Abstract
Background
Serum cytokines correlate with tuberculosis (TB) progression and are predictors of TB recurrence in people living with HIV. We investigated whether serum cytokine biosignatures could diagnose TB among HIV-positive inpatients.
Methods
We recruited HIV-positive inpatients with symptoms of TB and measured serum levels of inflammation biomarkers including IL-2, IL-4, IL-6, IL-10, tumour necrosis factor-alpha (TNF-α) and interferon-gamma (IFN-γ). We then built and tested our TB prediction model.
Results
236 HIV-positive inpatients were enrolled in the first cohort and all the inflammation biomarkers were significantly higher in participants with microbiologically confirmed TB than those without TB. A binary support vector machine (SVM) model was built, incorporating the data of four biomarkers (IL-6, IL-10, TNF-α and IFN-γ). Efficacy of the SVM model was assessed in training (n=189) and validation (n=47) sets with area under the curve (AUC) of 0.92 (95% CI 0.88 to 0.96) and 0.85 (95% CI 0.72 to 0.97), respectively. In an independent test set (n=110), the SVM model yielded an AUC of 0.85 (95% CI 0.76 to 0.94) with 78% (95% CI 68% to 87%) specificity and 85% (95% CI 66% to 96%) sensitivity. Moreover, the SVM model outperformed interferon-gamma release assay (IGRA) among advanced HIV-positive inpatients irrespective of CD4+ T-cell counts, which may be an alternative approach for identifying Mycobacterium tuberculosis infection among HIV-positive inpatients with negative IGRA.
Conclusions
The four-cytokine biosignature model successfully identified TB among HIV-positive inpatients. This diagnostic model may be an alternative approach to diagnose TB in advanced HIV-positive inpatients with low CD4+ T-cell counts.
Keywords: Tuberculosis, Bacterial Infection
WHAT IS ALREADY KNOWN ON THIS TOPIC
Diagnosing tuberculosis (TB) among HIV-positive inpatients remains challenging, mainly as they often present non-specifical clinical manifestations and are unable to provide sputum samples for microbiological testing.
WHAT THIS STUDY ADDS
We built a binary TB predictive model, support vector machine (SVM), based on the combination of serum cytokines including IL-6, IL-10, tumour necrosis factor-alpha and interferon-gamma, which can identify TB among HIV-positive inpatients with symptoms of TB.
HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY
Our SVM model outperformed interferon-gamma release assay among advanced HIV-positive inpatients with low CD4+ T-cell counts. This will be a rapid approach to screen and assist in diagnosis of TB among HIV-positive inpatients using non-sputum samples.
Introduction
Tuberculosis (TB), caused by Mycobacterium tuberculosis (Mtb), is the common opportunistic infection and leading cause of hospital admission in people living with HIV (PLHIV). TB accounts for an estimated 0.7 million newly diagnosed cases and 0.2 million deaths among PLHIV in 2021.1 HIV-positive inpatients are typically severely immune suppressed with various infectious diseases and have non-specific clinical symptoms.2 In addition, almost half of the hospitalised patients with HIV are unable to provide sputum samples for diagnostic testing.3 Therefore, how to diagnose TB accurately among HIV-positive inpatients remains challenging.
Although conventional diagnostic assays, such as sputum smear microscopy and culture, have been widely used for decades, they have limitations with low sensitivity and low efficiency to yield the results. Despite GeneXpert MTB/Rifampicin (RIF) and GeneXpert MTB/RIF Ultra having high sensitivity and specificity; they are not affordable in resource-limited countries.4 5 Another diagnostic test based on the detection of plasma or urine lipoarabinomannan (LAM) offers a fast and low-cost method to identify TB in PLHIV but its low sensitivity in PLHIV with high CD4+ T-cell counts limits its widespread use.6 7 Host biomarkers have been extensively explored as an alternative solution for diagnosis of TB. The current gold standard for detecting Mtb infection is Mtb-specific interferon-gamma release, as measured by interferon-gamma release assay (IGRA). However, IGRA cannot distinguish TB from latent TB infection (LTBI) and has diminished diagnostic accuracy in advanced HIV-positive patients with decreased T-cell functions.8 9
HIV and TB coinfection have been associated with immune activation and systemic inflammation. T-cell activation and elevated levels of plasma interleukin 6 (IL-6) and sCD14 were observed in PLHIV with TB, which were correlated with disease progression.10 Additionally, in PLHIV receiving antiretroviral therapy (ART), inflammation markers including IL-6, C-X-C motif chemokine ligand 9 (MIG) and C-X-C motif chemokine ligand 10 (IP10) predicted the risk of TB recurrence.11 Recent studies have also suggested that plasma or urine cytokines are differentially expressed in PLHIV with and without TB,12–14 indicating that cytokine biosignatures may be useful for the diagnosis of TB in PLHIV.
To explore this potential, our study measured serum levels of six cytokines, including IL-2, IL-4, IL-6, IL-10, tumour necrosis factor-alpha (TNF-α) and interferon-gamma (IFN-γ), in HIV-positive inpatients with and without TB. A combination of multiple markers to form specific biosignatures has substantially improved the diagnostic accuracy of the individual marker.15 Therefore, we then developed a prediction model that incorporated four of these cytokines, and evaluated its diagnostic accuracy in both training and validation sets of patients. Finally, we tested the model’s performance in a separate clinical cohort. Our findings suggest that the serum cytokine biosignatures could be a valuable tool for identifying TB among HIV-positive inpatients.
Methods
Study cohort
The HIV-positive inpatients were enrolled at Guangzhou Eighth People’s Hospital from January 2019 to October 2022. All participants were admitted to the hospital for the signs or symptoms of TB irrespective of ART. Signs and symptoms of TB were defined as abnormal chest X-ray results, fever, cough, weight loss, haemoptysis or signs of extrapulmonary TB. The participants underwent clinical assessment that including HIV serology, CD4+ T-cell count, chest X-ray or CT, IGRA, sputum or lymph node tissue microbiological examinations by microscopy test, culture, GeneXpert MTB/RIF or Mtb-DNA test. Two cohorts of participants were recruited in the study. The first cohort consisted of 236 HIV-positive inpatients, including 91 with definite TB and 145 without TB. A total of 189 (80%) participants in the first cohort were randomly split as the training set, while the other 47 (20%) participants served as the validation set. In the second independent cohort, 126 participants were prospectively recruited. After exclusion of 1 with probable TB and 15 with incomplete data, the remaining 110 individuals were subjected for further analysis (table 1, figure 1).
Table 1.
Demographic characteristic of the HIV-positive inpatients
| First cohort (n=236) | Second cohort (n=110) | |||
| Parameter | Without TB | With TB | Without TB | With TB |
| No | 145 | 91 | 83 | 27 |
| Sex | ||||
| Male | 119 (82%) | 78 (86%) | 69 (83%) | 22 (81%) |
| Female | 26 (18%) | 13 (14%) | 14 (17%) | 5 (19%) |
| Age | 48±14 | 46±13 | 47±15 | 47±14 |
| CD4+ T cell counts | ||||
| >350 | 28 (19%) | 1 (1%) | 15 (18%) | 1 (4%) |
| 350–200 | 24 (17%) | 8 (9%) | 20 (24%) | 1 (4%) |
| <200 | 78 (54%) | 76 (84%) | 47 (57%) | 23 (85%) |
| Not tested | 15 (10%) | 6 (6%) | 1 (1%) | 2 (7%) |
| IGRA | ||||
| Positive | 0 (0) | 48 (53%) | 0 (0) | 21 (78%) |
| Negative | 65 (45%) | 35 (38%) | 35 (42%) | 6 (22%) |
| Not tested | 80 (55%) | 8 (9%) | 48 (58%) | 0 (0) |
| AFB | ||||
| Positive | 0 (0) | 34 (37%) | 0 (0) | 3 (11%) |
| Negative | 97 (67%) | 40 (44%) | 62 (75%) | 21 (78%) |
| Not tested | 48 (33%) | 17 (19%) | 21 (25%) | 3 (11%) |
| Culture | ||||
| Positive | 0 (0) | 79 (87%) | 0 (0) | 14 (52%) |
| Negative | 123 (85%) | 7 (8%) | 70 (84%) | 12 (44%) |
| Not tested | 22 (15%) | 5 (5%) | 13 (16%) | 1 (4%) |
| TB-DNA | ||||
| Positive | 0 (0) | 37 (41%) | 0 (0) | 12 (44%) |
| Negative | 74 (51%) | 24 (26%) | 43 (52%) | 12 (44%) |
| Not tested | 71 (49%) | 30 (33%) | 40 (48%) | 3 (12%) |
| GeneXpert | ||||
| Positive | 0 (0) | 51 (56%) | 0 (0) | 14 (52%) |
| Negative | 74 (51%) | 20 (22%) | 43 (52%) | 12 (44%) |
| Not tested | 71 (49%) | 20 (22%) | 40 (48%) | 1 (4%) |
| Extrapulmonary TB diagnosis | NA | 18 (20%) | NA | 3 (11.1%) |
Data in parentheses represent percentage (%). All participants were hospitalised patients with signs or symptoms, including abnormal chest X-ray results, fever, cough, weight loss or haemoptysis.
AFB, acid-fast bacilli; IGRA, interferon-gamma release assay; NA, not applicable; TB, tuberculosis.
Figure 1.
Study design and numbers of participants included in the cohorts. In the first cohort, all HIV-positive inpatients with tuberculosis (TB) were classified with definite TB. In the second cohort, probable TB or participants with incomplete data were excluded from the analysis. IGRA, interferon-gamma release assay; LR, logistic regression; MLP, multilayer perceptron; ROC, receiver operation curve; SVM, support vector machine.
Case definitions
Participants were diagnosed by two experienced specialists based on a combination of clinical and laboratory findings. Definite TB was microbiologically confirmed by any culture, GeneXpert MTB/RIF or Mtb-DNA test from any samples positive for Mtb during admission. Non-TB patients were defined as all microbiological examinations, IGRA and chest radiography negative for TB diagnosis. Probable TB was defined as only one positive microscopy test from any samples for Mtb, or no sputum but positive IGRA, chest radiography evidence and symptoms responding to TB treatment.
Laboratory procedures
Sputum and lymph node tissue samples were collected from participants and detected using microbiological examinations, as previously described.16
For IGRA, ELISA-based QFT-GIT (TB102) tests were performed in fully accredited diagnostic laboratory according to the manufacturer’s instructions. Whole blood samples (5 mL) were collected using lithium heparin tubes (BD), and then 1 mL of blood was transferred into each tube of the kits including a nil control tube, a positive control tube (phytohaemagglutinin), and a TB antigen tube (containing ESAT-6 protein EsxA (ESAT-6) and ESAT-6-like protein EsxB (CFP10)). Each tube was incubated for 16 to 24 hours at 37°C, followed by centrifuged for 10 min at 2000 g. IFN-γ levels in the supernatants were measured by ELISA.
For cytokine detection, whole blood samples (5 mL) were collected using vacuum blood collection tubes (BD Vacutainer). Serum was separated by centrifugation at 500 g for 5 min within 2 hours of collection. A cytometric bead array (CBA) assay kit (Cellgene Biotech, China) was used to detect the concentration of serum cytokines, according to the manufacturer’s instructions. Data collection was performed on BD Canto II flow cytometer (BD Biosciences), and data process and analysis were carried out using the FCAP Array Software (BD Biosciences). Standard dilutions were set to determine the concentration of the cytokines.
Model development and validation
Binary support vector machine (SVM), multilayer perceptron (MLP) and logistic regression (LR) predictive models were built to distinguish between HIV-positive inpatients with TB and without TB based on the four cytokines using the scikit-learn (V.1.0.2; https://www.scikit-learn.org/) library. To tune the best parameters for the models, fivefold cross-validation with a grid search method was conducted in the training set. The performance of the models was assessed by the area under the curve (AUC) method with the corresponding 95% CI from the receiver operation curve (ROC), as well as accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and followed by calibration using the Hosmer-Lemeshow goodness-of-fit test, as previous described.15 The performance of SVM model compared with IGRA was measured by Kolmogorov-Smirnov test and lift score using python libraries stats and mlxtend.
Statistical analysis
Continuous features were expressed as the means±SDs. Statistical differences were assessed using unpaired Student’s t-test after log2 transformation. ROC analysis was used to assess the performance of each cytokine on HIV-positive inpatients with TB and without TB using ‘pROC’ package. Gini importance of each cytokine was conducted using random forest algorithm. All statistical analyses were conducted in R (V.4.1.2; https://www.r-project.org/).
Results
Serum cytokines are differentially expressed in HIV-positive inpatients with and without TB
Two cohorts of HIV-positive inpatients with symptoms of TB were recruited in the study and their demographic information and clinical findings were collected. In the first cohort, we enrolled 91 HIV-positive patients with definite TB and 145 patients without TB, and then measured the levels of serum cytokines, including IL-2, IL-4, IL-6, IL-10, TNF-α and IFN-γ, in the patient’s sera. The levels of all cytokines we detected were significantly higher in HIV-positive inpatients with definite TB compared with non-TB patients. Following further analysis, we found that IFN-γ is a promising biomarker to distinguishing TB among HIV-positive patients, producing an AUC of 0.89 (95% CI 0.84 to 0.93) with 86% (95% CI 80% to 91%) specificity and 82% (95% CI 73% to 90%) sensitivity. In addition, the AUCs of IL-6, IL-10 and TNF-α were all over 0.75, indicating they can be individually used to diagnose TB among HIV-positive inpatients. However, IL-2 and IL-4 were unsuitable for TB classification (figure 2, online supplemental table 1).
Figure 2.
Cytokine levels in serum samples of HIV-positive inpatients in the first cohort and ROC plots for the diagnosis of TB among these patients. Representative plots showing IL-2 (A, G), IL-4 (B, H), IL-6 (C, I), IL-10 (D, J), TNF-α (E, K), and IFN-γ (F, L) levels (pg/mL) between HIV-positive inpatients with and without TB. Data are presented as means±SD, **p<0.01, ****p<0.0001, by unpaired Student’s t-test. AUC, area under the curve; IFN-γ, interferon-gamma; ROC, receiver operation curve; TB, tuberculosis; TNF-α, tumour necrosis factor-alpha.
thorax-2023-220782supp001.pdf (56.1KB, pdf)
An SVM model that combines four-cytokine biosignature accurately identifies TB among HIV-positive inpatients
To develop and validate a TB diagnostic model, we randomly split the first cohort into training and validation sets at a ratio of 4:1. The training set comprised 74 HIV-positive inpatients with definite TB and 115 patients without TB, while the validation set included 17 HIV-positive inpatients with definite TB and 30 patients without TB. Three supervised models (SVM, MLP, LR) were established to build the TB diagnostic model based on the serum cytokines using the training set. The SVM was identified as the best model for diagnosis of TB among HIV-positive inpatients, exhibiting superior diagnostic accuracy (online supplemental table 2). Moreover, the Hosmer-Lemeshow test result (p>0.05) indicated that the expected probabilities from the SVM model were good estimates of the true probabilities. The overall importance of the selected cytokines to the SVM model was determined using the Gini index, with the highest of IFN-γ, followed by IL-6, IL-10 and TNF-α (figure 3A).
Figure 3.
ROC curve analysis of different cohorts. (A) Gini importance of the selected cytokines in the four-cytokine biosignature to the SVM model. ROC curves generated from the training set (B), validation set (B) and independent test cohort (C). AUCs in different cohorts are shown. AUC, area under the curve; IFN-γ, interferon-gamma; ROC, receiver operation curve; SVM, support vector machine; TNF-α, tumour necrosis factor-alpha.
For the diagnosis of TB among HIV-positive inpatients, the SVM model showed sufficient efficacy in both training and validation sets. Specifically, with the optimum threshold (0.44) determined by Youden’s index, the performance of the model in training set: 0.92 (95% CI 0.88 to 0.96) AUC, 87% (95% CI 82% to 92%) accuracy, 95% (95% CI 89% to 98%) specificity, 76% (95% CI 64% to 85%) sensitivity, 86% (95% CI 79% to 91%) NPV and 90% (95% CI 80% to 96%) PPV (figure 3B, table 2); and in validation set: 0.85 (95% CI 0.72 to 0.97) AUC, 83% (95% CI 69% to 92%) accuracy, 77% (95% CI 58% to 90%) specificity, 94% (95% CI 71% to 100%) sensitivity, 96% (95% CI 79% to 100%) NPV and 70% (95% CI 47% to 87%) PPV (figure 3C, table 2).
Table 2.
Diagnostic efficiency of the four-cytokine biosignature in diagnosing TB among HIV-positive inpatients
| Model | Clinical diagnosis (n) | Classified as TB (n)/ non-TB (n) | AUC (95% CI) | Accuracy (95% CI) | Specificity (95% CI) | Sensitivity (95% CI) | NPV (95% CI) | PPV (95% CI) |
| Training | Without TB (115) | 6/109 | 0.92 (0.88 to 0.96) | 87% (82% to 92%) | 95% (89% to 98%) | 76% (64% to 85%) | 86% (79% to 91%) | 90% (80% to 96%) |
| With TB (74) | 56/18 | |||||||
| Validation | Without TB (30) | 7/23 | 0.85 (0.72 to 0.97) | 83% (69% to 92%) | 77% (58% to 90%) | 94% (71% to 100%) | 96% (79% to 100%) | 70% (47% to 87%) |
| With TB (17) | 16/1 | |||||||
| Test | Without TB (83) | 18/65 | 0.85 (0.76 to 0.94) | 80% (71% to 87%) | 78% (68% to 87%) | 85% (66% to 96%) | 94% (86% to 98%) | 56% (40% to 72%) |
| With TB (27) | 23/4 |
Data in parentheses represent 95% CI.
AUC, area under the curve; NPV, negative predictive; PPV, positive predictive value; TB, tuberculosis.
The TB prediction model performs well in an independent test cohort
In order to assess the performance of the TB diagnostic model in a real clinical setting, we tested the SVM model incorporating the four-cytokine biomarker signature using a prospective study. For this, 126 HIV-positive inpatients with symptoms of TB were prospectively recruited, with 15 excluded due to incomplete data. In order to confine the analysis using the best-characterised participant cohort, 1 with probable TB was also excluded from the analysis. The remaining 110 participants, including 27 (25%) with definite TB and 83 (75%) without TB, were subjected to test analysis. Notably, the SVM model performed well for identification of TB in the independent test cohort, with 0.85 (95% CI 0.76 to 0.94) AUC, 80% (95% CI 71% to 87%) accuracy, 78% (95% CI 68% to 87%) specificity, 85% (95% CI 66% to 96%) sensitivity, 94% (95% CI 86% to 98%) NPV and 56% (95% CI 40% to 72%) PPV (figure 3D, table 2). Together, these results indicate that the SVM model combined with the four-cytokine biomarker could successfully diagnose TB among HIV-positive inpatients.
The TB prediction model shows better performance compared with IGRA, regardless of low CD4+ T-cell counts
IGRA is useful in the diagnosis of Mtb infection in PLHIV, although its performance is not as reliable as used in the general population.17 Analysis of the IGRA for diagnosing Mtb infection in 210 participants who underwent the assays revealed an AUC of 0.81 (95% CI 0.76 to 0.86), with 80% (95% CI 74% to 85%) accuracy, 99% (95% CI 95% to 100%) specificity and 63% (95% CI 53% to 72%) sensitivity (data not shown).
Since the cytokine release and IGRA rely on functional CD4+ T-cell responses, we next investigated whether CD4+ T-cell counts affect the performance of the SVM model and IGRA for diagnosing Mtb infection among HIV-positive inpatients. More than 80% of hospitalised HIV-positive inpatients coinfected with Mtb in our study showed severe immune suppression with CD4+ T-cell counts under 200 cells/µL (table 1). Therefore, we evaluated the diagnostic accuracy of the two approaches in these advanced HIV-positive inpatients. In line with a previous report,18 IGRA performed worse in the diagnosis of Mtb infection in HIV-positive inpatients with a low CD4+ T-cell count, showing an AUC of 0.80 (95% CI 0.75 to 0.85), with 77% (95% CI 69% to 83%) accuracy, 100% (95% CI 94% to 100%) specificity and 61% (95% CI 50% to 71%) sensitivity. In contrast, the SVM model outperformed IGRA in identifying Mtb infection, with 0.85 (95% CI 0.76 to 0.94) AUC, 79% (95% CI 67% to 87%) accuracy, 74% (95% CI 60% to 86%) specificity, 87% (95% CI 66% to 97%) sensitivity, 92% (95% CI 79% to 98%) NPV and 62% (95% CI 44% to 79%) PPV (table 3). Moreover, the performance of SVM model in HIV-positive inpatients with negative IGRA showed an AUC of 0.84 (95% CI 0.77 to 0.92), with 66% (95% CI 58% to 74%) accuracy, 61% (95% CI 50% to 70%) specificity and 80% (95% CI 65% to 91%) sensitivity (table 4). Thus, our TB prediction model provides a more reliable diagnosis of Mtb infection in advanced HIV-positive inpatients, regardless of low CD4+ T-cell counts, compared with IGRA.
Table 3.
Comparison of the SVM model and IGRA in diagnosing Mtb infection among HIV-positive inpatients with low CD4+ T cell count
| Parameter | SVM | IGRA |
| AUC | 0.85 (0.76 to 0.94) | 0.80 (0.75 to 0.85) |
| Accuracy | 79% (67% to 87%) | 77% (69% to 83%) |
| Specificity | 74% (60% to 86%) | 100% (94% to 100%) |
| Sensitivity | 87% (66% to 97%) | 61% (50% to 71%) |
| NPV | 92% (79% to 98%) | 64% (53% to 73%) |
| PPV | 62% (44% to 79%) | 100% (94% to 100%) |
| K-S/lift score | 0.64/2.12 | 0.61/1.68 |
Data in parentheses represent 95% CI. The performance of SVM model compared with IGRA was measured by K-S test and lift score (p<0.001).
AUC, area under the ROC curve; IGRA, interferon-gamma release assay; K-S, Kolmogorov-Smirnov test; Mtb, Mycobacterium tuberculosis; NPV, negative predictive value; PPV, positive predictive value; SVM, support vector machine.
Table 4.
The performance of the SVM model in diagnosing Mtb infection among HIV-positive inpatients with negative IGRA
| Parameter | SVM (n=41) |
| AUC | 0.84 (0.77 to 0.92) |
| Accuracy | 66% (58% to 74%) |
| Specificity | 61% (50% to 70%) |
| Sensitivity | 80% (65% to 91%) |
| NPV | 88% (78% to 95%) |
| PPV | 46% (34% to 58%) |
Data in parentheses represent 95% CI.
AUC, area under the curve; IGRA, interferon-gamma release assay; NPV, negative predictive value; PPV, positive predictive value; SVM, support vector machine.
Discussion
TB is the primary cause of hospital admission and in-hospital deaths among PLHIV.19 20 Therefore, the prompt and accurate identification of TB among HIV-positive inpatients is imperative, as it continues to be a major strategy in controlling TB. In the present study, we enrolled hospitalised HIV-positive patients who had TB symptoms and analysed the levels of serum cytokines. Subsequently, we incorporated data from four cytokines (IL-6, IL-10, TNF-α and IFN-γ) and developed a series of mathematical models for diagnosing TB among HIV-positive inpatients. We further validated and tested the diagnostic accuracy of these models. The optimal model, SVM, was successfully to identify TB among HIV-positive inpatients in both training and validation set. We also confirmed the diagnostic performance of the SVM model in an independent prospective test cohort. Our findings suggest that the biosignature that integrates serum cytokines has potential for diagnosing TB among HIV-positive inpatients.
For decades, significant efforts have been devoted to identifying both pathogen and host biomarkers for the diagnosis of TB among PLHIV. GeneXpert and LAM are commercially available and are the most successful pathogen-specific markers. However, the diagnosis of paucibacillary specimens from HIV-positive inpatients is frequently hindered by low sensitivity and substantial variation in accuracy due to the low CD4+ T-cell levels, thereby limiting their widespread application among PLHIV.21 22 Although host responses-based detection methods, such as tuberculin skin test (TST) and IGRA, provide alternative solutions for diagnosing Mtb infection, they share the limitation of being unable to distinguish active TB from LTBI and exhibit reduced efficacy in advanced PLHIV with low CD4+ T-cell counts.8 23 24 Other host biomarkers, apart from TST and IGRA, are in earlier stages of development for TB diagnostics.
Previous studies have found a positive correlation between the increased serum cytokine levels, severe TB progression and a heightened risk of recurrence in PLHIV.10 11 25 The significant differences in serum cytokine levels between PLHIV with and without TB infection suggest the potential usage of these biomarkers for TB diagnostic. Attempts have been made to evaluate the diagnostic performance of a wide range of cytokines for diagnosing TB in PLHIV. For instance, a study involved a small sample size of 86 participants demonstrated that cytokines, including IFN-γ, TNF-α, IL-2, IL-6 and IL-10, could identify TB in patients coinfected with HIV.26 In addition, high levels of IL-6 were detected in patients with HIV and TB coinfection before initiation of ART and anti-TB therapy, and persisted after 6 months of treatment, suggesting its potential as a marker of Mtb infection in PLHIV.27 More recently, a study showed that PLHIV with LTBI had elevated levels of TNF, IL-6, IL-4 compared with the PLHIV without TB infection, independent of CD4+ T-cell count and ART duration.28 However, these studies have limitations. First, some studies have small sample sizes and lack validation in the real clinical settings, which limits the ability of cytokines to diagnose TB in PLHIV. Second, cytokine biomarker-based diagnostics cannot differentiate TB from other infections, as certain studies recruited healthy PLHIV as controls. Many inflammatory markers can distinguish infection from healthy state but fail to identify TB among advanced PLHIV with associated comorbidities. Third, the detection of cytokines using antigen-stimulated cell culture supernatants makes it difficult to standardise blood collection and sample processing. In comparison, the present study enrolled a sufficient sample size of hospitalised HIV-positive patients with signs or symptoms of TB. With validation and independent test, we provide a fast, reliable diagnostic model consisting of four cytokines for the diagnosis of TB among HIV-positive inpatients with high sensitivity and specificity.
The association between lower CD4+ T-cell counts and increased risk of TB in PLHIV has been well established in previous studies.29 30 Consistent with this concept, we observed that most of the HIV-positive inpatients coinfected with TB in the present study showed a CD4+ T-cell count lower than 200 cells/µL. Consequently, assays based on T cell immune responses do not work well and underestimate TB diagnosis in advanced PLHIV. In agreement with previous reports,8 31 32 stratifying participants in the study by CD4+ T cell revealed a decrease in the diagnostic accuracy of IGRA, the current gold standard immunodiagnostic for Mtb infection, in those with lower CD4+ T-cell counts. However, the performance of the TB prediction model incorporating the four-cytokine biosignature was not affected by low CD4+ T-cell counts. Instead, the SVM model demonstrates superior performance compared with IGRA and could serve as an alternative approach for identifying Mtb infection among HIV-positive inpatients who test negative using IGRA.
The major limitations of the present study include that all participants recruited in the study were hospitalised HIV-patients with symptoms suggestive of TB. The use of narrow patient spectrums will impede the translation of the four-cytokine biosignature in PLHIV, particularly those who are asymptomatic or not receiving hospital-level care. Additionally, although the SVM model was both internally and externally validated using two cohorts in the study, the findings based on a single centre may not represent the status of patientswith TB in PLHIV globally. Validation with small sample size and excluding participants with incomplete data may be vulnerable to selection bias. Finally, cytokine detection is not routinely performed even in major clinical laboratories. There is still a long way to develop the multiplex cytokine assay into an available at the point-of-care, affordable and stable TB test. Future research is needed to assess the utility of the SVM model identifying TB in broader PLHIV populations from multiple clinical centres.
Conclusion
Our findings demonstrate that the SVM model with a four-cytokine biosignature as a valuable alternative approach for identifying TB among HIV-positive inpatients. With further validation and development, the TB prediction model may be applied in the following clinical situations. First, because the model was established and validated in inpatients with symptoms of TB, we anticipate that it may discriminate TB from diverse pulmonary diseases, including lung cancer, silicosis, sarcoidosis and other opportunistic infections, among HIV-positive patients. Second, the fact that the NPV of the model in validation and test set is higher than its PPV suggests that the model may be more useful for ruling out TB instead of confirming the diagnosis. With an appropriate cut-off value, this test alone, or in combination with IGRA, can be reliably used as a rule-out test in HIV-TB high-prevalence settings. Finally, it may be helpful for patients with suspected TB in whom standard tests are not diagnostic or who are unable to provide sputum samples for TB diagnosis.
Acknowledgments
The authors would like to thank all participants and clinical specialists at Guangzhou Eighth People’s Hospital for their contributions to the study.
Footnotes
HZ and LL contributed equally.
Contributors: YD is responsible for the overall content as guarantor. HZ and YD had full access to all of the data in the study and took responsibility for the content of the manuscript, integrity of the data, and the accuracy of the data analysis. YD, LL and XT designed the study. YL, RX, ML, WH and YG enrolled the participants and performed the laboratory procedures. HZ, WX and YD analysed the data and performed the statistical analysis and modelling. HZ, WX and YD drafted the initial manuscript. All authors reviewed and drafted the manuscript for critical content. All authors approved the final version of the manuscript.
Funding: This study was supported by the Science and Technology Project of Guangdong Province (grant no. 2023A1515010351); the Science and Technology Project of Guangzhou (grant no. SL2023A03J01070); the Guangdong Science Fund for Distinguished Young Scholars.
Competing interests: None declared.
Provenance and peer review: Not commissioned; externally peer reviewed.
Supplemental material: This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.
Data availability statement
Data are available on reasonable request.
Ethics statements
Patient consent for publication
Not applicable.
Ethics approval
This study involves human participants and the study was approved by the Ethics Committees of Guangzhou Eighth People’s Hospital (reference number: 202307244). Participants gave informed consent to participate in the study before taking part.
References
- 1. World Health organization . Global tuberculosis report. Geneva: World Health organization; 2022. [Google Scholar]
- 2. Cribbs SK, Crothers K, Morris A. Pathogenesis of HIV-related lung disease: immunity, infection, and inflammation. Physiol Rev 2020;100:603–32. 10.1152/physrev.00039.2018 [DOI] [PubMed] [Google Scholar]
- 3. Gupta-Wright A, Corbett EL, van Oosterhout JJ, et al. Rapid urine-based screening for tuberculosis in HIV-positive patients admitted to hospital in Africa (STAMP): a pragmatic, Multicentre, parallel-group, double-blind, randomised controlled trial. Lancet 2018;392:292–301. 10.1016/S0140-6736(18)31267-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Penn-Nicholson A, Gomathi SN, Ugarte-Gil C, et al. A prospective multicentre diagnostic accuracy study for the Truenat tuberculosis assays. Eur Respir J 2021;58:2100526. 10.1183/13993003.00526-2021 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5. Trébucq A, Enarson DA, Chiang CY, et al. Xpert® MTB/RIF for national tuberculosis programmes in low-income countries: when, where and how Int J Tuberc Lung Dis 2011;15:1567–72. 10.5588/ijtld.11.0392 [DOI] [PubMed] [Google Scholar]
- 6. Lawn SD, Kerkhoff AD, Vogt M, et al. Diagnostic accuracy of a low-cost, urine antigen, point-of-care screening assay for HIV-associated pulmonary tuberculosis before antiretroviral therapy: a descriptive study. Lancet Infect Dis 2012;12:201–9. 10.1016/S1473-3099(11)70251-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Huerga H, Bastard M, Lubega AV, et al. Novel Fujilam assay to detect tuberculosis in HIV-positive ambulatory patients in four African countries: a diagnostic accuracy study. Lancet Glob Health 2023;11:e126–35. 10.1016/S2214-109X(22)00463-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8. Mthembu M, Bowman KA, Davies LRL, et al. Discrepancy between Mtb-specific IFN-Γ and IgG responses in HIV-positive people with low Cd4 counts. EBioMedicine 2023;90:104504. 10.1016/j.ebiom.2023.104504 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Chaw L, Chien LC, Wong J, et al. Global trends and gaps in research related to latent tuberculosis infection. BMC Public Health 2020;20:352. 10.1186/s12889-020-8419-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Sullivan ZA, Wong EB, Ndung’u T, et al. Latent and active tuberculosis infection increase immune activation in individuals co-infected with HIV. EBioMedicine 2015;2:334–40. 10.1016/j.ebiom.2015.03.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Sivro A, McKinnon LR, Yende-Zuma N, et al. Plasma cytokine predictors of tuberculosis recurrence in antiretroviral-treated human immunodeficiency virus-infected individuals from Durban, South Africa. Clin Infect Dis 2017;65:819–26. 10.1093/cid/cix357 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Eribo OA, Leqheka MS, Malherbe ST, et al. Host urine immunological biomarkers as potential candidates for the diagnosis of tuberculosis. Int J Infect Dis 2020;99:473–81. 10.1016/j.ijid.2020.08.019 [DOI] [PubMed] [Google Scholar]
- 13. Adankwah E, Nausch N, Minadzi D, et al. Interleukin-6 and Mycobacterium tuberculosis dormancy antigens improve diagnosis of tuberculosis. J Infect 2021;82:245–52. 10.1016/j.jinf.2020.11.032 [DOI] [PubMed] [Google Scholar]
- 14. de Sá NBR, de Souza NCS, Neira-Goulart M, et al. Inflammasome genetic variants are associated with tuberculosis, HIV-1 infection, and TB/HIV-immune reconstitution inflammatory syndrome outcomes. Front Cell Infect Microbiol 2022;12:962059. 10.3389/fcimb.2022.962059 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Dai Y, Shan W, Yang Q, et al. Biomarkers of iron metabolism facilitate clinical diagnosis in Mycobacterium tuberculosis infection. Thorax 2019;74:1161–7. 10.1136/thoraxjnl-2018-212557 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Dai Y, Cai Y, Wang X, et al. Autoantibody-mediated erythrophagocytosis increases tuberculosis susceptibility in HIV patients. mBio 2020;11:e03246-19. 10.1128/mBio.03246-19 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Klautau GB, da Mota NVF, Salles MJC, et al. Interferon-Γ release assay as a sensitive diagnostic tool of latent tuberculosis infection in patients with HIV: a cross-sectional study. BMC Infect Dis 2018;18:585. 10.1186/s12879-018-3508-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Goletti D, Delogu G, Matteelli A, et al. The role of IGRA in the diagnosis of tuberculosis infection, differentiating from active tuberculosis, and decision making for initiating treatment or preventive therapy of tuberculosis infection. Int J Infect Dis 2022;124:S12–9. 10.1016/j.ijid.2022.02.047 [DOI] [PubMed] [Google Scholar]
- 19. Gupta RK, Lucas SB, Fielding KL, et al. Prevalence of tuberculosis in post-mortem studies of HIV-infected adults and children in resource-limited settings: a systematic review and meta-analysis. AIDS 2015;29:1987–2002. 10.1097/QAD.0000000000000802 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Shen Y, Lu H. Prevalence and predictors of tuberculosis among adults with newly diagnosed HIV/AIDS. Int J Infectious Dis 2016;45:412. 10.1016/j.ijid.2016.02.879 [DOI] [Google Scholar]
- 21. Penn-Nicholson A, Georghiou SB, Ciobanu N, et al. Detection of isoniazid, fluoroquinolone, ethionamide, amikacin, kanamycin, and capreomycin resistance by the Xpert MTB/XDR assay: a cross-sectional multicentre diagnostic accuracy study. Lancet Infect Dis 2022;22:242–9. 10.1016/S1473-3099(21)00452-7 [DOI] [PubMed] [Google Scholar]
- 22. Mwaura M, Engel N. Constructing confidence: user perspectives on Alerelam testing for tuberculosis. Int J Infect Dis 2021;112:237–42. 10.1016/j.ijid.2021.09.018 [DOI] [PubMed] [Google Scholar]
- 23. Cobelens FG, Egwaga SM, van Ginkel T, et al. Tuberculin skin testing in patients with HIV infection: limited benefit of reduced cutoff values. Clin Infect Dis 2006;43:634–9. 10.1086/506432 [DOI] [PubMed] [Google Scholar]
- 24. Aabye MG, Ravn P, PrayGod G, et al. The impact of HIV infection and Cd4 cell count on the performance of an interferon gamma release assay in patients with pulmonary tuberculosis. PLoS One 2009;4:e4220. 10.1371/journal.pone.0004220 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Pillay K, Lewis L, Rambaran S, et al. Plasma biomarkers of risk of tuberculosis recurrence in HIV Co-infected patients from South Africa. Front Immunol 2021;12:631094. 10.3389/fimmu.2021.631094 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. Kisuya J, Chemtai A, Raballah E, et al. The diagnostic accuracy of Th1 (IFN-Γ, TNF-Α, and IL-2) and Th2 (IL-4, IL-6 and IL-10) Cytokines response in AFB microscopy smear negative PTB- HIV Co-infected patients. Sci Rep 2019;9:2966. 10.1038/s41598-019-39048-x [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27. Nosik M, Ryzhov K, Rymanova I, et al. Dynamics of plasmatic levels of pro- and anti-inflammatory cytokines in HIV-infected individuals with M. tuberculosis co-infection. Microorganisms 2021;9:11.:2291. 10.3390/microorganisms9112291 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Temu TM, Polyak SJ, Wanjalla CN, et al. Latent tuberculosis is associated with heightened levels of pro-and anti-inflammatory cytokines among Kenyan men and women living with HIV on long-term antiretroviral therapy. AIDS 2023;37:1065–75. 10.1097/QAD.0000000000003523 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Mendelsohn SC, Fiore-Gartland A, Awany D, et al. Clinical predictors of pulmonary tuberculosis among South African adults with HIV. EClinicalMedicine 2022;45:101328. 10.1016/j.eclinm.2022.101328 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Liu E, Makubi A, Drain P, et al. Tuberculosis incidence rate and risk factors among HIV-infected adults with access to antiretroviral therapy. AIDS 2015;29:1391–9. 10.1097/QAD.0000000000000705 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Kroon EE, Kinnear CJ, Orlova M, et al. An observational study identifying highly tuberculosis-exposed, HIV-1-positive but persistently TB, tuberculin and IGRA negative persons with M. tuberculosis specific antibodies in Cape town, South Africa. EBioMedicine 2020;61:103053. 10.1016/j.ebiom.2020.103053 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Aichelburg MC, Tittes J, Breitenecker F, et al. Prognostic value of indeterminate IFN-Γ release assay results in HIV-1 infection. J Clin Microbiol 2012;50:2767–9. 10.1128/JCM.01054-12 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
thorax-2023-220782supp001.pdf (56.1KB, pdf)
Data Availability Statement
Data are available on reasonable request.