### CFG Parser Driver Module Example

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Illustrative example of a driver module for the CFG parser. Applications should adapt this structure for their specific needs.

```go
package main

import (
	"fmt"
	"os"

	"github.com/facebookresearch/Clinical-Trial-Parser/src/ct/cfg"
)

func main() {
	if len(os.Args) < 3 {
		fmt.Fprintf(os.Stderr, "usage: %s <config_file> <input_file>\n", os.Args[0])
		os.Exit(1)
	}

	cfg.Parse(os.Args[1], os.Args[2])
}

```

--------------------------------

### Install PyText NLP

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Install the PyText NLP library using pip within an Anaconda environment. This is a prerequisite for the IE parser.

```shell
pip install pytext-nlp
```

--------------------------------

### Ingest Clinical Trials Data

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Example script for ingesting clinical trial data. This script samples a limited number of trials and uses a PostgreSQL database. Modifications to 'where' clauses are often necessary.

```shell
./script/ingest.sh
```

--------------------------------

### Represent Parsed Criteria Structures

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Example JSON output for numerical and ordinal relations extracted by the CFG parser.

```json
// CFG Parser Output - Numerical/Ordinal Relations
{
  "id": "200",
  "name": "age",
  "unit": "year",
  "lower": {"incl": true, "value": "18"},
  "upper": {"incl": true, "value": "65"},
  "variableType": "numerical",
  "score": 1.0
}

{
  "id": "100",
  "name": "ecog",
  "value": ["0", "1"],
  "variableType": "ordinal",
  "score": 1.0
}

{
  "id": "203",
  "name": "bmi",
  "unit": "kg/m2",
  "lower": {"incl": true, "value": "18.5"},
  "upper": {"incl": false, "value": "25"},
  "variableType": "numerical",
  "score": 1.0
}

// IE Parser Output - Nominal Relations with MeSH Concepts
// TSV columns: nct_id, eligibility_type, criterion, label, term, ner_score, concepts, tree_numbers, nel_score
// NCT04346355  exclusion  Diverticulitis or intestinal perforation  word_scores:chronic_disease  diverticulitis  0.996  Diverticulitis  C06.405.205.282.500  1.000

// Variable types:
// - nominal: disease presence (yes/no), e.g., "must have COVID-19"
// - ordinal: discrete values, e.g., "ECOG 0 or 1"
// - numerical: continuous ranges, e.g., "age 18-65 years"
```

--------------------------------

### Download Clinical Trials Data

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Example script for downloading clinical trial data from AACT. Applications may need to modify this script or use alternative methods for data ingestion.

```shell
./script/aact.sh
```

--------------------------------

### Load and Use Variables Dictionary (Go)

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Demonstrates how to load a variables dictionary from a CSV file, set it globally, and then look up variables by ID, alias, or retrieve their associated questions. Ensure the CSV file path is correct.

```go
package main

import (
    "fmt"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/variables"
)

func main() {
    // Load variable definitions
    dict, _ := variables.Load("src/resources/variables/variables.csv")
    variables.Set(dict)

    catalog := variables.Get()

    // Look up variable by ID
    ageVar := catalog.Variable("200")
    fmt.Printf("Name: %s, Display: %s, Unit: %s\n",
        ageVar.Name, ageVar.Display, ageVar.UnitName)

    // Look up variable ID by alias
    id, found := catalog.ID("hemoglobin a1c")
    if found {
        fmt.Printf("Found variable ID: %s\n", id) // Output: 400
    }

    // Get the question for a variable
    question := catalog.Question("203")
    fmt.Println(question) // Output: What is your BMI?
}
```

--------------------------------

### Manage Machine-Readable Relations in Go

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Demonstrates creating, serializing, parsing, and negating eligibility relations.

```go
package main

import (
    "fmt"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/relation"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/variables"
)

func main() {
    // Create a numerical relation (age requirement)
    ageRelation := &relation.Relation{
        ID:           "200",
        Name:         "age",
        Unit:         "year",
        Lower:        &relation.Limit{Incl: true, Value: "18"},
        Upper:        &relation.Limit{Incl: true, Value: "65"},
        VariableType: variables.Numerical,
        Score:        1.0,
    }

    // Output: {"id":"200","name":"age","unit":"year","lower":{"incl":true,"value":"18"},"upper":{"incl":true,"value":"65"},"variableType":"numerical","score":1}
    fmt.Println(ageRelation.JSON())

    // Create an ordinal relation (ECOG score)
    ecogRelation := &relation.Relation{
        ID:           "100",
        Name:         "ecog",
        Value:        []string{"0", "1"},
        VariableType: variables.Ordinal,
        Score:        1.0,
    }

    // Output: {"id":"100","name":"ecog","value":["0","1"],"variableType":"ordinal","score":1}
    fmt.Println(ecogRelation.JSON())

    // Parse relation from JSON string
    jsonStr := `{"id":"203","name":"bmi","unit":"kg/m2","lower":{"incl":true,"value":"18.5"},"upper":{"incl":false,"value":"25"},"variableType":"numerical","score":1}`
    parsedRelation := relation.Parse(jsonStr)
    fmt.Printf("Variable: %s, Lower: %s, Upper: %s\n",
        parsedRelation.Name,
        parsedRelation.Lower.Value,
        parsedRelation.Upper.Value)

    // Negate a relation (for exclusion criteria)
    exclusionRelation := &relation.Relation{
        ID:           "405",
        Name:         "platelet_count",
        Unit:         "cells/ul",
        Upper:        &relation.Limit{Incl: false, Value: "50000"},
        VariableType: variables.Numerical,
        Score:        1.0,
    }
    exclusionRelation.Negate(nil)  // Negates: <50000 becomes >=50000
    fmt.Println(exclusionRelation.JSON())
}
```

--------------------------------

### Build and test the CFG parser

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/README.md

Commands to compile and verify the CFG parser module.

```bash
go build ./...
go test ./...
```

--------------------------------

### Create and Parse Clinical Studies in Go

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Initializes a study with raw eligibility text, loads required dictionaries, and extracts machine-readable relations.

```go
package main

import (
    "fmt"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/studies"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/units"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/variables"
)

func main() {
    // Load variable and unit dictionaries (required before parsing)
    variableDictionary, _ := variables.Load("src/resources/variables/variables.csv")
    variables.Set(variableDictionary)

    unitDictionary, _ := units.Load("src/resources/units/units.csv")
    units.Set(unitDictionary)

    // Create a study with eligibility criteria
    eligibilityCriteria := `
        Inclusion Criteria:
          - Age 18 to 65 years
          - BMI between 18.5 and 30.0 kg/m2
          - ECOG performance status 0 or 1
        Exclusion Criteria:
          - Platelets <50,000 /mmc
          - Pregnant or breastfeeding
    `

    study := studies.NewStudy(
        "NCT12345678",                    // NCT ID
        "Example Clinical Trial",          // Title
        []string{"COVID-19", "Diabetes"}, // Conditions
        eligibilityCriteria,              // Raw eligibility text
    )

    // Parse the eligibility criteria
    study.Parse()

    // Get parsed relations as TSV string
    relations := study.Relations()
    fmt.Println(relations)

    // Access parsed criteria directly
    inclusions, exclusions := study.Criteria()
    fmt.Printf("Inclusion criteria: %d\n", len(inclusions))
    fmt.Printf("Exclusion criteria: %d\n", len(exclusions))

    // Get statistics
    fmt.Printf("Total criteria: %d\n", study.CriteriaCount())
    fmt.Printf("Parsed criteria: %d\n", study.ParsedCriteriaCount())
    fmt.Printf("Relations extracted: %d\n", study.RelationCount())
}
```

--------------------------------

### Parse Clinical Trial Criteria with Go

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Initializes dictionaries and uses the Parser and Interpreter to extract relations from criterion text.

```go
package main

import (
    "fmt"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/parser"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/units"
    "github.com/facebookresearch/Clinical-Trial-Parser/src/ct/variables"
)

func main() {
    // Initialize dictionaries
    variableDictionary, _ := variables.Load("src/resources/variables/variables.csv")
    variables.Set(variableDictionary)
    unitDictionary, _ := units.Load("src/resources/units/units.csv")
    units.Set(unitDictionary)

    // Create parser and parse a criterion
    p := parser.NewParser()

    // Parse individual criterion text
    criteria := p.Parse("age between 18 and 65 years")
    fmt.Printf("Parsed %d criterion segments\n", len(criteria))

    // Use the interpreter for full relation extraction
    interpreter := parser.Get()

    orRelations, andRelations := interpreter.Interpret("bmi >= 18.5 and < 25 kg/m2")
    orRelations.Process()
    andRelations.Process()

    fmt.Printf("OR relations: %d\n", len(orRelations))
    fmt.Printf("AND relations: %d\n", len(andRelations))

    for _, r := range andRelations {
        fmt.Printf("Variable: %s, JSON: %s\n", r.Name, r.JSON())
    }
}
```

--------------------------------

### Train and test the NER model

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/README.md

Commands to train and evaluate the NER model using PyText configuration files.

```bash
pytext train < src/resources/config/ner.json
```

```bash
pytext test < src/resources/config/ner.json
```

--------------------------------

### Execute clinical trial parsing scripts

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/README.md

Scripts to run the CFG and IE parsers on clinical trial data.

```bash
./script/cfg_parse.sh
```

```bash
./script/ie_parse.sh
```

--------------------------------

### Run CFG Parser for Eligibility Criteria

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Executes the CFG parser to extract numerical and ordinal requirements from clinical trial data.

```bash
# Run the CFG parser on clinical trial data
./script/cfg_parse.sh

# Or run directly with Go
go run src/cmd/cfg/main.go \
  -conf src/resources/config/cfg.conf \
  -i data/input/clinical_trials.csv \
  -o data/output/cfg_parsed_clinical_trials.tsv \
  -logtostderr

# Input CSV format (clinical_trials.csv):
# nct_id,title,has_us_facility,conditions,eligibility_criteria
# NCT04346355,Study Title,false,COVID-19,"
#   Inclusion Criteria:
#     - age > 18 years
#     - BMI between 18.5 and 30.0
#   Exclusion Criteria:
#     - Neutrophils <500 /mmc
# "

# Output TSV format (cfg_parsed_clinical_trials.tsv):
# nct_id  eligibility_type  variable_type  criterion_index  criterion  question  relation
# NCT04346355  inclusion  numerical  0  age > 18 years  How old are you?  {"id":"200","name":"age","unit":"year","lower":{"incl":false,"value":"18"},"variableType":"numerical","score":1}
```

--------------------------------

### NEL Configuration Settings

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Configuration file for Named Entity Linking (NEL) settings. Adjust vocabulary sources, NER thresholds, matching parameters, and valid NER labels to tune precision and recall for specific clinical domains. LSH parameters are used for vocabulary search optimization.

```ini
# src/resources/config/nel.conf

# Vocabulary source files
vocabulary_file = data/mesh/descriptor.xml
custom_vocabulary_file = data/mesh/custom_mesh_concepts_p1.tsv;custom_mesh_concepts_p2.tsv
vocabulary_source = mesh

# NER confidence threshold (entities below this are ignored)
ner_threshold = 0.7

# NEL matching thresholds
match_threshold = 0.75  # Minimum similarity score to accept a match
match_margin = 0.02     # Margin for considering near-matches

# Valid NER labels to process
valid_labels = word_scores:treatment,word_scores:chronic_disease,word_scores:clinical_variable,word_scores:cancer,word_scores:gender,word_scores:pregnancy,word_scores:allergy_name,word_scores:contraception_consent,word_scores:language_fluency,word_scores:technology_access,word_scores:ethnicity

# Locality-sensitive hashing parameters for vocabulary search
lsh_rows = 3
lsh_bands = 16
```

--------------------------------

### Upgrade ONNX and Torch with Conda

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Upgrade ONNX and PyTorch using Conda. These commands ensure compatibility with PyText.

```shell
conda install onnx -c conda-forge
```

```shell
conda install pytorch torchvision -c pytorch
```

--------------------------------

### Run CFG Parser Script

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Execute the CFG parser using the provided shell script. This script parses clinical trial eligibility criteria and outputs relations to a TSV file. Configuration can be adjusted via command-line arguments or a config file.

```shell
./script/cfg_parse.sh
```

--------------------------------

### Train Custom NER Model (Bash)

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Commands to train and test a custom Named Entity Recognition (NER) model using PyText. The model extracts 13 entity types, including demographics, diseases, treatments, and clinical variables. Training data should be in TSV format with entity spans and normalized text.

```bash
# Train a new NER model
pytext train < src/resources/config/ner.json

# Test the NER model
pytext test < src/resources/config/ner.json

# The NER model extracts 13 entity types:
# - age, bmi: demographic measurements
# - cancer: cancer/tumor mentions
# - chronic_disease: disease conditions
# - clinical_variable: lab values, vitals
# - treatment: medications, procedures
# - pregnancy: pregnancy status
# - gender: male/female mentions
# - allergy_name: allergy mentions
# - contraception_consent: contraception requirements
# - language_fluency: language requirements
# - technology_access: technology requirements
# - ethnicity: ethnicity mentions
# - lower_bound, upper_bound: numerical constraints

# Training data format (data/ner/train_processed_medical_ner.tsv):
# column1: entity spans (start:end:label,...)
# column2: normalized text
# Example:
# 1:13:chronic_disease,17:30:treatment	major trauma or major surgery within @NUMBER days
```

--------------------------------

### Run IE Pipeline for Medical Entity Extraction

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Executes the three-stage IE pipeline to extract criteria, perform NER, and link entities to MeSH concepts.

```bash
# Run the complete IE pipeline
./script/ie_parse.sh

# The pipeline executes three stages:

# Stage 1: Extract inclusion/exclusion criteria
go run src/cmd/extract/main.go \
  -i data/input/clinical_trials.csv \
  -o data/output/ie_extracted_clinical_trials.tsv \
  -logtostderr

# Stage 2: Run NER to detect medical entities
export PYTHONPATH="$(pwd)/src"
python src/ie/ner.py \
  -m bin/ner.c2 \
  -i data/output/ie_extracted_clinical_trials.tsv \
  -o data/output/ie_ner_clinical_trials.tsv

# Stage 3: Link entities to MeSH concepts (NEL)
go run src/cmd/nel/main.go \
  -conf src/resources/config/nel.conf \
  -i data/output/ie_ner_clinical_trials.tsv \
  -o data/output/ie_parsed_clinical_trials.tsv \
  -logtostderr

# Output includes matched MeSH concepts:
# nct_id  eligibility_type  criterion  label  term  ner_score  concepts  tree_numbers  nel_score
# NCT04346355  exclusion  Diverticulitis or intestinal perforation  word_scores:chronic_disease  diverticulitis  0.996  Diverticulitis  C06.405.205.282.500  1.000
```

--------------------------------

### Run IE Parser Script

Source: https://github.com/facebookresearch/clinical-trial-parser/blob/main/doc/developer_guide.md

Execute the IE parser using the provided shell script. This script identifies medical terms and matched concepts, writing them to a TSV file. Ensure word embeddings are unzipped and MeSH vocabulary is downloaded.

```shell
./script/ie_parse.sh
```

--------------------------------

### Ingest Clinical Trial Data from AACT

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Scripts for downloading and ingesting clinical trial data, including the underlying SQL query used for extraction.

```bash
# Prerequisites: PostgreSQL 11+ installed

# Download and restore AACT database snapshot
./script/aact.sh

# Ingest clinical trials from AACT to CSV
./script/ingest.sh

# The ingest script executes SQL queries like:
# SELECT
#   s.nct_id,
#   s.brief_title,
#   CASE WHEN f.nct_id IS NOT NULL THEN true ELSE false END AS has_us_facility,
#   string_agg(DISTINCT c.name, '|') AS conditions,
#   e.criteria AS eligibility_criteria
# FROM studies s
# LEFT JOIN facilities f ON s.nct_id = f.nct_id AND f.country = 'United States'
# LEFT JOIN conditions c ON s.nct_id = c.nct_id
# LEFT JOIN eligibilities e ON s.nct_id = e.nct_id
# WHERE s.overall_status = 'Recruiting'
# GROUP BY s.nct_id, s.brief_title, e.criteria
# ORDER BY s.nct_id;

# Output format matches clinical_trials.csv:
# nct_id,title,has_us_facility,conditions,eligibility_criteria
```

--------------------------------

### Custom MeSH Vocabulary (TSV)

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Format for extending the MeSH vocabulary with custom medical concepts. Each line represents a concept with its name, synonyms, and MeSH tree number. This improves entity linking for domain-specific terms.

```tsv
# data/mesh/custom_mesh_concepts_p1.tsv
# Format: Concept Name<TAB>Synonyms<TAB>Tree Number

COVID-19	sars-cov2 infection|coronavirus disease 2019|covid	C01.925
Smartphone Access	smartphone|iphone|android phone|mobile phone	TA01.01
Email Access	email|e-mail|electronic mail	TA01.02

# data/mesh/custom_mesh_concepts_p2.tsv
# Additional custom concepts for contraception, language fluency, etc.

Intrauterine Devices, Consent	intrauterine device|iud	CC01.03.01
Language Fluency, English	fluent english|english fluency|speak english	LF01.01
```

--------------------------------

### Variables Dictionary CSV Format

Source: https://context7.com/facebookresearch/clinical-trial-parser/llms.txt

Header definition for the medical variables CSV file.

```csv
# src/resources/variables/variables.csv
# variable_id,variable_type,variable_name,display_name,aliases,bounds,default_unit_name,question
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.