Business Context
A hospital wants to automatically structure unstructured patient notes by identifying symptoms and medications using a pre-trained CRF model.
Data Preparation
Creation of patient notes and simulation of the 5 required CRF model tables (Labels, Attributes, Features, Attr-Features, Templates) adapted for medical terms.
Copied!
| 1 | DATA mycas.med_docs; LENGTH doc_id $10 text $200; INFILE DATALINES delimiter='|'; INPUT doc_id $ text $; DATALINES; |
| 2 | DOC1|Patient takes Aspirin for headache. |
| 3 | DOC2|No allergies reported. |
| 4 | ; |
| 5 | RUN; |
| 6 | |
| 7 | /* Simulate Model Tables */ |
| 8 | DATA mycas.m_label; LENGTH _label_ $20 _type_ $20; INPUT _label_ $ _type_ $; DATALINES; |
| 9 | B-DRUG MEDICATION |
| 10 | I-DRUG MEDICATION |
| 11 | B-SYMP SYMPTOM |
| 12 | O OTHER |
| 13 | ; RUN; |
| 14 | |
| 15 | DATA mycas.m_attr; LENGTH _attr_ $50 _value_ $50; INPUT _attr_ $ _value_ $; DATALINES; |
| 16 | WORD[0] Aspirin |
| 17 | WORD[0] headache |
| 18 | WORD[0] Patient |
| 19 | ; RUN; |
| 20 | |
| 21 | DATA mycas.m_feat; LENGTH _feature_ $50; INPUT _feature_ $; DATALINES; |
| 22 | U00:Aspirin |
| 23 | U00:headache |
| 24 | ; RUN; |
| 25 | |
| 26 | DATA mycas.m_attr_feat; INPUT _attrid_ _featureid_ _weight_; DATALINES; |
| 27 | 1 1 1.5 |
| 28 | 2 2 1.2 |
| 29 | ; RUN; |
| 30 | |
| 31 | DATA mycas.m_temp; LENGTH _template_ $100; INPUT _template_ $; DATALINES; |
| 32 | U00:%w[0] |
| 33 | ; RUN; |
Étapes de réalisation
1
Scoring the patient notes using the medical CRF model.
Copied!
| 1 | PROC CAS; |
| 2 | conditionalRandomFields.crfScore |
| 3 | TABLE={name='med_docs'}, |
| 4 | model={ |
| 5 | attr={name='m_attr'}, |
| 6 | attrfeature={name='m_attr_feat'}, |
| 7 | feature={name='m_feat'}, |
| 8 | label={name='m_label'}, |
| 9 | template={name='m_temp'} |
| 10 | }, |
| 11 | casOut={name='med_scored', replace=true}, |
| 12 | target='ner_tags'; |
| 13 | RUN; |
2
Verification of the results.
Copied!
| 1 | PROC PRINT DATA=mycas.med_scored; RUN; |
Expected Result
The 'med_scored' table is created. It contains the original text and a new column 'ner_tags' where 'Aspirin' is tagged as B-DRUG and 'headache' as B-SYMP (based on the simulated weights).