| Title: | Developing a look a like model for the prediction of premature mortality and chronic disease |
| Lead Institution: | University of Edinburgh |
| Principal investigator: | Dr Filippo Menolascina |
Application 54045
WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
About
Identifying patients mortality risks and in particular the impact of chronic disease is key to preventative medicine. In recent years a number of machine learning algorithms have been applied to this problem. In this study, we will attempt to validate look-a-like modelling as an alternative approach. Look-a-like modelling has been hugely successful in other industries. It enables new inputs to be categorised based on their similarities to known positive examples. Unlike other machine learning models, look-a-like models are not considered black boxes (that is to say they can be easily interpreted). There is every reason to believe that they should be able to identify patients at risk of premature death or chronic disease. We propose a systematic comparison of look-a-like modelling and machine learning approaches to risk prediction (Deep learning, Random Forest and Cox regression). Machine learning techniques have already achieved superhuman performance in many fields of diagnostic medicine, De Fauw et al (2018). The continual evaluation of novel approaches is essential to this process. Accurate risk prediction algorithms should facilitate early and effective interventions, when treating disease. We require access to clinical, medical history and sociodemographic data from the UK Biobank. We will then perform feature selection using the predictive potential of these data, in particular with regard to specific disease onset. To evaluate these models predictive capabilities we would require access to the full UK Biobank cohort (approx. n=500,000). This would enable us to develop robust models which could be validated on previously unseen data (essential in machine learning).1 Publication
| Pub ID | Title | Author(s) | Year | Journal |
|---|---|---|---|---|
| 12626 | XGBoost, A Novel Explainable AI Technique, in the Prediction of Myocardial Infarction: A UK Biobank Cohort Study | Alexander Moore (+1) | 2022 | Clinical Medicine Insights Cardiology |
Enabling scientific discoveries that improve human health