Clinical, microbiologic, and immunologic determinants of mortality in hospitalized patients with HIV-associated tuberculosis: A prospective cohort study

Abstract

Background: In high-burden settings, case fatality rates are reported to be between 11% and 32% in hospitalized patients with HIV-associated tuberculosis, yet the underlying causes of mortality remain poorly characterized. Understanding causes of mortality could inform the development of novel management strategies to improve survival. We aimed to assess clinical and microbiologic determinants of mortality and to characterize the pathophysiological processes underlying death by evaluating host soluble inflammatory mediators and determined the relationship between these mediators and death as well as biomarkers of disseminated tuberculosis.

Methods and findings: Adult patients with HIV hospitalized with a new diagnosis of HIV-associated tuberculosis were enrolled in Cape Town between 2014 and 2016. Detailed tuberculosis diagnostic testing was performed. Biomarkers of tuberculosis dissemination and host soluble inflammatory mediators at baseline were assessed. Of 682 enrolled participants, 576 with tuberculosis (487/576, 84.5% microbiologically confirmed) were included in analyses. The median age was 37 years (IQR = 31-43), 51.2% were female, and the patients had advanced HIV with a median cluster of differentiation 4 (CD4) count of 58 cells/L (IQR = 21-120) and a median HIV viral load of 5.1 log10 copies/mL (IQR = 3.3-5.7). Antituberculosis therapy was initiated in 566/576 (98.3%) and 487/576 (84.5%) started therapy within 48 hours of enrolment. Twelve-week mortality was 124/576 (21.5%), with 46/124 (37.1%) deaths occurring within 7 days of enrolment. Clinical and microbiologic determinants of mortality included disseminated tuberculosis (positive urine lipoarabinomannan [LAM], urine Xpert MTB/RIF, or tuberculosis blood culture in 79.6% of deaths versus 60.7% of survivors, p = 0.001), sepsis syndrome (high lactate in 50.8% of deaths versus 28.9% of survivors, p < 0.001), and rifampicin-resistant tuberculosis (16.9% of deaths versus 7.2% of survivors, p = 0.002). Using non-supervised two-way hierarchical cluster and principal components analyses, we describe an immune profile dominated by mediators of the innate immune system and chemotactic signaling (interleukin-1 receptor antagonist [IL-1Ra], IL-6, IL-8, macrophage inflammatory protein-1 beta [MIP-1β]/C-C motif chemokine ligand 4 [CCL4], interferon gamma-induced protein-10 [IP-10]/C-X-C motif chemokine ligand 10 [CXCL10], MIP-1 alpha [MIP-1α]/CCL3), which segregated participants who died from those who survived. This immune profile was associated with mortality in a Cox proportional hazards model (adjusted hazard ratio [aHR] = 2.2, 95%CI = 1.9-2.7, p < 0.001) and with detection of biomarkers of disseminated tuberculosis. Clinicians attributing causes of death identified tuberculosis as a cause or one of the major causes of death in 89.5% of cases. We did not perform longitudinal sampling and did not have autopsy-confirmed causes of death.

Conclusions: In this study, we did not identify a major contribution from coinfections to these deaths. Disseminated tuberculosis, sepsis syndrome, and rifampicin resistance were associated with mortality. An immune profile dominated by mediators of the innate immune system and chemotactic signaling was associated with both tuberculosis dissemination and mortality. These findings provide pathophysiologic insights into underlying causes of mortality and could be used to inform the development of novel treatment strategies and to develop methods to risk stratify patients to appropriately target novel interventions. Causal relationships cannot be established from this study.

Fig 1. Study flowchart.

Showing disposition of screened patients. Patients in the emergency room and medical wards were screened daily on weekdays, and potentially eligible patients were identified. Potentially eligible patients were randomly selected and approached for enrolment in the study. A total of 682 patients were enrolled, and 659 had HIV infection with a CD4 count <350 cells/μL. Tuberculosis was diagnosed in 576 participants (microbiologically confirmed and probable tuberculosis), and these participants are included in the analysis. CD4, cluster of differentiation 4.

Fig 2. Association between tuberculosis dissemination and 12-week mortality.

(A) Kaplan Meier curve showing percentage survival over 84 days for participants who tested positive for Mycobacterium tuberculosis blood culture versus those who tested negative. Curves were compared using log-rank (Mantel-Cox) test. (B) Frequencies of individuals who died, stratified by tuberculosis dissemination score values, were compared using the Pearson chi-squared test with linear trend. TB, tuberculosis.

Fig 3. Host soluble inflammatory mediators associated with mortality in participants with hospitalized HIV-associated tuberculosis.

(A) Values of inflammatory mediators were log transformed and z-score normalized. A non-supervised two-way hierarchical cluster analysis (Ward method with 100× bootstrap) was employed to test if simultaneous assessment of indicated mediators could group separately individuals that died from those who survived. Only mediators that were statistically different between the study groups after adjustment for multiple measurements (Holm-Bonferroni method) are shown. Data on other mediators are shown in S4 Table and S2 and S3 Figs. (B) Bars represent fold-difference values between participants that died versus those who survived, with Holm-Bonferroni p-values. Yellow bars indicate mediators that were significantly higher, whereas blue bars highlight mediators that were lower in participants who died compared with those who survived. FGF, basic fibroblast growth factor; IFNγ, interferon gamma; IL, interleukin; IL-1Ra, IL-1 receptor antagonist; IP-10, interferon gamma-induced protein; CXCL10, C-X-C motif chemokine ligand 10; MIP-1α, macrophage inflammatory protein-1 alpha; CCL3, C-C motif chemokine ligand 3; MIP-1β, macrophage inflammatory protein-1 beta; CCL4, C-C motif chemokine ligand 4; PDGF, platelet-derived growth factor; RANTES, regulated on activation, normal T-cell expressed and secreted; CCL5, C-C motif chemokine ligand 5; TGF-β1, transforming growth factor beta 1.

Fig 4. Principal components analysis of host soluble inflammatory mediators.

(A) Principal components analysis with rotation was used to analyze the inflammatory mediators and explain the variance of the data distribution in the cohort. Participants are represented by dots and colored by outcome. The two axes represent principal components 1 (PC1 on the x-axis) and 2 (PC2 on the y-axis), and their contribution to the total data variance is shown as a percentage. PC3 contributed 19% of total variance and is not shown. (B) Variables contributing to PC1, PC2, and PC3 are shown with red bars indicating positive weighting and blue bars indicating negative weighting. CCL, C-C motif chemokine ligand; CSF2, colony stimulating factor 2; CSF3, colony stimulating factor 3; CXCL, C-X-C motif chemokine ligand; FGF, basic fibroblast growth factor; G-CSF, granulocyte-colony stimulating factor; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFNγ, interferon gamma; IL, interleukin; IL-1RA, IL-1 receptor antagonist; IP-10, interferon gamma-induced protein; MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein; PC, principal component; PDGF, platelet-derived growth factor; RANTES, regulated on activation, normal T-cell expressed and secreted; TGF-β1, transforming growth factor beta 1; TNFα, tumor necrosis factor alpha; VEGF, vascular endothelial growth factor.

Fig 5. Cox-regression analysis to evaluate association between principal components score and 28-day mortality.

Cox regression analysis was conducted with each variable individually (unadjusted) and then all variables were included in a multivariable model (adjusted). Age, sex, and HIV viral load were incorporated a priori to adjust for patient specific variance and HIV-related factors. The model was censored at 28 days to meet the proportional hazards assumption, and the global proportional hazards test for the multivariable model result was p = 0.43. PC, principal component.

Fig 6. Associations between leukocyte counts in peripheral blood and systemic inflammation.

(A) Absolute counts of indicated leukocytes were compared between the participants who died versus those who survived. Data are represented with violin plots, with scatter dots and box and whiskers indicating median values, IQRs, and maximum and minimum values excluding outliers, respectively. Groups were compared using the Mann-Whitney U test. (B) Spearman correlations between indicated cell counts are shown. Linear curve fit (with 95% CI) was used to illustrate trends of data distribution. (C) Spearman correlation analysis between cell counts and inflammatory mediator values in plasma. (D) Spearman correlations between indicated cell counts and PC1 score values are shown. Linear curve fit (with 95% CI) was used to illustrate trends of data distribution. CCL, C-C motif chemokine ligand; CSF, colony stimulating factor; CXCL, C-X-C motif chemokine ligand; FGF, basic fibroblast growth factor; G-CSF, granulocyte-colony stimulating factor; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFNγ, interferon gamma; IL, interleukin; IL-1RA, IL-1 receptor antagonist; IP-10, interferon gamma-induced protein; MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein; PC1, principal component 1; PDGF, platelet-derived growth factor; RANTES, regulated on activation, normal T-cell expressed and secreted; TGF-β1, transforming growth factor beta 1; TNFα, tumor necrosis factor alpha; VEGF, vascular endothelial growth factor.

Fig 7. Associations between tuberculosis dissemination score, components thereof, and host soluble mediators of inflammation.

(A) Mean values of log-transformed value of each plasma mediator per TB dissemination score values were calculated for all participants who had all three tests performed (urine LAM test, urine Xpert MTB/RIF test, mycobacterial blood culture), n = 457. Inflammatory mediator values were z-score normalized and illustrated in a heatmap in which inflammatory mediators were grouped using hierarchical clustering (Ward method with 100× bootstrap). Dendrograms represent Euclidean distance. (B) PC1 score values were compared between those who tested positive and negative for each of the three tests used to calculate the tuberculosis dissemination score using the Mann-Whitney U test. (C) PC1 score values were compared between participants presenting with increasing tuberculosis dissemination scores from 0 to 3 in both outcome groups, and values were compared using the Kruskal-Wallis test with the nonparametric linear trend ad hoc test. Lines connect median values. CCL, C-C motif chemokine ligand; CSF, colony stimulating factor; CXCL, C-X-C motif chemokine ligand; FGF, basic fibroblast growth factor; G-CSF, granulocyte-colony stimulating factor; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFNγ, interferon gamma; IL, interleukin; IP-10, interferon gamma-induced protein; LAM, lipoarabinomannan; MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein; MTB, Mycobacterium tuberculosis; PC, principal component; PDGF, platelet-derived growth factor; RANTES, regulated on activation, normal T-cell expressed and secreted; TB, tuberculosis; TGF-β1, transforming growth factor beta 1; TNFα, tumor necrosis factor alpha; VEGF, vascular endothelial growth factor.

Fig 8. Inverse correlations between host soluble mediators of inflammation and time to death.

(A) Histogram shows the frequency of participants who died over time. (B) Pie chart shows the frequency of participants who survived, those who died within 7 days of admission (early death), and those who died after 7 days (late death). (C) Left panel: data were log transformed and ranked and colored in a heatmap from minimum to maximum values detected for each inflammatory mediator. Participants were ordered based on time to death (in days), and plasma inflammatory mediators were clustered (Ward method with 100× bootstrap) according to the distribution profile in the study population. Dendrograms represent Euclidean distance. Right panel: Spearman correlations for each mediator and time to death. Blue bars indicate statistically significant correlations (which were all negative) after corrections for multiple measurements (Holm-Bonferroni method). (D) Spearman correlations between PC1 score values and IL1-Ra mean fluorescence intensity values and time to death are shown. Nonlinear curve fit (quadratic, with 95% CI) was used to illustrate trends of data distribution. CCL, C-C motif chemokine ligand; CXCL, C-X-C motif chemokine ligand; FGF, basic fibroblast growth factor; G-CSF, granulocyte-colony stimulating factor; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFNγ, interferon gamma; IL, interleukin; IL1-Ra, IL-1 receptor antagonist; IP-10, interferon gamma-induced protein; max., maximum; MCP, monocyte chemoattractant protein; MFI, mean fluorescence intensity; min., minimum; MIP, macrophage inflammatory protein; no., number; PC1, principal component 1; PDGF, platelet-derived growth factor; RANTES, regulated on activation, normal T-cell expressed and secreted; TGF-β1, transforming growth factor beta 1; TNFα, tumor necrosis factor alpha; VEGF, vascular endothelial growth factor.

Fig 9. Participants who died early after admission exhibited a distinct inflammatory profile in plasma.

(A) Left panel: mean values of log-transformed values of each soluble inflammatory mediator were calculated for early deaths (within 7 days of enrolment) and late deaths (after 7 days, within 12 weeks). Values were z-score normalized and illustrated in a heatmap in which inflammatory mediators were grouped using hierarchical clustering (Ward method with 100× bootstrap). Dendrograms represent Euclidean distance. Right panel: bars represent fold-difference values between participants who died early or late versus those who survived. Yellow bars indicate mediators that were significantly higher, whereas blue bars highlight mediators that were significantly lower in the groups of participants who died compared with those who survived, after adjustments for multiple measurements (Holm-Bonferroni method). Arrows indicate values higher than the upper limit of the axis. (B) Venn diagrams illustrate the inflammatory mediators that were significantly different between participants who died and survived. Mediators indicated in red were higher, whereas those in blue were lower in the groups of participants who died versus participants who survived. (C) ROC curve analyses of the combination of uniquely expressed mediators (module 1 or module 2) were used to test the power to predict early or late mortality versus survival. AUC, area under the curve; CCL, C-C motif chemokine ligand; CXCL, C-X-C motif chemokine ligand; FGF, basic fibroblast growth factor; G-CSF, granulocyte-colony stimulating factor; GM-CSF, granulocyte-macrophage colony-stimulating factor; IFNγ, interferon gamma; IL, interleukin; IL-1RA, IL-1 receptor antagonist; IP-10, interferon gamma-induced protein; MCP, monocyte chemoattractant protein; MIP, macrophage inflammatory protein; PDGF, platelet-derived growth factor; RANTES, regulated on activation, normal T-cell expressed and secreted; ROC, receiver operating characteristic; TGF-β1, transforming growth factor beta 1; TNFα, tumor necrosis factor alpha; VEGF, vascular endothelial growth factor.