Therapy duration and long-term outcomes in extra-pulmonary tuberculosis

Tuberculosis (TB) affects 8.7 million people each year; while 1 in every 3 human beings on earth has been exposed to Mycobacterium tuberculosis and has latent tuberculosis infection [1]. The vast majority of research on tuberculosis has focused on pulmonary disease, with efforts to shorten therapy duration. Meanwhile, the rate of “extra-pulmonary” tuberculosis (EPTB) has increased from one in every 25 TB patients in 1995 to about one in every five TB patients in 2002, and one in every three TB patients in 2011 [2, 3]. More than 50% of HIV-infected patients with tuberculosis present with EPTB. Patients using tumor necrosis factor-α inhibitors for autoimmune diseases have a 4-5 fold increase in the rate of active tuberculosis compared with non-users, and have EPTB rates similar to those with AIDS [4]. The recommended treatment for drug-susceptible EPTB is with isoniazid, rifampin, ethambutol and pyrazinamide for 6 months, with the exception of tuberculous meningitis which is treated with 9 to 12 months of therapy [5]. In general, extra-pulmonary tuberculosis in adult patients has been lumped aside in a wastebasket category. Meningitis is a notable exception [6–10]. Dutt, Moers and Stead studied the outcomes over an average period of 3 years post treatment in 219 EPTB the early 1980s [11]. However, this cohort of patients was treated with a regimen of isoniazid and rifampin alone, not the standard four drug regimen used today. In a second study, Kwara et al examined data on 126 patients with EPTB at the dawn of the AIDS pandemic, and found that the short-term mortality during treatment of EPTB was associated with HIV-infection, meningeal tuberculosis and disseminated disease [12]. Long-term outcomes were not reported. Thus, the long-term outcome in EPTB patients “adequately treated” with the current standard regimen is unknown, so that it is unclear if standard short course therapy is effective in reducing long-term mortality. We have developed methods to use the Texas electronic database to examine long-term outcomes such as death in tuberculosis patients [6]. This study is focused on the evaluation of long-term outcomes in patients with EPTB, and the demographic and therapy factors predictive of poor long-term outcome.

There were 9,441 patients diagnosed with tuberculosis in Texas during the entire study period. Of these 837 (16%) had EPTB, and 438 (52%) fulfilled study criteria (Figure 1). In 328 (75%) patients the diagnosis was microbiologically proven. The initial drug regimens used and reported for all study patients were consistent with standard CDC recommendations. The patient characteristics are shown in Table 1. Data on patient weight was missing in many records, and when available not verifiable. Thus, although this factor had been one of the several that prompted the study, it was not further analyzed. Homelessness and substance abuse were common social factors associated with tuberculosis in this cohort (Table 1). The table shows that patients’ median age significantly differed across the type of EPTB (p = 0.02). In addition, EPTB syndrome (i.e., anatomic site of tuberculosis) significantly differed by self-identified “race”: 83% of genitourinary system-tuberculosis patients were “white” while 17% were distributed among the remainder. The median study follow-up period for the entire patient cohort was 7.8 (inter-quartile range: 6.0-10.1) years, and was not significantly different between major disease sites (p = 0.11); patients were evenly distributed across the surveillance period. The median therapy duration was 8 (interquartile range: 6-11) months. However, therapy duration differed significantly between EPTB syndromes, especially due to the longer duration of therapy in patients with bone/joint disease (Figure 2).

One hundred twenty-four (28%) patients died during the observation period. The overall all-cause mortality rate was 0.025 (95% confidence interval [CI]: 0.021-0.030) per 100 person-year of follow-up; 103 (83%) died within 70.5 months of diagnosis, only 40 (9%) died during therapy, 346 (79%) were recorded as having completed therapy. In the remaining 12% of patients, the reason for failing to document completion of therapy was unknown in 30 patients, 14 were lost to follow up, five moved out of the Texas jurisdiction, and three were uncooperative with healthcare workers. Thus, further analysis was confined to the 346 patients with documented evidence of having completed treatment. Since life expectancy is higher in women than men, and long-term survival is affected by age, the long-term survival in patients with EPTB is shown for different genders and for different ages in Figure 3.

CART analyses identified age-at-diagnosis (age), therapy duration, EPTB syndrome (i.e., anatomic site of disease), and HIV-status as significant predictors of survival. The highest decision node (i.e., root node) ranked by CART was age, with a cut-off point of 64.5 years. Altogether 35/70 (50%) patients >64.5 years died versus 36/376 (13%) patients ≤64.5 years. In patients >64.5 years, the next and last decision node was treatment duration, with a cut-off value of 6 months, with survival of 100% in those with therapy duration ≥6 months versus 43.5% for therapy duration <6 months. Thus therapy duration was the next most important predictor of outcome in patients >64.5 years. On the other hand, for patients ≤ 64.5 years the next decision node was type of EPTB syndrome, with a split into those with genitourinary tuberculosis-TB showing 100% survival versus the rest who had an 83.3% survival. Further down the tree in this younger group of patients, the next most important node was treatment duration among patients without genitourinary tuberculosis, with a cut-off value of 7 months which classified patients into those treated for ≤7 months who had a higher survival of 90.4% versus 81% in those treated for longer than 7 months. In other words, therapy duration versus survival was described by a “V” shaped response. The predictive accuracy for the decision tree, in cross-validation in a test sample, was 92%.

Next, we examined if each of the four parameters identified by CART analyses were also significant when standard parametric multivariate analysis was utilized, with time-to-death as an outcome. The analysis confirmed CART findings, and in addition demonstrated that gender was also a significant risk factor (Table 2). The relationship between gender and mortality is shown in Figure 4A. As regards to age, the hazard ratio for each year of age-at-diagnosis versus time-to-death was 1.05 (CI: 1.04-1.06; p < 0.001). Figure 4B shows the survival curves of HIV-infected versus non-HIV-infected among the 280 patients who were tested for HIV. Figure 4C and D show that when the relationship between EPTB syndrome and time-to-death was examined, genitourinary tuberculosis differed significantly from all other syndromes. When tuberculous meningitis, which had the sharpest early decline in survival after completion of therapy, was examined as the referent, the mortality also differed from some of the non-genitourinary tuberculosis patients (Table 2). Thus, there are 3 significantly different survival categories, namely (a) patients with genitourinary tuberculosis who did as well as those with latent tuberculosis treated with isoniazid prophylaxis, (b) patients with tuberculous meningitis and peritoneal tuberculosis who had very poor long-term outcomes, and (c) an intermediate mortality group comprised of bone/joint, lymphatic, and “other” tuberculosis syndromes.

Variable	Hazard ratio (95% CI)	P-value
Age (years)
18 – 24	Referent
25 – 34	2.75 (0.60, 12.59)	0.19
35 – 44	2.08 (0.46, 9.44)	0.34
45 – 54	2.42 (0.54, 10.85)	0.25
55 – 64	2.02 (0.42, 9.66)	0.38
65 – 74	3.71 (0.77, 17.79)	0.10
>75	7.43 (1.54, 35.93)	0.01
HIV-infected versus non-infected	2.16 (1.22, 3.83)	<0.01
Male	1.52 (1.01, 2.72)	0.04
TB disease site
Meningeal	Referent
Lymphatic	0.50 (0.27, 0.91)	0.02
Bone/joint	0.42 (0.21, 0.81)	0.01
Peritoneal	0.59 (0.27, 1.31)	0.20
Other sites	0.47 (0.23, 0.97)	0.04
Therapy duration
6-months	Referent
<6-months	2.72 (1.34, 5.50)	<0.01
7 – 10 months	1.63 (0.85, 3.12)	0.14
>10-months	1.52 (0.75, 3.10)	0.25

*Multivariate regression analysis drops patients with missing variables, so the final model comprised of 402 instead of 438 patients.

The relationship between treatment duration and mortality was examined in greater detail for each EPTB syndrome, since treatment duration is potentially modifiable. Genitourinary tuberculosis was excluded since only one patient with that diagnosis died. The CART-derived threshold for treatment duration versus long-term mortality is shown in Figure 5. It can be seen in Figure 5, that the relationship between treatment duration for those who completed therapy versus mortality was a “V” shaped curve. As therapy duration increased towards the nadir, survival improved to a maximum at the vertex, after which longer therapy was associated with worse survival. The nadir of the relationship is the therapy duration associated with lowest mortality, which differed between disease syndromes. The therapy duration associated with lowest mortality was 7 months for peritoneal tuberculosis, 8 months for meningitis, 9 months for lymphadenitis, and 10 months for all “other”-tuberculosis syndromes.

The pulmonary system has always been considered the “seat” of tuberculosis since the early days of systematic medicine [28–30]. Detailed studies of the metabolic and biochemical status of M. tuberculosis, as enumerated in the three population model hypothesis, with direct consequences on therapeutics, have been painstakingly mapped for pulmonary tuberculosis [29, 31–36]. Nothing of a similar scope has been performed for most “EPTB”. Our first major finding is that the use of this wastebasket term conceals radically different types of tuberculosis in terms of long-term outcomes and duration of therapy. Lumping these diagnoses together as “EPTB” is no more useful than lumping Staphylococcus aureus vertebral osteomyelitis, carbuncles, and brain abscesses as “non-bacteremic Staphylococcal disease”. We found a high case fatality rate for some syndromes during the observation period, despite “adequate” chemotherapy. To put this in perspective, the 5 year case fatality rate for untreated pulmonary tuberculosis that is smear negative but culture positive is 12-15%, while that for sputum positive is 42-70% [37]. Thus for microbiologically proven and adequately treated meningitis long-term mortality of 46.2% and 42.6% for peritonitis, long-term outcomes are reminiscent of untreated “consumption.” We propose that each EPTB syndrome is important in its own right, so that each EPTB syndrome’s pathology, M. tuberculosis physiology, and various pharmacological factors should be separately delineated from the others.

It is unclear why TB infection in different organs is associated with different mortality. This could reflect the impact of tuberculosis damage on different organs. An example are the cerebrovascular accidents in tuberculous meningitis [38, 39]. Alternatively, this could reflect the differential penetration of antibiotics into different anatomic sites. Poor penetration would reduce both peak concentration and 24 hr area under the concentration-time curve (AUC) to minimum inhibitory concentration ratios, which are known to correlate with microbial kill and suppression of acquired drug resistance in pre-clinical hollow fiber system of tuberculosis studies and in patients [21, 40–44]. As an example, in tuberculous meningitis, the blood-brain barrier variably and unpredictably decreases drug levels depending on compartment and drug selected, with such drugs as rifampin achieving only 2-20% those achieved in the serum, while isoniazid AUC is 80% that in serum [10, 45, 46]. Indeed, when drugs that penetrate well into the meninges such as quinolones are included, or rifampin’s dose is increased so that the AUC is increased, tuberculous meningitis patients have decreased mortality [8, 10]. Similarly, in bone/joint disease, the percentage of drug in foci in sclerotic bone is 0-8% for rifampin, 0-6% for pyrazinamide, and 0-23% for isoniazid [47]. These drug-concentration scenarios are known to lead to acquired drug resistance and poor efficacy [21, 48]. Finally, it could be that for some syndromes such as meningitis, the M. tuberculosis genotypes that preferentially cause such disease are more difficult to kill and have a higher propensity to fail and develop drug resistance [49–52].

A third important finding was that the relationship between therapy duration for each syndrome versus outcome as “V” shaped. This implies that there is a combination therapy duration point associated with the best long-term outcome, and that therapy duration beyond this point leads to poorer survival. Thus, unnecessarily long therapy could be harmful to patients. In studies of antibiotics used to treat other bacterial infections, it has been suggested that better outcomes occur with shorter rather than longer duration therapy [53]. However, this is a matter of ongoing debate. In our hollow fiber studies, longer duration of isoniazid and pyrazinamide therapy was an important driver of acquired drug resistance by M. tuberculosis[41, 42]. Thus, longer therapy duration than necessary could worsen microbiologic outcomes. Nevertheless, given the retrospective nature of the data, we cannot definitively ascribe causation of poor long-term outcomes to the longer therapy duration. Another possibility is that patients who were not doing well could have had their therapy duration extended by clinicians, so that the longer therapy duration may be an indication of patients who were having poor outcome rather than being the cause of poor outcome, a post hoc ergo propter hoc fallacy. These possibilities cannot be teased out in a retrospective study, and thus causation cannot be ascribed at the present moment. Our findings should thus be viewed merely as generating a hypothesis, which must then be tested in prospective studies.

Our study has several limitations. First, the most important limitation is the retrospective nature of the design. Additionally, since the data were routinely collected for surveillance purposes some data points were missing, while some could have been entered erroneously. Second, the study patients’ age distribution was skewed in favor of an elderly population with naturally higher co-morbid conditions and consequently poorer long-term survival. Third, variations in the combinations of anti-tuberculosis drugs used to treat each patient could affect outcomes, given comorbid conditions with potential drug interaction, or regimen changes to accommodate acquired drug resistance, all of which limit interpretation of the data. Fourth, a more detailed exploration of the factors identified as associated with overall mortality in treated tuberculosis patients was precluded by the incompleteness of data in the Report of Verified Case of Tuberculosis. Finally, in the current analysis, we chose to use CART analyses. However, CART is prone to over fitting, and slight changes in dataset may lead to selection of different predictors. This problem led to extension of the procedure into random forests, where instead of the forest instead of a single tree, is grown [54]. We utilized CART because it is more intuitive to clinical decision making by clinicians, is familiar to clinicians, and is much easier to interpret. In order to make sure the predictors chosen by CART were indeed correct and not due to instability, standard parametric multivariate analysis was utilized to examine the CART-derived predictors and thresholds, which were all found to be statistically significant.

In summary, different EPTB syndromes have different long-term outcomes and different therapy durations associated with lowest mortality. The long-term outcomes for such syndromes as peritoneal and meningeal tuberculosis adequately treated with modern chemotherapy are sufficiently poor as to be indistinguishable from some types of tuberculosis before the advent of chemotherapy. The different syndromes should not be lumped into single wastebasket diagnosis of EPTB.