OUP user menu

Diagnosing tuberculous pericarditis

H. Reuter, L. Burgess, W. van Vuuren, A. Doubell
DOI: http://dx.doi.org/10.1093/qjmed/hcl123 827-839 First published online: 22 November 2006

Abstract

Background: Definitive diagnosis of tuberculous pericarditis requires isolation of the tubercle bacillus from pericardial fluid, but isolating the organism is often difficult.

Aim: To improve diagnostic efficiency for tuberculous pericarditis, using available tests.

Design: Prospective observational study.

Methods: Consecutive patients (n = 233) presenting with pericardial effusions underwent a predetermined diagnostic work-up. This included (i) clinical examination; (ii) pericardial fluid tests: biochemistry, microbiology, cytology, differential white blood cell (WBC) count, gamma interferon (IFN-γ), adenosine deaminase (ADA) levels, polymerase chain reaction testing for Mycobacterium tuberculosis; (iii) HIV; (iv) sputum smear and culture; (v) blood biochemistry; and (vi) differential WBC count. A model was developed using ‘classification and regression tree’ analysis. The cut-off for the total diagnostic index (DI) was optimized using receiver operating characteristic (ROC) curves.

Results: Fever, night sweats, weight loss, serum globulin (>40 g/l) and peripheral blood leukocyte count (<10 × 109/l) were independently predictive. The derived prediction model had 86% sensitivity and 84% specificity when applied to the study population. Pericardial fluid IFN-γ ⩾50 pg/ml, concentration had 92% sensitivity, 100% specificity and a positive predictive value (PPV) of 100% for the diagnosis of tuberculous pericarditis; pericardial fluid ADA ⩾40 U/l had 87% sensitivity and 89% specificity. A diagnostic model including pericardial ADA, lymphocyte/neutrophil ratio, peripheral leukocyte count and HIV status had 96% sensitivity and 97% specificity; substituting pericardial IFN-γ for ADA yielded 98% sensitivity and 100% specificity.

Discussion: Basic clinical and laboratory features can aid the diagnosis of tuberculous pericarditis. If available, pericardial IFN-γ is the most useful diagnostic test. Otherwise we propose a prediction model that incorporates pericardial ADA and differential WBC counts.

Introduction

Prompt treatment of tuberculous pericarditis can save lives.1 Effective treatment requires a rapid and accurate diagnosis, but this is often difficult.2 Ziehl-Neelsen (ZN)-stained smears of pericardial fluid have poor sensitivity for detecting Mycobacterium tuberculosis, while culture is both slow and insensitive.1–3 Pericardial biopsy is invasive, requires technical skills and is often not diagnostic.2, 4, 5 Clinicians thus have to rely heavily on the clinical features of pericardial tuberculosis (TB) to initiate therapy,6–8 but in view of the potential for toxic effects and the duration of anti-tuberculous chemotherapy, it is important to identify which clinical and basic laboratory features should be used. Multivariate logistic regression has been used to model the clinical predictors of tuberculous meningitis in 251 adults, producing a simple diagnostic rule with 86% sensitivity and 79% specificity for the diagnosis of tuberculous meningitis.9 A similar algorithm or scoring system for pericardial TB could improve diagnostic accuracy.

Where the diagnostic infrastructure and resources are available, suspected cases of tuberculous pericarditis may be diagnosed using polymerase chain reaction (PCR) analysis,10, 11 adenosine deaminase (ADA) activity2, 12, 13 and pericardial interferon gamma (IFN-γ) levels.13 The effect of human immunodeficiency virus (HIV) on ADA levels in patients with TB is controversial,13, 14 and the impact of HIV infection on the production of IFN-γ has not been fully evaluated. While cytology has been used to diagnose pleural TB,15–17 its use for the diagnosis of pericardial TB has not been studied.

We aimed to develop a strategy that would optimize the diagnostic efficiency of available diagnostic tests and test these prospectively in a population with a high prevalence of HIV.

Methods

All patients presenting to the Cardiology Unit, Tygerberg Academic Hospital, South Africa, with large pericardial effusions between February 1995 and June 2001 were prospectively enrolled and followed-up for a minimum of 12 months. The study complied with the Declaration of Helsinki, and all patients gave written informed consent for participation in the study, which was approved by the Ethics Committee of Stellenbosch University.

Patients underwent clinical evaluation, 12-lead surface electrocardiogram (ECG), chest radiography and 2D echocardiographic studies (Hewlett Packard, Sonos 2000 Phased Array Imaging System). Only patients with large pericardial effusions (⩾10 mm epi-pericardial separation during diastole) and/or echocardiographic evidence of tamponade (inversion of >30% of the right atrial wall during late diastole and/or early systole, and/or inward motion of the right ventricular wall in early diastole persisting after mitral valve opening) were included. A pericardial tap was performed under echocardiographic guidance via a pigtail catheter, and fluid sent for biochemistry (protein, lactate dehydrogenase [LDH], ADA), microbiology (including TB culture and ZN stain), cytology and differential white-cell count. In addition, each patient underwent tests for HIV, sputum smear and culture, blood culture, blood biochemistry (total protein, globulin, albumin, LDH), as well as serological testing for antinuclear factor (ANF), rheumatoid factor (RF), antistreptolysin-O titre (ASOT) and C-reactive protein (CRP). ADA activity (U/l) was measured in all pericardial fluid specimens according to the method described by Giusti.18 Differential leukocyte counts were performed on peripheral blood and pericardial fluid samples (3–5 ml of blood in separate EDTA tubes).

Pericardial fluid specimens (5–10 ml) for cytology were fixed by adding an equal amount of 50% ethanol. Smears were prepared from sediments by routine methods, and stained according to the Papanicolaou method. The commercially available PCR assay (Roche Amplicor PCR for M. tuberculosis) was used for the detection of the IS6110 sequence of M. tuberculosis in pericardial fluid specimens, applying standard techniques and procedures. A 5 ml sample of the pericardial fluid was sent for Gram and ZN staining and microscopic examination. A 7 ml sample of pericardial fluid was injected into BACTEC medium immediately after completion of the pericardiocentesis procedure, and cultured routinely in an automated radiometric BACTEC MGIT 960 system (Becton Dickenson and Co.). At the time of pericardiocentesis, pericardial fluid was collected on ice and frozen within 30 minutes at −70°C for the analysis of IFN-γ by using an enzyme-linked immunosorbent assay (ELISA) according to manufacturer's instructions (Amersham Pharmacia Biotech).

Samples of expectorated sputum were stained with auramine and examined by fluoromicroscopy. During the first 18 months of the study, pericardial biopsy was performed if no aetiological diagnosis had been made within one week. Biopsy tissue was sent for histology, as well as bacterial and TB culture. A specimen of pericardial tissue was also formalin-fixed (10% buffered formalin) and processed using routine technology. Each sample was sectioned and stained using routine histochemical techniques with haematoxylin-eosin (H&E) and ZN stains.

Diagnostic criteria

Definite tuberculous pericarditis was diagnosed by one or more of the following criteria: (i) isolation of M. tuberculosis from the drained pericardial effusion or pericardial biopsy specimen (positive ZN stain and/or positive TB culture); (ii) demonstration of granulomatous inflammation on histological examination of the pericardial biopsy sample; and/or (iii) isolation of M. tuberculosis from sputum or non-pericardial exudates in the presence of clinical and/or radiological evidence of TB, associated with a positive response to anti-tuberculous therapy, and in the absence of any other obvious cause for pericardial effusions.

Probable tuberculous pericarditis was diagnosed in patients who presented with compatible clinical features (at least three of the following five symptoms: cough, weight loss, fever, night sweats, anorexia) in the absence of an alternative diagnosis, and associated with a sustained response to anti-tuberculous chemotherapy.

Non-tuberculous pericardial effusion was diagnosed in patients who were effusion/sputum ZN- and TB-culture-negative, for whom an alternative diagnosis was established, and in whom no evidence of TB was detected for 6 months after initial presentation.

Patients with clinical evidence of TB were started empirically on therapy, according to the South African National TB Control Programme,19 and followed up for at least 12 months in order to assess therapeutic response. HIV-positive patients were given cotrimoxazole prophylaxis, and referred to the Infectious Diseases Clinic for further staging and treatment.

Statistical analysis

Statistical analysis used the Mann-Whitney U, the Wilcoxon two-sample, the Kruskall-Wallis one-way ANOVA, the Bonferroni two-way ANOVA, and the χ2 test. The correlation between two variables was plotted on a scatter plot, and Pearson product moment method or Spearman rank coefficient was used to express the relationship. A p value <0.05 was considered statistically significant. All statistical analyses used Statistica version 7.0.

The results of the various laboratory and diagnostic tests were compared for patients with tuberculous and non-tuberculous effusions, and the diagnostic value of the observations was assessed in terms of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic efficiency (DE). Sensitivity was defined as TP/(TP + FN) × 100, specificity as TN/(TN + FP) × 100, PPV as TP/(TP + FP) × 100, NPV as TN/(TN + FN) × 100 and DE was defined as (TP + TN)/(TP + FP + TN + FN) × 100, where TP = true positive, TN = true negative, FP = false positive and FN = false negative.

A TB prediction model was developed using a ‘classification and regression tree’ (CART) analysis.20 Classification trees were developed after separate consideration of all variables. The range of each variable was divided into two groups so as to obtain the best separation between patients with tuberculous pericarditis and those with non-tuberculous pericarditis. The resulting subsets of cases were then partitioned independently in turn. This process was done recursively until a stopping condition was satisfied. Node deviance, which measures node heterogeneity, was set at 0.1 to stop the tree-growing process, and subsets <10 were not partitioned further. The sample was randomly divided into a training set and a test set. The prediction model was then derived from the training set, and used on the test set to determine sensitivity and specificity values. The optimum cut-off for the total diagnostic index (DI) by which to classify a patient as having tuberculous pericarditis was found by use of receiver operating characteristic (ROC) curves.21

Results

During the study period, 233 patients (101 females, 43%) presented with large pericardial effusions requiring pericardiocentesis. Tuberculous pericarditis accounted for 162 effusions (69.5%), malignant effusions 22 (9.4%), effusions associated with connective tissue diseases 12 (5.2%), septic pericarditis 5 (2.1%), and ‘other’ effusions 32 (13.7%). In total, 84 patients were HIV-positive, including 81 patients who had pericardial TB (50.0% of TB patients). The epidemiology of these effusions has been described previously.22 Of the 162 tuberculous patients, 11 had been on anti-tuberculous therapy for >48 h at the time of pericardial aspiration; all of these were HIV-negative.

Microbiological/histopathological diagnosis

Definite TB was diagnosed in 118 (73%) of the 162 patients classified as having tuberculous pericarditis. These included three patients with a positive pericardial fluid ZN smear, 91 with a positive pericardial effusion TB culture, 16 patients with a positive pericardial biopsy, 32 patients with a positive sputum ZN smear and/or culture, and 16 patients with a positive TB culture and/or histology in one or more extra-cardiac sites. Positive TB cultures were obtained from pleural fluid (n = 8), peritoneal fluid (n = 2), blood (n = 3), lymph node aspirate (n = 3) and skin biopsies (n = 2). Twenty-three patients had TB demonstrated in more than one site. Thirty-six pericardial biopsy specimens were evaluated histologically: 15 biopsies demonstrated granulomatous inflammation; 12 were accompanied by caseating necrosis. A biopsy specimen from an HIV-positive patient resembled acute purulent pericarditis; a diagnosis of pericardial TB was based on the presence of numerous acid-fast bacilli (AFB). Of the remaining 20 biopsies, 17 were non-diagnostic (non-specific features), and three were diagnostic for non-tuberculous disease.

Clinical and echocardiographic features

The mean±SD age of patients presenting with tuberculous pericarditis (35.8 ± 12.2 years) was significantly lower (p < 0.001) than that of patients presenting with non-tuberculous pericardial disease (44.7 ± 17.8 years). The clinical and echocardiographic features observed at the time of admission are summarized in Table 1. Significant differences (p < 0.05) between patients with tuberculous and non-tuberculous effusions were observed with regard to the presence of fever, night sweats, weight loss, cough, dyspnoea, and lymphadenopathy. Univariate analysis demonstrated an increased frequency of fever (p = 0.02), weight loss (p < 0.0001), and lymphadenopathy (p < 0.001) in the HIV-positive (vs. HIV-negative) TB group. The echocardiographic features of tuberculous vs. non-tuberculous pericarditis were different with respect to the presence of tamponade (p = 0.04), pericardial thickness ⩾5 mm (p = 0.02), and the presence of fibrinous strands in the pericardial space (p = 0.02).

View this table:
Table 1

Univariate analysis of clinical and echocardiographic features at admission

FeatureTB/HIV (n = 81)TB/HIV+ (n = 81)pTB (n = 162)Non-TB (n = 71)p
Fever65%85%0.0275%52%0.0006
Night sweats56%68%0.1662%30%<0.0001
Weight loss64%94%<0.000179%44%<0.0001
Cough87%93%0.3290%69%0.0002
Dyspnoea80%93%0.2586%73%0.03
Orthopnoea44%32%0.0838%41%0.71
Chest pain30%23%0.2227%51%0.001
Lymphadenopathy22%51%<0.00136%20%0.009
Pleural effusion42%34%0.2538%41%0.71
Tachycardia72%76%0.9174%54%0.003
Soft heart sounds75%74%0.9975%55%0.003
↑JVP ⩾4 cm80%76%0.1778%70%0.20
Pulsus paradoxus24%30%0.3427%16%0.15
Hypotension5%7%0.256%6%0.92
Hepatomegaly63%60%0.7862%45%0.02
Ankle oedema42%34%0.3238%47%0.24
Tamponade90%90%1.090%78%0.04
Pericardium >5 mm63%72%0.3467%46%0.02
Fibrin strands60%68%0.4465%46%0.02
  • JVP, elevated jugular venous pressure; HIV, human immunodeficiency virus negative; HIV+, human immunodeficiency virus positive; Non-TB, non-tuberculous.

Biochemistry and haematology results

Biochemistry and haematology results are presented in Table 2. Application of Light's criteria23 correctly classified all tuberculous pericardial aspirates as exudates. Patients with tuberculous pericarditis had significantly higher serum protein, serum globulin and pericardial protein levels than patients with non-tuberculous pericarditis (p < 0.001). Using a serum globulin concentration >40 g/l as a cut-off level resulted in an OR of 15.1 (p < 0.001), whereas a pericardial total protein level >39 g/l resulted in an OR of 3.42 (p = 0.005). Serum protein, serum globulin and pericardial fluid protein levels were also significantly higher in HIV-positive patients with pericardial TB than in HIV-negative tuberculous pericarditis patients (p < 0.01).

View this table:
Table 2

Comparison of biochemistry and haematology results

MeasurementTB/HIV (n = 64)TB/HIV+ (n = 78)pTB (n = 142)Non-TB (n = 61)p
S protein (g/l)72 (9.8)78 (8.6)<0.00175 (9.0)65 (11.4)<0.001
S globulin (g/l)43 (7.7)51 (6.4)<0.00147 (7.1)35 (8.2)<0.001
S albumin (g/l)30 (2.1)27 (2.2)0.4628 (2.2)30 (2.4)0.09
S albumin/S globulin0.67 (0.25)0.48 (0.21)<0.0010.56 (0.23)0.83 (0.29)<0.001
Pc protein (g/l)52.1 (11.9)60.2 (14.9)0.00156.2 (8.0)47.2 (17.9)<0.001
Pc protein/S protein0.72 (0.16)0.79 (0.15)0.080.76 (0.15)0.73 (0.18)0.27
Pc LDH/S LDH3.86 (5.06)2.40 (1.62)0.013.2 (3.31)4.0 (6.2)0.26
CRP (mg/l)109 (84.9)121 (79.0)0.46115 (82)90 (85.7)0.04
PB leukocytes (× 109/l)7.9 (3.37)6.7 (4.15)0.017.33 (3.76)12.6 (4.8)<0.001
PB neutrophils (× 109/l)5.68 (1.34)4.92 (0.91)0.015.38 (1.31)8.87 (1.56)<0.001
PB lymphocytes (× 109/l)1.33 (0.61)0.93 (0.53)0.031.13 (0.57)2.01 (2.23)<0.001
PB monocytes (× 109/l)0.55 (0.19)0.24 (0.12)0.010.40 (0.16)0.69 (0.25)0.001
Pc leukocytes (× 109/l)2.69 (2.33)1.85 (1.52)0.032.27 (1.92)4.62 (3.81)0.001
Pc% neutrophils28.4 (22.5)35.9 (25.5)0.0532.2 (24.0)55.8 (24.6)0.01
Pc% lymphocytes52.1 (25.5)39.4 (22.5)0.0345.5 (25.2)25.6 (22.4)0.01
  • Data are means (SD). HIV, human immunodeficiency virus negative; HIV+, human immunodeficiency virus positive; Non-TB, non-tuberculous; S, serum; PB, peripheral blood; Pc, pericardial fluid; LDH, lactate dehydrogenase; CRP, C-reactive protein.

Peripheral blood total leukocyte counts, neutrophil and lymphocyte numbers were significantly lower in patients diagnosed with TB, compared to patients presenting with non-tuberculous pericarditis (p < 0.001). Significant differences were also demonstrated between HIV-positive and HIV-negative pericardial TB patients for each of these variables (p < 0.05). Using a peripheral blood total leukocyte count <10 × 109 cells/l as a cut-off resulted in an OR of 12.6 (p < 0.01). Pericardial fluid total leukocyte counts and pericardial fluid neutrophil counts were significantly higher (p < 0.05) in non-tuberculous effusions than in tuberculous exudates, which in turn were characterized by significantly higher lymphocyte counts than in non-tuberculous cases. The best results were obtained using a pericardial lymphocyte/neutrophil (pc-L/N) ratio of ⩾1.0. This resulted in sensitivity, specificity, PPV, NPV and DE of 73%, 79%, 86%, 61%, and 75%, respectively.

Clinical prediction model for TB diagnosis

The sample set used for deriving the prediction model consisted of 164 patients, who had valid test results for all the variables as well as a definite diagnosis of either tuberculous pericarditis (n = 110) or definitely non-tuberculous aetiology (n = 54). Univariate analysis of the results of the variables used for the clinical prediction model did not, however, find any significant difference between the patients with definite tuberculous pericarditis (included) and those diagnosed with probable tuberculous pericarditis (not included). Five admission variables were identified that were independently predictive for tuberculous pericarditis, and a diagnostic index (DI) was calculated for each (Table 3).

View this table:
Table 3

Odds ratios and weighted diagnostic index for admission variables

Admission variableOdds ratioWeightDiagnostic index
Weight loss6.150.131
Night sweats4.160.091
Fever7.710.172
Serum globulin >40 g/l15.090.333
Leukocyte count <10 × 109/l12.760.283

The first three variables (weight loss, night sweats, fever) are yes/no variables, whereas the remaining two (serum globulin and peripheral blood leukocyte count) are measurements for which optimal threshold values were derived using CART analysis. The OR for TB was calculated for each variable, and used to calculate its weight and DI. Weight was defined as the OR for that variable divided by the sum of all ORs; DI as weight × 10.

Total DI was calculated for each patient according to the formula: DI (weight loss) + DI (night sweats)+DI (fever) + DI (serum globulin) + DI (peripheral blood leukocyte count). The optimum cut-off for the total DI was found using a ROC curve.

The prediction model was derived from the training set (n = 101) and predictions were made for the test set (n = 63) to determine sensitivity and specificity. Cases were randomly assigned to either the training set or the test set. The best DE for diagnosing TB corresponded to a total DI of 6. Results for the prediction model applied to the training data demonstrated 86% sensitivity and 87% specificity for the diagnosis of TB, whereas application to the test set resulted in 82% sensitivity and 76% specificity. Our suggested diagnostic rule was therefore that a patient with a total DI score ⩾6 has tuberculous pericarditis, and a patient with total DI <6 has non-tuberculous pericarditis. Applying this diagnostic rule to all those patients with valid results for all variables (n = 203), including 142 patients with tuberculous and 61 patients with non-tuberculous pericarditis, resulted in a sensitivity of 86% and a specificity of 84% for the diagnosis of pericardial TB.

Pericardial fluid ADA activity

Mean ± SD ADA activities for patients with tuberculous (HIV-negative), tuberculous (HIV-positive), malignancy, septic pericarditis, connective tissue disease and other non-tuberculous pericardial effusions are presented in Table 4. Various levels of pericardial fluid ADA activity were evaluated as cut-off levels, and based on ROC curves,21 the best results were obtained at 40 U/l.

View this table:
Table 4

Pericardial adenosine deaminase (ADA) activity in various diagnostic groups of pericarditis

Diagnostic groupsnADA (U/l)p
MeanSD
Pericardial effusions (overall)21265.848.1
Tuberculous pericarditis HIV7555.1
Tuberculous pericarditis HIV+7638.11.00
Malignant pericardial effusion2036.10.007
Uraemic pericarditis922.40.004
Septic pericarditis438.61.00
Connective tissue disease829.50.06
Other pericardial effusions2018.40.0002

In total, 13/71 patients with non-tuberculous effusions had ADA activity >40 U/l. These included patients with septic pericarditis (n = 4), one case of systemic lupus erythematosus, rheumatoid arthritis (n = 1), post-traumatic pericarditis (n = 1), pericarditis of unknown origin (n = 1, corresponding ADA 50.4 U/l), non-haematological malignancies (n = 3); and haematological malignancies (n = 2). The malignancies included pericardial adenocarcinoma, squamous cell carcinoma, undifferentiated carcinoma, lymphoblastic T-cell lymphoma, and chronic myelomonocytic leukaemia.

The cut-off level of 40 U/l resulted in 22 false-negative pericardial effusions: 12 in HIV-negative and 10 in HIV-positive patients. Of the 12 HIV-negative patients, 9 were on anti-tuberculous therapy at the time of pericardiocentesis, whereas no HIV-positive patient was on active anti-tuberculous therapy at the time of pericardial aspiration. In three patients with tuberculous pericarditis, no specific cause could be identified for the low pericardial ADA activities (22 U/l, 24 U/l and 34 U/l). After excluding all patients on anti-tuberculous therapy and all those patients categorized as pericarditis of unknown cause, an ADA cut-off of 40 U/l yielded sensitivity, specificity, PPV, NPV and DE of 87%, 89%, 95%, 72% and 86%, respectively.

Pericardial fluid IFN-γ concentration

The mean ± SD IFN-γ concentrations for HIV-negative tuberculous, HIV-positive tuberculous and non-tuberculous effusions were 787 ± 115, 624 ± 103 and 27 ± 19 pg/ml, respectively. The difference in IFN-γ concentration between tuberculous and non-tuberculous effusions was highly significant (p < 0.0001), whereas no difference was seen between HIV-positive and HIV-negative tuberculous groups (p = 0.89). IFN-γ levels were detectable in only three of the non-tuberculous effusions, including a case of staphylococcal sepsis (28.9 pg/ml), a case of metastatic adenocarcinoma (42.9 pg/ml), and a case of diffuse large-cell lymphoma (39.4 pg/ml).

Various levels of pericardial fluid IFN-γ concentration were evaluated as a cut-off level, and based on ROC curves, the best results were obtained at a cut-off of 50 pg/ml. This resulted in 92% sensitivity, 100% specificity and a PPV of 100% for pericardial TB.

Polymerase chain reaction for Mycobacterium tuberculosis

PCR for M. tuberculosis was performed in pericardial fluid samples of 48 consecutive patients, including 33 patients with tuberculous pericarditis and 15 with non-tuberculous effusions. Positive PCR results were obtained for four ‘definite’ tuberculous effusions and for six ‘probable’ tuberculous effusions, but for none of the non-tuberculous effusions. This resulted in sensitivity, specificity, PPV, NPV and DE of 30%, 100%, 100%, 31%, and 52%, respectively.

Cytopathology

Cytopathology results were available for 192 patients, including 136 with pericardial TB (not currently receiving anti-tuberculous therapy) and 56 patients with non-tuberculous pericarditis. Each pericardial fluid specimen was classified into one specific cytopathological category on the basis of cytodiagnostic criteria15–17, 24, 25 (Table 5).

View this table:
Table 5

Cytopathological classifications for various diagnostic groups

DiagnosisnBenignPurulentNon-purulentTuberculousMalignant
Tuberculous1367 (5%)10 (8%)108 (79%)11 (8%)0
HIV-negative613 (5%)4 (7%)48 (79%)6 (10%)0
HIV-positive754 (5%)6 (8%)60 (80%)5 (7%)0
Non-tuberculous5611 (20%)6 (11%)25 (45%)2 (4%)12 (21%)
Malignant191 (5%)05 (26%)1 (5%)12 (63%)
CNTD93 (33%)1 (11%)4 (44%)1 (11%)0
Uraemia82 (25%)4 (100%)6 (75%)00
Septic41 (14%)1 (25%)6 (86%)00
Traumatic72 (40%)03 (60%)00
Idiopathic52 (50%)01 (25%)00
Other400000
  • CNTD, connective tissue disease.

Three of the 192 effusions were ZN-positive, including one purulent and two non-purulent inflammatory effusions. All three were from HIV-positive individuals, and results were verified by positive TB culture. None of these effusions fulfilled the cytodiagnostic criteria for the ‘tuberculous’ category. Non-purulent inflammatory effusions were demonstrated in 79% of TB patients. Pc lymphocyte/neutrophil (L/N) ratios >3 : 1 were demonstrated in 53% of HIV-negative pericardial effusions, and in 20% of HIV-positive tuberculous effusions. In addition, significant numbers of mesothelial cells were present in the majority of pericardial aspirates obtained from pericardial TB patients. Pericardial aspirates were classified as tuberculous in 13 patients (Figure 1a), including 11 with a diagnosis of pericardial TB, one case of SLE, and one case of histologically-confirmed lymphoblastic T-cell lymphoma. Cytopathological review of the lymphoma specimen revealed features suggestive of a haematological malignancy (Figure 1b).

Figure 1.

Cytopathology smears of patients with large pericardial effusions.

Thirty-six pericardial biopsy samples were available for comparative analysis between histopathological and cytopathological diagnosis. Fifteen of these biopsies demonstrated granulomatous inflammation and were classified as non-purulent inflammatory (n = 13) and tuberculous effusions (n = 2). Two cases of serofibrinous pericarditis and one case of pericardial lymphoma were classified as tuberculous effusions. Using the cytodiagnostic category ‘tuberculous effusion’ as a measure of diagnosing pericardial TB resulted in 11% sensitivity and 85% specificity.

Tuberculin skin testing

Tuberculin skin testing (TST) was performed in 52 consecutive patients, including 36 with tuberculous pericarditis (12 HIV-positive) and 16 patients with non-tuberculous pericarditis. The aetiological causes for the non-tuberculous group included malignancy (n = 6), sepsis (n = 2), uraemia (n = 2), post-trauma (n = 2), rheumatoid arthritis (n = 1), scleroderma (n = 1) and systemic lupus erythematosus (n = 1). Different cut-off levels for the diameter of skin induration were tested and the best diagnostic efficiency for diagnosing TB (including HIV-positive patients) was obtained at a diameter ⩾10 mm. At this level, tuberculin skin tests were positive in 32/36 tuberculous pericarditis patients (89% ‘true positive‘) and 7/16 patients with non-tuberculous effusions (44% ‘false positive‘), yielding sensitivity, specificity, PPV, NPV and DE of 89%, 56%, 82%, 69%, and 79%, respectively. At a cut-off of diameter ⩾15 mm, the sensitivity, specificity, PPV, NPV and DE were 43%, 93%, 93%, 38%, and 57%.

Diagnostic classification tree

Based on their sensitivity and specificity for the diagnosis of TB, we identified four parameters (ADA level, HIV serology, peripheral blood leukocyte counts and pericardial fluid differential leukocyte counts) that could be used in a classification tree for the diagnosis of pericardial TB (Figure 2). Application to the patient data set resulted in 96% sensitivity and 97% specificity for the diagnosis of pericardial TB. A classification tree using IFN-γ levels instead of ADA activity resulted in 98% sensitivity and 100% specificity (Figure 3).

Figure 2.

Classification tree developed for the diagnosis of pericardial TB. Pc-ADA, pericardial adenosine deaminase; PB-WCC, peripheral blood white cell count; Pc L/N ratio, pericardial lymphocyte/neutrophil ratio.

Figure 3.

Classification tree based on the determination of IFN-gamma levels for the diagnosis of pericardial TB. Pc-IFN, pericardial interferon-gamma; PB-WCC, peripheral blood white-cell count; Pc L/N ratio, pericardial lymphocyte/neutrophil ratio.

The utility of the various tests for the diagnosis of TB, including pericardial effusion culture, pericardial histology, PCR, IFN-γ, TST and ADA activity, is summarized in Table 6.

View this table:
Table 6

Utility of various diagnostic tests

nSensitivity(%)Specificity (%)PPV (%)NPV (%)DE (%)
Pericardial effusion culture positive203521001004856
Histopathology (granulomatous and/or ZN+)36641001005575
Pericardial ADA activity ⩾40 U/l2128789957286
Pericardial lymphocyte/neutrophil ratio ⩾11927379866175
IFN-γ ⩾50 pg/ml118921001008594
Tuberculin skin test ⩾10 mm diameter of skin induration528956826979
Diagnostic rule using total DI score ⩾62038684887885
PCR for M. tuberculosis positive48301001003152
Diagnostic classification tree (using ADA >40 U/l)2129697989296
Diagnostic classification tree (using IFN-γ >50 pg/ml)118981001009699
  • IFN, interferon; PPV, positive predictive value; NPV, negative predictive value; DE, diagnostic efficiency; DI, diagnostic index.

Discussion

The accurate diagnosis of tuberculous pericarditis is important, because without specific treatment, the mean survival rate is 3.7 months, with a mortality rate approaching 85% at 6 months.26 Our study confirms the insensitivity of traditional diagnostic tools for diagnosing pericardial TB. Pericardial biopsy specimens taken from the tuberculous group demonstrated caseating granulomatous inflammation in about 50% of samples, and non-specific histopathological findings were common, even when M. tuberculosis was cultured in the pericardial fluid. Although mycobacterial culture was more sensitive than fluid smear, it yielded positive results in only 56% of patients with pericardial TB. Our microbiological diagnoses were significantly augmented by culturing M. tuberculosis from extracardiac sites, including sputum (n = 32), pleural fluid (n = 8) and lymph node aspirate (n = 3). Pleural TB and peripheral tuberculous lymphadenitis are the two most common forms of extra-pulmonary TB,27 and both are commonly seen in patients with pericardial TB. In one series, 50% of patients with tuberculous pericarditis had necropsy evidence of tuberculous pleuritis,28 and peripheral lymphadenopathy affecting the cervical glands has been reported in 13%–28% of patients with tuberculous pericarditis.3, 26 Persistent generalized lymphadenopathy is a common manifestation of HIV-positive patients,29, 30 but when the enlarged lymph nodes are >1 cm in diameter and their distribution is asymmetric, disseminated TB should be considered. Tuberculous lymphadenopathy is characterized by caseation and ‘matting’ on palpation, and the adenopathy clears or regresses on specific therapy.31 The diagnostic work-up of patients with suspected pericardial TB should thus include sputum smears in patients with a productive cough and lymph node aspiration. However, culture results take a long time, and rarely influence decision-making.

In 2003, the WHO reported 8.8 million new cases of TB and an estimated 1.7 million TB deaths, including 229 000 co-infected with HIV.32 The majority of TB cases came from South East Asia, but most cases of HIV and TB co-infection were reported in sub-Saharan Africa.32 These regions are characterized by a serious lack of financial resources and almost complete absence of diagnostic infrastructure.6, 8, 9 Our study demonstrates the potential usefulness of a basic diagnostic rule to assist clinical decision-making, that could be applied in settings with poor resources. In our study, the features of pericardial effusion and cardiac compression (jugular distension, soft heart sounds, pulsus paradoxus, Kussmauls's sign, ankle oedema, hypotension) were similarly distributed among tuberculous patients and non-tuberculous patients; this concurs with previous reports in the literature.5, 26, 28, 33–37

Univariate analysis of the admission variables suggested a set of potentially discriminative clinical features, including cough, fever, night sweats, weight loss and lymphadenopathy. In addition, patients with tuberculous pericarditis have higher serum globulin levels than non-tuberculous patients and will usually not present with peripheral blood leukocytosis. Multivariate logistical regression and CART analyses identified five features that were independently predictive of tuberculous (vs. non-tubercuous) pericarditis. These were: fever, night sweats, weight loss, globulin level and peripheral leukocyte count (Table 3). Based on the odds ratios for each of these variables, we developed a weighted score or diagnostic index that when added together would amount to a potential maximum score of 10 (Table 3). Our suggested diagnostic rule was a total score ⩾6 indicated tuberculous pericarditis, and <6, non-tuberculous pericarditis. Applying this rule to all 203 patients with valid results for these five variables resulted in 86% sensitivity, 85% specificity and 85% diagnostic efficiency for the diagnosis of pericardial TB, which in spite of its simplicity, is better than the diagnostic efficiency for culture or pericardial histology observed in this study and published elsewhere.1, 5, 32 Major advantages of this proposed diagnostic rule include non-invasiveness, availability, cost-effectiveness and rapidity of the laboratory tests. However, all five variables used for the rule were significantly influenced by underlying HIV infection. In a setting with a substantially different TB and HIV prevalence than that of the Western Cape, these predictors should not be used without prospective evaluation in that setting.

Pericardiocentesis is not always feasible, and non-invasive tests that are indicative of tuberculous aetiology are thus highly desirable. We found chest radiography to be useful in identifying patients with large pericardial effusions.33 Features compatible with pulmonary TB were noted in 30% of tuberculous pericarditis cases, and mediastinal lymphadenopathy was suggestive of HIV and TB co-infection.33 Echocardiography is the definitive investigation for pericardial effusion and tamponade, and is particularly valuable for distinguishing an effusion from a subacute constriction.34, 35 We found the echocardiographic detection of fibrinous strands and the presence of pericardial thickening to be suggestive of tuberculous aetiology. Liu et al reported these features as being diagnostic of pericardial TB,36 but in our study, pericardial thickening and the presence of fibrinous strands was also seen in septic pericarditis (n = 1), rheumatoid arthritis (n = 1), pericardial malignancy (n = 2), and post-traumatic pericardial effusions (n = 2). Cherian et al. demonstrated the usefulness of computerized tomography of the chest (chest CT) for detecting mediastinal lymphadenopathy in virtually 100% of patients with pericardial TB, and enlarged nodes disappeared or regressed with specific anti-tuberculous therapy.3, 31 Unfortunately, in sub-Saharan Africa, CT is not widely available outside academic hospitals and the private health care sector.

Widely available, but less specific than CT tomography, is the tuberculin skin test (TST), where a T-cell-mediated immune response to an intradermally injected dose of purified protein derivative (PPD) is quantified by measuring the diameter of skin induration 48–72 h after injection. Our study confirmed its lack of specificity for the diagnosis of active TB. At a cut-off level of 10 mm, we observed 89% sensitivity and 56% specificity for correct diagnostic classification. The high proportion of ‘false positive’ tests (44% of patients with non-tuberculous effusions) is similar to previously reported rates of 30–40%.5 A strongly positive TST increases the suspicion of pericardial TB, a negative result (11% of TB patients in our study) does not exclude tuberculous disease; the response to tuberculin may be affected by HIV infection, malnutrition and time of presentation.7, 28, 33, 43

The strength of association between HIV infection and tuberculous pericarditis cannot be overemphasized. Pericardial effusion is one of the early presenting features of HIV infection in sub-Saharan Africa,6, 44, 45 and the disease has been ascribed to TB in as many as 100% of cases.6, 46 In our series, TB caused 94% of cases and only 3/84 HIV-positive patients were diagnosed with non-tuberculous aetiology, including two cases of septic pericarditis. Univariate analysis demonstrated significantly higher frequency of lymphadenopathy, weight loss and oral candidiasis in HIV-positive TB patients compared with the HIV-negative group. The differential diagnosis of pericarditis in HIV-positive individuals includes other opportunistic infections (bacterial, fungal, protozoal or viral), Kaposi's sarcoma, lymphoma, uraemia, and idiopathic pericarditis.47–49 In our experience, septic pericarditis was the most likely non-tuberculous cause and it was characterized by peripheral blood leukocyte counts >10 × 109/l, peripheral blood neutrophil counts >7 × 109/l, pericardial leukocyte count >2 × 109/l, and a pericardial fluid lymphocyte/neutrophil ratio <0.5.

In some patients, invasive tests are required to establish the aetiological diagnosis. In our study, ADA activity was significantly elevated in tuberculous pericardial effusions compared to non-tuberculous pericardial effusions. Pericardial ADA levels were not affected by HIV infection, and contrary to a previous report,14 the diagnostic utility of ADA activity was not diminished by underlying HIV. The best diagnostic results were obtained at a cut-off of 40 U/l, which yielded sensitivity, specificity, PPV, NPV, and DE of 90%, 74%, 90%, 76% and 86%, respectively. In HIV-negative TB patients, the most notable cause for low pericardial ADA levels was the concomitant use of anti-tuberculous chemotherapy at the time of pericardiocentesis.50 The use of pericardial ADA levels in a patient with suggestive clinical features provides a rapid and accurate means of diagnosing tuberculous pericarditis, especially in high-prevalence areas. If a high ADA level is found, septic pericarditis and haematological neoplastic diseases need to be excluded by differential white blood cell, direct microscopy, Gram stain, bacterial culture, and cytological analysis, all of which should be performed routinely on patients with large pericardial effusions. If these tests are negative, the diagnosis will in all likelihood be TB, especially in countries where this infection is endemic or in those individuals co-infected with HIV. If a low level of ADA is found, non-tuberculous causes are likely, and further tests such as cytology or pericardial biopsy are indicated to determine the cause.

The cytopathological analysis of tuberculous pericardial exudates demonstrated two distinct differences between pericardial and pleural tuberculous exudates. First, mesothelial cells were present in the majority of tuberculous pericardial aspirates, and second, almost 30% of tuberculous pericardial exudates displayed neutrophilic dominance, whereas tuberculous pleural exudates are characterized by prominent lymphocytosis and virtually complete absence of mesothelial cells.15–17 Consequently, cytodiagnostic criteria that had been developed for the diagnosis of pleural TB resulted in a low diagnostic yield for pericardial TB. The finding of pericardial lymphocytosis was nevertheless 73% sensitive and 79% specific for a diagnosis of pericardial TB.

The literature is vague about pericardial lymphocytosis; it is mentioned, but to the best of our knowledge, has not been quantified.3, 51, 52 Our study illustrates the useful role of pericardial cytology for the diagnosis of malignant effusions53–55 and as a screening tool for septic pericarditis. Finding a purulent exudate does not, however, exclude TB, and has also been described in SLE, rheumatoid arthritis and various fungal infections.17, 56 Further investigations such as Gram stain, bacterial culture and TB culture are necessary.

If possible, determination of IFN-γ levels would be the ideal investigation. Significantly elevated levels of IFN-γ were demonstrated in tuberculous pericarditis compared to non-tuberculous pericardial effusions and, importantly, concentrations never exceeded 50 pg/ml in any of the non-tuberculous effusions. A cut-off of 50 pg/ml resulted in 92% sensitivity, 100% specificity and a diagnostic efficiency of 95%. Diagnostic efficiency was not influenced by HIV infection. The underlying immunological principle has resulted in the development of the highly sensitive and accurate enzyme-linked immunospot (ELISPOT) test that detects IFN-γ -producing T-cells specific for M. tuberculosis antigen.57, 58 Using peripheral blood, the test is very useful for detecting tuberculous infection, but does not differentiate between latent infection and active disease, which is a major problem in TB-endemic areas. It is unclear what role the ELISPOT test could play when used on pericardial aspirates. The diagnostic utility of currently available IFN-γ assays is unfortunately limited by technical and financial constraints.

The use of PCR for the detection of M. tuberculosis provides a test with high specificity for the diagnosis of pericardial TB; however, sensitivity was low at 32%. This surprisingly poor sensitivity has also been reported in tuberculous pleural effusion59 and pericardial effusion studies,10 where sensitivities have ranged between 15–20%, and specificities between 96–100%. Explanations for the poor sensitivity of PCR in tuberculous exudates include poor specimen preparation, the presence of inhibitors such as fibrin and haemoglobin, as well as low numbers of tubercle bacilli or their DNA in the specimens.60, 61 Sensitivity is much better when pericardial tissue is used for analysis by PCR.10, 62 Recently, concerns have been raised about false-positive results with PCR.11 Overall, PCR has not yet provided a sensitive, specific and cost-effective test for the diagnosis of TB.

By combining the predictive attributes of four widely available laboratory tests into a diagnostic classification tree (Figure 2), we achieved our aim of developing a tool that is sensitive, specific, rapid and relatively inexpensive. HIV serology, peripheral blood leukocyte counts and pericardial differential leukocyte counts do not require sophisticated technology, and are usually routinely performed in rural hospitals. ADA is an enzyme that is stable for at least 24 h at 25°C, 7 days at 4°C and 3 months at −20°C.63, 64 It can be determined within hours by a simple hand method requiring only a spectrophotometer.13 It is thus possible to perform this test in basic rural laboratories.

Application of the entire patient data set to the classification tree resulted in a sensitivity of 96%, a specificity of 97% and a diagnostic efficiency of 96%. The diagnostic utility was not influenced by inclusion or exclusion of patients with ‘probable TB’. In settings where it is not possible to measure ADA activity and where pericardiocentesis cannot be done, we suggest the use of the diagnostic rule, with total score ⩾6 or more indicating pericardial TB. The likelihood of TB increases with an increase in the total diagnostic index. Echocardiographic detection of fibrinous strands and the presence of pericardial thickening would strongly support a diagnosis of tuberculous pericarditis. Where available, chest CT may play an important role as a non-invasive diagnostic test.3, 31 However, the predictive value of clinical features and diagnostic tests for TB is influenced by the prevalences of both HIV and TB in the population. Our diagnostic tree and acoompanying rule should not be used in areas of substantially different TB and HIV prevalence to those of Southern Africa, until prospectively assessed for diagnostic accuracy.

Where available, IFN-γ is the diagnostic tool of choice. and should be used to aid in the rapid diagnosis of pericardial TB. It can be used effectively in combination with ADA and pericardial cytology to confirm or exclude contentious cases. A classification tree using IFN-γ levels >50 pg/ml instead of ADA activity resulted in 98% sensitivity and 100% specificity (Figure 3). For resource-poor areas, however, the diagnostic utility of currently available IFN-γ assays is limited by technical and financial constraints.

References

View Abstract