Variables to operationalise constructs included in the statistical models
| Node (construct) | Covariates included in the model | Covariates excluded from the model (missing data threshold) | Covariates excluded from the model (no available measure) |
| State | State | ||
| Race | Race, indigenous, quilombola | ||
| Local area | Urbanicity, running water, sewage, electricity, water store, garbage collection | House type | Transit access |
| Education | Years of education, literacy | ||
| Socioeconomic vulnerability | Child work, institutionalisation, work-acquired TB | Employment, pension receipt, unemployment benefit, alimony receipt | Food security, adequate nutrition, perception of poverty |
| Age and sex | Age, sex | Gender identity | |
| Comorbidities | AIDS, alcohol use disorder, diabetes, HIV, mental disorder, other chronic illness | General mental health, stress | |
| Income | Income | ||
| Expenditure | (on) Food, energy, gas, water | (on) Rent, transport | Medical costs |
| Health-seeking behaviour | Directly observed treatment | Engagement with primary care | |
| TB form and severity | Chest X-ray, initial sputum smear, pulmonary/extrapulmonary, throat culture, tuberculin skin test | MDR-TB (is included in outcome as non-successful treatment) | |
| Drug regimen | Rifampicin, isoniazid, ethambutol, streptomycin, pyrazinamide, ethionamide, other drugs |
Not all covariates included under one of the constructs in the directed acyclic graph (DAG) were included in the propensity score model. Table 1 summarises which covariates were included and which were excluded. Some covariates that might reasonably be part of the pathways encoded in this DAG were excluded as there was no adequate measure of them in these linked administrative data. Other covariates were excluded by the missing data threshold, which itself was chosen to balance measurability of each of the constructs with the loss of sample size from undertaking a complete case analysis.
The housing quality node was not included in the model as it was not associated with outcome (TB mortality) or exposure. The housing node included measurable covariates of roof, floor, and wall material, number of people in the home, and the number of bedrooms and bathrooms, as well as the unmeasurable covariate of indoor air pollution.
MDR-TB, multidrug-resistant tuberculosis; TB, tuberculosis.