Abstract
Clinical prediction models are statistical tools that can be used to estimate the probability of a patient to either have a specific outcome or to develop an outcome in time. This
probability is estimated based on patient or disease-specific input variables. It provides insights into the diagnosis (e.g. disease status) or prognosis (e.g. 5-year survival probability) of a patient, and can subsequently be used to support (shared) decision-making regarding the optimal management of the disease. Prediction models are developed and evaluated using data from patients that can be classified in similar patient groups (e.g. diagnosed with estrogen receptor positive breast cancer), but with varying disease characteristics (e.g. tumor stage, treatment received, nodal involvement etc.).
Before the available models are used to support in routine healthcare decision-making some challenges on the identification of currently existing models (accessibility), review
of the quality of the models (transparency), assessment how well they perform on external validation (generalizability), and investigation of the potential benefit of recalibrating the validated models (updating). Subsequently, models showing adequate performance will be ready for implementation in clinical practice after clearly defined intended model use is described (interpretation), and the intended model use is substantiated by evidence regarding added value (impact assessment).
In this thesis, multiple studies aiming to overcome the challenges are described using examples on breast and prostate cancer. Since breast and prostate cancer are among
the top three most commonly diagnosed cancers in women and men, respectively, there is a large amount of data available to establish clinical prediction models for patients
diagnosed with breast or prostate cancer. Currently available models for breast and prostate cancer are required to be critically assessed to demonstrate which models are
valuable and which information is still lacking when used in Dutch care.
probability is estimated based on patient or disease-specific input variables. It provides insights into the diagnosis (e.g. disease status) or prognosis (e.g. 5-year survival probability) of a patient, and can subsequently be used to support (shared) decision-making regarding the optimal management of the disease. Prediction models are developed and evaluated using data from patients that can be classified in similar patient groups (e.g. diagnosed with estrogen receptor positive breast cancer), but with varying disease characteristics (e.g. tumor stage, treatment received, nodal involvement etc.).
Before the available models are used to support in routine healthcare decision-making some challenges on the identification of currently existing models (accessibility), review
of the quality of the models (transparency), assessment how well they perform on external validation (generalizability), and investigation of the potential benefit of recalibrating the validated models (updating). Subsequently, models showing adequate performance will be ready for implementation in clinical practice after clearly defined intended model use is described (interpretation), and the intended model use is substantiated by evidence regarding added value (impact assessment).
In this thesis, multiple studies aiming to overcome the challenges are described using examples on breast and prostate cancer. Since breast and prostate cancer are among
the top three most commonly diagnosed cancers in women and men, respectively, there is a large amount of data available to establish clinical prediction models for patients
diagnosed with breast or prostate cancer. Currently available models for breast and prostate cancer are required to be critically assessed to demonstrate which models are
valuable and which information is still lacking when used in Dutch care.
| Original language | English |
|---|---|
| Qualification | Doctor of Philosophy |
| Awarding Institution |
|
| Supervisors/Advisors |
|
| Award date | 22 Jun 2022 |
| Place of Publication | Enschede |
| Publisher | |
| Print ISBNs | 978-90-365-5394-0 |
| DOIs | |
| Publication status | Published - 22 Jun 2022 |
