ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology

National University of Singapore Obstetrics and Gynecology Artificial Intelligence (NUS OBGYN-AI) Collaborative Group

doi:10.1016/j.ajog.2023.04.020

ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology

National University of Singapore Obstetrics and Gynecology Artificial Intelligence (NUS OBGYN-AI) Collaborative Group

Research output: Contribution to journal › Article › peer-review

9 Scopus citations

Abstract

Background: Natural language processing is a form of artificial intelligence that allows human users to interface with a machine without using complex codes. The ability of natural language processing systems, such as ChatGPT, to successfully engage with healthcare systems requiring fluid reasoning, specialist data interpretation, and empathetic communication in an unfamiliar and evolving environment is poorly studied. This study investigated whether the ChatGPT interface could engage with and complete a mock objective structured clinical examination simulating assessment for membership of the Royal College of Obstetricians and Gynaecologists. Objective: This study aimed to determine whether ChatGPT, without additional training, would achieve a score at least equivalent to that achieved by human candidates who sat for virtual objective structured clinical examinations in Singapore. Study Design: This study was conducted in 2 phases. In the first phase, a total of 7 structured discussion questions were selected from 2 historical cohorts (cohorts A and B) of objective structured clinical examination questions. ChatGPT was examined using these questions and responses recorded in a script. Of note, 2 human candidates (acting as anonymizers) were examined on the same questions using videoconferencing, and their responses were transcribed verbatim into written scripts. The 3 sets of response scripts were mixed, and each set was allocated to 1 of 3 human actors. In the second phase, actors were used to presenting these scripts to examiners in response to the same examination questions. These responses were blind scored by 14 qualified examiners. ChatGPT scores were unblinded and compared with historical human candidate performance scores. Results: The average score given to ChatGPT by 14 examiners was 77.2%. The average historical human score (n=26 candidates) was 73.7 %. ChatGPT demonstrated sizable performance improvements over the average human candidate in several subject domains. The median time taken for ChatGPT to complete each station was 2.54 minutes, well before the 10 minutes allowed. Conclusion: ChatGPT generated factually accurate and contextually relevant structured discussion answers to complex and evolving clinical questions based on unfamiliar settings within a very short period. ChatGPT outperformed human candidates in several knowledge areas. Not all examiners were able to discern between human and ChatGPT responses. Our data highlight the emergent ability of natural language processing models to demonstrate fluid reasoning in unfamiliar environments and successfully compete with human candidates that have undergone extensive specialist training.

Original language	English
Pages (from-to)	172.e1-172.e12
Journal	American Journal of Obstetrics and Gynecology
Volume	229
Issue number	2
DOIs	https://doi.org/10.1016/j.ajog.2023.04.020
State	Published - Aug 2023

Bibliographical note

Funding Information:
We acknowledge the Obstetrical and Gynaecological Society of Singapore for providing the mock objective structured clinical examination questions and anonymized historical scores. In addition, we would like to acknowledge the Royal College of Obstetricians and Gynaecologists examinations department for providing us with results for the Member of the Royal College of Obstetricians and Gynaecologists Part 3 examination for Singapore candidates over the past 2 years. Lastly, we sincerely appreciate the efforts of all who volunteered as candidates and examiners in this study. The members of the NUS OBGYN-AI Collaborative Group are as follows: • Department of Obstetrics and Gynaecology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore: Mary Rauff, FRCOG; Arijit Biswas, FAMS; Jeffrey J.H. Low, FAMS; Joseph S. Ng, FAMS; Arundhati T. Gosavi, FRCOG; Zhongwei Huang, PhD; and Preethi Rajendran, BSc • Department of Obstetrics and Gynaecology, National University Hospital, Singapore, Singapore: Kwok Weng Roy Ng, FAMS; Arundhati T. Gosavi, FRCOG; Jeslyn J.L. Wong, MMED; Grace M.F. Chan, MBBS; Kelvin Zhi Xing Lee, MD; Jeannie J.Y. Yap, MBChB; Nurulhuda Ahmad, MBBS; and Shwetha Shanmugam, MRCOG • Department of Obstetrics and Gynaecology, KK Women's and Children's Hospital, Singapore, Singapore: Rajeswari Kathirvel, FRCOG, and Manisha Mathur, FRCOG • Bia-Echo Asia Centre for Reproductive Longevity and Equality, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore: Zhongwei Huang, PhD • Department of Obstetrics and Gynaecology, Academia, Singapore General Hospital, Singapore, Singapore: Whui Whui Lim, MRCOG • Obstetrical and Gynaecological Society of Singapore (OGSS), Singapore, Singapore and Singapore International Representative Committee, Royal College of Obstetricians and Gynaecologists: Min Yu Lim, FRCOG. This study received no financial support.

Publisher Copyright:
© 2023 The Author(s)

Keywords

Chat Generative Pre-trained Transformer
artificial intelligence
objective structured clinical examination
obstetrics and gynecology
postgraduate specialty training
reasoning

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

Access to Document

10.1016/j.ajog.2023.04.020

Cite this

@article{cd1d12f4f87b43a5b484beb4207c8fc5,

title = "ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology",

abstract = "Background: Natural language processing is a form of artificial intelligence that allows human users to interface with a machine without using complex codes. The ability of natural language processing systems, such as ChatGPT, to successfully engage with healthcare systems requiring fluid reasoning, specialist data interpretation, and empathetic communication in an unfamiliar and evolving environment is poorly studied. This study investigated whether the ChatGPT interface could engage with and complete a mock objective structured clinical examination simulating assessment for membership of the Royal College of Obstetricians and Gynaecologists. Objective: This study aimed to determine whether ChatGPT, without additional training, would achieve a score at least equivalent to that achieved by human candidates who sat for virtual objective structured clinical examinations in Singapore. Study Design: This study was conducted in 2 phases. In the first phase, a total of 7 structured discussion questions were selected from 2 historical cohorts (cohorts A and B) of objective structured clinical examination questions. ChatGPT was examined using these questions and responses recorded in a script. Of note, 2 human candidates (acting as anonymizers) were examined on the same questions using videoconferencing, and their responses were transcribed verbatim into written scripts. The 3 sets of response scripts were mixed, and each set was allocated to 1 of 3 human actors. In the second phase, actors were used to presenting these scripts to examiners in response to the same examination questions. These responses were blind scored by 14 qualified examiners. ChatGPT scores were unblinded and compared with historical human candidate performance scores. Results: The average score given to ChatGPT by 14 examiners was 77.2%. The average historical human score (n=26 candidates) was 73.7 %. ChatGPT demonstrated sizable performance improvements over the average human candidate in several subject domains. The median time taken for ChatGPT to complete each station was 2.54 minutes, well before the 10 minutes allowed. Conclusion: ChatGPT generated factually accurate and contextually relevant structured discussion answers to complex and evolving clinical questions based on unfamiliar settings within a very short period. ChatGPT outperformed human candidates in several knowledge areas. Not all examiners were able to discern between human and ChatGPT responses. Our data highlight the emergent ability of natural language processing models to demonstrate fluid reasoning in unfamiliar environments and successfully compete with human candidates that have undergone extensive specialist training.",

keywords = "Chat Generative Pre-trained Transformer, artificial intelligence, objective structured clinical examination, obstetrics and gynecology, postgraduate specialty training, reasoning",

author = "{National University of Singapore Obstetrics and Gynecology Artificial Intelligence (NUS OBGYN-AI) Collaborative Group} and Li, {Sarah W.} and Kemp, {Matthew W.} and Logan, {Susan J.S.} and Dimri, {Pooja Sharma} and Navkaran Singh and Mattar, {Citra N.Z.} and Pradip Dashraath and Harshaana Ramlal and Mahyuddin, {Aniza P.} and Suren Kanayan and Carter, {Sean W.D.} and Thain, {Serene P.T.} and Fee, {Erin L.} and Illanes, {Sebastian E.} and Choolani, {Mahesh A.} and Mary Rauff and Arijit Biswas and Low, {Jeffrey J.H.} and Ng, {Joseph S.} and Gosavi, {Arundhati T.} and Zhongwei Huang and Preethi Rajendran and Ng, {Kwok Weng Roy} and Wong, {Jeslyn J.L.} and Chan, {Grace M.F.} and Lee, {Kelvin Zhi Xing} and Yap, {Jeannie J.Y.} and Nurulhuda Ahmad and Shwetha Shanmugam and Rajeswari Kathirvel and Manisha Mathur and Lim, {Whui Whui} and Lim, {Min Yu}",

note = "Funding Information: We acknowledge the Obstetrical and Gynaecological Society of Singapore for providing the mock objective structured clinical examination questions and anonymized historical scores. In addition, we would like to acknowledge the Royal College of Obstetricians and Gynaecologists examinations department for providing us with results for the Member of the Royal College of Obstetricians and Gynaecologists Part 3 examination for Singapore candidates over the past 2 years. Lastly, we sincerely appreciate the efforts of all who volunteered as candidates and examiners in this study. The members of the NUS OBGYN-AI Collaborative Group are as follows: • Department of Obstetrics and Gynaecology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore: Mary Rauff, FRCOG; Arijit Biswas, FAMS; Jeffrey J.H. Low, FAMS; Joseph S. Ng, FAMS; Arundhati T. Gosavi, FRCOG; Zhongwei Huang, PhD; and Preethi Rajendran, BSc • Department of Obstetrics and Gynaecology, National University Hospital, Singapore, Singapore: Kwok Weng Roy Ng, FAMS; Arundhati T. Gosavi, FRCOG; Jeslyn J.L. Wong, MMED; Grace M.F. Chan, MBBS; Kelvin Zhi Xing Lee, MD; Jeannie J.Y. Yap, MBChB; Nurulhuda Ahmad, MBBS; and Shwetha Shanmugam, MRCOG • Department of Obstetrics and Gynaecology, KK Women's and Children's Hospital, Singapore, Singapore: Rajeswari Kathirvel, FRCOG, and Manisha Mathur, FRCOG • Bia-Echo Asia Centre for Reproductive Longevity and Equality, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore: Zhongwei Huang, PhD • Department of Obstetrics and Gynaecology, Academia, Singapore General Hospital, Singapore, Singapore: Whui Whui Lim, MRCOG • Obstetrical and Gynaecological Society of Singapore (OGSS), Singapore, Singapore and Singapore International Representative Committee, Royal College of Obstetricians and Gynaecologists: Min Yu Lim, FRCOG. This study received no financial support. Publisher Copyright: {\textcopyright} 2023 The Author(s)",

year = "2023",

month = aug,

doi = "10.1016/j.ajog.2023.04.020",

language = "English",

volume = "229",

pages = "172.e1--172.e12",

journal = "American Journal of Obstetrics and Gynecology",

issn = "0002-9378",

publisher = "Mosby Inc.",

number = "2",

}

National University of Singapore Obstetrics and Gynecology Artificial Intelligence (NUS OBGYN-AI) Collaborative Group 2023, 'ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology', American Journal of Obstetrics and Gynecology, vol. 229, no. 2, pp. 172.e1-172.e12. https://doi.org/10.1016/j.ajog.2023.04.020

ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology. / National University of Singapore Obstetrics and Gynecology Artificial Intelligence (NUS OBGYN-AI) Collaborative Group.
In: American Journal of Obstetrics and Gynecology, Vol. 229, No. 2, 08.2023, p. 172.e1-172.e12.

Research output: Contribution to journal › Article › peer-review

TY - JOUR

T1 - ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology

AU - National University of Singapore Obstetrics and Gynecology Artificial Intelligence (NUS OBGYN-AI) Collaborative Group

AU - Li, Sarah W.

AU - Kemp, Matthew W.

AU - Logan, Susan J.S.

AU - Dimri, Pooja Sharma

AU - Singh, Navkaran

AU - Mattar, Citra N.Z.

AU - Dashraath, Pradip

AU - Ramlal, Harshaana

AU - Mahyuddin, Aniza P.

AU - Kanayan, Suren

AU - Carter, Sean W.D.

AU - Thain, Serene P.T.

AU - Fee, Erin L.

AU - Illanes, Sebastian E.

AU - Choolani, Mahesh A.

AU - Rauff, Mary

AU - Biswas, Arijit

AU - Low, Jeffrey J.H.

AU - Ng, Joseph S.

AU - Gosavi, Arundhati T.

AU - Huang, Zhongwei

AU - Rajendran, Preethi

AU - Ng, Kwok Weng Roy

AU - Wong, Jeslyn J.L.

AU - Chan, Grace M.F.

AU - Lee, Kelvin Zhi Xing

AU - Yap, Jeannie J.Y.

AU - Ahmad, Nurulhuda

AU - Shanmugam, Shwetha

AU - Kathirvel, Rajeswari

AU - Mathur, Manisha

AU - Lim, Whui Whui

AU - Lim, Min Yu

N1 - Funding Information: We acknowledge the Obstetrical and Gynaecological Society of Singapore for providing the mock objective structured clinical examination questions and anonymized historical scores. In addition, we would like to acknowledge the Royal College of Obstetricians and Gynaecologists examinations department for providing us with results for the Member of the Royal College of Obstetricians and Gynaecologists Part 3 examination for Singapore candidates over the past 2 years. Lastly, we sincerely appreciate the efforts of all who volunteered as candidates and examiners in this study. The members of the NUS OBGYN-AI Collaborative Group are as follows: • Department of Obstetrics and Gynaecology, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore: Mary Rauff, FRCOG; Arijit Biswas, FAMS; Jeffrey J.H. Low, FAMS; Joseph S. Ng, FAMS; Arundhati T. Gosavi, FRCOG; Zhongwei Huang, PhD; and Preethi Rajendran, BSc • Department of Obstetrics and Gynaecology, National University Hospital, Singapore, Singapore: Kwok Weng Roy Ng, FAMS; Arundhati T. Gosavi, FRCOG; Jeslyn J.L. Wong, MMED; Grace M.F. Chan, MBBS; Kelvin Zhi Xing Lee, MD; Jeannie J.Y. Yap, MBChB; Nurulhuda Ahmad, MBBS; and Shwetha Shanmugam, MRCOG • Department of Obstetrics and Gynaecology, KK Women's and Children's Hospital, Singapore, Singapore: Rajeswari Kathirvel, FRCOG, and Manisha Mathur, FRCOG • Bia-Echo Asia Centre for Reproductive Longevity and Equality, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore: Zhongwei Huang, PhD • Department of Obstetrics and Gynaecology, Academia, Singapore General Hospital, Singapore, Singapore: Whui Whui Lim, MRCOG • Obstetrical and Gynaecological Society of Singapore (OGSS), Singapore, Singapore and Singapore International Representative Committee, Royal College of Obstetricians and Gynaecologists: Min Yu Lim, FRCOG. This study received no financial support. Publisher Copyright: © 2023 The Author(s)

PY - 2023/8

Y1 - 2023/8

N2 - Background: Natural language processing is a form of artificial intelligence that allows human users to interface with a machine without using complex codes. The ability of natural language processing systems, such as ChatGPT, to successfully engage with healthcare systems requiring fluid reasoning, specialist data interpretation, and empathetic communication in an unfamiliar and evolving environment is poorly studied. This study investigated whether the ChatGPT interface could engage with and complete a mock objective structured clinical examination simulating assessment for membership of the Royal College of Obstetricians and Gynaecologists. Objective: This study aimed to determine whether ChatGPT, without additional training, would achieve a score at least equivalent to that achieved by human candidates who sat for virtual objective structured clinical examinations in Singapore. Study Design: This study was conducted in 2 phases. In the first phase, a total of 7 structured discussion questions were selected from 2 historical cohorts (cohorts A and B) of objective structured clinical examination questions. ChatGPT was examined using these questions and responses recorded in a script. Of note, 2 human candidates (acting as anonymizers) were examined on the same questions using videoconferencing, and their responses were transcribed verbatim into written scripts. The 3 sets of response scripts were mixed, and each set was allocated to 1 of 3 human actors. In the second phase, actors were used to presenting these scripts to examiners in response to the same examination questions. These responses were blind scored by 14 qualified examiners. ChatGPT scores were unblinded and compared with historical human candidate performance scores. Results: The average score given to ChatGPT by 14 examiners was 77.2%. The average historical human score (n=26 candidates) was 73.7 %. ChatGPT demonstrated sizable performance improvements over the average human candidate in several subject domains. The median time taken for ChatGPT to complete each station was 2.54 minutes, well before the 10 minutes allowed. Conclusion: ChatGPT generated factually accurate and contextually relevant structured discussion answers to complex and evolving clinical questions based on unfamiliar settings within a very short period. ChatGPT outperformed human candidates in several knowledge areas. Not all examiners were able to discern between human and ChatGPT responses. Our data highlight the emergent ability of natural language processing models to demonstrate fluid reasoning in unfamiliar environments and successfully compete with human candidates that have undergone extensive specialist training.

AB - Background: Natural language processing is a form of artificial intelligence that allows human users to interface with a machine without using complex codes. The ability of natural language processing systems, such as ChatGPT, to successfully engage with healthcare systems requiring fluid reasoning, specialist data interpretation, and empathetic communication in an unfamiliar and evolving environment is poorly studied. This study investigated whether the ChatGPT interface could engage with and complete a mock objective structured clinical examination simulating assessment for membership of the Royal College of Obstetricians and Gynaecologists. Objective: This study aimed to determine whether ChatGPT, without additional training, would achieve a score at least equivalent to that achieved by human candidates who sat for virtual objective structured clinical examinations in Singapore. Study Design: This study was conducted in 2 phases. In the first phase, a total of 7 structured discussion questions were selected from 2 historical cohorts (cohorts A and B) of objective structured clinical examination questions. ChatGPT was examined using these questions and responses recorded in a script. Of note, 2 human candidates (acting as anonymizers) were examined on the same questions using videoconferencing, and their responses were transcribed verbatim into written scripts. The 3 sets of response scripts were mixed, and each set was allocated to 1 of 3 human actors. In the second phase, actors were used to presenting these scripts to examiners in response to the same examination questions. These responses were blind scored by 14 qualified examiners. ChatGPT scores were unblinded and compared with historical human candidate performance scores. Results: The average score given to ChatGPT by 14 examiners was 77.2%. The average historical human score (n=26 candidates) was 73.7 %. ChatGPT demonstrated sizable performance improvements over the average human candidate in several subject domains. The median time taken for ChatGPT to complete each station was 2.54 minutes, well before the 10 minutes allowed. Conclusion: ChatGPT generated factually accurate and contextually relevant structured discussion answers to complex and evolving clinical questions based on unfamiliar settings within a very short period. ChatGPT outperformed human candidates in several knowledge areas. Not all examiners were able to discern between human and ChatGPT responses. Our data highlight the emergent ability of natural language processing models to demonstrate fluid reasoning in unfamiliar environments and successfully compete with human candidates that have undergone extensive specialist training.

KW - Chat Generative Pre-trained Transformer

KW - artificial intelligence

KW - objective structured clinical examination

KW - obstetrics and gynecology

KW - postgraduate specialty training

KW - reasoning

UR - http://www.scopus.com/inward/record.url?scp=85160042351&partnerID=8YFLogxK

U2 - 10.1016/j.ajog.2023.04.020

DO - 10.1016/j.ajog.2023.04.020

M3 - Article

C2 - 37088277

AN - SCOPUS:85160042351

SN - 0002-9378

VL - 229

SP - 172.e1-172.e12

JO - American Journal of Obstetrics and Gynecology

JF - American Journal of Obstetrics and Gynecology

IS - 2

ER -

ChatGPT outscored human candidates in a virtual objective structured clinical examination in obstetrics and gynecology

Abstract

Bibliographical note

Keywords

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this