def _generate_icd9_lookup(self):
self._diag_to_desc = {}
tree = ICD9('../lib/icd9/codes.json')
for d in self._diags:
try:
self._diag_to_desc[d] = tree.find(d[2:]).description
except:
if d[2:] == "008":
self._diag_to_desc[
d] = "Intestinal infections due to other organisms"
elif d[2:] == "280":
self._diag_to_desc[d] = "Iron deficiency anemias"
elif d[2:] == "284":
self._diag_to_desc[
d] = "Aplastic anemia and other bone marrow failure syndrome"
elif d[2:] == "285":
self._diag_to_desc[d] = "Other and unspecified anemias"
elif d[2:] == "286":
self._diag_to_desc[d] = "Coagulation defects"
elif d[2:] == "287":
self._diag_to_desc[
d] = "Purpura and other hemorrhagic conditions"
elif d[2:] == "288":
self._diag_to_desc[d] = "Diseases of white blood cells"
else:
self._diag_to_desc[d] = "Not Found"
def generate_icd9_lookup():
"""Generate description from ICD9 code"""
tree = ICD9('../lib/icd9/codes.json')
for ud in uniq_diag:
try:
diag_to_desc[ud] = tree.find(ud[2:]).description
except:
if ud[2:] == "008":
diag_to_desc[
ud] = "Intestinal infections due to other organisms"
elif ud[2:] == "280":
diag_to_desc[ud] = "Iron deficiency anemias"
elif ud[2:] == "284":
diag_to_desc[
ud] = "Aplastic anemia and other bone marrow failure syndrome"
elif ud[2:] == "285":
diag_to_desc[ud] = "Other and unspecified anemias"
elif ud[2:] == "286":
diag_to_desc[ud] = "Coagulation defects"
elif ud[2:] == "287":
diag_to_desc[ud] = "Purpura and other hemorrhagic conditions"
elif ud[2:] == "288":
diag_to_desc[ud] = "Diseases of white blood cells"
else:
diag_to_desc[ud] = "Not Found"
def rollup_mimic(mimic_path, mimic_filename, codes_path):
"""
Rolls up all the codes in mimic data from our preprocessing
"""
mimic_data = pd.read_csv(os.path.join(mimic_path, mimic_filename))
tree = ICD9(codes_path)
mimic_data["DIAG_ICD_ALL"] = mimic_data.ICD_DIAG.apply(
rollup_multiple_codes, icd_tree=tree)
mimic_data = mimic_data.assign(**mimic_data.DIAG_ICD_ALL.apply(pd.Series))
mimic_data.to_csv(os.path.join(mimic_path, "mimic_processed.csv"))
def _generate_icd9_lookup(self):
self._diag_to_desc = {}
tree = ICD9('../../lib/icd9/codes.json')
for d in self._diags:
try:
self._diag_to_desc[d] = tree.find(d[2:]).description
except:
if d[2:] == "285.9":
self._diag_to_desc[d] = "Anemia"
elif d[2:] == "287.5":
self._diag_to_desc[d] = "Thrombocytopenia"
elif d[2:] == "285.1":
self._diag_to_desc[d] = "Acute posthemorrhagic anemia"
else:
self._diag_to_desc[d] = "Not Found"
import os
import ipdb
from nltk.tokenize import word_tokenize
from icd9 import ICD9
from transformers import AutoConfig, AutoModel, AutoTokenizer
import torch
from tqdm import tqdm
import numpy as np
import sys
sys.path.append("../../pretrain")
from load_umls import UMLS
tree = ICD9('codes.json')
device = torch.device("cuda:1" if torch.cuda.is_available() else "cpu")
log_list = 1 / np.log2(list(range(2, 1001, 1)))
batch_size = 512
max_seq_length = 32
def get_icd9_pairs(icd9_set):
icd9_pairs = {}
with open('icd9_grp_file.txt', 'r', encoding="utf-8") as infile:
data = infile.readlines()
for row in data:
codes, name = row.strip().split('#')
name = name.strip()
codes = codes.strip().split(' ')
new_codes = set([])
for code in codes:
if code in icd9_set:
out_directory = 'data'
# In[330]:
data_train.sort_index().to_pickle(out_directory + '/data_train.pkl')
data_test.sort_index().to_pickle(out_directory + '/data_test.pkl')
# In[331]:
target_train.sort_index().to_pickle(out_directory + '/target_train.pkl')
target_test.sort_index().to_pickle(out_directory + '/target_test.pkl')
# In[332]:
from icd9 import ICD9
tree = ICD9('codes_pretty_printed.json')
# In[333]:
labels_icd9 = []
ids = []
for k in range(390, 459, 1):
ids.append(k - 390)
if tree.find(str(k)) is not None:
#print(tree.find(str(k)).description)
labels_icd9.append(tree.find(str(k)).description)
else:
#print(tree.find('390-459').description+str(k))
labels_icd9.append(tree.find('390-459').description + str(k))
np.random.seed(12345)
# In[4]:
reload(eu)
reload(meu)
reload(sdu)
# In[6]:
sys.path.append('icd9')
from icd9 import ICD9
# feel free to replace with your path to the json file
tree = ICD9('icd9/codes.json')
# ## Configure pandas and matplot lib for nice web printing
# In[7]:
pd.options.display.max_rows = 1000
pd.options.display.max_columns = 50
pd.options.display.max_colwidth = 100
# In[8]:
get_ipython().run_line_magic('matplotlib', 'inline')
# ## Load config files, configure logging
import csv
from icd9 import ICD9 # not python3 compatible
tree = ICD9('./codes.json')
def cluster_codes(dataset_path, level=2):
"""
levels
- 1 top (eg 'Endocrine, Nutritional And Metabolic Diseases, And Immunity Disorders')
- 2 super (eg 'Diseases Of Other Endocrine Glands')
- 3 basic (eg 'Diabetes mellitus')
- 4 sub (eg 'Diabetes mellitus without mention of complication')
- 5 specific (eg 'Diabetes mellitus without mention of complication, type II or unspecified type, not stated as uncontrolled')
"""
with open(dataset_path) as data_f:
data = csv.reader(data_f, delimiter=',')
diagnosis_codes = {}
results = []
for i, line in enumerate(data):
if i == 0:
diag_index = line.index('diag_1')
line.append('diag_1_desc')
line.append('diag_2_desc')
line.append('diag_3_desc')
results.append(line)
continue
diag_1 = line[diag_index]
diag_2 = line[diag_index + 1]
diag_3 = line[diag_index + 2]
for j, diag in enumerate([diag_1, diag_2, diag_3]):
def rollup_mimic(mimic_path, mimic_filename, codes_path):
"""
Rolls up all the codes in mimic data from our preprocessing
"""
mimic_data = pd.read_csv(os.path.join(mimic_path, mimic_filename))
tree = ICD9(codes_path)
mimic_data["DIAG_ICD_ALL"] = mimic_data.ICD_DIAG.apply(
rollup_multiple_codes, icd_tree=tree)
mimic_data = mimic_data.assign(**mimic_data.DIAG_ICD_ALL.apply(pd.Series))
mimic_data.to_csv(os.path.join(mimic_path, "mimic_processed.csv"))
if __name__ == "__main__":
import pdb
pdb.set_trace()
tree = ICD9("codes.json")
# mimic_rollup_test = pd.read_csv("mimic_rollup_test.csv")
# mimic_rollup_test["ICD_3"] = mimic_rollup_test.ICD9_CODE.apply(rollup, level = 3, icd_tree = tree)
# mimic_rollup_test["ALL_ICD"] = mimic_rollup_test.ICD9_CODE.apply(rollup_all_levels, icd_tree = tree)
# mimic_rollup_test.assign(**mimic_rollup_test.ALL_ICD.apply(pd.Series))
# for i, row in mimic_rollup_test.iterrows():
# row.ICD_rolled_up = rollup(code = str(row.ICD9_CODE), level = 1, icd_tree = tree)
rollup_mimic(
"/Users/michalmalyska/Desktop/Projects/semantic_health_public/deep-patient-cohorts/data",
"mimic_real_test.csv",
("/Users/michalmalyska/Desktop/Projects/semantic_health_public/deep-patient-cohorts"
"/scripts/icd_rollup/codes.json"),
)
def create_descriptive_recordset(input_data, one_hot_map):
# Create IC9 map of diagnoses codes & Record
descriptors = {
"admloc_": "admission_location",
"disloc_": "discharge_location",
"ins_": "insurance",
"lang_": "language",
"rel_": "religion",
"mar_": "marital_status",
"eth_": "ethnicity",
"gen_": "gender",
"dia_": "diagnoses"
}
enabled_descriptors = set()
tree = ICD9()
diagnostic_codes_map = {}
for key in one_hot_map.keys():
for descriptor in descriptors.keys():
if key.startswith(descriptor):
enabled_descriptors.add(descriptors[descriptor])
if key.startswith("dia_"):
try:
condition = key[len("dia_D_"):]
diagnostic_codes_map[key] = "{}-{}".format(
condition,
tree.find(condition).description)
except:
diagnostic_codes_map[key] = key[len("dia_D_"):]
DescriptiveRecord = namedtuple("DescriptiveRecord",
list(sorted(enabled_descriptors)))
# Actually process stuff
sparse_descriptive_records = []
for i, record in enumerate(input_data):
record_lists = {key: [] for key in enabled_descriptors}
# Process record
for column in record:
for column_mask, column_description in descriptors.items():
if column.startswith(column_mask):
if column_mask == "dia_":
condition = column[len("dia_D_"):]
condition_description = "{}-{}".format(
condition, diagnostic_codes_map[column])
record_lists[column_description].append(
condition_description)
else:
record_lists[column_description].append(
column[len(column_mask):])
# Package record
packaged_record = [
tuple(record_lists[key]) for key in sorted(enabled_descriptors)
]
sparse_descriptive_records.append(DescriptiveRecord(*packaged_record))
# Report progress
if (i + 1) % 1000 == 0:
print("{} records processed so far.".format(i + 1))
return sparse_descriptive_records, list(sorted(enabled_descriptors))
