Data source and study population
The data from first wave of the Longitudinal Ageing Study of India (LASI, 2017–18) was used for this study which is a survey of national representative sample of 72,250 adults aged 45 years and above (including spouses irrespective of their age) from all states and union territories in India, except Sikkim. Only older adults aged 45 years and above were included in the present study making the effective sample size 65,562. Sampling design adopted by LASI was a multistage stratified area probability cluster sampling design. More in-depth information on methodology, the survey design and data collection, was published in the survey report [22].
Measures
Dependent variable
Probable psychiatric diagnosis of major depression is the dependent variable of the study. LASI has adopted the Short Form Composite International Diagnostic Interview (CIDI-SF) is used for this purpose. The scale has been validated in field settings and is commonly used in population-based surveys. A score greater than or equal to three is used as an indicator of probable major depression [22].
Early life conditions
Childhood socioeconomic status. Socioeconomic status during childhood was assessed in LASI by asking: “Compared to other families in your community, would you say your family during that time was pretty well off financially, about average, or poor”. The possible responses were: “pretty well off financially”, “about average”, “poor”, and “varied”. The variable was dichotomized by collapsing “pretty well off financially”, “about average”, and “varied” responses into the one category, the other category being “poor” childhood SES.
Childhood health. Health during childhood was assessed by asking: “In general, would you say your childhood health was very good, good, fair, poor or very poor on the basis of what you remember, or what you heard or perceived from your parents?” The variable was dichotomized by grouping those reporting ‘very good’, ‘good’ or fair into one category and ‘poor’ or ‘very poor’ into the other.
Life-course factors
Education. The variable representing the highest educational qualification of the respondents was recoded into the following categories: No formal education, primary school, secondary school, and higher education. For mediation analysis, the responses were dichotomized into ‘no formal education’ and ‘any level of formal education’.
Current economic status. The monthly per capita consumption expenditure (MPCE) as used as a measure of current economic status. Respondents were asked sets of 11 and 29 questions on food and non-food expenses, respectively. Food expenditure was collected over a seven-day reference period, whereas non-food expenditure was collected over 30-day and 365-day reference periods. The 30-day reference period was used to standardise the two types of expenditures. MPCE is computed as a summary measure of consumption and household were categorised into five quintiles based on it. These quintiles were used as categories in logistic regression. For mediation analysis, dichotomization was done by using poorest as one category and collapsing the rest into other category.
Chronic physical morbidity. Status of physical comorbidity was assessed with the question “Has any health professional ever diagnosed you with the following chronic conditions or diseases?” Responses included yes/no for the following chronic physical conditions: ” Hypertension or high blood pressure”, “Diabetes or high blood sugar”, “Cancer or a malignant tumor”, “Chronic lung disease such as asthma ,chronic obstructive pulmonary disease/Chronic bronchitis or other chronic lung problems”, “Chronic heart diseases such as Coronary heart disease (heart attack or Myocardial Infarction), congestive heart failure, or other chronic heart problems”, “Stroke”, “Arthritis or rheumatism, Osteoporosis or other bone/joint diseases” , “High cholesterol”. The variable used to assess the status of physical comorbidity was dichotomous. Those who responded “yes” one or more of the options mentioned in the above mentioned questioned were grouped together, and those responded “no’ to all the options were grouped into other category.
Discrimination. Discrimination was assessed using the six-item Everyday Discrimination Scale (EDS) (Short version). Responses ranged from 1 = “never” to 6 = “almost every day” for 31 the following items "You are treated with less courtesy or respect than other people", "You receive poorer service than other people at restaurants or stores", "People act as if they think you are not smart", "People act as if they are afraid of you", "You are threatened or harassed", "You receive poorer service or treatment than other people from doctors or hospitals ". For this study, the response for each of the six were first dichotomized to ‘never’ = 0 and ‘ever’ (for which responses ‘less than once a year’ to 'almost every day’ were collapsed into one category) = 1. The scores (0 or 1) were summed. The sum of total score ranged from 0 to 6. For logistic regression, these scores were then categories into no discrimination (sum of score=0), moderate discrimination (sum of scores 1-2), and high discrimination (sum of scores 3-6) [23]. For mediation analysis, the sum of scores was dichotomized into no discrimination (sum of score=0), and any level of discrimination experienced (sum of scores 1-6).
Health behaviours. Two health behaviours- tobacco use and alcohol consumption were included in the analysis. In the survey, the respondents were asked whether they have ever used tobacco (smoking or smokeless). It was a yes/no question. The same categories were used in this paper for the variable to assess use of tobacco. Alcohol consumption status was also assessed with a yes/no question.
Confounders
To adjust for the effect of other sociodemographic variables on the analysis, all the models were controlled for individual and household sociodemographic factors. Factors at individual level included were age (45–54, 55–64, 65–74, or 75+ years) and gender (male, female). Factors controlled for at household level were caste (scheduled tribe, scheduled caste, other backward class, or other), religion (Hindu, Muslim, Christian, Sikh, or other). At community level place of residence (rural or urban), and region (North, Central, East, Northeast, West, and South) was controlled for. The models were also adjusted for living arrangement (living alone, living with spouse and/or others, living with spouse and children, living with children and others, living with others only) which can be a potential confounder [24,25].
Statistical Analysis
The statistical analysis was done in three steps. In the first step, bivariate analysis using simple logistic regression was performed to examine association between depression and each of the independent variables in the early life as well as during life-course (childhood SES, childhood health, education, employment, tobacco use, alcohol consumption, chronic physical morbidity) to find the odds ratio (OR). In the second step, the models for testing the association between each of the independent variables with depression was adjusted for age, gender, caste, religion, region, place of residence, and living arrangement, to test the associations with adjusted odds ratio (AOR). The 95 percent confidence interval (CI) for both OR and AOR were calculated. The independent variables which were found to be significantly associated with depression were included in the mediation analysis, which is the third step of our empirical analysis. Mediation analysis was performed using KHB method [26,27]. KHB method allows to decompose the total effect of independent variables on dependent variables into direct and indirect components. The indirect component corresponds to extent of mediation, and the direct component represents the proportion of effect of independent variable that was not mediated. It allows to assess the extent to which each of the mediator mediates the relationship. For mediation analysis also alpha was set at 0.05. Analysis were performed after assigning survey weight which was given in LASI. Statistical analysis was performed using statistical software STATA, version 14.0 (StataCorp, Texas).