The DOI matches its Crossref record; 43 of the 43 references it cites were found in a registry or an index; 1 of the 5 sources its claims cite were checked for support.
A Curated Dataset for Question Answering on Laboratory Test Reference Ranges and Interpretation
Open accessAbstract
LabQAR is a manually curated lab QA dataset with 550 reference ranges covering 363 unique lab tests and two QA sets for reference-range prediction and low/normal/high classification.
Overview
CoChat Summary
## TL;DR
LabQAR is a manually curated question-answering dataset built from 550 laboratory test reference ranges covering 363 unique tests, designed to evaluate both reference-range retrieval and lab-result interpretation. In baseline experiments, GPT-4o performed best on the reported tasks, reaching 63.3% accuracy on multiple-choice classification.
## Key findings
- The dataset contains 550 lab test reference ranges spanning 363 unique lab tests and includes contextual variation such as gender-specific (26), age group-specific (7), women-related condition (15), and other condition (33) cases.
- LabQAR was turned into two QA sets, each with 550 questions: Set 1 asks for the correct reference range, and Set 2 asks whether a value is Low, Normal, or High.
- In baseline evaluation, GPT-4o achieved 63.3% accuracy on Set 2, compared with 43.1% for GPT-4 and 26.7% for LLaMA 3.1.
- For Set 1, GPT-4 Turbo improved from 27.8% exact match to 49.8% within ±20%, and GPT-4o from 30.9% exact match to 50.5% within ±20%.
## Methods
- The dataset was manually curated from authoritative sources, including Laposata's Laboratory Medicine, ABIM, and Stanford Medicine; the paper says the 550 reference ranges were extracted, standardized to include traditional and SI units, and mapped to LOINC identifiers.
- Four undergraduate annotators extracted metadata such as test name, specimen type, age/sex specificity, clinical condition, and reference intervals; disagreements were resolved by consensus under supervision.
- The authors built two JSON QA formats: open-ended range retrieval (Set 1) and multiple-choice classification (Set 2), then benchmarked LLaMA 3.1, GatorTronGPT, GPT-3.5 Turbo, GPT-4 Turbo, and GPT-4o using exact match plus ±10%/±20% tolerance for Set 1 and exact-match accuracy for Set 2.
## Limitations
- The paper explicitly says the benchmarking experiments are only baseline demonstrations and do not constitute technical validation of the dataset itself or clinical deployment of the models.
- The authors note that because GPT-4o pretraining details are not publicly disclosed, they cannot attribute its performance to specific architectural or data factors; this is stated in the paper.
- Evident from the text: the evaluation uses synthetically generated Set 2 values and tolerance-based scoring for Set 1, so reported performance is tied to the dataset construction and scoring choices rather than real-world clinical use.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
1 Valid · 1 Not valid · 2 Not checked · 76 Own results and interpretations
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The findings indicate that GPT-4o outperforms other models, showcasing the potential of LLMs in clinical decision support.”
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“encompassing 363 unique lab tests”
- Also stated in: § Data Records The LabQAR dataset is available at Figshare (https://doi.org/10.6084/m9.figshare.29189894) and is stored in JSON format. It consists of an array of question-answer pairs, each identified by a unique question ID. Each record represents a specific laboratory test scenario and includes fields for the question ID, the full ARTICLE IN PRESS text of the lab-related question (detailing the test name, specimen type, demographic attributes such as sex and age group, and relevant clinical condition), and the expected answer. The dataset is organized into two main sets of question-answer pairs. Set 1 consists of reference range questions, where each record presents a lab test scenario specifying the test name, specimen type, demographic attributes (such as sex and age group), and clinical condition. The question asks for the correct lower and upper bound values for the test's SI reference range, and the answer provides these values directly. Set 2 includes lab test classification questions, in which a specific lab value is given for the same detailed scenario. The question asks whether the provided value is normal, low, or high, and respondents select the answer from predefined choices (e.g., 'A: Normal, B: Low, C: High'). The dataset entries include fields for the question ID, the question text, answer choices (for Set 2), and the correct answer. A comprehensive README accompanies the dataset, detailing the annotation schema, including lab tests, specimen types, clinical conditions, answer formats, and providing representative examples for both sets. An example of the data format for one record can be found in the README file. The annotation guidelines (annotation\_guidelines.pdf) and the README (LabQAR\_README.md) file are included with the Figshare dataset. The dataset is available at Figshare, with this section serving as the primary source of information on the availability and content of the data described here. The code used for preparing the LabQAR dataset and conducting benchmarking experiments is available at (https://github.com/balubhasuran/LabQAR ) . Data Overview The LabQAR dataset comprises 550 lab test reference ranges derived from authoritative medical sources, encompassing 363 unique lab tests. Among these, 282 tests are general-purpose with no specific subgroup or condition, such as Albumin , which applies broadly across populations. A subset of 26 tests is tailored to gender-specific reference ranges, exemplified by Apolipoprotein A , with separate thresholds for males and females. Additionally, 7 tests exhibit age-specific variations, including cholesterol levels that differ between adults and children. The dataset also captures women-specific physiological contexts for 15 tests, such as 17α-Hydroxyprogesterone , with stratification across the follicular, luteal, and postmenopausal phases. Furthermore, 33 tests are linked to other clinical conditions, like Carboxyhemoglobin , which is annotated with contextual values for nonsmokers and toxic exposure cases. This rich stratification enables precise, context-aware evaluation of large language models in clinical reasoning tasks. The 550 lab test reference ranges map to 363 unique laboratory tests because a single test may appear multiple times under · plain
“encompassing 363 unique lab tests”
- Also stated in: § ARTICLE IN PRESS · plain
“These reference ranges encompassed 363 unique lab tests”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Also stated in: § Data Records The LabQAR dataset is available at Figshare (https://doi.org/10.6084/m9.figshare.29189894) and is stored in JSON format. It consists of an array of question-answer pairs, each identified by a unique question ID. Each record represents a specific laboratory test scenario and includes fields for the question ID, the full ARTICLE IN PRESS text of the lab-related question (detailing the test name, specimen type, demographic attributes such as sex and age group, and relevant clinical condition), and the expected answer. The dataset is organized into two main sets of question-answer pairs. Set 1 consists of reference range questions, where each record presents a lab test scenario specifying the test name, specimen type, demographic attributes (such as sex and age group), and clinical condition. The question asks for the correct lower and upper bound values for the test's SI reference range, and the answer provides these values directly. Set 2 includes lab test classification questions, in which a specific lab value is given for the same detailed scenario. The question asks whether the provided value is normal, low, or high, and respondents select the answer from predefined choices (e.g., 'A: Normal, B: Low, C: High'). The dataset entries include fields for the question ID, the question text, answer choices (for Set 2), and the correct answer. A comprehensive README accompanies the dataset, detailing the annotation schema, including lab tests, specimen types, clinical conditions, answer formats, and providing representative examples for both sets. An example of the data format for one record can be found in the README file. The annotation guidelines (annotation\_guidelines.pdf) and the README (LabQAR\_README.md) file are included with the Figshare dataset. The dataset is available at Figshare, with this section serving as the primary source of information on the availability and content of the data described here. The code used for preparing the LabQAR dataset and conducting benchmarking experiments is available at (https://github.com/balubhasuran/LabQAR ) . Data Overview The LabQAR dataset comprises 550 lab test reference ranges derived from authoritative medical sources, encompassing 363 unique lab tests. Among these, 282 tests are general-purpose with no specific subgroup or condition, such as Albumin , which applies broadly across populations. A subset of 26 tests is tailored to gender-specific reference ranges, exemplified by Apolipoprotein A , with separate thresholds for males and females. Additionally, 7 tests exhibit age-specific variations, including cholesterol levels that differ between adults and children. The dataset also captures women-specific physiological contexts for 15 tests, such as 17α-Hydroxyprogesterone , with stratification across the follicular, luteal, and postmenopausal phases. Furthermore, 33 tests are linked to other clinical conditions, like Carboxyhemoglobin , which is annotated with contextual values for nonsmokers and toxic exposure cases. This rich stratification enables precise, context-aware evaluation of large language models in clinical reasoning tasks. The 550 lab test reference ranges map to 363 unique laboratory tests because a single test may appear multiple times under · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“including multiple-choice questions with annotations on reference ranges, specimen types, and other factors impacting interpretation”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“we present LabQAR ( Lab oratory Q uestion A nswering with R eference Ranges), a manually curated dataset comprising 550 lab test reference ranges derived from authoritative medical sources”
- Also stated in: § Data Records The LabQAR dataset is available at Figshare (https://doi.org/10.6084/m9.figshare.29189894) and is stored in JSON format. It consists of an array of question-answer pairs, each identified by a unique question ID. Each record represents a specific laboratory test scenario and includes fields for the question ID, the full ARTICLE IN PRESS text of the lab-related question (detailing the test name, specimen type, demographic attributes such as sex and age group, and relevant clinical condition), and the expected answer. The dataset is organized into two main sets of question-answer pairs. Set 1 consists of reference range questions, where each record presents a lab test scenario specifying the test name, specimen type, demographic attributes (such as sex and age group), and clinical condition. The question asks for the correct lower and upper bound values for the test's SI reference range, and the answer provides these values directly. Set 2 includes lab test classification questions, in which a specific lab value is given for the same detailed scenario. The question asks whether the provided value is normal, low, or high, and respondents select the answer from predefined choices (e.g., 'A: Normal, B: Low, C: High'). The dataset entries include fields for the question ID, the question text, answer choices (for Set 2), and the correct answer. A comprehensive README accompanies the dataset, detailing the annotation schema, including lab tests, specimen types, clinical conditions, answer formats, and providing representative examples for both sets. An example of the data format for one record can be found in the README file. The annotation guidelines (annotation\_guidelines.pdf) and the README (LabQAR\_README.md) file are included with the Figshare dataset. The dataset is available at Figshare, with this section serving as the primary source of information on the availability and content of the data described here. The code used for preparing the LabQAR dataset and conducting benchmarking experiments is available at (https://github.com/balubhasuran/LabQAR ) . Data Overview The LabQAR dataset comprises 550 lab test reference ranges derived from authoritative medical sources, encompassing 363 unique lab tests. Among these, 282 tests are general-purpose with no specific subgroup or condition, such as Albumin , which applies broadly across populations. A subset of 26 tests is tailored to gender-specific reference ranges, exemplified by Apolipoprotein A , with separate thresholds for males and females. Additionally, 7 tests exhibit age-specific variations, including cholesterol levels that differ between adults and children. The dataset also captures women-specific physiological contexts for 15 tests, such as 17α-Hydroxyprogesterone , with stratification across the follicular, luteal, and postmenopausal phases. Furthermore, 33 tests are linked to other clinical conditions, like Carboxyhemoglobin , which is annotated with contextual values for nonsmokers and toxic exposure cases. This rich stratification enables precise, context-aware evaluation of large language models in clinical reasoning tasks. The 550 lab test reference ranges map to 363 unique laboratory tests because a single test may appear multiple times under · plain
“The LabQAR dataset comprises 550 lab test reference ranges derived from authoritative medical sources, encompassing 363 unique lab tests.”
- Also stated in: § Input Data Sources · plain
“The LabQAR dataset was created by compiling information from 550 laboratory tests gathered from authoritative medical resources”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Rests on an assumption not shown here.
- Also stated in: § Data Records The LabQAR dataset is available at Figshare (https://doi.org/10.6084/m9.figshare.29189894) and is stored in JSON format. It consists of an array of question-answer pairs, each identified by a unique question ID. Each record represents a specific laboratory test scenario and includes fields for the question ID, the full ARTICLE IN PRESS text of the lab-related question (detailing the test name, specimen type, demographic attributes such as sex and age group, and relevant clinical condition), and the expected answer. The dataset is organized into two main sets of question-answer pairs. Set 1 consists of reference range questions, where each record presents a lab test scenario specifying the test name, specimen type, demographic attributes (such as sex and age group), and clinical condition. The question asks for the correct lower and upper bound values for the test's SI reference range, and the answer provides these values directly. Set 2 includes lab test classification questions, in which a specific lab value is given for the same detailed scenario. The question asks whether the provided value is normal, low, or high, and respondents select the answer from predefined choices (e.g., 'A: Normal, B: Low, C: High'). The dataset entries include fields for the question ID, the question text, answer choices (for Set 2), and the correct answer. A comprehensive README accompanies the dataset, detailing the annotation schema, including lab tests, specimen types, clinical conditions, answer formats, and providing representative examples for both sets. An example of the data format for one record can be found in the README file. The annotation guidelines (annotation\_guidelines.pdf) and the README (LabQAR\_README.md) file are included with the Figshare dataset. The dataset is available at Figshare, with this section serving as the primary source of information on the availability and content of the data described here. The code used for preparing the LabQAR dataset and conducting benchmarking experiments is available at (https://github.com/balubhasuran/LabQAR ) . Data Overview The LabQAR dataset comprises 550 lab test reference ranges derived from authoritative medical sources, encompassing 363 unique lab tests. Among these, 282 tests are general-purpose with no specific subgroup or condition, such as Albumin , which applies broadly across populations. A subset of 26 tests is tailored to gender-specific reference ranges, exemplified by Apolipoprotein A , with separate thresholds for males and females. Additionally, 7 tests exhibit age-specific variations, including cholesterol levels that differ between adults and children. The dataset also captures women-specific physiological contexts for 15 tests, such as 17α-Hydroxyprogesterone , with stratification across the follicular, luteal, and postmenopausal phases. Furthermore, 33 tests are linked to other clinical conditions, like Carboxyhemoglobin , which is annotated with contextual values for nonsmokers and toxic exposure cases. This rich stratification enables precise, context-aware evaluation of large language models in clinical reasoning tasks. The 550 lab test reference ranges map to 363 unique laboratory tests because a single test may appear multiple times under · plain
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Set 2: 550 multiple-choice questions asking whether a provided lab value is 'normal', 'low', or 'high' based on those same contextual factors.”
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
28 more claims
Create a free account to see all 80 claims and every source verdict.
- Probably not a substantial claim. 29
- It may state two separate claims as one. 30
- The quote only partly states this claim. 1
Citations
Cites
The publisher hasn’t deposited this paper’s reference list.
Research works printed but not linked: 9
- Pal, A., Umapathi, L. K. & Sankarasubbu, M. MedMCQA: A Large-scale Multi-Subject Multi-Choice Dataset for Medical domain Question Answering. in Proceedings of the Conference on Health, Inference, and Learning (eds Flores, G., Chen, G. H., Pollard, T., Ho, J. C. & Naumann, T.) vol. 174 248–260 (PMLR, 2022).
- Vaswani, A. et al. Attention is All you Need. in Advances in Neural Information Processing Systems (eds Guyon, I. et al.) vol. 30 (Curran Associates, Inc., 2017).
- Laposata, M. Clinical Laboratory Reference Values. in Laposata’s Laboratory Medicine: Diagnosis of Disease in the Clinical Laboratory, 4th Edition (McGraw Hill, New York, NY, 2025).
- Clinical Guide to Laboratory Tests. (W.B. Saunders, Philadelphia London, 1995).
- AMA Manual of Style: A Guide for Authors and Editors. (Oxford Univ. Pr, Oxford, 2007).
- Clinical Diagnosis and Management by Laboratory Methods. (Saunders, Philadelphia, Pa. London, 2001).
- Tietz Textbook of Clinical Chemistry and Molecular Diagnostics (Saunders, St. Louis, Mo, 2013).
- Laposata, M. Clinical Laboratory Reference Values. in Laposata’s Laboratory Medicine: Diagnosis of Disease in the Clinical Laboratory, 3e (McGraw-Hill Education, New York, NY, 2019).
- Brown, T. et al. Language Models are Few-Shot Learners. In Advances in Neural Information Processing Systems. (edsLarochelle, H., Ranzato, M., Hadsell, R., Balcan, M. F. & Lin, H.) vol. 33 1877–1901 (Curran Associates, Inc., 2020).
Metadata
- Authors
- Bhasuran, B., Jin, Q., Deville, A., Wu, Y., Hanna, K., Lu, Z., & He, Z.
- Affiliations
- Florida State University (US)
- United States National Library of Medicine (US)
- University of Florida Health (US)
- University of South Florida (US)
- Corresponding author
- He, Zhe
- Venue
- Scientific Data
- Published
- 2026-06-10
- Volume
- 13
- Issue
- 1
- ISSN
- 2052-4463 (ISSN-L)
- PMC
- PMC13720059
- OpenAlex
- W7164148867
- Semantic Scholar
- 38a868a7be9a75c18a1c13d21fc7e7135db216e2
- Cited by
- 0
- References
- 0
- Type
- data paper
- Language
- en
- Open access
- gold
- doi.org
- Licence
- cc-by-nc-nd
- creativecommons.org/licenses/by-nc-nd/4.0 (tdm)
- creativecommons.org/licenses/by-nc-nd/4.0 (vor, 141 d)
- Locations
- Scientific Data (publishedVersion, cc-by-nc-nd)
- PubMed (publishedVersion)
- PubMed Central (publishedVersion, cc-by-nc-nd)
- Funders
- U.S. Department of Health & Human Services | Agency for Healthcare Research and Quality (R21HS029969)
- U.S. Department of Health & Human Services | NIH | National Center for Advancing Translational Sciences (UM1TR005128)
- École Nationale d'Ingénieurs de Saint-Etienne (P30AG073105)
- U.S. Department of Health & Human Services | NIH | U.S. National Library of Medicine
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Health Sciences › Health Professions › Medical Laboratory Technology › Clinical Laboratory Practices and Quality Control
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
- Keywords
- laboratory test interpretation, question answering, clinical decision support, large language models, GPT-4o, multiple-choice questions, specimen type
- MeSH terms
- Large Language Models, Humans, Reference Values, Clinical Laboratory Techniques, Decision Support Systems, Clinical
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: A Curated Dataset for Question Answering on Laboratory Test Reference Ranges and Interpretation. https://app.cochat.ai/papers/10.1038/s41597-026-07554-z
- The paper
- Bhasuran, B., Jin, Q., Deville, A., Wu, Y., Hanna, K., Lu, Z., & He, Z. (2026). A Curated Dataset for Question Answering on Laboratory Test Reference Ranges and Interpretation. Scientific Data. https://doi.org/10.1038/s41597-026-07554-z
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.