added the ability to link visits by their ids instead of just their timestamps, which should work better in case dates are wrong or noisy in mimic3 (#547)

Co-authored-by: John Wu <johnwu3@sunlab-work-01.cs.illinois.edu>

Honestly, this is really bad practice, but we need to push this forward, because I need this merge for another PR lol

Author: jhnwu3

pyhealth/datasets/configs/mimic3.yaml

@@ -54,14 +54,22 @@ tables:
           - "dischtime"
     timestamp: "dischtime"
     attributes:
+      - "hadm_id"
       - "icd9_code"
       - "seq_num"
 
   prescriptions:
     file_path: "PRESCRIPTIONS.csv.gz"
     patient_id: "subject_id"
     timestamp: "startdate"
+    join:
+      - file_path: "ADMISSIONS.csv.gz"
+        "on": "hadm_id"
+        how: "inner"
+        columns:
+          - "dischtime"
     attributes:
+      - "hadm_id"
       - "drug"
       - "drug_type"
       - "drug_name_poe"
@@ -88,6 +96,7 @@ tables:
           - "dischtime"
     timestamp: "dischtime"
     attributes:
+      - "hadm_id"
       - "icd9_code"
       - "seq_num"
 
@@ -116,9 +125,16 @@ tables:
   noteevents:
     file_path: "NOTEEVENTS.csv.gz"
     patient_id: "subject_id"
+    join:
+      - file_path: "ADMISSIONS.csv.gz"
+        "on": "hadm_id"
+        how: "inner"
+        columns:
+          - "dischtime"
     timestamp:
       - "charttime"
     attributes:
+      - "hadm_id"
       - "text"
       - "category"
       - "description"