus man), more than high school degreed adults (versus low school degree), <25 kg/m² (versus >25 kg/m²), never cigarette (versus former and current smokers), lifetime abstainer (versus former drinker and current drinker) were associated with greatest risk of all-cause mortality.

In order to confirm the results of the study, we performed two sensitivity analyses. Firstly, participants with chronic diseases had little effect on risk estimates of all-cause mortality were excluded (little trouble: 1.19 (1.16–1.23); a lot of trouble: 1.59 (1.61–1.67); deaf: 1.66 (1.42–1.94). Secondly, all deaths within 2 years of follow-up were excluded [little trouble: 1.17 (1.14–1.20); a lot of trouble: 1.44 (1.38–1.50); deaf: 1.61 (1.43–1.82)] (Supplementary Tables S2A,B).

In our study, in addition to previous studies in which HL was associated with all-cause mortality, we determined whether HL was associated with mortality due to CVD; heart disease; stroke; CLRD; cancer; diabetes; Alzheimer’s disease; influenza and pneumonia; and accidents. Our study confirmed that hearing loss was associated not only with all-cause mortality, but also with specific mortality for each cause. Our subgroup analysis found that for the same hearing condition, in older adults, women, high school education and above, and lifetime alcohol abuse were associated with an increased risk of all-cause mortality, and these important findings could be important for our prevention of HL. Our findings have important public health implications as they suggest that future recommendations could be expanded to preventing or inhibiting the pathogenic factors of hearing loss is important for reducing the risk of death.

Our study has several advantages. First, this study is a prospective cohort study, and the data come from a large sample of representative US adults. NHIS aims to represent the US population and the survey response rate is extremely high (95%–98%). In addition, we constantly adjusted the potential confounding factors in the model and conduct subgroup and sensitivity analysis to ensure the reliability of this study. Some restrictions should be noted. Firstly, the state of HL is obtained self-reported questionnaires of participants, which may cause recall bias. We still cannot completely exclude the remaining confounding factors. 7.5% of the total cohort was excluded (n = 37,038/493,036 participants) were excluded owing to missing data on exposures or covariates, as well as hearing loss who are answered by other family members, which might have led to bias if those excluded differed from those not excluded. The results might not be adapted to children. Besides, age available only is a three-level ordinal variable in the US dataset. Secondly, this study did not include non-fatal events, which limited our ability to estimate the risk of incident diseases. Thirdly, I do not have data on the hearing aids and/or cochlear implants available, poor communication between doctors and hearing loss patients results in a variety of adverse outcomes.

In the present study, elderly participants with extreme HL (deaf patients) are associated with increased risk of all-cause and cause-specific mortality. The findings of our research and the growing literature have also revealed the potential burden of HL and consequences, and it is necessary to improve the screening and management of HL among U.S. adults, especially among the elderly.