Risk factors for recurrent tuberculosis after successful treatment in a high burden setting: a cohort study

Patients who have recently completed treatment for tuberculosis (TB) are at elevated risk of recurrent TB disease, either as a result of relapse or reinfection [1, 2]. When individuals are diagnosed with recurrent TB, they are less likely to complete treatment and suffer higher mortality than those with first episodes of TB [3]. While there is a growing appreciation of the risks associated with recurrent TB, a clearer picture of which covariates place individuals at greater risk of recurrent disease can inform more targeted post-tuberculosis care and monitoring.

Several covariates have been found to be associated with recurrence of TB after completion of treatment. Drug-resistance [4], smoking [5], HIV infection with low CD4 lymphocyte counts [6], substance use [7], chronic lung disease [8], sputum smear-positive disease [9], and cavitary pulmonary disease [10] have each been found to be associated with increased risk of recurrent tuberculosis. These factors may increase risk of recurrence because they make it more likely for some small numbers of mycobacteria to persist beyond treatment or because they are associated with immune incompetence that places individuals at greater risk of disease following reinfection [11].

Over the last few years, molecular genetic tools have provided a new view on the extent of within-host genotypic heterogeneity of Mycobacterium tuberculosis infections. These “mixed-strain” or “heterogeneous” infections arise either through sporadic mutation (clonal heterogeneity) or host acquisition of more than one strain of M. tuberculosis (polyclonal heterogeneity), either at the time of infection or through serial re-infection [12]. While detection of mixed-strain infection is limited by the biological samples collected as well as the laboratory processes and sensitivity of sequencing, the frequency at which clonal heterogeneity and polyclonal heterogeneity are reported is nearly 20% in some settings [13]. While several studies have found that within-host heterogeneity, especially polyclonal infections, are associated with poorer treatment response [14], the effects on risk of recurrence are less clear [13].

Studies of recurrent TB are challenging to conduct because the risk of recurrent TB can remain elevated for up to ten years [2]. In this analysis, we combined a research database with electronic notification records to study the long term risk of recurrent tuberculosis among a cohort of ambulatory pulmonary TB patients with a high prevalence of HIV co-infection after successful completion of anti-tubercular therapy.

During the study enrollment and follow-up period, 110,859 cases of tuberculosis were recorded in the Umgungundlovu Health District ETR. Of the 500 patients initially enrolled, we matched 478 (96%) within the ETR (Fig. 1). 135 (28%) had a treatment outcome of “not evaluated” due to clinic transfer or loss to follow-up. Further searches allowed discovery of the treating clinic of 89 of these 135 (66%) to determine a final treatment outcome. 421 (88%) of the matched participants had a positive baseline mycobacterial culture with successful MIRU-VNTR genotyping.

333 (79%) of the original study participants with a MIRU-VNTR genotype were recorded as achieving treatment completion or cure for their initial tuberculosis episode (Table 1). Those with a successful treatment outcome were less likely to be living with HIV (65.8 vs 79%) and less likely to have sputum negative for AFB (40.8% vs 56.3%) (Supplemental Table 1). In this cohort, median age was 35 years, and 57% were men. 228 (69%) were HIV positive and 77 (23%) had evidence of heterogeneous M. tuberculosis infection by MIRU-VNTR genotyping. When utilizing the ClassTR algorithm to interpret MIRU-VNTR results, 30 (9%) had clonally heterogeneous M. tuberculosis disease, and 47 (14%) had polyclonal heterogeneous M. tuberculosis disease. 59 (18%) had a history of tuberculosis prior to enrollment. 7 (2%) had multidrug-resistant TB (MDR-TB).

The time from initial tuberculosis diagnosis to searching in the District ETR was a median of 2125 days (interquartile range 2032 to 2271 days). Fourty-five participants (14%) were diagnosed with recurrent tuberculosis, of which 35 participants (11%) had bacteriologically proven tuberculosis, by either sputum smear, Gene Xpert, or culture, after a median of 504 days. Kaplan-Meier curves for bacteriologically proven recurrent tuberculosis after treatment for either simple, clonal or polyclonal M. tuberculosis tuberculosis infection were not statistically significantly different by log-rank test (Fig. 2).

In univariable analysis with Cox proportional hazards models, younger age was significantly associated with increased risk of bacteriologically proven tuberculosis recurrence with a hazard ratio of 0.94 (95% confidence interval (CI) 0.91–0.98) for every additional year of age (Table 2). In our multivariable model, a 3+ AFB sputum smear grade was significantly associated with recurrent tuberculosis with a hazard ratio of 3.33 (95% CI 1.44–7.7, p = 0.005) (Fig. 3). The presence of clonally heterogeneous M. tuberculosis infection at baseline had a hazard ratio for recurrence of 0.92 (95% CI 0.27–3.1), while polyclonal heterogeneous M. tuberculosis infection at baseline had a hazard ratio for recurrence of 1.96 (95% CI 0.86–4.5). None of those with recurrent tuberculosis had MDR-TB at baseline, which made inclusion of MDR-TB in our models impossible.

After following a cohort of largely HIV-coinfected ambulatory individuals successfully treated for pulmonary tuberculosis disease in a high burden setting for a median of 6 years, we found that 11% (95% CI 7.2 to 13.8%) had bacteriologically proven recurrent tuberculosis and an additional 3% (95% CI 1.2 to 4.8%) had empirically-diagnosed recurrent tuberculosis. This is consistent with other studies performed in high burden settings [2, 27].

Our finding of an association between higher sputum AFB smear grade during the initial tuberculosis episode and recurrent episodes of tuberculosis may be due higher mycobacterial load disease and cavitary disease, leading to a higher risk of relapse after treatment completion. The association between smear grade and poor treatment outcomes, including recurrent disease, has been seen in other studies [28].

In this long-term follow-up, the point estimate for the odds of recurrence for individuals with polyclonal M. tuberculosis infection was 1.96, but the 95% confidence interval did not exclude 1. The relatively small number of outcomes recorded limits our ability to make conclusive statements about whether such polyclonal infections make it more likely for patients to experience relapse after successful treatment. An association of recurrent tuberculosis and heterogeneous M. tuberculosis infection is plausible as heterogeneous infection leads to the potential for heteroresistance. Additionally, Shin and colleagues recently found an association between heterogeneous M. tuberculosis infection and treatment failure in Botswana that was independent of heteroresistance [14]. A high mycobacterial burden within a host or repeated exposures to tuberculosis in a high TB prevalence environment could be other mechanisms for an association of heterogeneous M. tuberculosis infection and poor treatment outcomes or recurrence.

HIV co-infection was not associated with the risk of recurrence in univariable or multivariable analysis. This could reflect rising rates of antiretroviral therapy in our setting, and that South African guidelines at the time of our initial study recommended placing patients living with HIV and tuberculosis co-infection on antiretrovirals. Antiretrovirals reduce the risk of developing active tuberculosis, but how that risk compares to the HIV uninfected population is not currently known [6, 29, 30]. Alternatively, participants living with HIV could have suffered higher mortality rates in subsequent years that may have biased our results given we did not have access to vital statistics.

There were limitations in linking our study dataset with ETR data to identify recurrent tuberculosis episodes as there is currently no national health identifier in South Africa. Fuzzy matching of patient identifiers identified many instances of likely matches that were initially missed due to misspellings and transpositions, but we do not know if other matches were missed due to larger discrepancies and whether this introduced bias into our findings. If risk factors of recurrence are also associated with death, lack of access to vital statistics may have biased results by not properly censoring participants who died during the follow-up period.

This limitation leads to our finding of 11% recurrence as a lower bound on the burden of recurrent tuberculosis. Additionally, even though the ETR encompasses a health district that serves over 1,000,000 people, the population of KZN is highly mobile [31] and the ETR database does not capture recurrences outside of the health district. 135 (28%) of the participants in this study transferred clinics or were lost to follow-up during their initial TB treatment and we were able to locate 66% of them at their final treating clinic within the health district. It is conceivable that migration out of the health district may have introduced some bias into our study.

Another limitation is that our sampling was at one timepoint from one anatomical source (sputum). Our group’s work with post-mortem studies in the same population as represented in this study has found significant proportions of multi-organ disseminated tuberculosis with strain heterogeneity that was not captured when sampling sputum alone [32]. Genotyping with MIRU-VNTR also has limited sensitivity for determining M. tuberculosis strain diversity and may have underestimated the true proportion of heterogeneous infection [12]. We also were unable to genotype M. tuberculosis strains from recurrent episodes to distinguish reinfection from relapse. In other studies with matched fingerprinting of initial and recurrent tuberculosis episodes, reinfection accounted for at least half of recurrent disease, reflecting the hyperendemic communities to which people living with tuberculosis return to after treatment completion. We suspect similarly high rates of reinfection occurred in our population as well, but we do not have genotypes from recurrent tuberculosis episodes to test this hypothesis.