Читать книгу Diabetic Retinopathy and Cardiovascular Disease - Группа авторов - Страница 26

Increasingly large datasets are being utilised for the development of risk prediction tools due to advancing technology in a number of fields, such as electronic health records, genomics, proteomics and metabolomics. Such data have posed challenges to traditional regression-based methods for creating predictive models. Analytic challenges include assumptions of linear relationships between risk factors and outcomes, interactions between covariates and large numbers of predictor variables. Machine learning methods have been developed to aid in harnessing the potential predictive capabilities of so-called big data. Machine learning algorithms perform automated stochastic (random) or deterministic searches for the predictive model with optimal fit within certain specified constraints [53]. Typically, the resulting models are less useful for understanding associations or causal relationships than traditional regression analyses, but can have stronger predictive capabilities.

Electronic health records have been rapidly adopted in recent years and potentially offer a wealth of data for use in risk prediction research. However, much of this information is stored as unstructured data. This has led to the development of information extraction and data mining systems which search electronic records and identify relevant cardiovascular risk factor data [54]. These systems utilise a combination of machine learning and rule based clinical text mining techniques.

Cardiovascular risk prediction models utilising machine learning are yet to be adopted into routine clinical practice and international guidelines. Nevertheless, there is ever increasing research using non-traditional data sources and potential novel prognostic biomarkers which is likely to impact clinical risk prediction and treatment decision making in the coming years.