While the PATIENT table contains basic administrative information about each patient, our assignment is to predict the mortality risk on the basis of each visit. The VISIT table contains one observation for each patient visit, along with some clinical information about each visit:

Notice that the patient ID is no longer the primary identifier of this table, since Patient #1 had two visits; instead, there is a Visit_id field that is numbered from 10001 to 10006 in this example, with one distinct ID per visit.

This table also contains Visit_date. Since the cardiology practice indicated they want to know the mortality risk within 6 months of the patient visit, we will have to use this field later when we compute the target variable.

Two of the fields in this table contain ICD (diagnosis) codes. Actual tables may contain dozens of codes for each visit. For each coded field, there is a corresponding name field that contains the name of the condition that the code represents. A popular approach in healthcare is to make, in the final table, a column for every clinical code we are interested in tracking (Futoma et al., 2015; Rajkomar et al., 2018). We will adopt this approach later in the chapter.

Finally, we note that the name of the attending physician is included, which can be used to measure physician performance.