Rethinking input complexity in transformer-based clinical prediction: implications for feature dimensionality and sequence length in longitudinal electronic health record data
Wansu Chen, Botao Zhou, Robert S Zeiger, William Crawford, Michael Schatz, Eric Puttock, Stanley Xu, Matthew T Slaughter, Fagen Xie · 2026
Read Rethinking input complexity in transformer-based clinical prediction: implications for feature dimensionality and sequence length in longitudinal electronic health record data by Wansu Chen, Botao Zhou, Robert S Zeiger, William Crawford, Michael Schatz, Eric Puttock, Stanley Xu, Matthew T Slaughter, Fagen Xie (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/PMC13625019.txt.
Description
Wansu Chen and colleagues question a common belief in medical prediction software, that more patient data always helps. Working with health records of adults with mild asthma, they trained transformer models to forecast sudden asthma attacks. They then trimmed the input two ways, cutting the feature list by several selection methods and shortening each patient's history, and weighed accuracy against computing cost.
Leaner models forecast nearly as well as the full eighty-feature model, and they flagged much the same high-risk patients. Shortening histories cut training time by more than seventy percent with little loss of sharpness. The catch sits at the top end, where lean models softened the highest risk estimates, a trade-off builders of hospital prediction tools should weigh.
Catalog record
| CREATOR | Wansu Chen, Botao Zhou, Robert S Zeiger, William Crawford, Michael Schatz, Eric Puttock, Stanley Xu, Matthew T Slaughter, Fagen Xie |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |