Osteoporosis, Bisphosphonate Use, and Risk of Moderate or Worse Hearing Loss in Women

Sharon G. Curhan MD, ScM; Konstantina Stankovic MD, PhD; Christopher Halpin PhD; Molin Wang PhD; Roland D. Eavey MD, SM; Julie M. Paik MD, ScD; Gary C. Curhan MD, ScD


J Am Geriatr Soc. 2021;69(11):3103-3113. 

In This Article


Study Population

The Conservation of Hearing Study (CHEARS) examines risk factors for hearing loss among participants in the NHS and NHS II, two large ongoing prospective cohort studies. NHS was established in 1976 and enrolled 121,700 female registered nurses ages 30–55 years. NHS II began in 1989 and enrolled 116,430 female registered nurses ages 25–42 years. In this study, NHS participants were followed up to 34 years, from 1982 to 2016; hence, the women were aged 36–61 years at baseline and aged 70–95 years at the end of follow-up. NHS II participants were followed up to 22 years, from 1995 to 2017; hence, the women were aged 31–48 years at baseline and aged 53–70 years at the end of follow-up. Participants completed questionnaires at baseline and every 2 years regarding numerous health outcomes, medications, diet, and lifestyle factors. Both cohorts have >90% follow-up response rates and only 5% of person-time has been lost to follow-up.[15] We limited our study to women who provided information on their hearing on the biennial questionnaires. Of the 65,855 women in NHS and 90,059 women in NHS II who answered the questionnaires, we excluded those who reported a hearing problem that began before the baseline for each analysis, did not report a date of onset, or reported cancer other than non-melanoma skin cancer (due to possible exposure to potentially ototoxic chemotherapy). In our primary analyses, 60,821 NHS and 83,078 NHS II women were included in the analytic samples. The study protocol was approved by the Institutional Review Board of Brigham and Women's Hospital.

Assessment of Exposures

Information on history of osteoporosis or low bone density (LBD) and date of diagnosis was collected by self-report on biennial questionnaires beginning in 1982 (NHS) and 2001 (NHS II). In NHS, women were asked whether they had clinician-diagnosed osteoporosis and date of diagnosis. In NHS II, women were asked about clinician-diagnosed osteoporosis or LBD and date of diagnosis, with osteoporosis and LBD queried separately in some questionnaire cycles and jointly in others. Therefore, in NHS II we examined osteoporosis/LBD as a combined category. Participants were also asked about regular use of bisphosphonates (e.g., Fosamax, Boniva, Actonel) beginning in 1998 (NHS) and 1999 (NHS II) and updated every 2 years. Self-reported diagnosis of osteoporosis and LBD was demonstrated to be valid; medical records for bone density tests were obtained in a random sample and self-reported diagnoses were confirmed in 114/120 (95%) using World Health Organization criteria for osteopenia/osteoporosis.[16]

Assessment of VF and HF has been described previously.[17] In NHS, participants were asked on the 2012 questionnaire about lifetime history of a clinician-diagnosed "vertebral (spine) fracture, X-ray confirmed" and year of diagnosis. Participants were asked again about VF diagnosis in 2014. In NHS II, participants were asked about VF and date of diagnosis in 2013. In NHS, participants who reported a VF in 2002 or afterward were mailed supplemental questionnaires and medical records were requested; VF was confirmed by clinical notes or radiology reports among those for whom sufficient information was available. In NHS, we examined only confirmed VF. In NHS II, few women had confirmed VF, thus we examined self-reported VF in the primary analysis; a sensitivity analysis was conducted examining only confirmed VF.

In NHS, participants reported previous HFs on the 1982 questionnaire and incident fractures were reported on subsequent biennial questionnaires. In NHS II, participants were first asked about HF and date in 2005. Self-reported HF was found to be highly reliable in these registered nurses.[18] Thus, we examined self-reported HF in our analyses of HF and risk of hearing loss.

Ascertainment of Outcome

The primary outcome, self-reported hearing loss that was moderate or worse in severity, was determined based on responses to questions regarding hearing status, and the date when a change in hearing was first noticed, on biennial questionnaires (https://www.nurseshealthstudy.org/participants/questionnaires). We a priori chose to examine moderate or worse hearing loss as the primary outcome to focus on hearing loss that is likely to be the most clinically meaningful and to minimize misclassification. The use of questionnaires to assess hearing loss in large populations was found to be reasonably reliable in previous studies,[19,20] and was effective in detecting significant relations in these cohorts.[21,22] In a validation study of self-reported hearing loss compared with audiometric hearing loss, sensitivity of a single question to assess hearing loss among women aged <70 was 95% for detecting moderate hearing loss (better ear pure-tone average at 0.5, 1, 2, 4 kHz [PTA0.5,1,2,4] >40 dB) and 100% for detecting "marked" hearing loss (PTA0.5,1,2,4 >60 dB); specificity was 65% and 64%, respectively.[20]

In secondary analyses among a subcohort in the CHEARS Audiometry Assessment Arm (AAA), we evaluated audiometric thresholds among women with and without osteoporosis/LBD, and with and without bisphosphonate use. Audiometric thresholds are a measure of hearing sensitivity based on the intensity (loudness) of sound. Pure-tone audiometry assesses the threshold of sound intensity reliably detected by each ear, ranging from −10 dB HL (decibels hearing level) to 120 dB HL, at each individual frequency across a range of frequencies (0.5–8 kHz). A higher audiometric threshold indicates poorer hearing sensitivity. The methods for the CHEARS AAA are described elsewhere.[23,24] Briefly, we assessed pure-tone air conduction hearing thresholds at 19 geographically diverse testing sites, conducted by licensed audiologists using standardized protocols and equipment calibrated to meet American National Standards Institute standards. We a priori invited participants who reported excellent or good hearing and no history of otologic disease to examine early threshold changes; the mean (SD) worse ear audiometric thresholds were 12.3 (8.5) at 0.5 kHz; 11.9 (8.6) at 1 kHz; 13.7 (9.8) at 2 kHz; 16.9 (11.2) at 3 kHz; 21.1 (12.9) at 4 kHz; 29.4 (15.5) at 6 kHz; and 32.6 (18.6) at 8 kHz. Multivariable-adjusted logistic regression models were used to examine differences in mean audiometric thresholds among the 3749 participants who completed baseline testing. Characteristics of AAA participants did not differ appreciably from the main cohort[24] (Table S1).

Assessment of Covariates

We considered potential confounders previously suggested to be associated with osteoporosis or hearing loss in these cohorts or previous literature. In our primary analyses, factors included age, race/ethnicity, body mass index, waist circumference, physical activity, Dietary Approaches to Stop Hypertension adherence score, alcohol intake, hypertension, diabetes mellitus, type II, smoking, thiazide use, menopausal status, and ibuprofen, acetaminophen, and oral hormone use. In additional models, we included furosemide use and intakes of specific nutrients, including dietary and supplemental calcium, magnesium, potassium, sodium, phosphorus, animal protein, sucrose, vitamin C, and vitamin D. Covariate information was obtained from validated biennial questionnaires[15] and updated every 2 years; dietary information was updated every 4 years. Indicator variables were created for missing information for each covariate and included in the multivariable models.

Statistical Analysis

The study design was prospective, with information collected before the reported onset of hearing loss. Person-time of follow-up was calculated from the date of return of the baseline questionnaire for that analysis until the date of onset of self-reported hearing loss or end of follow-up. Participants with cancers other than non-melanoma skin cancer were censored when reported during follow-up. Women with no history of osteoporosis or LBD served as the referent group. Person-months of follow-up were allocated according to exposure status at the start of each follow-up period. If a participant reported hearing loss that was mild in severity, that participant did not contribute person-time for that time period but could re-enter the analysis if moderate or worse hearing loss was reported in a subsequent time period. We used Cox proportional hazards regression models to estimate relative risks and 95% confidence intervals (95% CIs) in age- and multivariable-adjusted analyses. We used the Anderson–Gill data structure,[25] with a new data record created for each biennial questionnaire, to manage time-varying covariates efficiently. To control as finely as possible for confounding by age, calendar time, and any possible two-way interactions between these two time scales, we stratified the analysis jointly by age in months at start of follow-up and calendar year of the current questionnaire cycle. The time scale for the analysis was then measured as months since the start of the current questionnaire cycle, which is equivalent to age in months. All p values are two-tailed and considered statistically significant at p < 0.05. Statistical tests were performed with SAS statistical software, version 9.4 (SAS Institute Inc., Cary, NC).