Travel distance to rifampicin-resistant tuberculosis treatment and its impact on loss to follow-up: the importance of continued RR-TB treatment decentralization in South Africa

This retrospective analysis used secondary data from a cluster-randomized trial investigating the impact of a nurse case management intervention on RR-TB outcomes in the KZN and EC provinces of SA. People aged 13 years and older receiving RR-TB care at one of the 10 participating South African TB hospitals between 2014 through 2020 who provided informed consent were consecutively enrolled in the parent study. People with extensively drug-resistant TB (XDR-TB) or pre-XDR-TB (as defined by the World Health Organization (WHO) during the study timeframe) were excluded [13]. The 10 participating South African TB hospitals were selected in collaboration with the provincial TB Program Managers based on the following criteria: 1) use of standard programmatic RR-TB treatment, 2) willingness to participate at a study site, and 3) located in the provinces of KZN and EC. Participants enrolled at five intervention sites had their care coordinated by a nurse case manager. Details of that intervention are described elsewhere [14]. Participants at the five control sites received care according to the South African RR-TB treatment guidelines at the time. This analysis includes a subset of those participants enrolled in the parent study who were assigned an outcome of cure, treatment complete, or LTFU based on the WHO definitions for multidrug-resistant TB outcome [6]. The parent study operationalized this WHO definition of LTFU by assigning patients an outcome of LTFU on the date that participants were without RR-TB medications for two consecutive months. Participants who transferred treatment site, died, or failed treatment in the parent study were excluded. This analytical approach is consistent with the current LTFU literature [8].

Distance to treatment was calculated as the shortest one-way travel distance by road between the hospital where a patient received RR-TB treatment and the centroid of the ward (small administrative subdivisions within SA) in which they resided at treatment initiation. The participant’s residential ward was manually linked from their National Health Laboratory Service (NHLS) [15] record using name and date of birth from the start and end dates of their RR-TB treatment episode.

The residential location data that participants reported to NHLS was linked to its corresponding ward using the 2016 Ward Delimitation Maps available through South Africa’s Municipal Demarcation Board [16], Google Maps [17], and HereWeGo [18]. Although some participant location data was collected in the parent study, it was too geographically broad to be used in meaningful distance to treatment calculations; however, it was used to substantiate the location data extracted from NHLS lab reports. Ultimately, 95.7% of (1475 out of 1541) participants were successfully linked to their ward of residence.

After participants were linked to their ward of residence, distance to treatment from ward centroid to RR-TB treatment site was calculated using the GeoRoute command for Stata, which calculates travel distance using the HERE API [19, 20]. Per the recommendations of the GeoRoute command developers, the command was run twice and the results compared [19]. No discrepancies were identified between the results of the first and second runs. For 5% of the participants selected at random, the distances calculated with GeoRoute were compared to distances calculated with Google Maps using the same inputs and again no discrepancies were identified. This distribution of travel distance to each treatment site was examined, and for each outlier the location extracted from NHLS, ward of residence, and the distance calculation were verified. Distance was categorized for this analysis based on comparable prior research in SA investigating the impact of distance to drug-sensitive TB (DS-TB) referral hospitals (i.e., 0–20 km, > 20–40 km, > 40–60 km, and > 60 km) [21].

Covariates include age in years at the time of RR-TB treatment initiation; sex (male or female); HIV-status; education level (less than secondary school or completed secondary school); employment status (employed, unemployed, or other); housing stability (informal housing/unhoused or house/flat); residential locale (village/farm or urban/township); RR-TB treatment site; and treatment regimen received (injectable regimen, all-oral regimen, or individualized regimen). Demographic information was collected via patient report during parent study enrollment, while treatment regimen was extracted from medical records at the end of treatment. The covariates used in this analysis were chosen due to their potential impact on transportation access, road quality, and care-seeking behavior (i.e., residential locale and treatment site) or an association with LTFU demonstrated in previous research [5, 8, 22]. Additionally, inclusion of the treatment site variable controls for any impact of the parent study intervention, as randomization occurred at the treatment site level.

We conducted a complete-case analysis (N = 1436), dropping any case with missing data points. This analytical method resulted in < 7% of participants being excluded from the reported data. Imputation was not performed because most of the missing data was within the independent variable – travel distance. When the missing distances were explored, missingness was not significantly associated with LTFU. Descriptive statistics were examined and compared across the four distance categories using \({\chi }^{2}\) for categorical variables and the Kruskal–Wallis H test for age.

Using multivariable logistic regression, we developed a ‘base model’ to analyze the relationship between distance to treatment and LTFU, compared to treatment success (i.e., cure and treatment complete). This ‘base model’ also controlled for two other covariates that we a priori decided to force into the model – treatment site and residential locale. Each of the other covariates was then individually added to this ‘base model’. Any covariate that was significant at a level of 0.05 when added to the ‘base model’ was included in the final multivariable logistic regression. With a sample size of 1436, power of 0.8, and alpha of 0.05, we can detect significant odds ratios equal to 1.5 or greater between our reference and smallest group. All statistical analyses were performed using Stata16.0.

The final sample consisted of 1436 people living in 481 wards, 38 local municipalities, and 14 districts across KZN and EC, South Africa. Among those participants, 821 (57.2%) were male, 1061 (73.9%) were living with HIV, and the median age was 35 (IQR: 29 – 43). The majority of people were unemployed (n = 789, 54.9%), living in a village or farm (n = 891, 62.0%), and had less than a secondary school education (n = 1034, 72.0%). Looking at clinical factors, 1152 (80.2%) received a standardized all-oral or injectable treatment regimen, depending upon when they enrolled in RR-TB treatment, while 284 (19.8%) received individualized treatment regimens. Ultimately, 1086 (75.6%) successfully completed RR-TB treatment and, 350 (24.4%) were LTFU.

The overall median distance traveled to receive RR-TB treatment was 40.96 km (IQR: 17.12, 63.49). Across the four distance categories, there were significant differences observed in housing stability (p = 0.001), residential locale (p < 0.001), employment status (p < 0.001), and treatment regimen (p = 0.001). Most notably, unstable housing (i.e., informal housing/unhoused) is more prevalent among those living closer to treatment, while living in a farm or village was more common in the farther distance categories. There were no significant differences in age (p = 0.842), sex (p = 0.342), HIV status (p = 0.939), and education level (p = 0.317) observed. Additional demographic information is presented in Table 1.

Age, sex, HIV status, education level, employment status, and treatment regimen were selected for inclusion in the final model. In the final model, living more than 60 km from RR-TB treatment increased the odds of LTFU by 91% over those living within 20 km of RR-TB treatment (aOR: 1.91, 95% CI 1.28—2.84), when controlling for selected risk factors. Those living between > 20—40 km (aOR: 1.16, 95% CI 0.80—1.69) and > 40—60 km (aOR: 1.19, 95% CI 0.78—1.82) had slightly increased odds of LTFU, but these results were not significant. Other factors that increased risk of LTFU in the final model include male sex (aOR: 1.67, 95% CI 1.28—2.19) and living with HIV (aOR: 1.39, 95% CI 1.02—1.89). Older age (aOR: 0.98, 95% CI 0.97—0.99), completing secondary school (aOR: 0.69, 95% CI 0.51—0.93) and receiving an all-oral (aOR: 0.54, 95% CI 0.38—0.78) or individualized treatment regimen (aOR: 0.61, 95% CI 0.43—0.87) were found to protect against LTFU. (See Table 2 for full results.) Treatment site also had a significant impact on LTFU.