Article Text

Download PDFPDF

Predictive factors for treatment failure in patients with presumed ocular tuberculosis in an area of low endemic prevalence
  1. Rupesh Agrawal1,2,3,
  2. Julio J Gonzalez-Lopez1,
  3. João Nobre-Cardoso1,
  4. Bhaskar Gupta4,
  5. Robert Grant5,
  6. Peter KF Addison1,
  7. Mark Westcott1,
  8. Carlos E Pavesio1,2
  1. 1Department of Medical Retina, Moorfields Eye Hospital, NHS Foundation Trust, London, UK
  2. 2Biomedical Research Centre, UCL Institute of Ophthalmology, London, UK
  3. 3Department of Ophthalmology, National Healthcare Group Eye Institute, Tan Tock Seng Hospital, Singapore, Singapore
  4. 4Department of Medical Retina, Royal Devon and Exeter NHS Trust, Exeter, UK
  5. 5St George's, University of London & Kingston University, Kingston, UK
  1. Correspondence to Dr Carlos E Pavesio, Moorfields Eye Hospital, City Road, London EC1V 2PD, UK; carlos.pavesio{at}


Aim To assess the impact of antitubercular therapy (ATT), oral steroids and steroid sparing immunosuppressive treatment on treatment success in cases with presumed ocular tuberculosis in an area of low endemic prevalence.

Methods A retrospective cross-sectional study was performed for 213 patients with presumed ocular tuberculosis from a database from a tertiary referral eye hospital in the UK. A logistic regression model was constructed incorporating demographics, baseline characteristics and different cut-offs of QuantiFERON-TB Gold In-Tube test (QFT-G) to identify significant factors accounting for the variability of the response variable (‘failure’) across the whole group. Treatment failure was defined as the recurrence of inflammation or inability to taper steroids within 6 months of completion of ATT or after at least 6 months of treatment in the non-ATT group.

Results There were 126 patients who had at least 6 months of ATT. Patients with QFT-G values >1.50 (OR=0.20, 95% CI 0.09 to 0.48, p<0.001) had less risk of treatment failure as against those with QFT-G values between 0.35 and 1.50. Steroid sparing immunosuppressive agents reduced the chances of treatment success (OR=24.10, 95% CI 6.75 to 86.11, p<0.001). This effect persisted even after adjusting for potential confounding factors.

Conclusions Patients with higher values of QFT-G (>1.5) are more likely to have treatment success with ATT. In our model, steroid sparing immunosuppressive agents reduced the chances of success in both ATT and non-ATT-treated patients. It is unclear whether this effect reflects the intrinsic underlying severity of disease (ie, study bias), or whether steroid sparing immunosuppressive agents mitigate against successful ATT.

  • Infection
  • Inflammation

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Tuberculosis (TB) remains a major global health problem with an estimated 8.6 million people who have developed TB and 1.3 million deaths in 2012.1 Among the developed Western countries, the UK has one of the highest incidence rates of 13.9/100 000 reported in 2012. The risk factors include young age, male gender and being born outside of the UK. Increased and rising rates are seen in the population from the Indian subcontinent and in those who have been resident in the UK for long periods of time prior to their TB diagnosis.2

The clinical diagnosis of ocular TB is difficult since clinical signs and symptoms can mimic other conditions.3 ,4 The ophthalmologist relies on a combination of clinical history, ocular signs, systemic examination by a physician and screening investigations such as chest radiographs and the tuberculin skin test (TST).3 ,4 A definite diagnosis of TB uveitis can be confirmed by performing acid-fast smears, mycobacterial cultures or PCR-based assays on ocular fluid samples.

Interferon γ release assays (IGRAs) are new immune-based in vitro tests based on the detection of interferon γ (IFN-γ) released by sensitised T cells on stimulation with very specific antigens, early secretory antigen target-6 and culture filtrate protein-10, for detecting Mycobacterium tuberculosis infection.5 ,6 Two commercially available IGRAs include the T-SPOT.TB test (based on the ELISpot technology to directly count the number of IFN-γ-secreting T cells) and the QuantiFERON-TB Gold In-Tube test (QFT-G; based on the ELISA technology, which measures the concentration of IFN-γ secretion; Cellestis, Carnegie, Victoria, Australia).7 We examined the predictive failure related to treatment failure or success in a TB non-endemic, cosmopolitan population.


This was a retrospective review of health records of patients seen at a large tertiary uveitic centre in central London, UK. The study was approved by the institutional review board (ROAD 14/012), and all the research adhered to the tenets of the Declaration of Helsinki.

The clinical case definition of presumed intraocular TB was broad, but entailed all patients with uveitis where the diagnosis of TB uveitis could be considered based on one or more of the broad clinical signs suggested by Gupta et al in 2007.3 In addition, all patients had had a diagnostic work-up to exclude other diagnoses. Data collected included information on demographics, clinical findings, investigations and therapeutic regimens. Patients at presentation were examined for signs suggestive of ocular TB by the uveitis team. All patients with presumed TB with a positive QFT-G were included in the study. All patients with positive QFT-G and active inflammation suggestive of presumed ocular TB were further referred to a respiratory physician. Based on the physician's discretion and the ophthalmologists’ recommendations, antitubercular therapy (ATT) was initiated. We have taken a ‘novel’ approach to this problem of selection bias for instituting ATT by looking at an expanded dataset of patients where the diagnosis of TB uveitis could reasonably be considered, based on clinical features, demographic risk factors and where investigations have excluded other diagnoses. By adding patients who do, and do not get ATT treatment, we add power to our analysis in the form of a multivariate logistic regression analysis. The decision not to give ATT was not based on intolerance to medications.

Important definitions used were:

QFT-G: positivity was assessed according to the US Centers for Disease Control and Prevention (CDC) recommendations8 (positive >0.35 IU/m). The test was not adjusted or considered inappropriate if the patient was on concurrent oral corticosteroids as the CDC has not recommended any specific criteria for assessment of QFT-G results based on concurrent steroid therapy. However, we acknowledge the fact that oral corticosteroid treatment at the time of assessment can act as a potential confounder for the QFT-G value, and we, hence, performed the statistical test to study the impact of the oral steroids or steroid sparing immunosuppressive therapy at the time of QFT-G assessment.

Definition of failure: treatment failure was defined as the inability to taper oral corticosteroids to <10 mg/day or topical steroids to less than two times per day or the inability to stop the steroid sparing immunosuppressive agent or persistence or recurrence of inflammation within the first 6 months of completion of ATT. For patients not on ATT, treatment failure was defined by the inability to taper medications to the same levels of systemic and topical use at the same follow-up duration as the ATT-treated group. The definition was adapted based on the absence of current conventional guidelines for diagnosis and treatment of presumed ocular TB. The final follow-up for assessment of treatment failure or success was based on the inflammation or recurrence within the first 6 months of completion of ATT or within 6 months of tapering or stopping oral corticosteroids for the non-ATT group.

Statistical methods

Qualitative variables were expressed as percentages. Quantitative variables were expressed as mean values±SD or as median values (range). A binary logistic regression model was built with the treatment failure result as the dependent variable. The selection of variables in the final model was performed by a forward-conditional method, with significance levels of ≤0.05 for inclusion and ≥0.1 for exclusion, and in addition, putting age, gender and QFT-G values into the equation. Overall model fit was assessed by the Nagelkerke R2 and the C-index. A ROC (receiver operating characteristics) plot was constructed to assess the specificity and sensitivity of the logistic regression variables for treatment failure. Results were considered statistically significant when p<0.05. Data were analysed using Stata/SE V.13.0 (Stata, College Station, Texas, USA).


A total of 365 patients were identified from our uveitis database with a diagnosis of presumed ocular TB. Three hundred and nine (81.64%) patients underwent QFT-G; out of which, 220 (60.44%) patients were QFT-G positive, and were included in the study (figure 1). Seven patients had insufficient follow-up, and were excluded from further analysis. Two hundred and thirteen patients with positive QFT-G were further grouped into those who received ATT (n=136, 63.85%) or not received ATT (n=77, 36.15%) (figure 1A). Figure 1B summarises success and failure in each subgroup with and without oral steroids or steroid sparing immunosuppressive therapy.

Figure 1

(A) Distribution of patients with presumed ocular TB with QGold-positive QFT-G. (B) Flow diagram of outcome of cohort of patients with presumed ocular tuberculosis with positive QFT-G ATT, antitubercular therapy; TB, tuberculosis; QFT-G, QuantiFERON-TB Gold In-Tube test.

Thirty eight (18.27%) patients were on oral steroids or steroid sparing immunosuppressives at the time of QFT-G. The mean value of QFT-G was 7.32±6.71 in patients not on prior steroids and/or steroid sparing immunosuppressives and 6.47±5.09 for those on prior corticosteroid and/or steroid sparing immunosuppressives (p=0.816, Mann–Whitney U test). The mean age of the population was 45.99±15.61 years with a preponderance of males and bilateral presentation (table 1). Patients receiving ATT were significantly younger (p<0.001) and more frequently of Asian (p=0.016) descent. There was no statistically significant association between bilaterality and patients receiving ATT (p=0.269).

Table 1

Baseline demographics

Comparing the clinical phenotypes of patients who received ATT therapy versus those who did not, those with ocular features of retinal vasculitis (p=0.014), serpiginous choroiditis (p=0.015) and panuveitis (p=0.009) in conjunction with positive QFT-G were more likely to be treated with ATT (table 2) compared with those without such features. Patients with milder forms of uveitis, such as anterior (p<0.001) or intermediate uveitis (p<0.001), were proportionally under-represented in the treatment group. There was no specific preference to institute ATT based on qualitative changes suggestive of old healed pulmonary TB seen on chest radiography (p=0.115) or quantitative value of the QFT-G (p=0.90) (table 3).

Table 2

Clinical phenotypes and their likelihood of receiving ATT

Table 3

Diagnostic and treatment characteristics based on patients receiving ATT

Treatment details

For the patients in the ATT-treated group, ATT was instituted by the respiratory physicians. Based on the preference of the physician, different ATT regimens were instituted. Out of 136 patients in ATT-treated group, there were 53 (39.0%) patients with 6 months of ATT, and 19 (14.0%) and 54 (39.7%) with 9 and 12 months of ATT, respectively, highlighting the heterogeneity in the treatment preference by the physicians. Also, 10 (7.3%) patients had only 3 months of isoniazid and rifampicin prophylaxis (like directly observed therapy, a short-course therapy for TB) before initiating steroid or steroid sparing immunosuppressive therapy. Due to relentless clinical progression, steroid sparing immunosuppressive agents were given in 15 (11.0%) patients. While 93 (68.38%) patients received conventional ATT regimen comprising of 2 months of rifampicin (10 mg/kg/day), ethambutol (25 mg/kg/day), isoniazid (5 mg/kg/day) and pyrazinamide (25 mg/kg/day) followed by 6–10 months of rifampicin (10 mg/kg/day) and isoniazid (5 mg/kg/day), 43 (21.62%) patients had 2 months of rifampicin, isoniazid, pyrazinamide and moxifloxacin (7.5–10 mg/kg/day) followed by rifampicin, isoniazid and moxifloxacin as advised by one of the treating chest physicians. This was noted to be a change in practice from the regimen most commonly used in 2009–2010 when moxifloxacin was rarely used. The type of regimen did not have any statistically significant impact on the treatment outcome.

Bivariate and logistic regression analyses were done to identify the risk factors for treatment failure for the entire cohort of 213 patients. Oral corticosteroids did not have any influence on the failure or success of therapy on bivariate analysis (p=0.109), and hence, was not taken in the multivariate regression analysis. The model for logistic regression analysis after inputting selective variables (as listed in table 4) from bivariate analysis was accurate (C index=0.822, 95% CI 0.780 to 0.863, Hosmer–Lemeshow p=0.463) (table 4). Patients receiving steroid sparing immunosuppressive agents had statistically high chances of failure (OR=24.10, 95% CI 6.75 to 86.11, p<0.001) across the whole cohort.

Table 4

Results of binary logistic regression model for factors predicting failure (n=213, Nagelkerke R2=0.357, C-index=0.822, Hosmer–Lemeshow p=0.463)

Mean QFT-G value for ATT group was 7.13 ((±5.96), and for non-ATT group was 7.02 (±7.22). As the conventional cut-off for QFT-G is 0.35, we explored newer cut-off values starting from 1.00 to 3.00. After analysing different cut-off values at 1.00, 1.50, 2.00, 2.50 and 3.00 of QFT-G, patients with QFT-G values >1.50 (OR=0.20, 95% CI 0.09 to 0.48, p<0.001) had less risk of treatment failure as against those with QFT-G values between 0.35 and 1.50. Erythrocyte sedimentation rate was a significant factor in univariate analysis, but not in regression analysis possibly due to the difference in age of the patients in the treatment group (patients without ATT were older=50.91±17.62 years compared with those who had ATT 43.21±13.65, p=0.001).

We further compared the cohort of patients who received steroid sparing immunosuppressive agents with the group who did not (table 5). Patients with bilateral disease and panuveitis were more likely to receive steroid sparing immunosuppressive agents, while patients with anterior uveitis were less likely to receive oral corticosteroids therapy (p<0.067). Fifteen patients (11.03%) were on concurrent steroid sparing immunosuppressive agents and ATT.

Table 5

Subgroup classification by oral immunosuppressive therapy

ROC and sensitivity and specificity plots were performed after logistic regression analysis (figure 2) to assess the ability of the regression model to correctly classify patients into treatment success or failure and also to assess the ROC with a new cut-off value for QFT-G. Using different cut-off values for QFT-G as described above, the forward conditional model included a cut-off value of 1.5 as this was found to maximise the predictive ability of the model versus the other values tried (1.0, 2.0, 2.5 and 3.0). The predicted probability of the model described in table 4 had an area under ROC curve (C-index) of 0.796 (figure 2A). Patients who received ATT and had QFT-G >1.50 had an Area Under the Curve (AUC) (figure 2B) of 0.804 (95% CI 0.690 to 0.924) and those with QFT-G <1.50 had AUC (figure 2C) of 0.8469 (95% CI 0.659 to 1.000). For non-ATT group, patients with QFT-G of >1.50 had AUC (figure 2D) of 0.646 (95% CI 0.443 to 0.864) and those with QFT-G <1.50 had AUC (figure 2E) of 0.616 (95% CI 0.354 to 0.846). ATT was not a statistically significant predictor for failure or treatment success. Treatment with ATT or oral corticosteroids alone or in combination in patients with positive QFT-G and any combinations of clinical signs reduced the likelihood of treatment failure, but did not reach statistically significant sensitivity or specificity levels (figure 2).

Figure 2

ROC curve of the binary logistic regression model for the prediction of treatment failure (A). ROC curve of the binary logistic regression model for the prediction of treatment failure in cohort of patients who had ATT and had QFT-G >1.50 (B). ROC curve of the binary logistic regression model for the prediction of treatment failure in cohort of patients who had ATT, but QFT-G was <1.50 (C). ROC curve of the binary logistic regression model for the prediction of treatment failure in cohort of patients who did not have ATT, and QFT-G was >1.50 (D). ROC curve of the binary logistic regression model for the prediction of treatment failure in cohort of patients who did not have ATT, but QFT-G was <1.50 (E). ATT, antitubercular therapy; ROC, receiver operating characteristics; QFT-G, QuantiFERON-TB Gold In-Tube test.


Diagnosis and management of tuberculous uveitis remain a conundrum for most clinicians. It may be the manifestation of a true infection or a hypersensitivity reaction to an extraocular infection.3 Mycobacterium culture or histopathology remains as the gold standard, but has inherent limitations due to low test yield and difficulty in obtaining samples due to poor access to ocular tissue.9 PCR is a well-established technique, but its usefulness remains limited.3 Fundus fluorescein angiography, indocyanine green angiography, ocular coherence topography, ultrasonic biomicroscopy and CT are all used as diagnostic adjuncts to monitor the progress and complications of uveitis, but their utility as primary diagnostic tools for tuberculous uveitis remain limited.10–12

IGRA are considered highly specific for M tuberculosis because they are not confounded by prior vaccination for TB. However, concerns have been raised13 about the sensitivity of QFT-G for detection of latent TB infection and superiority to traditional TST.14 Kurup et al15 reported no demonstrable advantage of QFT-G over TST for detection of latent TB infection in patients with granulomatous uveitis. Recently, Ang et al16 recommended Gold In-Tube as the first-line test in preference to T-SPOT.TB, but also demonstrated that QFT-G was not superior to TST in sensitivity as a screening test or first-line study in TB-related uveitis.17 Similarly, Babu et al18 showed that QFT-G is not specific for intraocular TB. QFT-G was positive in 70% of our patients with presumed ocular TB. It remains a valuable tool in diagnosis and management of disease in regions with low prevalence rates of TB compared with Singapore and India with high prevalence rates of both pulmonary and extrapulmonary TB where other ancillary tests are likely to be positive.

The treatment of ocular TB is largely presumptive, and there are no randomised controlled clinical trials or locally recommended guidelines to guide the management of ocular TB. Specifically, there are no randomised controlled clinical trials looking at the role of ATT in treatment of ocular TB. This has led to a wide range of treatment practice patterns, which have been reported in the literature. Like in the UK, there are no national TB guidelines about ocular TB diagnosis and treatment in many non-endemic regions including Canada and USA. There are conflicting reports and literature on the use of mono versus multiple drug therapy, and there is also significant debate about treatment duration with ATT (6–18 months). Clinical outcomes are, therefore, inconsistent due to lack of a clinical definition for the exact diagnosis of ocular TB, concomitant corticosteroid therapy and variation in severity of ocular inflammation prior to initiating treatment.19 ,20

In our current study, diagnosis was made by the clinician, and only those cases that had positive QFT-G or strong clinical suspicion of intraocular TB were referred to respiratory physicians for further consultation and initiating ATT. Patients with stable or mild anterior uveitis and retinal vasculitis were not referred for ATT due to less visual morbidity; however, patients with posterior or panuveitis or recurrent intermediate uveitis were referred for a therapeutic trial of ATT. The final decision to initiate ATT was based on the discretion of the treating physician. Similar variation in practice occurs in regard to drugs, duration and use of concurrent local or systemic steroids. Systemic steroids have been used and recommended in other forms of extrapulmonary TB;21 ,22 they are used in ocular TB uveitis to control persistent inflammation or retinal vasculitis, but their role is unproven.

In our study, the use of steroid sparing immunosuppressive therapy increased the likelihood of treatment failure. It is difficult to ascertain the exact cause. One possible hypothesis is that these agents suppress host immune function, and this can be disadvantageous in TB infection. In our series, a significant proportion of patients were on prior corticosteroid therapy before ATT was commenced. Clinical phenotypes represented by positive QFT-G and those that failed to respond to ATT and oral corticosteroid therapy may be the ‘immune-mediated’ form and possibly require prolonged immunosuppression, and hence, steroid sparing immunosuppressives did not have a positive impact on the treatment outcome.

Second, the failure in patients treated with steroid sparing immunosuppressives could be attributed to undertreatment with corticosteroids or rapid tapering of the oral steroids before the cumulative effect of steroid sparing agents sets in. Also the effective dose of corticosteroid is half the actual dose in the presence of rifampicin, which affects the metabolism of corticosteroids. Rifampicin increases the plasma clearance of prednisolone by 45%, and reduces drug bioavailability in tissues by 66%,23 which needs to be considered when deciding the dose of prednisolone. In addition, drug resistance to ATT, reinfection or immunological response to tubercular antigens released during ATT are other possible reasons for worse outcomes when immunosuppressives were used. Recurrence of disease in up to 25% of cases is a recognised phenomenon, and certain clinical phenotypes like intermediate uveitis and retinal vasculitis represent increased risk.24–26 There is a reasonable possibility that these cases irrespective of QFT-G results were not associated with tubercular uveitis, and they represented non-infectious posterior uveitis requiring prolonged immunosuppression; hence, there was failure to respond to steroid therapy. Our findings that steroid sparing immunosuppressives in ATT groups increased the likelihood of treatment failure could be an artefactual finding of study bias, and may simply reflect the underlying severity of uveitis.

According to our model, treatment success as per our definition of successful steroid taper was optimal with a QFT-G cut-off of 1.5 IU/mL. A similar trend was seen by Gineys et al27 in their small cohort, but they concluded the cut-off value was >2 IU/mL. The authors demonstrated a higher success rate with ATT in patients with higher cut-off value (7.67 vs 1.22 IU/mL with p=0.026 by Wilcoxon test). The authors included patients with varying degree and types of inflammation in their case series as there were seven patients with scleritis, 34 patient with panuveitis, 15 patients with posterior uveitis, 14 patients with intermediate uveitis and 15 patients with anterior uveitis. There were 42 patients with positive QFT-G and 54 patients with negative QFT-G; 12 (29%) patients receiving concurrent oral steroids in the QFT-G-positive group and 14 (26%) patients receiving oral steroids in the QFT-G-negative group. The authors evaluated the treatment success based on a therapeutic trial with ATT. Thus, there is emerging evidence that laboratory cut-off values (0.35 IU/mL) set by the manufacturer are possibly too low for non-endemic areas. It may be time to revisit and set up higher cut-off values if one is going to recommend ATT treatment based on a positive IGRA in the context of presumptive clinical signs of tuberculous uveitis.

The ROC curves demonstrated that ATT was not a significant predictor of success in our models and that the ROC curves were comparable in the ATT-treated versus the non-ATT-treated group in respective subgroups of patients with QFT-G of <1.50 and QFT-G of >1.50. The success rate would have been better if we had treated only those patients who had QFT-G >1.5. This conclusion can only be inferred post analysis. However, the equivalence of success rates between the ATT-treated group and the non-ATT-treated group (presumed non infective) suggests that the clinicians are doing a reasonable job in identifying those patients who will do well with ATT therapy.

In summary, we present one of the largest clinical case cohort studies of presumptive ocular TB with positive QFT-G in a population with low endemic prevalence. Positive QFT-G provides useful information, and directs towards the use of ATT in the presence of other clinical signs suggestive of presumptive ocular TB. However, based on our results, it seems the cut-off value of QFT-G needs further investigation in low-endemic regions. Our finding that steroid sparing immunosuppressive therapy increased the likelihood of treatment failure is intriguing. While it could be an artefactual finding, and simply reflects the underlying severity of uveitis, we propose an alternative hypothesis, namely that immunosuppression increases the likelihood of treatment failure in TB uveitis. This is biologically plausible as immunosuppressives suppress host immune function. There are no randomised controlled trials explicitly looking at the benefit of oral steroids or steroid sparing immunosuppression in tuberculous uveitis. In fact, the evidence base for the benefit of oral steroids in TB-associated uveitis is extremely weak, largely resting on expert opinion and very low quality evidence.28

Indeed, this study has inherent limitations of being retrospective in nature with no defined treatment duration. Prospective and ideally randomised controlled trials are urgently needed in order to investigate this further. Such studies are likely to require a multicentre approach in order to compare results in TB endemic and non-endemic counties.


The authors would like to acknowledge support from Margaret Richards who helped in identifying case records with QFT-G at our centre. The authors will also acknowledge support from Dr Farzana Rahman, Dr Ioanna and Dr Sumita Phatak for their assistance with the data collection.



  • Contributors BG had written the first draft of the manuscript, and was involved in critical inputs and data analysis and interpretation. RA had edited the first draft of the manuscript, and was involved in study design, data collection, analysis and intellectual inputs. JJG-L had contributed to the first draft of the manuscript, and did statistical analysis. RG assisted in statistical analysis. CEP, PKFA and MW were directly involved in patient care and edited the draft and provided intellectual inputs for the study design and data collection. JC did revised data collection and checking the data, and did revise the manuscript.

  • Funding This work was supported by the National Institute for Health Research (NIHR) Biomedical Research Centre based at Moorfields Eye Hospital NHS Foundation Trust and UCL Institute of Ophthalmology.

  • Competing interests JJG-L has received study grants from Alcon, Novartis and Merck, and has provided unpaid consultancy to Bayer. RA is on NMRC overseas research training fellowship at Institute of Ophthalmology and Moorfields Eye Hospital, London.

  • Ethics approval Ethics approval was obtained from institute ethics committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement All the data are available on an encrypted and password-protected computer with the authors for this study.

Linked Articles

  • At a glance
    Keith Barton James Chodosh Jost Jonas