Dr Malcolm Kendrick recently discussed a paper in which computers analysed routine clinical data from UK GP practices to identify the factors that most accurately predicted a cardiovascular event over the next ten years. All the 378,256 people whose records were analysed were initially free of cardiovascular disease and 48 variables were identified.
The top ten things that were most likely to see you in hospital with a heart attack or stroke, in order, were:
Chronic Obstructive Pulmonary Disease
Prescribed oral steroids
Severe mental illness
South Asian ethnicity
Chronic Kidney Disease
The least predictive were LDL, Forced expiratory volume ( a measure of asthma) and AST/ALT ( a measure of liver function). Total cholesterol was 25th.
Can machine learning improve cardiovascular risk prediction using routine clinical data? http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0174944