Decision models which have not been rigorously derived and prospectively validated should be used with caution, if at all.  Many seemingly promising models fail the test of validation.   One reason is over modeling, which captures the idiosyncrasies of the derivation population, limiting transportability to different populations. 

Frequently, initial reports of clinical prediction tools include both a derivation set and a validation set.  Thus the rule may be derived from the first group of consecutive patients then tested on subsequent patients, within the same population.  Alternatively, a random sample of the derivation population may be tested.  These are termed "data splitting" methods of validation. 

The following data, from a decision model for predicting MI in patients with LBBB, shows a typical presentation for this type of data. 

More sophisticated, and complex, methods for validation utilizing the derivation set include jacknike and bootstrap methods. 
 

However, the best validation studies are conducted on independent samples in a different population.   Without this type of evidence, use caution.  To find validation studies try searching the SI Citation Database.  Using this search tool you can find all other references which have cited a particular reference