Another pub from one of the kings of networks.
Has a huge clinical database. Makes a network of most stated diseases in database, where a link indicates that the two diseases occur in the same patient at a higher than chance frequency.
He uses two measures of association which have opposite biases.
Relative Risk (RR)-- overestimates relationships involving rare diseases while discounting common ones
Phi-correlation (phi) accurate for diseases with the same frequency, discounts comorbidity between rare and common disease
The network's predicitive power is on the same order as family studies and/or some genetic studies.
Some discussion on inferring directionality
No comments:
Post a Comment