Methods for prediction in cancer: prognosis and treatment effects

This thesis focuses in Part 1 on predicting how cancer develops (prognosis) and in Part 2 on the effects of treatments in cancer patients using statistical models. In Part 1, life expectancy was estimated by analyzing nationwide population-based data from Dutch cancer patients diagnosed between 1989 and 2019. The results indicated that life expectancy has improved due to better detection and treatments, although certain cancers still have a poor prognosis. For melanoma patients who underwent sentinel lymph node biopsy, this thesis developed a predictive model to estimate the risk of recurrence or death and the risk of death due to melanoma within five years.

In Part 2, the systematic review shows that patients may benefit from making treatment decisions by considering personalized variation in treatment effects, rather than relying solely on the overall effect. Specifically, for patients with diffuse large B-cell lymphoma, this thesis demonstrated that intensified treatment is particularly beneficial for high-risk patients. Additionally, this thesis made methodological advances in quantifying individualized treatment effects and evaluating models that estimate these effects.

This thesis makes five recommendations: 1) use life expectancy as a measure for prognosis and treatment effects, 2) use simple models, 3) update existing prediction models before developing a new one but validate the new model if you do, 4) strive to make as many personalized predictions for patients as possible, and 5) critically evaluate data analysis to ensure that accurate information influences medical decisions. These insights support clinicians and patients in making better-informed decisions.