Association of multidimensional poverty and tuberculosis in India

Data from the fourth round of the National Family Health Survey (NFHS-4), 2015–16, India was used in the analysis. The NFHS-4 is a nation-wide, cross-sectional, and large-scale demographic health survey conducted under the stewardship of the Ministry of Health and Family Welfare (MoHFW), Government of India. It used multi-stage stratified systematic sampling to draw representative estimates at the national and subnational levels (state/district level). In the first stage, villages in the case of rural areas and census enumeration blocks in the case of urban areas were selected through the probability-proportional-to-size method. Households were selected through a systematic sampling scheme in the second stage of the sampling. The survey successfully collected information from 601,509 households (with a 98% response rate), covering 2,869,043 individuals throughout the country. The survey used three sets of schedules, namely, the household schedule, the woman schedule, and the man schedule. The household schedule provided information on household amenities (type of house, cooking fuel, drinking water, sanitation, electricity, assets ownership, etc) and basic demographic details for each member of the household (age, educational attainment, caste, religion and relationship to head of the household). The woman schedule was canvassed to women in the 15–49 age group in the domains of fertility, maternal health, child health, nutritional status, mortality, contraception, etc. Similarly, the man schedule collected information on health, marriage, contraception, etc. from men aged 15–54 years. The detailed methodology, sampling design, and survey instruments have been given in the report [15].

The NFHS-4 provides six types of separate data files, namely household, birth, kids, women, men, and person’s data file. The person’s data file contains all information at the individual level, which includes demographic, educational attainment and some health-related information for all members in a household. It may be mentioned that the household characteristics, such as electricity, type of house, cooking fuel and household assets, are the same for every member of a household, while the individual characteristics, such as age, sex, educational attainment, and some biomarkers, vary for each member of the household. The individual data file was used to estimate the extent of multidimensional poverty and tuberculosis.

Three specific questions related to tuberculosis were asked to the respondents in the household schedule. First, a screening question on ascertainment of tuberculosis was canvassed (Question No-23). It read as “Does any usual resident of your household suffer from tuberculosis?”. If anybody in the household reported tuberculosis, two subsequent questions (Q-24 and Q-24A) were asked that read as “Who suffers from TB in the household?” and “Has the person received medical treatment for it?”. Further, information was collected on the place of treatment (private/public facility). Tuberculosis was either self-reported or reported by any some other member of the household. Of the 8718 sample households that have at least one member with tuberculosis, 96.2% had only one member infected with tuberculosis. The unit data of NFHS-4 is publicly available and can be accessed by registering at www.​dhsprogram.​com/​data/​.

Descriptive statistics, multidimensional poverty indices, and logistic regression model were used in the analysis. Descriptive analyses were carried out to estimate the prevalence of tuberculosis across socio-economic and geographic characteristics. We used the Alkire and Foster (AF) method to estimate multidimensional poverty indices in the states of India. The AF method uses a dual cut-off methodology. First, the poor were identified in each weighted indicator based on pre-determined threshold/cut-off points of deprivation for each indicator (column three of Table 1). For instance, all household members were considered to be deprived in a particular indicator, say electricity if the household did not have electricity. In the second step, the aggregation of deprivation was made. Multidimensional poverty was defined at a threshold of 33% across composite deprivation score in line with global multidimensional poverty literature [18]. We used the estimates of both tuberculosis and multidimensional poverty at the individual level. This is logical as the data on tuberculosis was collected for all members of a household, while multidimensional poverty (similar to the poverty headcount ratio) was estimated at the population level (percentage of population below a threshold level). However, many of the indicators used for estimating multidimensional poverty were at the household level and remained the same for each member in the household. We also included two additional indicators, that is, number of households with at least one-member suffering from tuberculosis and percentage of households who are multidimensional poor.

We used three broad dimensions - education, health, and standard of living - and a total of ten indicators in these three domains to define multidimensional poverty. Each indicator was directly or indirectly related to the SDGs. The inclusion criteria of the indicators and the deprivation cut off-points for measuring multidimensional poverty have been described elsewhere [18]. The deprivation cut-off point for the school attainment indicator was revised to completion of 8 years of schooling to suit the Indian context. The details of the dimensions, indicators, cut-off points, and weights are given in Table 1. Equal weights were assigned to each dimension, and within a dimension, equal weights were assigned to each indicator. The AF method provides three types of indices, that is, head count ratio, intensity of poverty, and multidimensional poverty index. The present study focuses on using multidimensional poverty since its prime objective was to estimates the prevalence of tuberculosis among multidimensional poor and non-poor. A brief description of the indices is given below.

Head count ratio (H): It is defined as the proportion of the multidimensionally poor to the total population and can be calculated as:

Where q is the number of persons who are multidimensional poor and n is the total population. We have presented H as a percentage.

Intensity of poverty (A) is calculated as,

Where c is the deprivation score that the poor experienced. It is an average weighted count of deprivation experienced by the multidimensional poor.

Multidimensional poverty index (MPI) is the product of headcount ratio and intensity of poverty and calculated as:

A chi-square test was carried out to understand the significant association between the prevalence of tuberculosis and the various characteristics considered in the study. A two sample proportions test was used to find out whether the difference between the prevalence of tuberculosis among the multidimensional poor and the non-poor was significant. We have also estimated Moran’s I statistic for tuberculosis and multidimensional poverty to see the extent of spatial clustering.

We used a binary logistic regression analysis to understand the risk factors associated with tuberculosis. Tuberculosis (the outcome variable) was categorized as 0 (absence of the disease) and 1 (presence of the disease). The most common mathematical equation used for the logistic regression modelling is as follows:

Where p is the probability of having tuberculosis and β_i are the regression coefficients indicating the relative effect of changing a particular risk factor (the explanatory variable) on tuberculosis. The X_i are the independent (explanatory) variables considered in the study such as multidimensional poverty head count ratio, place of residence (rural or urban), religion, caste, smoking behaviour, overcrowding (defined as more than three persons living in one room), sex, and age. We did not included any such variable in the multivariate analysis that was used to estimate multidimensional poverty. We interpreted the results in terms of odds ratio (OR), with a 95% confidence interval and provided the robust standard error of the estimates. The confidence interval (CI) are based on robust standard error. All statistical analyses were performed in the STATA version 14 software package. All methods were carried out in accordance with relevant guidelines and regulations.