Abstract Background Hematological indices from routine complete blood counts, neutrophil-to-lymphocyte ratio (NLR), monocyte-to-lymphocyte ratio (MLR), and platelet-to-lymphocyte ratio (PLR), are inexpensive inflammatory markers that may aid early risk stratification in hospitalized tuberculosis (TB). Objective To evaluate the prognostic performance of admission NLR, MLR, and PLR for 28-day mortality among adults hospitalized with pulmonary TB. Methods This is a retrospective cohort study conducted at Persahabatan Hospital, a national referral for respiratory diseases in Jakarta. Adults with microbiologically or molecularly confirmed pulmonary TB and complete index data within 24 hours of admission were included. Discrimination for 28-day mortality was assessed using receiver operating characteristic (ROC) curves and area under the curve (AUC). A multivariable model was developed using a logistic regression approach to identify a better-performing predictive model that may include hematological indices and one or more of the confounding factors (age, sex, smoking, diabetes, steroid use, and immunosuppressant use) Results This study included 257 pulmonary TB patients, comprising 17 non-survivors and 240 survivors. NLR demonstrated good discrimination for 28-day mortality (AUC 0.769, SE 0.061, 95% CI 0.649-0.889; p 0.001). PLR showed weaker discrimination (AUC 0.625, SE 0.069, 95% CI 0.489-0.760; p = 0.071), while MLR had limited discriminative ability (AUC 0.557, SE 0.076, 95% CI 0.408-0.707; p = 0.452). Better-performing multivariable model, comprising NLR, PLR, and smoking, was found to have better discriminatory capability for 28-day mortality (AUC 0.841, SE 0.050, 95% CI 0.744-0.939; p 0.001) Conclusion Among adult TB inpatients, admission NLR outperformed MLR and PLR in identifying patients at risk of 28-day mortality. Given its low cost and universal availability, NLR is a pragmatic candidate to support early risk stratification in resource-constrained settings, especially when supported by other variables in a multivariable model This abstract is funded by: None
Wardoyo et al. (Fri,) studied this question.