Study design & aims
This is a cross-sectional analysis of routine TB surveillance data from Go Vap and District 8, Ho Chi Minh City, Viet Nam. The primary aim of the study was to determine if there is association between the government-defined residency status and WHO-defined treatment success and loss to follow-up. The secondary aim was to conduct a comparative impact evaluation on these two outcomes after initiation of a treatment adherence intervention carried out by community health workers in Go Vap. District 8 served as the concurrent control area providing only routine government TB program services.
Figure 1 shows the relative location of the two study districts within Ho Chi Minh City. Go Vap district housed a population of 685,000 people in 16 communes on an area of 20km2. The control was District 8, which housed 450,000 persons in 16 communes on 19km2 with a comparable demographic composition. Each district has one District TB Unit (DTU), which diagnoses TB and administers DOTS. The provincial TB control program recommended the intervention and control districts based on the comparability of their relative demographics and TB burden.
The intervention began in April 2014 and consisted of an intensified support program provided by incentivized community health workers. Individuals diagnosed with TB were contacted by the health workers within two weeks of diagnosis and received counseling at a location of their choice. Subsequent activities were determined by the support workers according to their perception of each patient’s need and included periodic in-person visits, phone calls or text messages with frequency and modality tailored to patient preferences and adherence patterns. Patients who missed scheduled appointments for directly observed therapy or follow-up sputum tests were contacted within 48 hours by phone followed by a home visit by the community health worker, if unreachable.
Community health workers
These activities were implemented by a cadre of 16 community health workers (CHW), who received a monthly salary (USD168) and performance-based incentives for case finding and treatment support. All CHWs were female with a median age of 56.5 (IQR: 54-58). The CHWs were recruited from local sociopolitical organizations, such as the Women’s Union and Red Cross Association, retired public health staff, and community health, population and family planning volunteers.
Data sources & processing
The study used digitized data from patient registers of the NTP’s routine surveillance system. The sample consisted of all drug-sensitive TB patients notified by the Go Vap and District 8 DTUs from 1 January 2011 to 31 March 2017. The intervention commenced 31 March 2014. We excluded cases with missing data in any of the primary exposure or outcome variables. For the primary exposure we used Viet Nam’s official, four-tier residency classification system: 1) permanent resident (abbreviated KT1); 2) long-term-intra-province migrant (KT2); 3) long-term-inter-province migrant (KT3); and 4) short-term-inter-province migrant (KT4). Other available patient covariates from the registers were used to estimate secondary risk factors. To assess the post-intervention impact, we used monthly treatment success and loss to follow-up rates aggregated by treatment initiation date.
All analyses were performed on Stata for Windows version 13. Descriptive statistics for study participants were cross-tabulated, and crude risk and odds ratios were calculated for primary and secondary exposures using univariate log-binomial and logistic regressions for treatment success and loss to follow-up, respectively.
Saturated, multivariate log-binomial and logistic maximum likelihood models were fitted onto the data to control for confounding to identify pre-intervention risk factors. For non-converging log-binomial models we used Poisson regression with robust standard errors. We measured associations between treatment success and loss to follow-up rates, and individual parameters plus a binary bifurcation of permanent and temporary (KT2-KT4) residency for crude analyses. For multivariate analyses we used the categorical residency parameter, which we tested for a dose-response effect, considering short-term inter province migrants to be the most intensely exposed to the vulnerabilities of migrant status, and long-term intra-province migrants the least exposed.
We conducted a comparative interrupted time series (ITS) analysis on aggregate monthly treatment success and loss to follow-up rates using segmented log-linear Poisson regression with robust standard errors.31 We modeled the ITS to include a step and a slope change. The step change aimed to reflect the instantaneous impact of the missed dose follow-up activities by the CHWs. The slope change estimated the impact of increased counseling and continuous case holding activities to cause gradual positive changes in treatment success and loss to follow-up rates. The parameters of the ITS were obtained for a segmented regression model with the following structure: . Here Yt is the outcome measure along time t; Tt is the monthly time counter; Xt indicates pre- and post-intervention periods, Z denotes the intervention cohort, and ZTt, ZXt, and ZXtTt 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.
The ITS analysis was conducted in two iterations. The first iteration included the complete sample, while the second iteration focused on temporary residents. We included all patients notified after 1 August 2013 in the exposed group based on the proportion of treatment outcomes reported in the post-intervention period. The Cumby-Huizinga test was used to identify serial autocorrelation in the intervention district and to adjust the ITS analysis using the generalized estimating equation (GEE) approach. We obtained model specifications from quasi-likelihood information criteria. Hypothesis tests were two-sided and point estimates included 95% confidence intervals.
The London School of Hygiene and Tropical Medicine Research Ethics Committee granted ethical approval for the epidemiologic analysis. The HCMC People’s Committee approved the implementation of the intervention. The Go Vap District Health Center approved use of the data. A consent waiver was granted based on the study’s use of routine surveillance data. We anonymized all patient data and removed identifying information prior to analysis.