-
Notifications
You must be signed in to change notification settings - Fork 0
/
01_prepare_study_sample_timestamps.py
472 lines (364 loc) · 38 KB
/
01_prepare_study_sample_timestamps.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
#### Master Script 1: Extract and prepare study sample timestamps from CENTER-TBI dataset ####
#
# Shubhayu Bhattacharyay
# University of Cambridge
# email address: [email protected]
#
### Contents:
# I. Initialisation
# II. Load and filter CENTER-TBI dataset
# III. Characterise ICU stay timestamps and fill in missing timestamp values
# IV. Prepare ICU stay windows for dynamic modelling
# V. Perform summary characteristic POLR analysis
### I. Initialisation
import os
import sys
import time
import tqdm
import glob
import random
import datetime
import warnings
import itertools
import numpy as np
import pandas as pd
import pickle as cp
import seaborn as sns
from scipy import stats
from pathlib import Path
from datetime import timedelta
import matplotlib.pyplot as plt
from collections import Counter
warnings.filterwarnings(action="ignore")
from pandas.api.types import CategoricalDtype
from statsmodels.miscmodels.ordinal_model import OrderedModel
from sklearn.preprocessing import OneHotEncoder, StandardScaler
### II. Load and filter CENTER-TBI dataset
# Load CENTER-TBI dataset to access ICU discharge date/times
CENTER_TBI_demo_info = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/DemoInjHospMedHx/data.csv',na_values = ["NA","NaN"," ", ""])
# Filter patients who were enrolled in the ICU
CENTER_TBI_demo_info = CENTER_TBI_demo_info[CENTER_TBI_demo_info.PatientType == 3]
# Filter patients who are or are above 16 years of age
CENTER_TBI_demo_info = CENTER_TBI_demo_info[CENTER_TBI_demo_info.Age >= 16]
# Filter patients who have non-missing GOSE scores
CENTER_TBI_demo_info = CENTER_TBI_demo_info[~CENTER_TBI_demo_info.GOSE6monthEndpointDerived.isna()]
### III. Characterise ICU stay timestamps and fill missing timestamp values
# Select columns that indicate ICU admission and discharge times
CENTER_TBI_ICU_datetime = CENTER_TBI_demo_info[['GUPI','ICUAdmDate','ICUAdmTime','ICUDischDate','ICUDischTime']].reset_index(drop=True)
# Compile date and time information and convert to datetime
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUAdmDate', 'ICUAdmTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'][CENTER_TBI_ICU_datetime.ICUAdmDate.isna() | CENTER_TBI_ICU_datetime.ICUAdmTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# To fill in missing timestamps, first check all available timestamps in nonrepeatable time info
non_repeatable_date_info = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/NonRepeatableDateInfo/data.csv',na_values = ["NA","NaN"," ", ""]).dropna(axis=1,how='all')
non_repeatable_date_info = non_repeatable_date_info[non_repeatable_date_info.GUPI.isin(CENTER_TBI_demo_info.GUPI)].reset_index(drop=True)
non_repeatable_date_info = non_repeatable_date_info[['GUPI']+[col for col in non_repeatable_date_info.columns if 'Date' in col]].select_dtypes(exclude='number')
non_repeatable_date_info = non_repeatable_date_info.melt(id_vars='GUPI',var_name='desc',value_name='date').dropna(subset=['date']).sort_values(by=['GUPI','date']).reset_index(drop=True)
non_repeatable_date_info['desc'] = non_repeatable_date_info['desc'].str.replace('Date','')
non_repeatable_time_info = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/NonRepeatableTimeInfo/data.csv',na_values = ["NA","NaN"," ", ""]).dropna(axis=1,how='all')
non_repeatable_time_info = non_repeatable_time_info[non_repeatable_time_info.GUPI.isin(CENTER_TBI_demo_info.GUPI)].reset_index(drop=True)
non_repeatable_time_info = non_repeatable_time_info[['GUPI']+[col for col in non_repeatable_time_info.columns if ('Time' in col)|('Hour' in col)]].select_dtypes(exclude='number')
non_repeatable_time_info = non_repeatable_time_info.melt(id_vars='GUPI',var_name='desc',value_name='time').dropna(subset=['time']).sort_values(by=['GUPI','time']).reset_index(drop=True)
non_repeatable_time_info['desc'] = non_repeatable_time_info['desc'].str.replace('Time','')
non_repeatable_datetime_info = pd.merge(non_repeatable_date_info,non_repeatable_time_info,how='outer',on=['GUPI','desc']).dropna(how='all',subset=['date','time'])
non_repeatable_datetime_info = non_repeatable_datetime_info.sort_values(by=['GUPI','date','time']).reset_index(drop=True)
os.makedirs('/home/sb2406/rds/hpc-work/timestamps/',exist_ok=True)
non_repeatable_datetime_info.to_csv('/home/sb2406/rds/hpc-work/timestamps/nonrepeatable_timestamps.csv',index=False)
# To fill in missing timestamps, second check all available timestamps in repeatable time info
repeatable_datetime_info = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/RepeatableDateTimeInfo/data.csv',na_values = ["NA","NaN"," ", ""]).dropna(axis=1,how='all')
repeatable_datetime_info = repeatable_datetime_info[repeatable_datetime_info.GUPI.isin(CENTER_TBI_demo_info.GUPI)].reset_index(drop=True)
repeatable_datetime_info['RowID'] = list(range(1,repeatable_datetime_info.shape[0]+1))
repeatable_datetime_info = repeatable_datetime_info.merge(CENTER_TBI_ICU_datetime,how='left',on='GUPI')
## Fix cases in which a missing date or time can be inferred from another variable
# Hourly values (HV) features
missing_HV_date = repeatable_datetime_info[(repeatable_datetime_info.HVDate.isna())&(~repeatable_datetime_info.HourlyValueTimePoint.isna())&(repeatable_datetime_info.HourlyValueTimePoint != 'None')].reset_index(drop=True)
missing_HV_date = missing_HV_date[['RowID','GUPI','ICUAdmDate','HourlyValueTimePoint','HVDate']]
missing_HV_date['ICUAdmDate'] = pd.to_datetime(missing_HV_date['ICUAdmDate'],format = '%Y-%m-%d')
missing_HV_date['HourlyValueTimePoint'] = missing_HV_date['HourlyValueTimePoint'].astype('int')
missing_HV_date['HVDate'] = (missing_HV_date['ICUAdmDate'] + pd.to_timedelta((missing_HV_date['HourlyValueTimePoint']-1),'days')).astype(str)
for curr_rowID in tqdm.tqdm(missing_HV_date.RowID,'Fixing missing HVDate values'):
curr_Date = missing_HV_date.HVDate[missing_HV_date.RowID == curr_rowID].values[0]
repeatable_datetime_info.HVDate[repeatable_datetime_info.RowID == curr_rowID] = curr_Date
repeatable_datetime_info.HVTime[(repeatable_datetime_info.HVTime.isna())&(~repeatable_datetime_info.HVHour.isna())] = repeatable_datetime_info.HVHour[(repeatable_datetime_info.HVTime.isna())&(~repeatable_datetime_info.HVHour.isna())]
# Daily values (DV) features (remove ward [non-ICU] observations)
ward_daily_vitals = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/DailyVitals/data.csv',na_values = ["NA","NaN"," ", ""])
ward_daily_vitals = ward_daily_vitals[ward_daily_vitals.PatientLocation == 'Ward'].reset_index(drop=True)
ward_daily_vitals = ward_daily_vitals[['GUPI','DVTimepoint','DVDate']]
repeatable_datetime_info = pd.merge(repeatable_datetime_info,ward_daily_vitals,how='left',on=['GUPI','DVTimepoint','DVDate'],indicator=True)
repeatable_datetime_info = repeatable_datetime_info[repeatable_datetime_info._merge != 'both'].drop(columns=['_merge']).reset_index(drop=True)
missing_DV_date = repeatable_datetime_info[(repeatable_datetime_info.DVDate.isna())&(~repeatable_datetime_info.DVTimepoint.isna())&(repeatable_datetime_info.DVTimepoint != 'None')].reset_index(drop=True)
missing_DV_date = missing_DV_date[['RowID','GUPI','ICUAdmDate','DVTimepoint','DVDate']]
missing_DV_date['ICUAdmDate'] = pd.to_datetime(missing_DV_date['ICUAdmDate'],format = '%Y-%m-%d')
missing_DV_date['DVTimepoint'] = missing_DV_date['DVTimepoint'].astype('int')
missing_DV_date['DVDate'] = (missing_DV_date['ICUAdmDate'] + pd.to_timedelta((missing_DV_date['DVTimepoint']-1),'days')).astype(str)
for curr_rowID in tqdm.tqdm(missing_DV_date.RowID,'Fixing missing DVDate values'):
curr_Date = missing_DV_date.DVDate[missing_DV_date.RowID == curr_rowID].values[0]
repeatable_datetime_info.DVDate[repeatable_datetime_info.RowID == curr_rowID] = curr_Date
# Remove follow-up information
repeatable_datetime_info = repeatable_datetime_info[repeatable_datetime_info.TimePoint.isna()].reset_index(drop=True)
# Outcomes
non_baseline_outcomes = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/Outcomes/data.csv',na_values = ["NA","NaN"," ", ""])
non_baseline_outcomes = non_baseline_outcomes[non_baseline_outcomes != 'Base'].reset_index(drop=True)
non_baseline_outcomes = non_baseline_outcomes[['GUPI','Timepoint']]
repeatable_datetime_info = pd.merge(repeatable_datetime_info,non_baseline_outcomes,how='left',on=['GUPI','Timepoint'],indicator=True)
repeatable_datetime_info = repeatable_datetime_info[repeatable_datetime_info._merge != 'both'].drop(columns=['_merge']).reset_index(drop=True)
# TIL
missing_TIL_date = repeatable_datetime_info[(repeatable_datetime_info.TILDate.isna())&(~repeatable_datetime_info.TILTimepoint.isna())&(repeatable_datetime_info.TILTimepoint != 'None')].reset_index(drop=True)
missing_TIL_date = missing_TIL_date[['RowID','GUPI','ICUAdmDate','TILTimepoint','TILDate']]
missing_TIL_date['ICUAdmDate'] = pd.to_datetime(missing_TIL_date['ICUAdmDate'],format = '%Y-%m-%d')
missing_TIL_date['TILTimepoint'] = missing_TIL_date['TILTimepoint'].astype('int')
missing_TIL_date['TILDate'] = (missing_TIL_date['ICUAdmDate'] + pd.to_timedelta((missing_TIL_date['TILTimepoint']-1),'days')).astype(str)
for curr_rowID in tqdm.tqdm(missing_TIL_date.RowID,'Fixing missing TILDate values'):
curr_Date = missing_TIL_date.TILDate[missing_TIL_date.RowID == curr_rowID].values[0]
repeatable_datetime_info.TILDate[repeatable_datetime_info.RowID == curr_rowID] = curr_Date
# Pivot repeatable date and time information to separate long format dataframes
repeatable_date_info = repeatable_datetime_info[['GUPI','RowID']+[col for col in repeatable_datetime_info.columns if 'Date' in col]].drop(columns=['ICUAdmDate']).dropna(axis=1,how='all')
repeatable_date_info['RowID'] = repeatable_date_info['RowID'].astype(str)
repeatable_date_info = repeatable_date_info.select_dtypes(exclude='number').melt(id_vars=['GUPI','RowID'],var_name='desc',value_name='date').dropna(subset=['date']).sort_values(by=['GUPI','date']).reset_index(drop=True)
repeatable_date_info['desc'] = repeatable_date_info['desc'].str.replace('Date','')
repeatable_time_info = repeatable_datetime_info[['GUPI','RowID']+[col for col in repeatable_datetime_info.columns if 'Time' in col]].dropna(axis=1,how='all')
repeatable_time_info['RowID'] = repeatable_time_info['RowID'].astype(str)
repeatable_time_info = repeatable_time_info.select_dtypes(exclude='number').melt(id_vars=['GUPI','RowID'],var_name='desc',value_name='time').dropna(subset=['time']).sort_values(by=['GUPI']).reset_index(drop=True)
repeatable_time_info['desc'] = repeatable_time_info['desc'].str.replace('Time','')
repeatable_datetime_info = pd.merge(repeatable_date_info,repeatable_time_info,how='outer',on=['RowID','GUPI','desc']).dropna(how='all',subset=['date','time'])
repeatable_datetime_info = repeatable_datetime_info.sort_values(by=['GUPI','date']).reset_index(drop=True)
repeatable_datetime_info.to_csv('/home/sb2406/rds/hpc-work/timestamps/repeatable_timestamps.csv',index=False)
## Fix missing ICU admission info based on non-repeatable datetime information
# Plan A: replace missing ICU admission timestamp with ED discharge times
missing_ICU_adm_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUAdmTimeStamp.isna()].reset_index(drop=True)
ED_discharge_datetime = non_repeatable_datetime_info[(non_repeatable_datetime_info.GUPI.isin(missing_ICU_adm_timestamps.GUPI))&(non_repeatable_datetime_info.desc == 'EDDisch')].fillna('1970-01-01').reset_index(drop=True)
ED_discharge_datetime = ED_discharge_datetime.merge(missing_ICU_adm_timestamps[['GUPI','ICUAdmDate']].rename(columns={'ICUAdmDate':'date'}),how='inner',on=['GUPI','date']).reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(ED_discharge_datetime.GUPI,'Imputing missing ICU admission timestamps with matching ED discharge information'):
curr_EDDisch_time = ED_discharge_datetime.time[ED_discharge_datetime.GUPI == curr_GUPI].values[0]
missing_ICU_adm_timestamps.ICUAdmTime[missing_ICU_adm_timestamps.GUPI == curr_GUPI] = curr_EDDisch_time
CENTER_TBI_ICU_datetime.ICUAdmTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_EDDisch_time
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUAdmDate', 'ICUAdmTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'][CENTER_TBI_ICU_datetime.ICUAdmDate.isna() | CENTER_TBI_ICU_datetime.ICUAdmTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan B: replace missing ICU admission timestamp with time of arrival to study hospital
missing_ICU_adm_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUAdmTimeStamp.isna()].reset_index(drop=True)
hosp_arrival_datetime = non_repeatable_datetime_info[(non_repeatable_datetime_info.GUPI.isin(missing_ICU_adm_timestamps.GUPI))&(non_repeatable_datetime_info.desc == 'PresSTHosp')].reset_index(drop=True)
hosp_arrival_datetime = hosp_arrival_datetime.merge(missing_ICU_adm_timestamps[['GUPI','ICUAdmDate']].rename(columns={'ICUAdmDate':'date'}),how='inner',on=['GUPI','date']).reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(hosp_arrival_datetime.GUPI,'Imputing missing ICU admission timestamps with matching hosptial arrival information'):
curr_hosp_arrival_time = hosp_arrival_datetime.time[hosp_arrival_datetime.GUPI == curr_GUPI].values[0]
missing_ICU_adm_timestamps.ICUAdmTime[missing_ICU_adm_timestamps.GUPI == curr_GUPI] = curr_hosp_arrival_time
CENTER_TBI_ICU_datetime.ICUAdmTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_hosp_arrival_time
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUAdmDate', 'ICUAdmTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'][CENTER_TBI_ICU_datetime.ICUAdmDate.isna() | CENTER_TBI_ICU_datetime.ICUAdmTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan C: replace missing ICU admission timestamp with time of initial study consent
missing_ICU_adm_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUAdmTimeStamp.isna()].reset_index(drop=True)
consent_datetime = non_repeatable_datetime_info[(non_repeatable_datetime_info.GUPI.isin(missing_ICU_adm_timestamps.GUPI))&(non_repeatable_datetime_info.desc == 'InfConsInitial')].reset_index(drop=True)
consent_datetime = consent_datetime.merge(missing_ICU_adm_timestamps[['GUPI','ICUAdmDate']].rename(columns={'ICUAdmDate':'date'}),how='inner',on=['GUPI','date']).reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(consent_datetime.GUPI,'Imputing missing ICU admission timestamps with matching study consent information'):
curr_consent_time = consent_datetime.time[consent_datetime.GUPI == curr_GUPI].values[0]
missing_ICU_adm_timestamps.ICUAdmTime[missing_ICU_adm_timestamps.GUPI == curr_GUPI] = curr_consent_time
CENTER_TBI_ICU_datetime.ICUAdmTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_consent_time
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUAdmDate', 'ICUAdmTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'][CENTER_TBI_ICU_datetime.ICUAdmDate.isna() | CENTER_TBI_ICU_datetime.ICUAdmTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan D: replace missing ICU admission timestamp with midday (no other information available on that day)
missing_ICU_adm_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUAdmTimeStamp.isna()].reset_index(drop=True)
CENTER_TBI_ICU_datetime.ICUAdmTime[CENTER_TBI_ICU_datetime.GUPI.isin(missing_ICU_adm_timestamps.GUPI)] = '12:00:00'
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUAdmDate', 'ICUAdmTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'][CENTER_TBI_ICU_datetime.ICUAdmDate.isna() | CENTER_TBI_ICU_datetime.ICUAdmTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
## Fix missing ICU discharge info based on (non)-repeatable datetime information
# Plan A: if patient died in the ICU, replace missing discharge information with death date
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDischTimeStamp.isna()].reset_index(drop=True)
death_datetime = non_repeatable_datetime_info[non_repeatable_datetime_info.GUPI.isin(missing_ICU_disch_timestamps.GUPI)&(non_repeatable_datetime_info.desc == 'Death')].reset_index(drop=True)
death_datetime = death_datetime.merge(missing_ICU_disch_timestamps[['GUPI','ICUDischDate']].rename(columns={'ICUDischDate':'date'}).dropna(subset=['date']),how='inner',on=['GUPI','date']).reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(death_datetime.GUPI,'Imputing missing ICU discharge timestamps with matching time of death information'):
curr_death_time = death_datetime.time[death_datetime.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_death_time
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_death_time
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# A_1: If neither date nor time is available, simply replace after a manual reasonability check
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDischTimeStamp.isna()].reset_index(drop=True)
death_datetime = non_repeatable_datetime_info[non_repeatable_datetime_info.GUPI.isin(missing_ICU_disch_timestamps.GUPI)&(non_repeatable_datetime_info.desc == 'Death')].reset_index(drop=True)
death_datetime = death_datetime[death_datetime.GUPI.isin(missing_ICU_disch_timestamps[missing_ICU_disch_timestamps.ICUDischTime.isna()]['GUPI'])].reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(death_datetime.GUPI,'Imputing missing ICU discharge timestamps with matching time of death information'):
curr_death_date = death_datetime.date[death_datetime.GUPI == curr_GUPI].values[0]
curr_death_time = death_datetime.time[death_datetime.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischDate[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_death_date
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_death_time
CENTER_TBI_ICU_datetime.ICUDischDate[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_death_date
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_death_time
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan B: Find transfer-out-of-ICU date and time if available
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDischTimeStamp.isna()].reset_index(drop=True)
transfer_datetime = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/TransitionsOfCare/data.csv',na_values = ["NA","NaN"," ", ""])
transfer_datetime = transfer_datetime[transfer_datetime.GUPI.isin(missing_ICU_disch_timestamps.GUPI)&(transfer_datetime.TransFrom == 'ICU')].dropna(subset=['DateEffectiveTransfer'])[['GUPI','DateEffectiveTransfer']].reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(transfer_datetime.GUPI,'Imputing missing ICU discharge timestamps with transfer out of ICU information'):
curr_transfer_date = transfer_datetime.DateEffectiveTransfer[transfer_datetime.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischDate[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_transfer_date
CENTER_TBI_ICU_datetime.ICUDischDate[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_transfer_date
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan C: For non-missing ICU discharge dates, replace time with latest time available among non-repeatable timestamps
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[(CENTER_TBI_ICU_datetime.ICUDischTime.isna())&(~CENTER_TBI_ICU_datetime.ICUDischDate.isna())].reset_index(drop=True)
viable_non_repeatables = non_repeatable_datetime_info.merge(missing_ICU_disch_timestamps.rename(columns={'ICUDischDate':'date'}),how='inner',on=['GUPI','date']).dropna(subset=['time']).reset_index(drop=True)
viable_non_repeatables = viable_non_repeatables.groupby(['GUPI','date'],as_index=False).time.aggregate(max)
for curr_GUPI in tqdm.tqdm(viable_non_repeatables.GUPI,'Imputing missing ICU discharge times with last available timestamp on the day'):
curr_disch_time = viable_non_repeatables.time[viable_non_repeatables.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_disch_time
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_disch_time
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan D: For non-missing ICU discharge dates, replace time with latest time available among repeatable timestamps
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[(CENTER_TBI_ICU_datetime.ICUDischTime.isna())&(~CENTER_TBI_ICU_datetime.ICUDischDate.isna())].reset_index(drop=True)
viable_repeatables = repeatable_datetime_info.merge(missing_ICU_disch_timestamps.rename(columns={'ICUDischDate':'date'}),how='inner',on=['GUPI','date']).dropna(subset=['time']).reset_index(drop=True)
viable_repeatables = viable_repeatables.groupby(['GUPI','date'],as_index=False).time.aggregate(max)
for curr_GUPI in tqdm.tqdm(viable_repeatables.GUPI,'Imputing missing ICU discharge times with last available timestamp on the day'):
curr_disch_time = viable_repeatables.time[viable_repeatables.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_disch_time
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_disch_time
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI.isin(missing_ICU_disch_timestamps.GUPI[missing_ICU_disch_timestamps.ICUDischTime.isna()])] = '23:59:00'
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.ICUDischTime.isna()] = '23:59:00'
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan E: Replace with the maximum timestamp within 30 days of admission
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDischTimeStamp.isna()].reset_index(drop=True)
viable_non_repeatables = non_repeatable_datetime_info[(non_repeatable_datetime_info.GUPI.isin(missing_ICU_disch_timestamps.GUPI))&(~non_repeatable_datetime_info.date.isna())]
viable_non_repeatables = viable_non_repeatables[viable_non_repeatables.date <= '1970-01-30'].sort_values(by=['GUPI','date','time'],ascending=False).reset_index(drop=True)
viable_non_repeatables = viable_non_repeatables.groupby('GUPI',as_index=False).first()
viable_repeatables = repeatable_datetime_info[(repeatable_datetime_info.GUPI.isin(missing_ICU_disch_timestamps.GUPI))&(~repeatable_datetime_info.date.isna())]
viable_repeatables = viable_repeatables[viable_repeatables.date <= '1970-01-30'].sort_values(by=['GUPI','date','time'],ascending=False).reset_index(drop=True)
viable_repeatables = viable_repeatables.groupby('GUPI',as_index=False).first()
viable_timepoints = pd.concat([viable_non_repeatables,viable_repeatables.drop(columns='RowID')],ignore_index=True)
viable_timepoints.time = viable_timepoints.time.str.replace('"','')
for curr_GUPI in tqdm.tqdm(viable_timepoints.GUPI,'Imputing missing ICU discharge times with last available timestamp on the day'):
if missing_ICU_disch_timestamps[missing_ICU_disch_timestamps.GUPI == curr_GUPI].ICUDischTime.isna().values[0]:
curr_disch_time = viable_timepoints.time[viable_timepoints.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_disch_time
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_disch_time
if missing_ICU_disch_timestamps[missing_ICU_disch_timestamps.GUPI == curr_GUPI].ICUDischDate.isna().values[0]:
curr_disch_date = viable_timepoints.date[viable_timepoints.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischDate[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_disch_date
CENTER_TBI_ICU_datetime.ICUDischDate[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_disch_date
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Plan F: Replace with hospital discharge timestamp if available
missing_ICU_disch_timestamps = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDischTimeStamp.isna()].reset_index(drop=True)
hosp_disch_timestamps = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/DemoInjHospMedHx/data.csv')
hosp_disch_timestamps = hosp_disch_timestamps[['GUPI','ICUDischDate','ICUDischTime','HospDischTime','HospDischDate']]
hosp_disch_timestamps = hosp_disch_timestamps[hosp_disch_timestamps.GUPI.isin(missing_ICU_disch_timestamps.GUPI)].reset_index(drop=True)
for curr_GUPI in tqdm.tqdm(hosp_disch_timestamps.GUPI,'Imputing missing ICU discharge times with hospital discharge'):
if missing_ICU_disch_timestamps[missing_ICU_disch_timestamps.GUPI == curr_GUPI].ICUDischTime.isna().values[0]:
curr_disch_time = hosp_disch_timestamps.HospDischTime[hosp_disch_timestamps.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischTime[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_disch_time
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_disch_time
if missing_ICU_disch_timestamps[missing_ICU_disch_timestamps.GUPI == curr_GUPI].ICUDischDate.isna().values[0]:
curr_disch_date = hosp_disch_timestamps.HospDischDate[hosp_disch_timestamps.GUPI == curr_GUPI].values[0]
missing_ICU_disch_timestamps.ICUDischDate[missing_ICU_disch_timestamps.GUPI == curr_GUPI] = curr_disch_date
CENTER_TBI_ICU_datetime.ICUDischDate[CENTER_TBI_ICU_datetime.GUPI == curr_GUPI] = curr_disch_date
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
## Inspect patients with less than 10 hours of ICU stay
inspection_batch = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDurationHours <= 10].reset_index(drop=True)
# Manual corrections based on plausible types
CENTER_TBI_ICU_datetime.ICUAdmDate[CENTER_TBI_ICU_datetime.GUPI == '6URh589'] = '1970-01-01'
CENTER_TBI_ICU_datetime.ICUAdmTime[CENTER_TBI_ICU_datetime.GUPI == '8uBs474'] = '05:19:00'
CENTER_TBI_ICU_datetime.ICUDischDate[CENTER_TBI_ICU_datetime.GUPI == '2CHw582'] = '1970-01-03'
CENTER_TBI_ICU_datetime.ICUDischDate[CENTER_TBI_ICU_datetime.GUPI == '3zmD494'] = '1970-01-02'
CENTER_TBI_ICU_datetime.ICUDischTime[CENTER_TBI_ICU_datetime.GUPI == '3zmD494'] = '13:00:00'
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUAdmDate', 'ICUAdmTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'][CENTER_TBI_ICU_datetime.ICUAdmDate.isna() | CENTER_TBI_ICU_datetime.ICUAdmTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUAdmTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = CENTER_TBI_ICU_datetime[['ICUDischDate', 'ICUDischTime']].astype(str).agg(' '.join, axis=1)
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'][CENTER_TBI_ICU_datetime.ICUDischDate.isna() | CENTER_TBI_ICU_datetime.ICUDischTime.isna()] = np.nan
CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] = pd.to_datetime(CENTER_TBI_ICU_datetime['ICUDischTimeStamp'],format = '%Y-%m-%d %H:%M:%S' )
CENTER_TBI_ICU_datetime['ICUDurationHours'] = (CENTER_TBI_ICU_datetime['ICUDischTimeStamp'] - CENTER_TBI_ICU_datetime['ICUAdmTimeStamp']).astype('timedelta64[s]')/3600
# Filter patients with at least 24 hours of ICU stay
CENTER_TBI_ICU_datetime = CENTER_TBI_ICU_datetime[CENTER_TBI_ICU_datetime.ICUDurationHours >= 24].sort_values(by=['GUPI']).reset_index(drop=True)
# Save timestamps as CSV
CENTER_TBI_ICU_datetime.to_csv('/home/sb2406/rds/hpc-work/timestamps/ICU_adm_disch_timestamps.csv',index = False)
### IV. Prepare ICU stay windows for dynamic modelling
# Define length in hours
WINDOW_HOURS = 2
# Define empty lists to store window information
STUDY_WINDOWS = []
# Iterate through study patients and append window timestamps to list
for i in tqdm.tqdm(range(CENTER_TBI_ICU_datetime.shape[0]),'Partitioning dataset into study windows'):
# Extract current patient information
curr_GUPI = CENTER_TBI_ICU_datetime.GUPI[i]
curr_ICUAdmTimeStamp = CENTER_TBI_ICU_datetime.ICUAdmTimeStamp[i]
curr_ICUDischTimeStamp = CENTER_TBI_ICU_datetime.ICUDischTimeStamp[i]
# Calculate total ICU duration and number of study windows
secs_ICUStay = (curr_ICUDischTimeStamp - curr_ICUAdmTimeStamp).total_seconds()
total_bin_count = int(np.ceil(secs_ICUStay/(WINDOW_HOURS*3600)))
# Derive start and end timestamps for study windows from admission
start_timestamps = [curr_ICUAdmTimeStamp + datetime.timedelta(hours=x*WINDOW_HOURS) for x in range(total_bin_count)]
end_timestamps = [ts - datetime.timedelta(microseconds=1) for ts in start_timestamps[1:]]
end_timestamps.append(curr_ICUDischTimeStamp+datetime.timedelta(microseconds=1))
# Compile current patient information into a dataframe
GUPI_windows = pd.DataFrame({'GUPI':curr_GUPI,
'TimeStampStart':start_timestamps,
'TimeStampEnd':end_timestamps,
'WindowIdx':[i+1 for i in range(total_bin_count)],
'WindowTotal':total_bin_count
})
# Append current GUPI dataframe to running compiled lists
STUDY_WINDOWS.append(GUPI_windows)
# Concatenate lists of dataframes into single study window dataframe
STUDY_WINDOWS = pd.concat(STUDY_WINDOWS,ignore_index=True)
# Save window dataframe lists into `timestamps` directory
STUDY_WINDOWS.to_csv('/home/sb2406/rds/hpc-work/timestamps/window_timestamps.csv',index = False)
### V. Perform summary characteristic POLR analysis
## Load CENTER-TBI information
# Load formatted study timestamps
CENTER_TBI_ICU_datetime = pd.read_csv('/home/sb2406/rds/hpc-work/timestamps/ICU_adm_disch_timestamps.csv')
# Load CENTER-TBI dataset to access summary characteristics
summary_characteristics = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/DemoInjHospMedHx/data.csv',na_values = ["NA","NaN"," ", ""])
IMPACT_characteristics = pd.read_csv('/home/sb2406/rds/hpc-work/CENTER-TBI/IMPACT/data.csv',na_values = ["NA","NaN"," ", ""]).rename(columns={'entity_id':'GUPI'})[['GUPI','marshall','tsah','EDH']]
summary_characteristics = summary_characteristics[summary_characteristics.GUPI.isin(CENTER_TBI_ICU_datetime.GUPI.unique())].reset_index(drop=True)
IMPACT_characteristics = IMPACT_characteristics[IMPACT_characteristics.GUPI.isin(CENTER_TBI_ICU_datetime.GUPI.unique())].reset_index(drop=True)
summary_characteristics = summary_characteristics[['GUPI','Age','Sex','GCSScoreBaselineDerived','EmplmtStatus','GOSE6monthEndpointDerived']].merge(IMPACT_characteristics,how='left')
summary_characteristics = summary_characteristics.merge(CENTER_TBI_ICU_datetime[['GUPI','ICUDurationHours']],how='left')
summary_characteristics['Severity'] = pd.cut(summary_characteristics['GCSScoreBaselineDerived'],bins=[2,8,12,15],labels=['Severe','Moderate','Mild'],ordered=False)
summary_characteristics['marshall'] = pd.cut(summary_characteristics['marshall'],bins=[0,1,2,3,4,6],labels=['I','II','III','IV','V_or_VI'],ordered=False)
summary_characteristics['Retired'] = (summary_characteristics.EmplmtStatus==8).astype(float)
summary_characteristics.Retired[summary_characteristics.EmplmtStatus.isna()] = np.nan
# Convert GOSE to ordered category type
summary_characteristics['GOSE6monthEndpointDerived'] = summary_characteristics['GOSE6monthEndpointDerived'].astype(CategoricalDtype(categories=['1', '2_or_3', '4', '5', '6', '7', '8'],ordered=True))
# Remove redundant columns and rename GOSE column
summary_characteristics = summary_characteristics.drop(columns=['GCSScoreBaselineDerived','EmplmtStatus']).rename(columns={'GOSE6monthEndpointDerived':'GOSE'})
# Remove rows with missing values
summary_characteristics = summary_characteristics.dropna().reset_index(drop=True)
# Convert categorical characteristics to proper type
cat_encoder = OneHotEncoder(drop = 'first',categories=[['M','F'],['I','II','III','IV','V_or_VI'],['Mild','Moderate','Severe']])
cat_column_names = ['Female','Marshall_II','Marshall_III','Marshall_IV','Marshall_V_or_VI','Severity_Moderate','Severity_Severe']
summary_categorical = pd.DataFrame(cat_encoder.fit_transform(summary_characteristics[['Sex','marshall','Severity']]).toarray(),
columns=cat_column_names)
summary_characteristics = pd.concat([summary_characteristics.drop(columns=['Sex','marshall','Severity']),summary_categorical],axis=1)
summary_POLR = OrderedModel(summary_characteristics['GOSE'],
summary_characteristics[summary_characteristics.columns[~summary_characteristics.columns.isin(['GUPI','GOSE'])]],
distr='logit')
res_summary_POLR = summary_POLR.fit(method='bfgs',
maxiter = 1000,
disp=False)
res_summary_POLR.save(os.path.join(summary_dir,'summary_polr.pkl'))