Data is usually organized in a database as separate tables that may be bound together by common patient or encounter identifiers. Machine learning algorithms usually work on a single data structure at a time. Therefore, combining and merging the data from several tables into one final table is an important task. Along the way, you'll have to make some decisions as to which data to preserve (demographic data is usually indispensable) along with which data you can safely forget (the exact timestamps of anti-asthmatic medication administrations may not be important if you are trying to predict cancer onset, for example).