Prognostic models: What the statistician wants the clinician to know.

Publisher: Elsevier Country of Publication: Netherlands NLM ID: 101120605 Publication Model: Print-Electronic Cited Medium: Internet ISSN: 1532-1916 (Electronic) Linking ISSN: 15216918 NLM ISO Abbreviation: Best Pract Res Clin Gastroenterol Subsets: MEDLINE

Publication: 2004- : Amsterdam : Elsevier
Original Publication: London : Baillière Tindall, c2001-

Prognosis* ; Models, Statistical*; Humans

Prognostic model building is a process that begins much earlier than data analysis and ends later than when a model is reached. It requires careful delineation of a clinical question, methodical planning of the approach and attentive exploration of the data before attempting model building. Once following these important initial steps, the researcher may postulate a model to describe the process of interest and build such model. Once built, the model will need to be checked, validated and the exercise may take the researcher back a few steps - for instance, to adapt the model to fit a variable that displays a 'curved' pattern - to then return to check and validate the model again. To interpret and report the results it is vital to relate the output to the original question, to be transparent in the methodology followed and to understand the limitations of the data and the approach.
Competing Interests: Declaration of competing interest None.
(Copyright © 2023 Elsevier Ltd. All rights reserved.)

Keywords: Cox proportional hazards; Logistic regression; Prognosis; Statistical modelling

Date Created: 20231216 Date Completed: 20231228 Latest Revision: 20231228

20231229

10.1016/j.bpg.2023.101872

38103928