Clinical Prediction Modelling (online)
**POST-PONED**
We are currently rescheduling the event, which will be online, soon.
Location: Remotely by invite. Contact use via rssleedsbradfordgroup@gmail.com
14:45-14:55 Login
14:55-15:00 Chair of the Leeds/Bradford Group
Introductions
15:00-15:40 Dr Glen Martin, Division of Informatics, Imaging and Data Science, University of Manchester
Opportunities and challenges in developing prediction models: time for a multivariate approach?
(View recording of presentation)
Clinical prediction models (CPMs) are used to predict the risk of clinically relevant outcomes or events (e.g. the risk of developing cardiovascular disease within ten years). They are increasingly used to support clinical decisions, yet they seldom reflect the interplay between developing multiple co-morbidities in terms of both pathophysiologic and treatment interactions. Specifically, the typical situation is to develop a CPM for one particular outcome, but this fails to capture multi-outcome patterns evolving over time. With rising emphasis on the prediction of multi-morbidity, there is a growing need for CPMs to simultaneously predict risks for each of multiple future outcomes.
A common approach to multi-outcome risk prediction is to derive a CPM for each outcome separately, then multiply the predicted risks. This approach is only valid if the outcomes are conditionally independent given the covariates, and it fails to exploit the potential relationships between the outcomes. In this talk, I will introduce the notion of a “CPM-Network”: an approach to clinical prediction, where the models can appropriately predict the risk of multiple outcomes simultaneously. In other words, I will advocate a move towards a more holistic approach to risk prediction that reflects the multi-dimensional aspects of diseases, co-morbidities and outcomes.
15:40-16:20 Dr David Jenkins, Division of Informatics, Imaging and Data Science, University of Manchester
Moving towards a learning prediction system
Over recent years clinical prediction models (CPMs) have become fundamental for risk stratification across the healthcare system. Commonly, the CPM pipeline (i.e. development and validation) is viewed as a one-time activity. However, healthcare is constantly evolving with changing practice and care pathways; since static CPMs ignore this natural evolution, their performance often worsens over time. CPMs need constant surveillance, embedded within appropriate feedback loops to generate an evolving model that retains high predictive performance.
Dynamic prediction models allow us to do this by constantly learning as new data arrives and generating a model that evolves with the data. Such a model can react to changes in the data in real-time, reducing the latency between the change in data and change in practice. This, however, presents challenges for how to validate models. Indeed, the notion of a validated model does not exist as this implies a one-time validation process: rather, the validation process must also be continuous. In this talk I will introduce dynamic modelling methods and propose the move away from static model development and validation, leading us to the notion of a learning prediction system
16:20-17:00 John Mbotwa, Malawi University of Science and Technology & University of Leeds
Predicting survival in patients with Chronic Heart Failure: A Latent Class Regression Modelling Approach
Most prediction models used in medical research fail to accurately predict health outcomes due to methodological limitations. The Cox Proportional Hazard (PH) model is one of the most popular model in survival analysis. The Cox PH model assumes that the ratio of the hazards for any two individuals within the sample remain constant over time (i.e. proportional hazard assumption). The proportional hazard assumption is often violated in the presence of unobserved heterogeneity. Risk prediction models that fail to account for the presence of heterogeneity tend to yield biased estimates about the underlying risk.
In my talk, I will introduce the use of a Cox PH model within a latent class framework to model survival of patients with chronic heart failure and demonstrate that the proposed latent class approach outperforms other modelling approaches that are commonly used.
17:00 - 17:15: Wrap-up
The committee and our speakers will stay online for a few more minutes to wrap up any discussions