Healthcare NLP v5.0.1 Release Notes

5.0.1

Highlights

We are delighted to announce a suite of remarkable enhancements and updates in our latest release of Spark NLP for Healthcare. This release comes with the first NER models that are augmented by LangTest library for robustness and bias as well as a support for RxHCC risk score calculation in latest versions.

• Integrated the Risk Adjustment for Prescription Drug Hierarchical Condition Categories (RxHCC) model into our risk adjustment score calculation engine
• Advanced entity detection for Section Headers and Diagnoses entities in clinical notes
• Augmented NER models by leveraging the capabilities of the LangTest library
• Enhanced Sentence Entity Resolver Models for associating clinical entities with LOINC
• Strengthen the performance of assertion status detection by reinforcing it with entity type constraints
• Entity blacklisting in AssertionFilterer to manage assertion status effectively
• Enhanced ChunkMergeApproach and ChunkFilterer with case sensitivity settings
• New feature for ChunkMergeApproach to enable filtering chunks according to confidence thresholds
• Included sentence ID information in Relation Extraction Model metadata
• Various core improvements; bug fixes, enhanced overall robustness and reliability of Spark NLP for Healthcare
  • Improved deidentification regex pattern for Romanian language
  • Fixed exploded sentences issue in RelationExtractionDLModel
• Updated notebooks and demonstrations for making Spark NLP for Healthcare easier to navigate and understand
• The addition and update of numerous new clinical models and pipelines continue to reinforce our offering in the healthcare domain

We believe that these enhancements will elevate your experience with Spark NLP for Healthcare, enabling more efficient, accurate, and streamlined analysis of healthcare-related natural language data.

Integrated The Risk Adjustment For Prescription Drug Hierarchical Condition Categories (RxHCC) Model Into Our Risk Adjustment Score Calculation Engine

We have integrated the RxHCC into our existing risk adjustment score calculation module. This means more accurate and comprehensive risk adjustment scores, especially for patients whose healthcare costs are significantly influenced by prescription drug usage. This enhancement brings a holistic view of a patient’s healthcare needs, further improving the precision of risk assessment.

We are pleased to introduce support for RxHCC risk score calculation in two new versions: v05 (applicable for 2020, 2021, 2022, and 2023) and v08 (applicable for 2022 and 2023). To utilize these versions with specific years, simply use the following formats: profileRxHCCV05YXX for v05 and profileRxHCCV08YXX for v08.

Example:

Input Data Frame:

# v08 year 2023
from sparknlp_jsl.functions import profileRxHCCV08Y23

df = df.withColumn("rxhcc_profile", profileRxHCCV08Y23(df.icd10_code, df.Age, df.Gender, df.eligibility, df.orec, df.esrd))

df = df.withColumn("rxhcc_profile", F.from_json(F.col("rxhcc_profile"), schema))
df = df.withColumn("risk_score", df.rxhcc_profile.getItem("risk_score"))\
       .withColumn("parameters", df.rxhcc_profile.getItem("parameters"))\
       .withColumn("details", df.rxhcc_profile.getItem("details"))\

Results (V08-Y23):

Advanced Entity Detection For Section Headers And Diagnoses Entities In Clinical Notes

We have a new state-of-the-art NER model that is specifically designed to extract vital data from clinical documents, focusing on two key aspects: Section Headers and Diagnoses. By accurately identifying and labeling various medical conditions like heart disease, diabetes, and Alzheimer’s disease, this model provides unparalleled insights into diagnosis and treatment trends.

Example:

clinical_ner = MedicalNerModel.pretrained("ner_section_header_diagnosis", "en","clinical/models")\
    .setInputCols(["sentence","token","embeddings"])\
    .setOutputCol("ner")\
    .setLabelCasing("upper")

text = """
Medical History:
Patient has a history of Chronic respiratory disease.
Clinical History:
Patient presented with shortness of breath and chest pain.
Chief Complaint:
Patient complained of chest pain and difficulty breathing.
History of Present Illness:
Patient has been experiencing chest pain and shortness of breath for the past week. Symptoms were relieved by medication at first but became worse over time.
Past Medical History:
Patient has a history of Asthma and was previously diagnosed with Bronchitis.
Medications:
Patient is currently taking Albuterol, Singulair, and Advair for respiratory issues.
Allergies:
Patient has a documented allergy to Penicillin.
"""

Result:

Please check: ner_section_header_diagnosis model card for more information.

Augmented NER Models Leveraging LangTest Library Capabilities

Newly introduced augmented NER models, namely ner_posology_langtest, ner_jsl_langtest, ner_ade_clinical_langtest, and ner_sdoh_langtest, are powered by the innovative LangTest library. This cutting-edge NLP toolkit is at the forefront of language processing advancements, incorporating state-of-the-art techniques and algorithms to enhance the capabilities of our models significantly.

Example:

clinical_ner = MedicalNerModel.pretrained("ner_sdoh_langtest", "en", "clinical/models")\
    .setInputCols(["sentence", "token", "embeddings"])\
    .setOutputCol("ner")


text = """Smith is 55 years old, living in New York, a divorced Mexcian American woman with financial problems. She speaks Spanish and Portuguese. She lives in an apartment. She has been struggling with diabetes for the past 10 years and has recently been experiencing frequent hospitalizations due to uncontrolled blood sugar levels. Smith works as a cleaning assistant and cannot access health insurance or paid sick leave."""

Result:

Enhanced Sentence Entity Resolver Models For Associating Clinical Entities With LOINC

Introducing the new sbiobertresolve_loinc_numeric model and improving the sbiobertresolve_loinc_augmented model, both offering enhanced accuracy for mapping medical laboratory observations and clinical measurements to their corresponding Logical Observation Identifiers Names and Codes (LOINC). The sbiobertresolve_loinc_numeric model is specialized in numeric LOINC codes, as it was trained without the inclusion of LOINC “Document Ontology” codes starting with the letter “L”. On the other hand, the sbiobertresolve_loinc_augmented model offers broader functionality, capable of returning both numeric and document ontology codes for enhanced versatility.

Example:

resolver = SentenceEntityResolverModel.pretrained("sbiobertresolve_loinc_numeric","en", "clinical/models") \
  .setInputCols(["sbert_embeddings"]) \
  .setOutputCol("loinc_code")\
  .setDistanceFunction("EUCLIDEAN")

sample_text = "The patient is a 22-year-old female with a history of obesity. She has a Body mass index (BMI) of 33.5 kg/m2, aspartate aminotransferase 64, and alanine aminotransferase 126."

Results:

Strengthen The Performance Of Assertion Status Detection By Reinforcing With Entity Type Constraints

Introducing the latest enhancements to our AssertionDLModel - the setEntityAssertion and setEntityAssertionCaseSensitive parameters. Now, you can effortlessly constrain assertions based on specific entity types using a convenient dictionary format: {"entity": [assertion_label1, assertion_label2, .. assertion_labelN]}. When an entity is not found in the dictionary, no constraints are applied, ensuring flexibility in your data processing. With the setEntityAssertionCaseSensitive parameter, you can control the case sensitivity for both entities and assertion labels. Unleash the full potential of your NLP model with these cutting-edge additions to the AssertionDLModel.

Example:

clinical_assertion = AssertionDLModel.pretrained("assertion_jsl_augmented", "en", "clinical/models") \
    .setInputCols(["sentence", "ner_chunk", "embeddings"]) \
    .setOutputCol("assertion")\
    .setEntityAssertionCaseSensitive(False)\
    .setEntityAssertion({
        "PROBLEM": ["hypothetical", "absent"],
        "treAtment": ["present"],
        "TEST": ["POssible"],
    })

text = '''
A 28-year-old female with a history of gestational diabetes mellitus diagnosed eight years prior to presentation and subsequent type two diabetes mellitus (T2DM), one prior episode of HTG-induced pancreatitis three years prior to presentation, and associated with an acute hepatitis, presented with a one-week history of polyuria, poor appetite, and vomiting.
She was on metformin, glipizide, and dapagliflozin for T2DM and atorvastatin and gemfibrozil for HTG. She had been on dapagliflozin for six months at the time of presentation.
Physical examination on presentation was significant for dry oral mucosa ; significantly , her abdominal examination was benign with no tenderness, guarding, or rigidity. Pertinent laboratory findings on admission were: serum glucose 111 mg/dl,  creatinine 0.4 mg/dL, triglycerides 508 mg/dL, total cholesterol 122 mg/dL, and venous pH 7.27.
'''

Result:

Entity Blacklisting In AssertionFilterer For Effective Assertion Status Management

With the setBlackList option in the AssertionFilterer annotator, you can now blacklist specific entities based on their assertion labels.

Example:

clinical_assertion = AssertionDLModel.pretrained("assertion_jsl_augmented", "en", "clinical/models") \
    .setInputCols(["sentence", "ner_chunk", "embeddings"]) \
    .setOutputCol("assertion")


assertion_filterer = AssertionFilterer()\
    .setInputCols("sentence","ner_chunk","assertion")\
    .setOutputCol("assertion_filtered")\
    .setBlackList(["Hypothetical"])\

text = """Patient has a headache for the last 2 weeks, needs to get a head CT, and appears anxious when she walks fast. No alopecia and pain noted"""

Without Filtering Results:

Filtered Results:

Enhanced ChunkMergeApproach And ChunkFilterer With Case Sensitivity Settings

The setCaseSensitive parameter now applies to the whitelist and blacklist functionalities. As part of the enhancement, this parameter has been included in the filtering feature, which serves as a superclass for, ChunkFilterer and ChunkMergeApproach. With this update, the caseSensitive setting can be conveniently utilized across these classes, offering improved control and consistency in the filtering process.

Example:

posology_ner = MedicalNerModel.pretrained("ner_posology", "en", "clinical/models") \
    .setInputCols(["sentence", "token", "embeddings"]) \
    .setOutputCol("ner")

ner_converter = NerConverterInternal()\
    .setInputCols(["sentence","token","ner"])\
    .setOutputCol("ner_chunk")

chunk_filterer = ChunkFilterer()\
    .setInputCols("sentence","ner_chunk")\
    .setOutputCol("chunk_filtered")\
    .setCriteria("isin")\
    .setWhiteList(['ADVIL','Metformin', 'Insulin Lispro'])\
    .setCaseSensitive(False)

text ="""The patient was prescribed 1 capsule of Advil for 5 days . She was seen by the endocrinology service and she was discharged on 40 units of insulin glargine at night , 12 units of insulin lispro with meals , metformin 1000 mg two times a day."""

Result:

# detected ner chunks
['1', 'capsule', 'Advil', 'for 5 days', '40 units', 'insulin glargine', 'at night', '12 units', 'insulin lispro', 'with meals', 'metformin', '1000 mg', 'two times a day']

# filtered ner chunks
['Advil', 'insulin lispro', 'metformin']

New Feature For ChunkMergeApproach To Enable Filtering Chunks According To Confidence Thresholds

We have added a new setEntitiesConfidence parameter to ChunkMergeApproach annotator that enables filtering the chunks according to the confidence thresholds. The only thing you need to do is provide a csv file that has the NER labels as keys and the confidence thresholds as values.

Example:

conf_dict = """DRUG,0.99
FREQUENCY,0.99
DOSAGE,0.99
DURATION,0.99
STRENGTH,0.99
"""
with open('conf_dict.csv', 'w') as f:
    f.write(conf_dict)

chunk_merger = ChunkMergeApproach()\
    .setInputCols("posology_ner_chunk")\
    .setOutputCol('merged_ner_chunk')\
    .setEntitiesConfidenceResource("conf_dict.csv")


sample_text = 'The patient was prescribed 1 capsule of Advil for 5 days. He was seen by the endocrinology service and she was discharged on 40 units of insulin glargine at night.'  

Detected chunks:

Filtered by confidence scores:

Included Sentence Id Information In RelationExtractionModel Metadata

Our Relation Extraction Models have been upgraded with the inclusion of sentence information in the metadata. This enhancement offers a deeper understanding of the extracted relationships and facilitates more precise analysis and interpretation of the results.

Example:

re_dl_model = RelationExtractionDLModel.pretrained('redl_bodypart_direction_biobert', "en", "clinical/models")\
    .setInputCols(["re_ner_chunks", "sentences"]) \
    .setOutputCol("relations_dl")\
    .setPredictionThreshold(0.5)

text = '''MRI demonstrated infarction in the upper brain stem , and  right basil ganglia.
No neurologic deficits other than some numbness in his left hand.
there is a problem at right chest.'''

Result:

Various Core Improvements: Bug Fixes, Enhanced Overall Robustness, And Reliability Of Spark NLP For Healthcare

• Improved deidentification regex pattern for Romanian language
• Fixed exploded sentences issue in Relation Extraction DL (when .setExplodeSentences(True) is used in SentenceDetector, RelationExtractionDLModel’s relation output has only the sentence#0 relations, other sentences’ relations are not displayed.)

Updated Notebooks And Demonstrations For making Spark NLP For Healthcare Easier To Navigate And Understand

• Updated Clinical_Named_Entity_Recognition_Model notebook according to latest improvement in ChunkFilterer
• Updated Clinical_Assertion_Model notebook according to latest improvement in AssertionFilterer
• Updated Clinical_NER_Chunk_Merger notebook according to latest improvement in ChunkMergerApproach
• Updated Clinical_Relation_Extraction notebook according to latest improvement in RelationExtractionModel’s metadata
• Updated Calculate_Medicare_Risk_Adjustment_Score notebook according to latest improvement in HCC implementation

We Have Added And Updated A Substantial Number Of New Clinical Models And Pipelines, Further Solidifying Our Offering In The Healthcare Domain.

• ner_section_header_diagnosis
• ner_posology_langtest
• ner_jsl_langtest
• ner_sdoh_langtest
• ner_ade_clinical_langtest
• sbiobertresolve_loinc_numeric
• sbiobertresolve_loinc_augmented

For all Spark NLP for Healthcare models, please check: Models Hub Page

Versions