Merge pull request #1447 from MIT-LCP/gcs_fix

Fix bug in calculation of first day GCS

Author: alistairewj

mimic-iv/concepts/firstday/first_day_gcs.sql

@@ -13,15 +13,25 @@
 WITH gcs_final AS
 (
     SELECT
-        gcs.*
+        ie.subject_id, ie.stay_id
+        , g.gcs
+        , g.gcs_motor
+        , g.gcs_verbal
+        , g.gcs_eyes
+        , g.gcs_unable
         -- This sorts the data by GCS
         -- rn = 1 is the the lowest total GCS value
         , ROW_NUMBER () OVER
         (
-            PARTITION BY gcs.stay_id
-            ORDER BY gcs.GCS
+            PARTITION BY g.stay_id
+            ORDER BY g.GCS
         ) as gcs_seq
-    FROM `physionet-data.mimiciv_derived.gcs` gcs
+    FROM `physionet-data.mimiciv_icu.icustays` ie
+    -- Only get data for the first 24 hours
+    LEFT JOIN `physionet-data.mimiciv_derived.gcs` g
+        ON ie.stay_id = g.stay_id
+        AND g.charttime >= DATETIME_SUB(ie.intime, INTERVAL '6' HOUR)
+        AND g.charttime <= DATETIME_ADD(ie.intime, INTERVAL '1' DAY)
 )
 SELECT
     ie.subject_id

mimic-iv/tests/test_first_day.py

@@ -0,0 +1,40 @@
+import pandas as pd
+from pandas.io import gbq
+
+def test_gcs_first_day_calculated_correctly(dataset, project_id):
+    """Verifies GCS first day values are calculated correctly."""
+    # almost every individual should have a GCS first day
+    query = f"""
+    SELECT COUNT(*) AS n, COUNT(g.gcs) AS n_gcs
+    FROM  {dataset}.first_day_gcs g
+    """
+    df = gbq.read_gbq(query, project_id=project_id, dialect="standard")
+    n, n_gcs = df.iloc[0, 0], df.iloc[0, 1]
+    frac = float(n_gcs) / n * 100.0
+    assert frac > 98, 'less than 98%% of stays have a first day GCS'
+
+
+    # verify a subset of values
+    known_values = {
+        37535507: {'gcs': 13, 'gcs_motor': 4, 'gcs_verbal': None, 'gcs_eyes': None},
+        38852627: {'gcs': None, 'gcs_motor': None, 'gcs_verbal': None, 'gcs_eyes': None},
+        32435143: {'gcs': 8, 'gcs_motor': 5, 'gcs_verbal': 1, 'gcs_eyes': 2},
+    }
+    query = f"""
+    SELECT g.stay_id
+    , g.gcs
+    , g.gcs_motor
+    , g.gcs_verbal
+    , g.gcs_eyes
+    , g.gcs_unable
+    FROM  {dataset}.first_day_gcs g
+    WHERE g.stay_id IN
+    (
+        {','.join([str(x) for x in known_values.keys()])}
+    )
+    """
+    df = gbq.read_gbq(query, project_id=project_id, dialect="standard")
+    df = df.sort_values(['stay_id']).set_index('stay_id')
+    for stay_id, row in df.iterrows():
+        for col, expected_val in known_values[stay_id].items():
+            assert row[col] == expected_val, f'first_day_gcs {col} value incorrect for stay_id={stay_id}'