Statistical analysis software version 9

Preliminary analyses included summaries of participant characteristics using frequencies and descriptive statistics. Interrelations between study variables were assessed with correlations. In addition, a preliminary multiple regression analysis was conducted to explore the relations between each sociodemographic variable and financial strain (while controlling for the other sociodemographic variables).
The effect of financial strain on self-rated health was assessed with a covariate-adjusted linear regression model. The indirect effects of financial strain on self-rated health through stress and depressive symptoms, respectively, were assessed using 2 single mediation models and a multiple mediator analysis that included both stress and depressive symptoms in the same model. Indirect effects were tested using a non-parametric, bias-corrected bootstrapping procedure.^{36 (link)} The bootstrapping procedure generated an empirical approximation of the sampling distribution of the product of the estimated coefficients in the indirect paths using 5000 resamples from the data set. All models were adjusted for sociodemographics, including age, sex, partner status, income, education, and employment status. All analyses were performed using Statistical Analysis Software version 9.3 (SAS Institute, Cary, NC).

Follow-up time was calculated from TB treatment initiation until the earliest of: last visit date for those who were LTFU, date of death, termination date or administrative censoring at 24 months. In the results published previously, ^{8 (link),14 (link),15 (link)} administrative censoring took place at 18 months post TB treatment initiation but in this analysis administrative censoring was at 24 months. Analysis was conducted at different follow-up times namely: during TB treatment, after the completion of TB treatment, before and after ART initiation as well as at 24 months post TB treatment initiation.
Multivariate proportional hazards regression models were used to identify predictors of LTFU at different follow-up time points. Baseline variables that were included in the multivariate analysis were: trial arm, age, gender, WHO stage of HIV disease, CD4+ cell count, presence or absence of history of TB and employment status. We used Poisson approximations to calculate 95% confidence interval (CI) for incidence rates and incidence rate ratio (IRR). Proportionality was tested by fitting the time dependent covariates created by interacting the baseline variables and a function of survival time. Statistical analyses were done using Statistical Analysis Software version 9.3 (SAS Institute, Cary, North Carolina, USA).

Chest CT scan scores, PFTs, and prednisone dose were assessed before and after RTX. Average values were calculated at baseline for all subjects with follow-up values at any time point (1, 2 or 3 years), and average values at subsequent time points were calculated only for subjects with baseline values for comparison. An improvement in CT severity scores was defined as a ≥10% decrease while improvement in PFTs was defined as a ≥10% increase in FVC. Stability in CT score was defined as a ≤10% increase while PFT stability was defined as a ≤10% decrease in FVC.[43 (link)] Paired univariate analyses were conducted with Wilcoxon signed rank tests where appropriate. Subgroup analyses were based on antisynthetase autoantibodies, ILD pattern, baseline mean HRCT score (194), PFT parameters indicating severe disease (FVC% < 50%, TLC% < 50%, and DLCO < 35%), some of which have been associated with increased ILD mortality,[44 ,45 (link)] use of RTX as initial or rescue therapy, and the presence of concurrent immunosuppression at time of starting RTX. A p-value of <0.05 was considered significant. All analyses were performed using Statistical Analysis Software version 9.3 (SAS Institute, Cary, NC). This project was granted IRB approval by the Partners Human Research Committee at the BWH (protocol number 2014P000110) and UPMC (IRB0409097).