How to develop a more accurate risk prediction model when there are few events.
Pavlou M, Ambler G, Seaman SR, Guttmann O, Elliott P, King M, Omar RZ. · 2015
Read How to develop a more accurate risk prediction model when there are few events. by Pavlou M, Ambler G, Seaman SR, Guttmann O, Elliott P, King M, Omar RZ. (2015) 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/europmc-pmc4531311.txt.
Description
Menelaos Pavlou and six co-authors explain how to build sound risk models from sparse data, in The BMJ in 2015. Doctors use risk models to forecast a patient's chance of an outcome and guide treatment choices. When the outcome is rare but many predictors are fed in, standard regression overfits: it memorizes the sample, then rates low-risk patients too low and high-risk ones too high.
Penalized regression, which shrinks the weight given to each predictor, counteracts that pull. The paper shows, with theory and examples, that penalized methods forecast better than standard ones when events are few. It also sets out how to judge whether a model is truly more accurate, so sparse-data models can be trusted on new patients.
Catalog record
| CREATOR | Pavlou M, Ambler G, Seaman SR, Guttmann O, Elliott P, King M, Omar RZ. |
|---|---|
| YEAR | 2015 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |