The Health 2000 Study, conducted in 2000–2001, is a cross-sectional survey comprising a comprehensive health interview and a detailed health examination in a population-based large sample of Finnish adults [25]. The sampling was carried out using two-stage stratified cluster sampling. The sample consisted of 8028 adults aged 30 and older, and was representative of the same-age Finnish general population. In the oldest age group, 80 and older, the sampling probability was doubled to ensure a sufficient number of the adults in this group were included in the sample. The present analysis was confined to adults aged 65 and older (range 65 to 99 years, N = 2 144). The sample participation rate for those aged 65 and over was 89 % for the interviews, and 83 % for the health examination.
Computer-assisted personal health interviews (questions available at http://www.terveys2000.fi/forms.html) were carried out at the participants’ homes or in residential institutions. In the interviews, the interviewer read the questions from the laptop screen and entered the answers directly into the laptop. Health examinations were performed in public health centers or in temporary examination facilities. If the participant was unable to travel to the health center or temporary facility, an abbreviated examination was carried out in the participant’s home [25].
Assessment of hearing
Pure-tone air-conduction hearing thresholds were assessed for both ears without a hearing aid using a screening audiometer (Micromate 304, Madsen Electronics) at the frequencies of 0.5, 1 and 2 kHz in a silent room [25]. Headphones with padded earpieces were used to minimize any environmental noise. The lowest signal intensity was 5 dB. The test started from the better-hearing ear or the right ear (if the participant reported no difference between the ears) at a frequency of 1 kHz (at 25 dB, or more for older people and those who seemed hard of hearing). Intensity was then reduced in decrements of 10 dB until the participant could no longer hear the signal. The intensity was then increased by in increments of 5 dB until the participant was able to hear the signal. The lowest intensity that the participant could hear was determined as the hearing threshold at 1 kHz. Next, hearing thresholds at frequencies of 2 and 0.5 kHz were similarly assessed, after which the other ear was assessed. If the participant could not hear at the intensity of 90 dB, 99 dB was marked as the hearing threshold. Better ear hearing level (BEHL0.5–2kHz) was calculated as the mean value over the measured frequencies. Test-retest repeatability was excellent (intraclass correlation coefficient = 0.97) [25]. Hearing level was dichotomized by categorizing those with BEHL0.5–2kHz > 40 dB as having hearing loss and those with BEHL0.5–2kHz ≤ 40 dB as having no hearing loss. A hearing threshold of 40 dB is the lower limit for moderate hearing loss [26] and was chosen as the cut-off since mild hearing loss (26 to 40 dB) is unlikely to markedly affect communication situations in health care, as these mostly take place between two persons in fairly quiet surroundings. Audiometric data were available for 1680 (78 %) persons.
Self-reported hearing difficulty was assessed with the question “Can you hear without difficulties what is said in a conversation between several people (with or without a hearing aid)?”. The response categories were 1) I can hear without difficulties 2) I can hear, but it causes difficulties and 3) I cannot hear at all. The latter two categories were combined as the number of men in the third category was too low to permit analysis of some of the outcome measures. Data on self-reported hearing difficulty was available for 1962 (92 %) persons.
Use of health services
In Finland, public health services, funded by local and central government, are available to all citizens. Some services (e.g. prevention, such as cancer screening) are free of charge and for some services (e.g. physician’s services, physiotherapy), small fees are charged. If a citizen uses private health services a minor proportion of the costs is usually borne by the Social Insurance Institution of Finland. Audiologic rehabilitation, including fitting a hearing aid, is provided free of charge by the public specialized health care service after referral from primary health care.
As a part of the home interview, the participants were asked how many times during the last 12 months they had visited a physician (not including hospitalizations) at a health center (primary care), hospital outpatient clinic (secondary care), occupational health care clinic, private clinic, in connection with a home visit, or elsewhere. The numbers of visits reported were summarized. Participants who reported hearing loss were also asked to state the number of physician visits within the last 12 months due to their hearing loss. This number was subtracted from the total number of physician visits to yield the number of visits not related to hearing loss. The total number of visits to a nurse within the last 12 months was obtained from three distinct questions (occupational nurse; other nurse; home visits). As physician visits (median 2, minimum 0, maximum 50), physician visits due to hearing loss (median 2, minimum 0, maximum 50) and nurse visits (median 0, minimum 0, maximum 1095) were non-normally distributed they were categorized. Physician visits and physician visits due to hearing loss were classified into categories no visits/1–4 visits/5 or more visits) and nurse visits were classified into categories no visits/1–5 visits/6 or more visits. The proportion of the participants reporting a high number of nurse visits was larger than the proportion reporting a high number of physician visits. Applying cut points of five for physician visits and six for nurse visits yielded comparable distributions between these variables. Participants were also asked whether they had received physical therapy (via a referral from a physician) and whether they had used mental health services within the last 12 months. They were further asked whether they had undergone any health examinations (organized within occupational health care; for war veterans; related to the driver’s license; in connection with unemployment; or for other any reasons) within the last 5 years. Participants were also asked to report any vision test, hearing test, mammography, or prostate-specific antigen (PSA) test taken during the last 5 years. The mammography analysis included women younger than 70, as they form the target group for the mammography screening arranged by municipal public health services. Participation in health promotion groups was defined as having attended a group targeting weight management, smoking cessation, neck/back rehabilitation, other physical exercise, mental wellbeing, support for patients’ relatives, or problems with alcohol or other addictions, parenthood, self-care/management of an illness, or other health promoting activities within the last 5 years. The participants were further asked: “Do you have a chronic illness for which you would like to get continuous treatment by a doctor but do not receive it?” and “Do you have a chronic condition for which you would like to get other type of care but do not receive it?”. Participants answering yes to either question were considered to experience unmet health care needs.
Potential confounders
Potential confounders were selected according to the criteria suggested by McNamee [27]. Accordingly, a confounder must be a cause of the outcome (use of health services), be correlated with the exposure (hearing loss/difficulty) and not be affected by the exposure. Sociodemographic factors included age, sex, mother tongue (Finnish/Swedish/other), income, years of education, and living alone (yes/no). Age and sex were obtained from the population register and income from taxation records. Household net income was divided by the number of consumption units in the household (first adult with weight 1, other adults 0.7 and children under 18 years 0.5) to yield the participant’s income. Self-reported diseases and health behavior that have been found to be associated with hearing loss, namely cardiovascular disease (myocardial infarction, angina pectoris, hypertension, lower limb arterial embolism) [5], stroke [28], arthritis (rheumatoid or osteoarthritis) [29], diabetes [30], alcohol use (8+ units/week vs. less) [31] and smoking (former or current vs. never) [32] were obtained from the home interview and self-administered questionnaire. Hearing aid use was defined as daily or almost daily use, and was based on two questions: “Do you have a hearing aid?” (yes/no) and “Do you use it daily or almost daily?” (yes/no). For calculation of body mass index (BMI), body weight and height were measured using standard procedures. If measured data were not available for a participant, self-reports were used to calculate BMI. Binocular far vision acuity was assessed on a decimal scale with eyeglasses on (if the participant usually wore them) using an illuminated (>350 lx) letter chart (Precision Vision Letter Chart Acuity Tests) [33]. Far vision acuity was dichotomized into the categories <0.5 (low vision), corresponding to <20/40 in 20/20 scale, and ≥0.5 (normal vision) [33].
Data analysis
The sampling design, i.e. stratification and clustering, was taken into account in all analyses. Observations were weighted to reduce bias due to non-response and to correct for oversampling of those aged >80 years using inverse probability weights constructed using register data on geographical area (university hospital and health center district), age, sex and mother tongue [25]. As age has a strong effect on hearing loss, background characteristics for those with and without a hearing loss were calculated controlling for the effect of age. P-values for comparisons were obtained from age-adjusted logistic and linear regression analyses using Stata version 14. Continuous variables were standardized for the analyses. Hearing loss was found to have significant interactions with sex on physician visits, physical therapy, and hearing test. Self-reported hearing difficulty showed a significant interaction with sex on physical therapy. Therefore, the results on these outcomes are reported separately for men and women.
First, we analyzed the proportions of health service users, adjusted for sex (in case of no interaction) and age, using logistic regression analysis and the predictive margins function in Stata. In the case of ordinal regression analysis, the odds ratio describes how likely persons with hearing loss are to have a certain value (versus all the lower values) of the ordinal outcome variable compared to those without hearing loss. Then, multivariable-adjusted logistic regression models were run using MPlus version 7 [34]. In the multivariable-adjusted models, sociodemographic variables and hearing aid use were used as covariates for all the outcome variables. In addition, diseases, smoking, alcohol use, and BMI were used as additional covariates for physician and nurse visits and participation in a health promotion group. For use of physical therapy, cardiac diseases, stroke, and rheumatoid arthritis/osteoarthritis were the additional covariates. For vision examination, far vision, diabetes and stroke were the additional covariates. Breast cancer and prostate cancer were used as additional covariates for mammography and the PSA test, respectively. The same analyses were repeated with self-reported hearing difficulty as the main predictor. Hearing aid use was not entered into the model since persons who had a hearing aid were asked to evaluate their hearing when wearing the hearing aid. The analyses employed the maximum likelihood estimator which automatically takes into account missing data in the dependent variables but not in the independent variables. Auxiliary variables were not used in the maximum likelihood estimation.
Next, the maximum likelihood method was applied in another way to further test that the results were not biased by missing data in the independent variables. This was done by repeating the above mentioned regression analyses and simultaneously requesting means and variances for the independent variables and using Monte Carlo integration without analysis weights in MPlus. In this procedure, the analysis takes into account missing data also in the independent variables. Maximum likelihood method does not fill in missing values but uses observed data to estimate the parameters of the variables with missing data. Based on the available data of the variables in the regression model (complete and incomplete), it identifies parameter estimates that have the highest likelihood of underlying the observed data. The maximum likelihood method, along with multiple imputation, is among the two missing data analysis techniques that are considered as efficient for accounting for missing data when the data are missing at random or missing completely at random [35]. Even if the data is not missing at random maximum likelihood yields less biased estimates than deletion techniques [35].
