Breaking the threshold: Developing multivariable models using computer-aided chest X-ray analysis for tuberculosis triage

The researchers pooled individual patient data from four studies, including 4,733 participants with complete covariate data. They found that multivariable models demonstrated excellent performance, with areas under the receiver operating characteristic curve of 0.91 for software A and 0.92 for software B. Compared to threshold-based CAD classification, the multivariable models increased specificity at 90% sensitivity. The conclusion suggests that using CAD scores in multivariable models outperforms the current practice of CAD-threshold-based chest X-ray classification for TB diagnosis.