Machine learning methodologies versus cardiovascular risk scores, in predicting disease risk

• DOI: 10.1186/s12874-018-0644-1
• Publication Date: 2018-12-29T10:49:59Z
• Authors: Alexandros C. Dim...; Mara Nikolaidou; Francisco Félix C...; Worrawat Engchuan; Albert Sanchez-Niubo; Holger Arndt; José Luis Ayuso-M...; Josep Maria Haro; Somnath Chatterji; Ekavi N. Georgous...; Christos Pitsavos; Demosthenes B. Pa...
• Keywords: Bootstrapping (finance); Predictive modelling

The use of Cardiovascular Disease (CVD) risk estimation scores in primary prevention has long been established. However, their performance still remains a matter concern. aim this study was to explore the potential using ML methodologies on CVD prediction, especially compared established tool, HellenicSCORE. Data from ATTICA prospective (n = 2020 adults), enrolled during 2001–02 and followed-up 2011–12 were used. Three different machine-learning classifiers (k-NN, random forest, decision tree) trained evaluated against 10-year incidence, comparison with HellenicSCORE tool (a calibration ESC SCORE). Training datasets, consisting 16 variables only 5 variables, chosen, or without bootstrapping, an attempt achieve best overall for machine learning classifiers. Depending classifier training dataset outcome varied efficiency but comparable between two methodological approaches. In particular, showed accuracy 85%, specificity 20%, sensitivity 97%, positive predictive value 87%, negative 58%, whereas methodologies, ranged 65 84%, 46 56%, 67 89%, 89 91%, 24 45%; forest gave results, while k-NN poorest results. alternative approach classification produced results that prediction and, thus, it can be used as method taking into consideration advantages may offer.