Assessing mortality risk in pulmonary tuberculosis and severe malnutrition: development of the IIR marker via artificial intelligence

The early identification of mortality risk in patients with tuberculosis (TB) and severe malnutrition (BMI <16 kg/m2) is critical for optimizing clinical outcomes. In this three-year ambispective study (October 1, 2021–September 30, 2024), conducted at Leamna Hospital, a reference center for the Oltenia Region, Romania, 216 patients with pulmonary tuberculosis were selected from a total of 3,547 TB cases for analysis. We assessed all-cause in-hospital mortality during the index admission only (from admission to discharge); deaths after discharge or during subsequent admissions were excluded, patients transferred without cross-facility linkage were right-censored at transfer, and all analyses used baseline hematological and biochemical parameters obtained before initiation of any treatment. We developed and validated the Immuno-Inflammatory Ratio (IIR), a novel machine-learning–assisted biomarker integrating neutrophils, lymphocytes, and eosinophils. The IIR demonstrated an apparent AUC of 0.9711 with an optimal threshold of 7.44 (sensitivity 99.40%, specificity 91.49%). In regression analyses, the IIR emerged as the strongest independent predictor of mortality (adjusted OR 13.98, p < 0.001), outperforming established indices such as the neutrophil-to-lymphocyte ratio (NLR) and the cumulative inflammatory index (IIC). Given its simplicity and strong discriminatory power, the IIR may support early risk stratification and prioritization of standard interventions (e.g., intensified monitoring, nutritional support, and timely optimization of anti-TB therapy). Prospective multicenter validation and longitudinal assessment of IIR dynamics are warranted to confirm clinical utility and define applications beyond the index admission, including in resource-limited settings.