Abstract Body

Background: Tuberculosis meningitis (TBM) diagnosis is notoriously difficult, new biomarkers are needed. We evaluated the diagnostic utility of 3 novel biomarkers, 1) delta-like 1 protein (DLL1), a Notch ligand, which selectively drives antigen-specific CD4 T helper type-1 (Th1) responses; 2) vitamin D binding protein (VDBP); 3) fetuin. VDBP and fetuin are promising biomarkers of bovine tuberculosis.

Methods: Biomarker concentrations were measured by ELISA in cryopreserved CSF from 139 HIV-infected Ugandan patients with meningitis. TBM was diagnosed by GeneXpert MTB/Rif assay and/or mycobacterial culture. Cohort diagnoses included TBM (n=21), cryptococcal meningitis (n=71), and aseptic ‘other’ meningitis (n=47) among those testing negative for TB, Cryptococcus, and bacteria. We evaluated the diagnostic performance of DLL1 (n=136), VDBP (n=130), and fetuin (n=130) for TBM.

Results: Patient characteristics were similar at diagnosis except for sex, CD4 count, CSF WBC count, and CSF protein. DLL, VDBP, and fetuin CSF concentrations were significantly higher in patients with TBM than cryptococcal or ‘other’ meningitis (Table 1). Using logistic regression, DLL1 alone provided 33% sensitivity, 98% specificity, 78% positive predictive value (PPV), and 88% negative predictive value (NPV). Combining DLL1 with interferon gamma (IFNg) and cryptococcal antigen testing improved sensitivity to 62% with slightly decreased specificity of 95%, PPV of 72% and NPV of 93%. For every two-fold increase in DLL1, the odds of TBM increased (adjusted Odds Ratio=2.6, 95%CI: 1.2-5.5; P=.011) in the multivariable logistic model (P<0.001 in univariate analysis). Further incorporating fetuin, VDBP, CSF white cell count, or CSF protein into a predictive model did not improve diagnostic performance. Among the 5 putative false positives as classified by the DLL1, IFNg, CRAG model, 2 had a clinical TBM diagnosis during hospitalization without microbiologic confirmation. Among 117 persons with known outcome, in-hospital mortality was associated with increasing DLL1 levels (Odds Ratio=2.0 per two-fold increase, 95% CI: 1.2-3.6; P=.008) and decreasing fetuin (Odds Ratio=0.67 per two-fold increase, 95% CI: 0.47-0.96; P=0.030) in CSF.

Conclusions: CSF DLL1 exhibited reasonable diagnostic performance, and may have a role as low cost adjunctive TBM diagnostic tools and may perform better in combination. Misclassification bias (of non-detection of TBM classified as ‘other’) hampers diagnostic studies, and future larger studies are required.

⇥

⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥

⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥

⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥

⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥⇥

⇥⇥

⇥

Biomarker concentration Tuberculosis meningitis Cryptococcal meningitis Other aseptic meningitis P value C statistic for TBM
DLL1, pg/mL 768 (522-1243) 421 (304-544) 350 (188-683) <.001 .790
VDBP, μg/mL 4.36 (2.74-8.45) 1.55 (0.74-2.71) 1.07 (0.43-6.15) <.001 .781
Fetuin, μg/mL 3.32 (2.15-4.60) 1.55 (0.82-2.37) 0.77 (0.43-2.81) <.001 .818

P values calculated from mean log2 transformed biomarker values analyzed by ANOVA.