Machine Learning as a Method
ML is a tool that is increasing in popularity in data science as it
identifies patterns in the data in order to predict outcomes. The use of
ML in medicine is rapidly expanding with the development of large,
scale, digitized medical data that could be used with ML for clinical
prediction. ML is a well-suited option for this STS-based study for many
reasons. First, ML is considered ”data-driven knowledge discovery” and
as such, no prior knowledge is needed of a predictor’s weight. The STS
database contains numerous low-granularity predictors that have unclear
significance. Using ML allowed the authors to use any variable that
exists within the database knowing that modelling will be able to
exclude those that are not significant for prediction. While, with
traditional statistical methods, the addition of numerous predictors
introduces collinearity, this is managed by ML as modelling is not based
on linear relationships. As such, in this study, 53 predictors of
unknown significance were used with 3 ML models to predict recurrent MR
and death with great success.