Tuberculosis among economic migrants: a cross-sectional study of the risk of poor treatment outcomes and impact of a treatment adherence intervention among temporary residents in an urban district in Ho Chi Minh City, Viet Nam

Table 5 Comparative interrupted time series analysis of monthly treatment success and loss to follow-up rates

	Treatment success			Loss to follow-up
	IRR┼	95% CI	p-value^Þ	IRR‡	95% CI	p-value^Þ
Baseline rate (β₀)¥	0.85	[0.83, 0.87]	< 0.001	0.05	[0.03, 0.09]	< 0.001
Pre-intervention trend, control (β₁)	1.00	[1.00, 1.00]	0.624	0.96	[0.92, 0.99]	0.024
Post-intervention step change, control (β₂)	1.00	[0.97, 1.03]	0.971	2.41	[0.97, 6.00]	0.059
Post-intervention trend, control (β₃)	1.00	[1.00, 1.00]	0.382	1.04	[1.00, 1.09]	0.050
Difference in baseline (β₄)	1.00	[0.95, 1.05]	0.909	1.91	[0.97, 3.76]	0.060
Difference in pre-intervention trends (β₅)	1.00	[1.00, 1.00]	0.541	1.02	[0.98, 1.07]	0.305
Difference in post-intervention step change (β₆)	1.07	[1.00, 1.15]	0.041	0.17	[0.04, 0.69]	0.013
Difference in post-intervention trends (β₇)	1.00	[1.00, 1.00]	0.435	0.90	[0.83, 0.98]	0.019

Notes
All patients in intervention and control districts, January 2011 to March 2017
^¥The parameters were obtained for a segmented regression model with the following structure: Y_t = β₀ + β₁T_t + β₂X_t + β₃X_tT_t + β₄Z + β₅ZT_t + β₆ZX_t + β₆ZX_tT_t;+ϵ_t. Here Y_t is the outcome measure along time t; T_t is the monthly time counter; X_t indicates pre- and post-intervention periods, Z denotes the intervention cohort, and ZT_t, ZX_t, and ZX_tT_t are interaction terms. β₀ to β₃ relate to the control group as follows: β₀, intercept; β₁, pre-intervention trend; β₂, post-intervention step change; β₃, post-intervention trend. β₄ to β₇ represent differences between the control and intervention districts: β₄, difference in baseline intercepts; β₅, difference in pre-intervention trends; β₆, difference in post-intervention step changes; β₇, difference in post-intervention trend
^┼IRR based on log-linear Poisson regression with robust standard error estimations;
^‡IRR based on log-linear GEE Poisson regression with an autoregressive correlation structure with lag order 2;
^ÞWald test;