Study Design
This was a secondary data analysis of a longitudinal study. Secondary data analysis refers to analysis of data that are collected by someone else for another primary purpose (37). Therefore, we report this study by following the guideline of The Reporting of studies Conducted using Observational Routinely-collected health Data (RECORD) Statement (38). The data were collected between January 2005 and December 2016. All methods were performed in accordance with the Declaration of Helsinki and approved by the Human Research Ethics Committee of The University of Hong Kong (Ref No: EA1904045).
Setting
The study was conducted in the 7 government-subsidized long-term care facilities, also known as Residential Care Services for the Elderly, operated by one of the largest non-governmental organizations in Hong Kong. Their services include residential care, meals, personal care and limited nursing care for elders who suffer from poor health or physical/mild mental disabilities with deficiency in daily living activities but are mentally suitable for communal living (39).
Participants
Only residents in the dataset who fulfilled the following eligibility criteria were selected for analysis:
Inclusion criteria
- Older people as defined by WHO at age ≥ 60 years at baseline (40),
Exclusion criteria
- Those who had no follow-up data,
- Those whose baseline cognitive impairment is severe, as defined by a Cognitive Performance Scale score of ≥ 5 (i.e., possible score range=0-6) (41), as there is limited room for them to decline further, and
- Those who have resided in the long-term care facilities for less than three years, because the extent of the cognitive decline in three years in people without dementia and with early dementia is relatively small (i.e., 0-2 MMSE points decline/year) (42). Cognitive Performance Scale is less likely to be sensitive enough to identify the minute cognitive decline because each Cognitive Performance Scale point difference varies from 0.8 to 6.3 MMSE points (43).
Data sources and measurement
The Minimum Data Set-Resident Assessment Instrument Version 2.0 (MDS-RAI 2.0), Hong Kong version, was utilized as the measurement tool (41, 44). It is a comprehensive tool measuring LTCF residents’ care needs with 22 sections (e.g., cognitive patterns, communication/hearing patterns, vision patterns, and disease diagnosis). The assessment drew on multiple data sources, which included direct questioning of care recipients and caregivers, observation of care recipients in the long-term care environment, and a review of related documents such as medical records. Various trained professionals (nurses, social workers, occupational therapists, and physiotherapists) collected the data following the standardized MDS-RAI 2.0 Users’ Manual (45). Nurses monitored the assessment and coordinated the care. The organization conducted in-house standardized training for each assessor. The MDS has been validated and proven to have good criterion validity on many of its sub-scales and good reliability in 80-90% of items (46). For example, the sub-scale Cognitive Performance Scale has substantial agreement with Mini-mental State Exam in identification of cognitive impairment (sensitivity=0.94, specificity=0.94, AUC=0.96).(47)
The data came from a dataset of a project entitled the Hong Kong Longitudinal Study on Long-Term Care Facility (LTCF) Residents, in which all residents in 11 long-term care facilities were repeatedly assessed (44). The exact period between two repeated-assessments could not be precisely controlled because of the availability of the residents (e.g., hospitalization) and the availability of the certified assessors. The period of two repeated-assessments of each resident varied but ranged from six to twelve months. In that project, health data were routinely collected from a cohort of Chinese long-term care residents in Hong Kong to review and improve clinical practices in long-term care facilities. The advantage of using regularly collected clinical data for analysis is that it allows dynamic relationships between variables to be examined over time.
Variables
Demographic and related clinical profile
Age, gender, cognitive function at baseline, follow-up year, and related co-morbidities were measured to describe residents’ demographic and clinical profiles.
Predictors
Hearing impairment was measured by the item entitled “Hearing” in the section on Hearing Patterns in the MDS-RAI 2.0. Residents’ hearing impairment was quantified to a score ranging from 0 to 3, which was in turn re-coded as a dichotomous variable. No impairment described being able to adequately hear normal talk, TV, and phone (i.e., Hearing score=0); impairment level (i.e., Hearing score=1-3) ranged from having minimal difficulty hearing (e.g., having difficulty hearing when not in a quiet setting) to being highly impaired (e.g., absence of useful hearing).
Visual impairment was measured by the item entitled “Vision” in the section on Vision Patterns in the MDS-RAI 2.0. Residents’ visual impairment was quantified to a score ranging from 0 to 4, which was then re-coded as a dichotomous variable. No impairment (i.e., Vision score=0-1) referred to seeing fine details adequately, including regular or large print in newspapers or books. Impairment (i.e., Vision score=2-4) either indicated moderately impaired vision (e.g., not able to see newspaper headlines but able to identify objects) or severe impairment (e.g., no vision or only able to see light, colors, or shapes).
Dual sensory impairment was measured by combining the re-coded hearing impairment and visual impairment scores. Residents were categorized as follows: 1) no impairment (i.e., no hearing or visual impairment), 2) one sensory impairment (i.e., having either hearing or visual impairment), and 3) dual impairment (i.e., being both audibly and visually impaired).
Mediators
Hearing aid use pattern was measured by the item entitled “Communication Devices” in the section on Hearing Patterns in the MDS-RAI 2.0. Residents’ use of hearing aids fell into one of these four categories: a) hearing aid present and used, b) hearing aid present and not used regularly, c) other receptive communication techniques used, and d) none of above. The use of hearing aids was re-coded into a 3-point categorical variable: 1) hearing aid present and used (i.e., category a), 2) hearing aid present and not used regularly (i.e., category b), and 3) having no / not using hearing aids (i.e., combining categories c and d).
Visual aid use was measured by the item entitled “Visual Appliances” in the section on Visual Patterns in the MDS-RAI 2.0. It is a dichotomous variable. Visual aid use denoted the use of glasses, contact lenses, or magnifying glasses. Not using visual aids referred to not using those devices.
Outcome
Cognitive decline was captured by the change in cognitive function from baseline to the last time point of observation (i.e., T0-T1). A higher score indicates more severe cognitive decline. Cognitive function was measured using the Cognitive Performance Scale (41), a hierarchical scale assessing cognitive function specifically in five areas: short-term memory, cognitive skills for daily decision-making, the ability to make oneself understood, comatose status, and dependence in eating (41). Scores range from 0 to 6; a higher score entails poorer cognitive function. Cognitive Performance Scale has good inter-rater reliability (Spearman ρ=0.85) (41), as well as good agreement with MMSE (r=-0.863, p<0.001) good criterion validity with MMSE to identify cognitive impairment (sensitivity=0.90-0.94, specificity=0.85-0.95) (47).
Confounders
Residents’ age and gender were measured at baseline in the database. Cognitive function at baseline, which is associated with the subsequent cognitive decline (48), and co-morbidities, including diabetes mellitus (DM), hypertension (HT), stroke, and dementia (49-51), known to confound the effect of sensory impairment on cognitive decline were controlled.
Bias
Sampling bias was probable since the median length of stay in the long-term-care facilities was 73.4 weeks (i.e., 1.4 years) (52). The majority of the residents were excluded because they were not eligible (i.e., 81.6%).
Study sample size
Only 2,233 residents in the 7 long-term care facilities who fulfilled the eligibility criteria were included in the sample.
Statistical methods
IBM SPSS Statistics 25 was adopted to conduct the statistical analysis. Mean with standard deviation and frequency with percentage were used to describe residents’ profiles and related variables in this study. To test the hypotheses #1, a univariate general linear model was employed, where cognitive decline served as the dependent variable and hearing impairment, visual impairment, and dual sensory impairment served as the independent variables. To test hypothesis #2, the test of linear moderated mediation using PROCESS macro for SPSS was employed (53). The dependent variable was cognitive decline, independent variables were hearing impairment and visual impairment, and the mediators were the use of hearing aids and visual aids. All models were adjusted for known confounders (i.e., age, gender, baseline cognitive function, DM, HT, dementia, and stroke), with 0.05 being the significance level. The estimates of the effects of the predictors were reported using either F-statistics or the estimated marginal mean difference of the outcome (i.e., the CPS change score between categories).
Data access and cleaning methods
The authors are team members of the “Well-being and Associated Factors of Vulnerable Populations in Long-term Care in Hong Kong” project, entailing our direct access to the database. After extracting data according to the eligibility criteria, we excluded participants with missing data in relation to the variables involved in this analysis.