Complications of miliary tuberculosis: low mortality and predictive biomarkers from a UK cohort
© The Author(s). 2017
Received: 6 August 2016
Accepted: 11 April 2017
Published: 20 April 2017
Untreated, miliary tuberculosis (TB) has a mortality approaching 100%. As it is uncommon there is little specific data to guide its management. We report detailed data from a UK cohort of patients with miliary tuberculosis and the associations and predictive ability of admission blood tests with clinical outcomes.
Routinely collected demographic, clinical, blood, imaging, histopathological and microbiological data were assessed for all patients with miliary TB identified from the London TB register from 2008 to 2012 from Northwest London Hospitals NHS Trust. Multivariable logistic regression was used to assess factors independently associated with the need for critical care intervention. Receiver operator characteristics (ROC) were calculated to assess the discriminatory ability of admission blood tests to predict clinical outcomes.
Fifty-two patients were identified with miliary tuberculosis, of whom 29% had confirmed central nervous system (CNS) involvement. Magnetic resonance imaging (MRI) was more sensitive than computed tomography (CT) or lumbar puncture for detecting CNS disease. Severe complications were frequent, with 15% requiring critical care intervention with mechanical ventilation. This was independently associated with admission hyponatraemia and elevated alanine aminotransferase (ALT). Having an admission sodium ≥125 mmol/L and an ALT <180 IU/L had 82% sensitivity and 100% specificity for predicting a favourable outcome with an area under the ROC curve (AUC) of 0.91. Despite the frequency of severe complications, one-year mortality was low at 2%.
Although severe complications of miliary tuberculosis were frequent, mortality was low with timely access to critical care intervention, anti-tuberculous therapy and possibly corticosteroid use. Clinical outcomes could accurately be predicted using routinely collected biochemistry data.
KeywordsTuberculosis Miliary tuberculosis Central nervous system infection Mortality Biomarkers
Tuberculosis is one of the leading causes of death from infectious diseases worldwide . Miliary tuberculosis results from widespread haematogenous spread of the Mycobacterium tuberculosis bacilli (TB). Clinical presentations are protean and vary from vague constitutional upset in ambulatory patients to abrupt multi-organ failure . Untreated, miliary TB has a mortality approaching 100% , reducing to 7.1-30% with treatment [2, 4–6].
Miliary TB is uncommon, comprising 1-2% of TB cases , thus specific data regarding its management, particularly the ability to identify patients at high risk of poor outcomes, are lacking . Current treatment guidelines are largely extrapolated from randomised controlled trials of pulmonary TB rather than specific trials of patients with miliary TB. As such, insights provided by cohort studies are important in guiding best practice. However, most cohorts are small and from the developing world, with limited neuroimaging and microbiological data. We sought to characterise the clinical course, investigation findings, and outcomes of a cohort with miliary TB from an ethnically diverse population with one of the highest TB incidences in the UK .
Patients with miliary TB were identified from the London TB register (LTBR) from 2008 to 2012 at Northwest London Hospitals NHS Trust. These records were cross-checked against the hospital database of all in-patient admissions.
Diagnosis of miliary TB was confirmed by identification of the characteristic miliary pattern on chest radiograph or computed tomography (CT) along with one or more of the following features: clinical features compatible with TB and response to anti-tuberculous therapy (ATT); histopathological evidence of TB; microbiological evidence of TB (positive smear or culture). Demographic, clinical, biochemical, radiological, histopathological evidence of TB (caseating or non-caseating granulomas with or without acid-alcohol fast bacilli on Ziehl-Neelson staining), and microbiological data were obtained from the LTBR and the hospital electronic patient records.
Statistical analysis and biomarker selection
Univariate analysis of continuous data was performed using t-tests and the Mann-Whitney U tests as appropriate. For subsequent biomarker analysis, the primary outcome of interest, defined as an ‘unfavourable outcome’, was the need for critical care intervention with mechanical intervention. Whereas, a ‘favourable outcome’ was defined as not requiring critical care intervention with mechanical ventilation.
Multivariable logistic regression (p < 0.1 for entry from univariable logistic regression) was used to assess factors associated with the need for critical care intervention. Biomarkers that were independently associated with the need for critical care intervention (i.e. an ‘unfavourable outcome’) in the multivariable model were then assessed for their predictive performance by determining their receiver operator characteristics (ROC). Biomarker classification thresholds were determined by optimising the balance of sensitivity and specificity. Biomarker classification performance was assessed and thresholds determined using the area under the ROC curve (AUC). This is equivalent to the balanced accuracy of the classifier (i.e. accuracy taking account of differences in prevalence of the outcome between groups) for both ‘favourable’ (i.e. not requiring critical care intervention) and ‘unfavourable’ outcomes. Additionally, sensitivity, specificity, positive and negative predictive values were calculated. Confidence intervals were determined using bootstrapping with 1000 replicates.
Biomarkers were combined using an exhaustive grid across the range of all possible values and combinations to determine the optimum threshold for each biomarker in combination so that the AUC was maximised.
Given the relatively small size of the dataset, splitting the data into separate training and testing datasets would be subject to bias. Therefore, to validate the predictive rule in this dataset, a leave-one-out-cross-validation (LOOCV) scheme was used. This entailed leaving one subject out of the dataset, then performing an exhaustive grid search, as before, to determine the optimal thresholds for the biomarker(s) of interest. This threshold(s) was then applied to the case that was left out. These steps were then repeated and summary statistics across the range of folds were calculated. Accuracy was determined by the percentage of correct responses across all the cross-validation folds.
Baseline characteristics, co-morbidities, admission blood tests, neurological tests and outcomes of a UK cohort of patients with miliary tuberculosis
n (%) or median (range)
Previous diagnosis of TB
Hepatitis B sAg positive
Hepatitis C positive
Vitamin D deficient
Admission blood tests
Alanine aminotransferase (IU/L)
Alkaline phosphatase (iU/L)
Erythrocyte sedimentation rate (mm/h)
C-reactive protein (mg/L)
White cell count (×109/L)
Neutrophil count (×109/L)
Platelet count (×109/L)
Vitamin D (nmol/L)
Assessment for CNS involvement
Number of patients assessed for CNS infection
- CT head consistent with CNS infection
- MRI brain consistent with CNS infection
- Lumbar puncture consistent with CNS infection
Total number with confirmed CNS infection
Outcomes and adverse events
Death (by one-year)
Nineteen patients (37%) had a CT brain scan performed on their initial admission. Of these, 6 (31%) had ring-enhancing lesions typical of tuberculomas. Eleven patients (21%) underwent magnetic resonance imaging (MRI) of their brain, of whom 10 (91%) demonstrated tuberculomas. Notably, 4 patients with normal/nonspecific admission CT had subsequent MRI because due to clinical concern of neurological disease, all of whom demonstrated tuberculomas. The overall prevalence of tuberculomas seen on MRI or CT was 12 of 23 (52%), but we may have missed lesions in the 29 patients who did not undergo neuroimaging, and in patients in whom tuberculomas developed subsequently.
Forty patients (77%) had positive cultures for M. tuberculosis (MTB) complex. Induced sputa lead to a positive culture in thirty (58%) patients. Only ten sputum samples (19%) revealed acid-alcohol fast bacilli (AFB) on microscopy. Seventeen patients (33%) had cerebrospinal fluid (CSF) examination. Of these, four (24%) had findings compatible with TB meningitis, all of whom subsequently cultured MTB (two of four [50%] had tuberculomas). Aside from sputa and CSF, MTB was cultured from joint (n = 2), pericardial (n = 2) and pleural (n = 1) fluids (n = 1), transbronchial biopsy (n = 1) and bone marrow (n = 1). Four (10%) were resistant to isoniazid, three (6%) to pyrazinamide, three (6%) to streptomycin, one to rifampicin and isoniazid (i.e. multidrug resistant – MDR) and none to ethambutol.
All patients were commenced on quadruple ATT with most receiving rifampicin, isoniazid, pyrazinamide and ethambutol. Median (range) duration of ATT was 12 (1.5 – 28) months. Ten (19%) patients stopped pyrazinamide due to drug-induced liver injury (15%) or intolerance (4%). 29 patients (56%) received corticosteroids during their ATT for a variety of indications including central nervous system (CNS) involvement, systemic inflammatory response syndrome and paradoxical reactions.
Follow-up to one-year was complete for 49 patients (94%). One 87-year-old patient with multiple co-morbidities, including chronic renal failure, died during treatment (after 1.5 months’ therapy). One patient died subsequently from constrictive pericarditis secondary to pericardial involvement, two years after successfully completing 12 months of ATT.
Univariable and multivariable logistic regression model results for the outcome of requiring critical care intervention
Odds ratio (95% CI)
Odds ratio (95% CI)
ALT (per 10 IU/L)
Bilirubin (per 10 μmol/L)
CRP (per 10 mg/L)
Platelet count (per 100 cells/mm3)
Sodium (per 10 mmol/L)
Classification performance of different biomarkers and their optimum threshold by outcome determined using bootstrapping
ALT ≥180 IU/L
Sodium <132 mmol/L
ALT <180 IU/L & sodium ≥125 mmol/L
Given the size of the dataset, biomarker prediction validity was assessed using LOOCV. For the prediction of an unfavourable outcome the median (range) ALT threshold was: ≥180 (180-191) IU/L with diagnostic accuracy of 92.3% across the cross-validation folds. For the prediction of a favourable outcome, the median (range) ALT threshold was <191 (180-200) IU/L; and for sodium was: ≥125 (106-126) mmol/L yielding a diagnostic accuracy of 76.9% across the cross-validation folds.
We report the lowest one-year mortality from miliary TB (2%), despite significant rates of severe complications. Additionally, we demonstrate how routinely collected biochemical measures can be used to accurately predict clinical outcomes.
The mortality rate in our cohort compares favourably to previously reported rates of 10-30% from the developing world and 7.1%-21% from the developed world [2, 4–6]. Critical care intervention and mechanical ventilation, which was required in 15% of our cohort, are likely to be responsible for the low mortality we report as these patients would almost certainly have died if these facilities were not available. Additionally, extensive experience in the management of complicated tuberculosis at a high-volume centre are likely to have contributed to our low mortality. Due to the retrospective nature of data collection it was not possible to assess the impact of corticosteroid usage on outcome, and whilst corticosteroid usage was higher in patients requiring mechanical ventilation this may reflect increased rates of CNS involvement and systemic inflammatory response syndrome in patients requiring critical care support. The impact of corticosteroids on mortality in CNS TB infection has been demonstrated, particularly in non HIV-infected individuals . Furthermore, meta-analysis of corticosteroid use in all forms of TB suggest a beneficial effect on mortality, however many of the 41 randomised controlled trials were identified as being at high risk of bias, and none were performed to assess corticosteroid usage specifically in miliary TB . Given the multisystem nature of miliary TB with frequent and potentially under-recognised CNS involvement, further research is warranted to clarify the role of corticosteroids.
Hyponatraemia and an elevated ALT on admission were associated with an unfavourable clinical course. The causes of hyponatraemia in those with miliary TB are numerous, but are likely to include brain injury as well as adrenal and pituitary dysfunction. Whereas, an increased ALT identifies those with significant liver injury and in this cohort a threshold of ≥180 IU/L was strongly predictive of an unfavourable outcome. Both factors are likely surrogates of disease burden, hence their association with clinical outcomes. Our data suggests these routinely collected biochemical measures could be used to identify patients with severe disease at risk of deterioration. However, these thresholds need verifying in an independent dataset. Pending further study, they should be interpreted with caution given the relatively low number of patients in our study. Furthermore, the influence of co-morbidity, in particular HIV-infection and type 2 diabetes given their increasing prevalence in regions with the highest burden of tuberculosis, on the relationship between predictive biomarkers and outcomes needs to be determined.
UK National Institute for Health and Care Excellence (NICE) guidelines  currently recommend neuroimaging only for miliary TB patients with CNS signs or symptoms. Given that miliary TB is caused by the widespread haematogenous dissemination of MTB, it is unsurprising that CNS involvement was proven in 29% of our cohort. However, as not all our patients underwent systematic neuroimaging or CSF sampling, this represents the lower bound of the true prevalence of CNS disease and confirms historical autopsy data documenting high rates of CNS disease . In keeping with other studies, MRI was more sensitive than CT in detecting tuberculomas . CSF examination demonstrated TB meningitis in a minority, suggesting that meningitis may be a late manifestation of disease, however, normal CSF biochemistry and microscopy did not exclude CNS involvement. In addition, CNS tubercles may initially be below the size of detection by CT or MRI, and may paradoxically appear during treatment.
There are no randomised trials to provide guidance on duration of therapy for miliary TB. A 6-month course is currently recommended for miliary TB without CNS involvement in UK, USA, Europe and WHO international guidelines [10, 12–14]. However, as we suspect CNS involvement in miliary TB is often under-recognised, a 6-month course of ATT may risk later relapse and complications. Generally, patients in our cohort were treated for 12-months. Whether this contributed to our low mortality is unclear. Given the ubiquity of CNS involvement, we advocate performing contrast-enhanced neuroimaging (ideally MRI) in all patients and CSF examination in those with symptoms or signs of meningitis. Ideally, this should be at the time of diagnosis and possibly repeated 6-12 weeks later to detect tuberculomas which appear during treatment . In settings where this is not possible, 9-12 months’ treatment to cover CNS disease may be prudent. Given the high a priori probability of CNS disease, we advocate consideration for longer empirical duration of therapy (9-12 months) if contrast-enhanced MRI and CSF examination cannot be undertaken, such as in the developing world.
The limitations of this case series include the retrospective and non-systematic nature of initial data collection and incomplete post-treatment follow-up. However, most mortality attributed to miliary TB occurs in the early stages of disease so it is unlikely that longer follow-up would significantly change our conclusions [16, 17]. Additionally, due to the low mortality in our cohort, where there was only one death, we lacked power to detect associations with this important outcome.
CNS involvement in miliary TB is common and is likely to be under-recognised and under-treated. Although severe complications occur frequently, they can be predicted by admission hyponatraemia and increased ALT. A low one-year mortality is achievable in a high resource setting with critical care support and mechanical ventilation. Further research is required to establish the role of corticosteroids and prolonged ATT in the management of miliary TB.
Central nervous system
Human immunodeficiency virus
Magnetic resonance imaging
We gratefully acknowledge the assistance of the TB nurses at Northwest London Hospitals NHS Trust for their assistance with this study.
No specific funding for this study was received.
Availability of data and materials
The raw data are not publicly available due to patients being potentially identifiable but are available (in modified form) from the corresponding author on reasonable request.
JU, FC, APS, LJ and RND conceived of and designed the study. JU, FC, APS and CC collected data from participating sites. JU conducted statistical analyses. JU, FC and AJK compiled the first draft of the study manuscript and all authors contributed to subsequent revisions. All authors read and approved the final manuscript.
The authors declare that they have no competing interests.
Consent for publication
Ethics approval and consent to participate
This study was undertaken with the retrospective analysis of routinely collected clinical data which was anonymised before analysis by the researchers. As such, formal Research Ethics Committee review was not required, as confirmed by the NHS Health Research Authority.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Murray CJ, Lopez AD. Global mortality, disability, and the contribution of risk factors: global burden of disease study. Lancet. 1997;349:1436–42.View ArticlePubMedGoogle Scholar
- Sharma SK, Mohan A, Sharma A, Mitra DK. Miliary tuberculosis: new insights into an old disease. Lancet Infect Dis. 2005;5:415–30.View ArticlePubMedGoogle Scholar
- Falk A. U. S. veterans administration-armed forces cooperative study on the chemotherapy of tuberculosis. Tuberculous meningitis in adults, with special reference to survival, neurologic residuals, and work status. Am Rev Respir Dis. 1965;91:823–31.View ArticlePubMedGoogle Scholar
- Venkatraman N, King T, Bell D, Woltmann G, Wiselka M, Abubakar I, et al. High levels of neurological involvement but low mortality in miliary tuberculosis: a 6-year case-series from the UK. Eur Respir J. 2016;47:1578–81.View ArticlePubMedGoogle Scholar
- Ormerod LP, Horsfield N. Miliary tuberculosis in a high prevalence area of the U.K.: Blackburn 1978-1993. Respir Med. 1995;89:555–7.View ArticlePubMedGoogle Scholar
- Touré NO, Cissé MF, Dia Kane Y, Diatta A, Bouker Bakioui B, Ndiaye EHM, et al. Miliary tuberculosis: a report of 49 cases. Rev Mal Respir. 2011;28:312–6.View ArticlePubMedGoogle Scholar
- Tuberculosis in the UK, 2014 report. Public Health England (https://www.gov.uk/government/uploads/system/uploads/attachment_data/file/360335/TB_Annual_report__4_0_300914.pdf); 2014.
- Prasad K, Singh MB, Ryan H. Corticosteroids for managing tuberculous meningitis. Cochrane Database Syst Rev. 2016;4:CD002244.PubMedGoogle Scholar
- Critchley JA, Young F, Orton L, Garner P. Corticosteroids for prevention of mortality in people with tuberculosis: a systematic review and meta-analysis. Lancet Infect Dis. 2013;13:223–37.View ArticlePubMedGoogle Scholar
- Tuberculosis. National Institute for Health and Care Excellence; 2016. (https://www.nice.org.uk/guidance/ng33/).
- American College of Physicians. Tuberculoma of the brain. Ann Intern Med. 1934;7:1141–5.
- American Thoracic Society - AJRCCM. American Thoracic Society/Centers for Disease Control and Prevention/Infectious Diseases Society of America. Am J Respir Crit Care Med. 2003;167:603–62.
- Migliori GB, Zellweger JP, Abubakar I, Ibraim E, Caminero JA, De Vries G, et al. European union standards for tuberculosis care. Eur Respir J. 2012;39:807–19.View ArticlePubMedPubMed CentralGoogle Scholar
- World Health Organization. Guidelines for treatment of tuberculosis. WHO. World Health Organization; 2010. (http://www.who.int/tb/publications/2010/9789241547833/en/).
- Afghani B, Lieberman JM. Paradoxical enlargement or development of intracranial tuberculomas during therapy: case report and review. Clin Infect Dis. 1994;19:1092–9.View ArticlePubMedGoogle Scholar
- Lin C-H, Lin C-J, Kuo Y-W, Wang J-Y, Hsu C-L, Chen J-M, et al. Tuberculosis mortality: patient characteristics and causes. BMC Infect Dis BioMed Central. 2014;14:5.View ArticleGoogle Scholar
- Gelb AF, Leffler C, Brewin A, Mascatello V, Lyons HA. Miliary Tuberculosis. American Review of Respiratory Disease. Am Lung Assoc. 1973;108(6):1327–33.