Clinical Prediction Models
Topics covered: statistical methods, machine-learning methods, developing prediction models, validation of prediction models, impact of prediction models (basic knowledge of R required)
Clinical prediction models are modern tools that provide the likely diagnosis or prognosis of an individual patient based on their data. This supports patients and doctors when they have to make evidence-based and personalized decisions regarding tests, treatments or lifestyle changes a patient may need. Clinical prediction models form a key aspect of “artificial intelligence in health”. Nowadays, researchers and private companies propose far more clinical prediction models than are actually used in clinical practice. For example, there are 408 models for COPD prognosis, 363 models for cardiovascular disease in the general population, and during the first 5 months of the covid-19 pandemic 236 models have been proposed. To understand which models could improve care, epidemiologists need to know about the key steps in prediction model development and assessment, potential pitfalls, and frequent causes of bias. In this elective, we cover the basics of statistical and popular machine learning methods to develop clinical prediction models. We also discuss how to evaluate the predictive performance (“validation”) and impact (on clinical practice and patient outcomes) of such models.
Assessment: group work, individual contribution to group work, individual paper