Longitudinal Changes in Hearing and Visual Impairments and Risk of Dementia in Older Adults in the United States

Key Points Question Is dual sensory impairment associated with risk of dementia, including Alzheimer disease and vascular dementia, among older adults? Findings In this cohort study that included 2927 adults aged 65 years and older, dual sensory impairment was associated with a 160% increased risk for all-cause dementia and a 267% increased risk for Alzheimer disease. Meaning These findings suggest that assessment of both hearing and vision may help to identify older adults who are at high risk of developing dementia.


Sensory domain Questions
Hearing • Can you hear well enough (with hearing aid if necessary) to use the telephone?
• Can you hear well enough (with hearing aid if necessary) to listen to a radio?
• Can you hear well enough (with hearing aid if necessary) to carry on a conversation?

Vision
• Can you see well enough (with glasses if needed) to drive?
• Can you see well enough (with glasses if needed) to watch TV?
• Can you see well enough (with glasses if needed) to recognize someone across the room?
• Can you see well enough (with glasses if needed) to read the newspaper?

Cox regression model diagnostics
Testing the proportionality assumption using the Schoenfeld and scaled Schoenfeld residuals: We verified the proportionality assumption by testing for a non-zero slope in a generalized linear regression of the scaled Schoenfeld residuals on functions of time. A non-zero slope is an indication of a violation of the proportional hazards assumption. We assessed the proportional hazards assumptions statistically and visually. The graph of the scaled Schoenfeld residuals indicate the absence of a nonzero slope and the statistical test of a non-zero slope is not significant (p-value = 0.87), which suggests there is no violation of the proportionality assumption.
Goodness of Fit using Cox-Snell residuals: We evaluated the fit of the model using the Cox-Snell residuals. We graphed the Nelson-Aalen cumulative hazard function and the Cox-Snell residuals as the time variable, and compared the hazard function to the diagonal line. The hazard function follows the 45 degree line, except towards the end of the line, which is to be expected when using censored data with large values of time and is not a cause of significant concern. Overall, we conclude that the model fits the data well.

3.
Verifying missing at random assumption for use of multiple imputation Information on APOE genotype was found to be missing in the most numbers and greatest frequency (nmissing = 239; 8.2%) compared to other variables (e.g., education, smoking status, alcohol intake, body mass index, physical activity, diabetes, hypertension, and total cholesterol level), which had very small proportions of missing data (<1%). Descriptions of how much missing information was present in these variables are provided in Table 1.
In order to appropriately use multiple imputation for participants missing information on APOE genotype, we checked whether the missing data for APOE genotype was missing at random. Below are the results using logistic regression to evaluate whether the missing data was missing at random by examining if any of the variables in the data predict missingness: Identifying potential predictors of APOE genotype missingness We found that education, smoking status, alcohol intake, body mass index, diabetes, hypertension, total cholesterol levels, and physical activity were significantly associated with missingness of APOE genotype, and the cases missing education, smoking status, alcohol intake, body mass index, diabetes, hypertension, total cholesterol level, or physical activity were also missing APOE genotype (as shown in the results below), suggesting that the data are missing at random.
Overlap between missingness in APOE genotype with other variables that have missing data  (1)