Statistical methods for prognostic factor and risk prediction research

Whittle, Rebecca Louise

Abstract

Prognosis research is an important part of medical research as it seeks to understand, predict, and improve future outcomes in people with a given disease or health condition. This thesis focuses on the application and development of statistical methods for prognosis research, with a particular focus on the identification of prognostic factors and the performance of risk prediction models.
The first part of the thesis considers the use of a single study for prognostic factor and prediction model research. Prognostic factors of adverse outcome in monochorionic diamniotic twin pregnancies are investigated and difference in nuchal translucency and crown-rump length were found to have prognostic value. The instability of developing a prediction model in small sample sizes is also illustrated. Then, a review of published prediction models is conducted which reveals potential concerns that measurement error may affect the predictors included in many models, and a lack of clarity about the timing of predictor measurements and the intended moment of using the proposed models. Recommendations for improved reporting are provided. A real example is then used to illustrate how displacing the collection of a time-varying predictor from the intended moment of model use leads to substantial differences in the predictor-outcome association, and the subsequent performance of the prediction model.
The second part of the thesis focuses on the synthesis of IPD from multiple studies. An IPD meta-analysis is used to validate existing stillbirth prediction models and demonstrates that the models should not be recommended for clinical practice due to poor predictive performance and insufficient clinical utility. Finally, a novel analytic method is developed to calculate the power of an IPD meta-analysis to examine prognostic factor effects with binary outcomes, based on published study aggregate data, to help researchers decide on the benefit of the IPD approach in advance of collecting IPD.

Citation

Whittle, R. L. (2023). Statistical methods for prognostic factor and risk prediction research. (Thesis). Keele University

• Thesis Type: Thesis
• Deposit Date: Jul 11, 2023
• Publicly Available Date: Jul 11, 2023
• Award Date: 2023-06