Accuracy Of Computer-Assisted Detection In Screening People With Diabetes Mellitus For Active Tuberculosis: A Systematic Review

ObjectivesDiabetes mellitus (DM) significantly increases the risk of tuberculosis (TB), and active TB screening of people with DM has been advocated. This systematic review aimed to evaluate the accuracy of computer-assisted detection (CAD) for identifying pulmonary TB among people living with DM. MethodsMedline, Embase, Scopus, Global Health and Web of science were searched from January 2010 to May 2024 supplemented with grey literature (Conference abstracts, Trial registries, MedRxiv.org). Studies evaluating CAD accuracy for identifying TB in populations living with diabetes were included. Two researchers independently assessed titles, abstracts, full texts, extracted data and assessed the risk of bias. Due to heterogeneity and a limited number of studies, a descriptive analysis was performed instead of statistical pooling. Forest plot and Summary Receiver Operating Curves (SROC) were generated using RevMan 5.4. ResultsFive eligible studies, all conducted in Asia between 2013 and 2023 were identified, including a total of 1879 individuals of whom 391 were diagnosed with TB. Four different Computer Assisted Detection (CAD) software algorithms were used. Sensitivities ranged from 0.73 (95%CI: 0.61-0.83) to 1.00 (95%CI:0.59-1.00), while specificities ranged from 0.60 (95%CI:0.53-0.67) to 0.88 (95%CI: 0.84-0.91). Area Under the receiver Operating Curve (AUC) values varied from 0.7 (95%CI: 0.68-0.75) to 0.9(95%CI: 0.91-0.96). False positive rates ranged from 0.24% to 30.5%, while false negative rates were 0-3.2%. The risk of bias assessment of the five studies was generally good to excellent. ConclusionsCAD tools show promise in screening people living with diabetes for active TB, but data are scarce, and performance varies across settings.