High AUROC can mask decision failure in sepsis transcriptomic classifiers: Preprocessing stability outweighs post hoc calibration across cohorts
Hongwei Zheng, Wenbiao Chen · 2026
Read High AUROC can mask decision failure in sepsis transcriptomic classifiers: Preprocessing stability outweighs post hoc calibration across cohorts by Hongwei Zheng, Wenbiao Chen (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0357585.txt.
Description
Hongwei Zheng and Wenbiao Chen warn that a popular score can make blood-based sepsis tests look better than they are. Their 2026 benchmark study checks gene-activity classifiers, which sort patients by patterns in blood, across four public patient datasets from different labs and machines. A test can rank patients well yet fail badly when a fixed cutoff decides who counts as sick.
That failure is exactly what they found for two common ways of scaling the data, which fell to coin-flip accuracy on one outside dataset despite glowing headline scores. A method that ranks genes within each single sample held up across all outside sets. Zheng and Chen argue that how data is prepared matters more than later fixes, a lesson for anyone building medical tests.
Catalog record
| CREATOR | Hongwei Zheng, Wenbiao Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52030 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |