This might be one of the most rudimentary scoring systems that I have ever seen, yet I love it; it's based on powerful predictors, and easy to interpret, measure and understand. However, it somewhat lacks cross-factor interactions (well, these interactions are modeled as additive and independent). Probably a good approximation though, I'd like to see how better more complex models do, compared with this one. Have you seen successful applications of rudimentary scoring systems like this one, in other fields? Of course, the apparent simplicity of this model hides the fact that the base metrics are carefully selected, and the binning (the six-bin selection, see below) done properly. This score results in 6 bins, and each bin each has certainly enough data to make the score a good predictor of stroke. Such simple scoring system, if proven accurate enough, are great examples of methods to be integrated in a a general automated, black-box data science package (we are working on this).
The following (explaining the score in question) is an extract from Wikipedia.
The ABCD2 score is a useful clinical prediction rule to determine the risk for stroke in the days following a transient ischemic attack (TIA, a condition in which temporary brain dysfunction results from oxygen shortage in the brain). Here's the score table:
For example, a person aged 60 (1 point) with normal blood pressure (0 point) and without diabetes (0 point) who experienced a TIA lasting 10 minutes (1 point) with a speech disturbance but no weakness on one side of the body (1 point) would score a total of 3 points.
The risk for stroke can be estimated from the ABCD2 score as follows: