Cell-of-origin classification is an established component of diffuse large B-cell lymphoma characterization, but the optimal testing approach remains debated. Gene expression profiling is often treated as the reference method, while the Hans immunohistochemistry algorithm is widely accessible and can be performed on routine formalin-fixed tissue. Conventional accuracy comparisons are difficult, however, because gene expression platforms themselves vary and there is no perfect gold standard against which every test can be judged.

The analysis assembled 19 publications comprising 2,247 tumors, including evidence comparing Hans immunohistochemistry with gene expression profiling and evidence comparing different gene expression profiling platforms. A Bayesian latent-class model was used to estimate an underlying cell-of-origin classification without assuming that any single platform was perfectly correct. Estimated diagnostic accuracy was 90% for the Hans algorithm and 95% for pooled gene expression profiling. The estimated accuracy difference was 4.5 percentage points, with the credible interval remaining within the prespecified 10% non-inferiority margin. Specificity and positive predictive value also met the margin. Sensitivity and negative predictive value narrowly crossed it at the upper credible-interval boundary.

The methodological contribution is as important as the pooled estimate. By combining immunohistochemistry-versus-gene-expression and gene-expression-versus-gene-expression evidence within a latent-class framework, the analysis addresses a major weakness of earlier comparisons: the absence of a single error-free reference standard. It does not treat a variable reference platform as error-free. The findings support the Hans algorithm as a potentially non-inferior approach for overall cell-of-origin classification accuracy, while preserving important nuance around individual performance metrics for which non-inferiority was not demonstrated as consistently.