Validation of prediction models in epidemiology and medicin (2024)

Welcome to Validation of prediction models in epidemiology and medicin 

PhD Program: Epidemiology and Biostatistics

Description:

While most medical studies aim to explain some phenomenon, a significant proportion does not have this as a primary goal. These studies instead aim to predict a certain event or measure as accurately as possible given a number of predictors. Although explaining and predicting are two separate goals, they are often interchanged in medical studies. More importantly, the statistical approaches used for these two types of data analyses are not the same. The course covers the following topics:

1. The basic differences between explanatory and predictive studies.
2. Model estimation: variable selection, variable predictive power and penalisation.
3. Model Performance: N-fold cross validation, bootstrap, geographical and temporal validation, external validation.
4. Model validity: Calibration and calibration slope, receiver operating characteristic (ROC) curve incl. area under the curve (AUC), Bland-Altman plots, prediction intervals and decision curve analysis.
5. Sample size estimation for development as well as for validation.
6. Examples from the scientific literature.

Upon completion of the course, the student will be able to distinguish between predictive and explanatory data analyses, as well as understand the basic statistical tools used in predictive studies.

Students will be evaluated based on an assignment with oral presentation in groups of 2-4 participants.

Literature (in order of priority):

Wynants L, Collins GS, Van Calster B. Key steps and common pitfalls in developing and validating risk models. BJOG. 2017 Feb;124(3):423-432. doi: 10.1111/1471-0528.14170. Epub 2016 Jun 30. PMID: 27362778.

Moons K, Royston P, Vergouwe Y, Grobbee D, Altman D. Prognosis and prognostic research: what, why and how? BMJ2008;b375.

Ewout W. Steyerberg, Yvonne Vergouwe; Towards better clinical prediction models: seven steps for development and an ABCD for validation, European Heart Journal, Volume 35, Issue 29, 1 August 2014, Pages 1925–1931, https://doi.org/10.1093/eurheartj/ehu207

Riley RD, Ensor J, Snell KIE, Harrell FE Jr, Martin GP, Reitsma JB, Moons KGM, Collins G, van Smeden M. Calculating the sample size required for developing a clinical prediction model. BMJ. 2020 Mar 18;368:m441. doi: 10.1136/bmj.m441. PMID: 32188600.