Access to diagnosis using liquid biopsy (ADLiB): Identifying lymphoma in a tuberculosis-endemic setting

In HIV- and tuberculosis (TB)-endemic regions, lymphoma diagnosis is often delayed because symptoms can overlap with TB, and access to biopsy and specialized pathology is limited. To address this, we developed and internally evaluated the Access to Diagnosis using Liquid Biopsy (ADLiB) platform-a plasma cell-free DNA (cfDNA)-based approach capturing molecular features of lymphoma and key pathogens. Plasma cfDNA undergoes ultra-low-pass whole genome sequencing and deep targeted sequencing (50,000× coverage). The combined output-including estimated tumor fraction, mutations, copy number alterations, structural variants, B- and T-cell receptor clonality, and pathogen detection-is integrated into a machine-learning classifier developed within this study to differentiate malignant from benign lymphadenopathy. In a prospective cohort of 124 South African patients with lymphadenopathy due to lymphoma (76.6%), benign causes (18.6%; TB adenitis [9.7%] and reactive [8.9%]), and disseminated solid tumors (4.8%), patients with lymphoma had higher cfDNA concentrations (mean 248.3 ng/mL), tumor fraction (mean 11.6%), and more mutations (mean 9.5), frequently in KMT2C, SOCS1, TP53, STAT3, and CREBBP. Targeted structural variants were 100% specific for lymphoma. The ADLiB machine-learning classifier distinguished malignant from benign lymphadenopathy within this cohort with 95% sensitivity, 75% specificity, and 92% accuracy (AUC 0.89). ADLiB provides a minimally invasive and feasible diagnostic approach to distinguish malignant from benign lymphadenopathy in resource-limited settings with high pathogen prevalence that can complicate lymphoma diagnosis. With further optimization and clinical validation, this cfDNA-based approach has the potential to accelerate and improve lymphoma diagnosis where traditional diagnostics are constrained.