Data and sample
We analyzed cross-sectional data sourced from the inaugural wave of the Longitudinal Ageing Study in India (LASI), 2017-18. The LASI is a large-scale nationally representative survey of health, economic, and social well-being of the Indian population aged 45 and older, as well as their spouses. The survey was conducted from April 2017 to December 2018, covering 35 states and union territories in the country. It included a sample of 31,464 persons aged 60 and above. The LASI is internationally harmonized with the U.S. Health and Retirement Study, and contains rich information on demographics, economic status, living environment, employment, chronic health conditions, family networks, and retirement. The survey also adopted a multistage stratified cluster sampling strategy to ensure representativeness and individual-level weights are available. Computer-assisted interviews were conducted face-to-face by trained interviewers. Response rates were generally high at around 85% on average.
Ethical approval for the LASI was obtained from the Indian Council of Medical Research and all participants provided written informed consent. A detailed description of LASI is given elsewhere [33, 34]. This paper is based on secondary data analyses and all methods were performed in accordance with the relevant research guidelines and regulations. After excluding 463 subjects with missing data on toileting facility, our final analytical sample comprised 20,789 community-dwelling adults aged ≥ 65 years spread across 35 states/territories. We included proxy respondents (who account for 11% of the analytical sample) to maximize sample size and also because it is fairly common to include proxy respondents in research involving older or disabled people. Missing values (including do not know or refuse) were minimal in the dataset and generally < 5% of the sample. Thus, we imputed missing values using mean imputation methodology. Individual-level weights (available from the dataset) are applied in all analyses in order to produce nationally representative population estimates. The weighted sample was N = 80,608,413.
Dependent variable
The outcome variable of interest was the presence of toileting disability. The LASI collates self-reported information on various ADL disabilities (e.g., dressing and bathing) using a standard question following the U.S. Health and Retirement Study. Specifically for the toileting ADL disability, respondents were asked: “Now, I will ask you about a few everyday activities. Please tell me if you have any difficulty with [using the toilet, including getting up and down] because of a physical, mental, emotional, or memory problem. Please exclude any difficulties you expect to last less than three months.” The answer categories provided were yes or no. Those who responded “yes” to the question were considered as having a toileting disability, as per prior studies [21,22,23,24].
Physical and social environmental factors
We draw on the disablement process model as the guiding framework to identify risk factors associated with toileting disability in this study. In the conceptual framework, environmental variables are grouped under the ‘extra-individual factors’ domain. This domain encompasses protective or exacerbating conditions occurring in the individual’s physical or social environment, which includes the presence of external supports that are critical especially in late-life disability [26]. The four other key domains in the framework are namely: (i) ‘pathologies’; (ii) ‘impairments’; (iii) ‘functional limitations’; and (iv) ‘intra-individual factors’, including both socio-demographic and health/lifestyle factors. Notably, this conceptual framework is aligned with WHO’s International Classification of Functioning, Health and Disability, which emphasizes that disability is the result of the interaction of the person with the environment [35]. In theoretical work, Fong and Feng [23] demonstrated how the disablement process model can be usefully deployed to classify risk factors in the context of toileting disability. Aspects of the physical environment may include measures of access to toilet facilities and the type of facility, while aspects of the social environment including external supports may include feeling safe at home or surroundings, trust in neighbors, reliance on mobility aids, and use of vision aids.
We adopt the same approach to identify potential risk factors of toileting disability in the LASI survey dataset (see Fig. 1). Under the ‘extra-individual factors’ domain, we identified a total of six variables related to the physical and social environment. For physical environment, we employed two separate indicator variables (1 = yes and 0 = no) for (i) lack of access to a toilet facility and (ii) sharing latrines with other households. Survey participants were asked: “What type of toilet facility does your household use?” Those who responded no facility (use open space or field) were classified as having no access to a safe sanitation system, which accounts for almost 30% of respondents in our weighted sample, confirming that this dichotomy is important to the present study.
A separate question was asked on whether respondents shared toilet facilities with other households, and those answering yes were classified as using shared latrines. The sharing of sanitation facilities is a common practice in many parts of India due to space and cost limitations. In fact, shared toilets are oftentimes the only option for many people living in the dense urban and rural areas throughout India [36,37,38]. For instance, Heijnen et al. [36] showed that showed that households in Orissa, India using communally-managed facilities (frequented by large numbers of households) were poorer and less educated, and the facilities were less clean and maintained as compared to households sharing a ‘localized’ sanitation facility with only their neighbors or landlord.
Risk factors of toileting disability based on LASI data and the disablement process conceptual framework
We account for social features of the toileting environment and usage of external supports using four distinct dummy variables. The first relates to trust in the neighborhood surroundings. Respondents were considered as having low neighborhood trust if they did not feel safe when they walked down their street/locality after dark. The second relates to safety in the immediate home environment. Specifically, LASI respondents were asked: “In general, how safe from crime and violence do you feel when you are alone at home?” Those who said not safe at all were categorized as not feeling safe at home. Use of assistive devices—external devices that are designed, made, or adapted to assist a person to perform a particular task—can influence the way that older adults operate within and interact with their physical and social environments. The LASI dataset contains a rich set of variables relating to aid or supportive devices that assist older adults in their daily life. Accordingly, we added a dummy variable for use of mobility devices (walker, walking stick, or wheelchair) and a separate dummy variable for use of vision aids (spectacles/contact lenses).
Control variables
Control variables were also systematically identified following the defined domains in the conceptual framework. For the ‘pathologies’ domain, we used seven indicator variables measuring of whether a respondent was ever diagnosed with a certain chronic condition, including chronic lung disease, diabetes, hypertension, stroke, angina/heart disease, arthritis, and cancer. The ‘impairments’ domain comprised three dummy variables indicating physical impairment (in lower or upper body), visual impairment, and hearing impairment, respectively. Consistent with past studies [21, 23], we used separate indicators in the ‘functional limitations’ domain to account for respondents’ difficulty in climbing stairs; stooping, kneeling or crouching; and extending arms. Additionally, we included difficulty carrying a heavy bag of groceries and difficulty walking 100 yards (approximately 92 meters) since these measures that were available in the LASI dataset also pertained to functional limitations.
The ‘intra-individual factors’ domain included age (65–69, 70–74, 75–79, and ≥ 80 years); sex (1 = female); marital status; (1 = married/cohabiting); education attainment (none, primary or less, at least some secondary education); residence (1 = rural); employment status (1 = currently working); and household wealth terciles (low, middle, high). Past studies have noted a high degree of variation in socioeconomic conditions across various regions in the Indian subcontinent [39, 40]. To capture regional differentiation in socioeconomic status and its potential effects on toileting disability, we thus added region dummies (North, Northeast, East, Central, West, and South). Health and lifestyle risk factors included poor self-rated health, health insurance coverage, currently smoke, currently drink, all measured dichotomously. The five BMI categories were: underweight (BMI < 18.5), normal (18.5 ≤ BMI < 23; reference), overweight (23 ≤ BMI < 25), obese (25 ≤ BMI < 30), and severely obese (BMI ≥ 30).
Statistical analysis
Our empirical approach began with an examination of the prevalence of toileting disability in the weighted sample. We also investigated graphically how toileting disability prevalence rate varies across states/territories in India in 2017-18. Although a correlation between state of residence and toileting disability is not a hypothesis or focus of this present paper, our descriptive result in the form of a heatmap provides a useful overview of the outcome variable (toileting disability) given that the LASI respondents are spread across 35 states/territories in India. Next, we compared the characteristics for persons with and without toileting disability, using standard statistical chi-square tests for the six environmental factors.
Hierarchical logistic regressions were implemented to analyze how the different extra-individual (environmental) factors were associated with toileting disability. Model 1 featured only the physical environment factors, model 2 only the social environment factors, and model 3 only the external supports environment factors. Model 4 contained all three types of environmental factors. The full set of control variables were used in all model specifications. We checked multicollinearity in the logistic models, and linearly combined control variables that were highly correlated (ρ > 0.5) following Fong and Feng [23]. The analyses were conducted using Stata version 17.0 (StataCorp LCC, College Station, United States of America), and accounted for the multistage stratified cluster sampling design and for potentially correlated data that occurs when respondents live in the same household. Sampling weights were used to ensure nationally representative population estimates, thus increasing the generalizability of the results.
Add Comment