add neuroblock to pinned data files

analysis/add-neuroblock.py

@@ -0,0 +1,161 @@
+# Import libraries
+from __future__ import print_function
+
+import pandas as pd
+import psycopg2
+import getpass
+import argparse
+
+from collections import OrderedDict
+
+# define the queries used to get neuromuscular blocks
+queries = {"eicu": """
+set search_path to public,eicu_crd;
+with has_vent as
+(
+  select
+        distinct p.patientunitstayid
+  FROM vent_unpivot_rc p
+  -- only include settings before 24 hours
+  WHERE p.chartoffset <= 24*60
+  AND p.peakpressure IS NOT NULL
+)
+, pt_in_hosp as
+(
+  select pt.hospitalid
+  , sum(neuroblock_day1) as num_neurblock
+  , sum(case when v.patientunitstayid is not null then 1 else 0 end) as num_vent
+  , sum(case when v.patientunitstayid is not null and neuroblock_day1 = 1 then 1 else 0 end) as num_vent_and_nb
+  , count(pt.patientunitstayid) as num_pat_total
+  from patient pt
+  inner join mp_neuroblock nb
+    on pt.patientunitstayid = nb.patientunitstayid
+  left join has_vent v
+    on pt.patientunitstayid = v.patientunitstayid
+  group by pt.hospitalid
+)
+, hosp_in_cohort as
+(
+  select distinct hospitalid
+  from mp_cohort
+  where exclusion_no_peak_pressure = 0
+)
+-- define what hospitals we keep:
+--  1) at least 10 patients admitted
+--  2) have good vent data (defined by hosp_in_cohort)
+--  3) at least 3% of patients are on neuromuscular blockade
+, hosp_keep as
+(
+    select
+        h.hospitalid
+        , ROUND(100.0*num_neurblock::NUMERIC/num_pat_total,2) as percent_nb
+    from pt_in_hosp h
+    LEFT JOIN hosp_in_cohort co
+      ON h.hospitalid = co.hospitalid
+    WHERE num_pat_total >= 10
+    AND co.hospitalid is not null
+    AND ROUND(100.0*num_neurblock::NUMERIC/num_pat_total,2) >= 3
+)
+select
+    d.patientunitstayid
+  , d.hospitalid
+  , d.neuroblock_day1
+  , d.neuroblock_day2
+FROM mp_data d
+INNER JOIN hosp_keep h
+    ON d.hospitalid = h.hospitalid
+WHERE patientunitstayid in
+(
+    select patientunitstayid
+    from mp_cohort
+    where GREATEST(exclusion_non_adult, exclusion_secondary_hospital_stay, exclusion_secondary_icu_stay, exclusion_by_apache, exclusion_no_rc_data, exclusion_trach, exclusion_not_vent_48hr, exclusion_no_peak_pressure) = 0
+)
+ORDER BY patientunitstayid
+""",
+'mimic': """
+set search_path to public,mimiciii;
+with vw1 as
+(
+  select co.icustay_id
+  , max(CASE WHEN coalesce(nb1.icustay_id, nb2.icustay_id) IS NOT NULL THEN 1 ELSE 0 END) as neuroblock_day1
+  from mpwr_cohort co
+  LEFT JOIN neuroblock_dose nb1
+    ON  nb1.icustay_id = co.icustay_id
+    and nb1.starttime >  co.starttime_first_vent - interval '1' hour
+    and nb1.starttime <= co.starttime_first_vent + interval '24' hour
+  LEFT JOIN neuroblock_dose nb2
+    ON  nb2.icustay_id = co.icustay_id
+    and nb2.starttime <  co.starttime_first_vent - interval '1' hour
+    and nb2.endtime   >= co.starttime_first_vent - interval '1' hour
+  GROUP BY co.icustay_id
+)
+-- day 2
+, vw2 as
+(
+  select co.icustay_id
+  , max(CASE WHEN coalesce(nb1.icustay_id, nb2.icustay_id) IS NOT NULL THEN 1 ELSE 0 END) as neuroblock_day2
+  from mpwr_cohort co
+  LEFT JOIN neuroblock_dose nb1
+    ON  nb1.icustay_id = co.icustay_id
+    and nb1.starttime >  co.starttime_first_vent + interval '24' hour
+    and nb1.starttime <= co.starttime_first_vent + interval '48' hour
+  LEFT JOIN neuroblock_dose nb2
+    ON  nb2.icustay_id = co.icustay_id
+    and nb2.starttime <  co.starttime_first_vent + interval '24' hour
+    and nb2.endtime   >= co.starttime_first_vent + interval '24' hour
+  GROUP BY co.icustay_id
+)
+-- data has exclusions applied
+select
+      d.icustay_id
+    , neuroblock_day1
+    , neuroblock_day2
+from mpwr_data d
+LEFT JOIN vw1
+  on d.icustay_id = vw1.icustay_id
+LEFT JOIN vw2
+  on d.icustay_id = vw2.icustay_id
+"""
+}
+
+parser = argparse.ArgumentParser(description='Add year to a dataset')
+parser.add_argument('data', default='mimic',
+                    help='filename of data that needs to have year')
+
+args = parser.parse_args()
+
+if args.data in ('mimic', 'data_mimic', 'data_mimic.csv'):
+    join_id = 'icustay_id'
+    dbname = 'mimic'
+    fn = '../data/pin_mimic.csv'
+    fn_out = '../data/mimic_neuroblock.csv'
+    query = queries['mimic']
+
+elif args.data in ('eicu', 'pin_eicu', 'pin_eicu.csv', 'data_eicu', 'data_eicu.csv'):
+    join_id = 'patientunitstayid'
+    dbname = 'eicu'
+    fn = '../data/pin_eicu.csv'
+    fn_out = '../data/eicu_neuroblock.csv'
+    query = queries['eicu']
+
+# Connect to local postgres version of mimic
+sqluser = getpass.getuser()
+con = psycopg2.connect(dbname=dbname, user=sqluser)
+print('Connected to postgres {}.{}.{}!'.format(int(con.server_version/10000),
+                                              int((con.server_version - int(con.server_version/10000)*10000)/100),
+                                              int(con.server_version - int(con.server_version/100)*100)))
+
+df = pd.read_csv(fn, header=0)
+nb = pd.read_sql_query(query, con)
+
+print('{} patients in original cohort.'.format(df.shape[0]))
+
+df = df.merge(nb, how='inner', on=join_id)
+for c in df.columns:
+    if c[-3:] == '_nb':
+        df.drop(c, axis=1, inplace=True)
+
+print('{} patients kept after exclusions.'.format(df.shape[0]))
+print('{} patients have neuroblock on day1.'.format(df['neuroblock_day1'].sum()))
+print('{} patients have neuroblock on day2.'.format(df['neuroblock_day2'].sum()))
+df.to_csv(fn_out, index=False)