Natural Language Inference for Patient Recruitment (NLI4PR)

Dataset Summary

This dataset card aims to be a base template for new datasets. It has been generated using this raw template.

Supported Tasks and Leaderboards

Natural Language Inference.

Language

English

Dataset Structure

Data Instances

Each instance of the dataset has the following fields and the following types of fields.

{
        "id": "621",
        "topic_id": "2",
        "statement_medical": "A 32-year-old woman comes to the hospital with vaginal spotting.  Her last menstrual period was 10 weeks ago. She has regular menses lasting for 6 days and repeating every 29 days. Medical history is significant for appendectomy and several complicated UTIs. She has multiple male partners, and she is inconsistent with using barrier contraceptives. Vital signs are normal.  Serum \u03b2-hCG level is 1800 mIU/mL, and a repeat level after 2 days shows an abnormal rise to 2100 mIU/mL.  Pelvic ultrasound reveals a thin endometrium with no gestational sac in the uterus.",
        "statement_pol": "I just turned 32 and last morning I woke up with strange blood stains on my underwear. My last periods were more than 2 months ago, which is unusual for me because I used to have regular periods lasting for 6 days every 29 days, more or less. I had several UTIs in the past. I also had appendicitis. I'm currently seeing several men and, to be honest, some of them do struggle to wear a condom. I went to the hospital to check myself up and they told me that my vitals were normal. I also had a blood test on Monday, and my \u03b2-hCG level was 1800 mIU/mL, and then on Wednesday, it went up to 2100 mIU/mL. The gynecologist also did an ultrasound and she told me that, hopefully, there was no ovule.",
        "premise": "Inclusion Criteria:\n\n          -  women with PUL\n\n        Exclusion Criteria:\nFemale\nAccepts Healthy Volunteers\n\n",
        "NCT_title": "Hysteroscopy for Pregnancy of Unknown Location",
        "NCT_id": "NCT02637739",
        "label": "Entailment"
}

Data Fields

Each instance has the following fields: id, topic_id, statement_medical, statement_pol, premise, NCT_title, NCT_id, label.

Data Splits

Train: 4904 instances Validation: 525 instances Test: 1578 instances

Dataset Creation

Source Data

Initial Data Collection and Normalization

premise (CTRs) and statement_medical taken from TREC-CT 2022.

Annotations

Annotation process

Automatic mapping of TREC-CT 2022's ranking to NLI annotations. eligible mapped as Entailment and excluded mapped as Contradiction.

Manual rephrasing of original statement_medical

Annotators

Paper's first author.

Dataset statistics

Split	# Entailment	# Contradiction
Train	2757	2147
Dev.	295	230
Test	887	691

Licensing Information

MIT

Citation Information

  @misc{aguiar2025ieligiblenaturallanguage,
        title={Am I eligible? Natural Language Inference for Clinical Trial Patient Recruitment: the Patient's Point of View}, 
        author={Mathilde Aguiar and Pierre Zweigenbaum and Nona Naderi},
        year={2025},
        eprint={2503.15718},
        archivePrefix={arXiv},
        primaryClass={cs.CL},
        url={https://arxiv.org/abs/2503.15718}, 
  }

Contributions