Lenard Cheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Lenard Cheng has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Attila Julius Hertelendy
Beth Israel Deaconess Medical Center
3 shared publications - Ryan Hata
Department of Emergency Medicine, Indiana University School of Medicine, Indianapolis, IN, United States
3 shared publications - Gregory R. Ciottone
Disaster Medicine Fellowship, Department of Emergency Medicine, Beth Israel Deaconess Medical Center, Boston, MA, USA
2 shared publications - Jeffrey Michael Franc
University of Alberta
2 shared publications - Manuela Verde
CRIMEDIM - Center for Research and Training in Disaster Medicine, Humanitarian Aid and Global Health, Università del Piemonte Orientale, Novara, Italy
2 shared publications Alexander Hart
Beth Israel Deaconess Medical Center Fellowship in Disaster Medicine, Harvard Medical School, Boston, Massachusetts, United States of America
1 shared publication
Latest Publications
- Open Access
- 18 November 2024
- Published by Korea Health Personnel Licensing Examination Institute in Journal of Educational Evaluation for Health Professions
- Vol. 21, 33
Comparison of virtual and in-person simulations for sepsis and trauma resuscitation training in Singapore: a randomized controlled trial
Purpose: This study aimed to compare cognitive, non-cognitive, and overall learning outcomes for sepsis and trauma resuscitation skills in novices with virtual patient simulation (VPS) versus in-person simulation (IPS).Methods: A randomized controlled trial was conducted on junior doctors in 1 emergency department from January to December 2022, comparing 70 minutes of VPS (n=19) versus IPS (n=21) in sepsis and trauma resuscitation. Using the nominal group technique, we created skills assessment checklists and determined Bloom’s taxonomy domains for each checklist item. Two blinded raters observed participants leading 1 sepsis and 1 trauma resuscitation simulation. Satisfaction was measured using the Student Satisfaction with Learning Scale (SSLS). The SSLS and checklist scores were analyzed using the Wilcoxon rank sum test and 2-tailed t-test respectively.Results: For sepsis, there was no significant difference between VPS and IPS in overall scores (2.0; 95% confidence interval [CI], -1.4 to 5.4; Cohen’s d=0.38), as well as in items that were cognitive (1.1; 95% CI, -1.5 to 3.7) and not only cognitive (0.9; 95% CI, -0.4 to 2.2). Likewise, for trauma, there was no significant difference in overall scores (-0.9; 95% CI, -4.1 to 2.3; Cohen’s d=0.19), as well as in items that were cognitive (-0.3; 95% CI, -2.8 to 2.1) and not only cognitive (-0.6; 95% CI, -2.4 to 1.3). The median SSLS scores were lower with VPS than with IPS (-3.0; 95% CI, -1.0 to -5.0).Conclusion: For novices, there were no major differences in overall and non-cognitive learning outcomes for sepsis and trauma resuscitation between VPS and IPS. Learners were more satisfied with IPS than with VPS (clinicaltrials.gov identifier: NCT05201950).
- 18 November 2024
- Published by Korea Health Personnel Licensing Examination Institute in Journal of Educational Evaluation for Health Professions
- Vol. 21, 33
Open AccessComparison of virtual and in-person simulations for sepsis and trauma resuscitation training in Singapore: a randomized controlled trial
Purpose: This study aimed to compare cognitive, non-cognitive, and overall learning outcomes for sepsis and trauma resuscitation skills in novices with virtual patient simulation (VPS) versus in-person simulation (IPS).Methods: A randomized controlled trial was conducted on junior doctors in 1 emergency department from January to December 2022, comparing 70 minutes of VPS (n=19) versus IPS (n=21) in sepsis and trauma resuscitation. Using the nominal group technique, we created skills assessment checklists and determined Bloom’s taxonomy domains for each checklist item. Two blinded raters observed participants leading 1 sepsis and 1 trauma resuscitation simulation. Satisfaction was measured using the Student Satisfaction with Learning Scale (SSLS). The SSLS and checklist scores were analyzed using the Wilcoxon rank sum test and 2-tailed t-test respectively.Results: For sepsis, there was no significant difference between VPS and IPS in overall scores (2.0; 95% confidence interval [CI], -1.4 to 5.4; Cohen’s d=0.38), as well as in items that were cognitive (1.1; 95% CI, -1.5 to 3.7) and not only cognitive (0.9; 95% CI, -0.4 to 2.2). Likewise, for trauma, there was no significant difference in overall scores (-0.9; 95% CI, -4.1 to 2.3; Cohen’s d=0.19), as well as in items that were cognitive (-0.3; 95% CI, -2.8 to 2.1) and not only cognitive (-0.6; 95% CI, -2.4 to 1.3). The median SSLS scores were lower with VPS than with IPS (-3.0; 95% CI, -1.0 to -5.0).Conclusion: For novices, there were no major differences in overall and non-cognitive learning outcomes for sepsis and trauma resuscitation between VPS and IPS. Learners were more satisfied with IPS than with VPS (clinicaltrials.gov identifier: NCT05201950).
- Open Access
- 30 September 2024
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 26 (1) , e55648
Accuracy of a Commercial Large Language Model (ChatGPT) to Perform Disaster Triage of Simulated Patients Using the Simple Triage and Rapid Treatment (START) Protocol: Gage Repeatability and Reproducibility Study
Background: The release of ChatGPT (OpenAI) in November 2022 drastically reduced the barrier to using artificial intelligence by allowing a simple web-based text interface to a large language model (LLM). One use case where ChatGPT could be useful is in triaging patients at the site of a disaster using the Simple Triage and Rapid Treatment (START) protocol. However, LLMs experience several common errors including hallucinations (also called confabulations) and prompt dependency. Objective: This study addresses the research problem: “Can ChatGPT adequately triage simulated disaster patients using the START protocol?” by measuring three outcomes: repeatability, reproducibility, and accuracy. Methods: Nine prompts were developed by 5 disaster medicine physicians. A Python script queried ChatGPT Version 4 for each prompt combined with 391 validated simulated patient vignettes. Ten repetitions of each combination were performed for a total of 35,190 simulated triages. A reference standard START triage code for each simulated case was assigned by 2 disaster medicine specialists (JMF and MV), with a third specialist (LC) added if the first two did not agree. Results were evaluated using a gage repeatability and reproducibility study (gage R and R). Repeatability was defined as variation due to repeated use of the same prompt. Reproducibility was defined as variation due to the use of different prompts on the same patient vignette. Accuracy was defined as agreement with the reference standard. Results: Although 35,102 (99.7%) queries returned a valid START score, there was considerable variability. Repeatability (use of the same prompt repeatedly) was 14% of the overall variation. Reproducibility (use of different prompts) was 4.1% of the overall variation. The accuracy of ChatGPT for START was 63.9% with a 32.9% overtriage rate and a 3.1% undertriage rate. Accuracy varied by prompt with a maximum of 71.8% and a minimum of 46.7%. Conclusions: This study indicates that ChatGPT version 4 is insufficient to triage simulated disaster patients via the START protocol. It demonstrated suboptimal repeatability and reproducibility. The overall accuracy of triage was only 63.9%. Health care professionals are advised to exercise caution while using commercial LLMs for vital medical determinations, given that these tools may commonly produce inaccurate data, colloquially referred to as hallucinations or confabulations. Artificial intelligence–guided tools should undergo rigorous statistical evaluation—using methods such as gage R and R—before implementation into clinical settings.
- 30 September 2024
- Published by JMIR Publications Inc. in Journal of Medical Internet Research
- Vol. 26 (1) , e55648
Open AccessAccuracy of a Commercial Large Language Model (ChatGPT) to Perform Disaster Triage of Simulated Patients Using the Simple Triage and Rapid Treatment (START) Protocol: Gage Repeatability and Reproducibility Study
Background: The release of ChatGPT (OpenAI) in November 2022 drastically reduced the barrier to using artificial intelligence by allowing a simple web-based text interface to a large language model (LLM). One use case where ChatGPT could be useful is in triaging patients at the site of a disaster using the Simple Triage and Rapid Treatment (START) protocol. However, LLMs experience several common errors including hallucinations (also called confabulations) and prompt dependency. Objective: This study addresses the research problem: “Can ChatGPT adequately triage simulated disaster patients using the START protocol?” by measuring three outcomes: repeatability, reproducibility, and accuracy. Methods: Nine prompts were developed by 5 disaster medicine physicians. A Python script queried ChatGPT Version 4 for each prompt combined with 391 validated simulated patient vignettes. Ten repetitions of each combination were performed for a total of 35,190 simulated triages. A reference standard START triage code for each simulated case was assigned by 2 disaster medicine specialists (JMF and MV), with a third specialist (LC) added if the first two did not agree. Results were evaluated using a gage repeatability and reproducibility study (gage R and R). Repeatability was defined as variation due to repeated use of the same prompt. Reproducibility was defined as variation due to the use of different prompts on the same patient vignette. Accuracy was defined as agreement with the reference standard. Results: Although 35,102 (99.7%) queries returned a valid START score, there was considerable variability. Repeatability (use of the same prompt repeatedly) was 14% of the overall variation. Reproducibility (use of different prompts) was 4.1% of the overall variation. The accuracy of ChatGPT for START was 63.9% with a 32.9% overtriage rate and a 3.1% undertriage rate. Accuracy varied by prompt with a maximum of 71.8% and a minimum of 46.7%. Conclusions: This study indicates that ChatGPT version 4 is insufficient to triage simulated disaster patients via the START protocol. It demonstrated suboptimal repeatability and reproducibility. The overall accuracy of triage was only 63.9%. Health care professionals are advised to exercise caution while using commercial LLMs for vital medical determinations, given that these tools may commonly produce inaccurate data, colloquially referred to as hallucinations or confabulations. Artificial intelligence–guided tools should undergo rigorous statistical evaluation—using methods such as gage R and R—before implementation into clinical settings.
- 7 March 2024
- Published by Cambridge University Press (CUP) in Disaster Medicine and Public Health Preparedness
- Vol. 18, e42
Determining the Impact of Hurricane Dorian and the Covid-19 Pandemic on Moral Distress in Emergency Medical Providers at the Rand Memorial Hospital: Moral distress in emergency medical personnel
Objectives:The aim of this work was to determine the impact of Moral Distress (MD) in emergency physicians, nurses, and emergency medical service staff at the Rand Memorial Hospital (RMH) in the Bahamas, and the impact of Hurricane Dorian and the COVID-19 pandemic on Moral Distress.Method:A cross-sectional study utilizing a 3-part survey, which collected sociodemographic information, Hurricane Dorian and COVID-19 experiences, as well as responses to a validated modified Moral Distress Scale (MDS).Results:Participants with 2 negatively impactful experiences from COVID-19 had statistically significantly increased MD compared to participants with only 1 negatively impactful experience (40.4 vs. 23.6, P = 0.014). Losing a loved one due to COVID-19 was associated with significantly decreased MD (B = - 0.42, 95% CI -19.70 to -0.88, P = 0.03). Losing a loved one due to Hurricane Dorian had a non-statistically significant trend towards higher MD scores (B = 0.34, 95% CI -1.23 to 28.75, P = 0.07).Conclusion:The emergency medical staff at the RMH reported having mild - moderate MD. This is one of the first studies to look at the impact of concurrent disasters on MD in emergency medical providers in the Bahamas.
- 7 March 2024
- Published by Cambridge University Press (CUP) in Disaster Medicine and Public Health Preparedness
- Vol. 18, e42
Determining the Impact of Hurricane Dorian and the Covid-19 Pandemic on Moral Distress in Emergency Medical Providers at the Rand Memorial Hospital: Moral distress in emergency medical personnel
Objectives:The aim of this work was to determine the impact of Moral Distress (MD) in emergency physicians, nurses, and emergency medical service staff at the Rand Memorial Hospital (RMH) in the Bahamas, and the impact of Hurricane Dorian and the COVID-19 pandemic on Moral Distress.Method:A cross-sectional study utilizing a 3-part survey, which collected sociodemographic information, Hurricane Dorian and COVID-19 experiences, as well as responses to a validated modified Moral Distress Scale (MDS).Results:Participants with 2 negatively impactful experiences from COVID-19 had statistically significantly increased MD compared to participants with only 1 negatively impactful experience (40.4 vs. 23.6, P = 0.014). Losing a loved one due to COVID-19 was associated with significantly decreased MD (B = - 0.42, 95% CI -19.70 to -0.88, P = 0.03). Losing a loved one due to Hurricane Dorian had a non-statistically significant trend towards higher MD scores (B = 0.34, 95% CI -1.23 to 28.75, P = 0.07).Conclusion:The emergency medical staff at the RMH reported having mild - moderate MD. This is one of the first studies to look at the impact of concurrent disasters on MD in emergency medical providers in the Bahamas.
- Open Access
- 19 December 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Accuracy of a Commercial Large Language Model (ChatGPT) to Perform Disaster Triage of Simulated Patients Using the Simple Triage and Rapid Treatment (START) Protocol: Gage Repeatability and Reproducibility Study (Preprint)
BACKGROUND The release of ChatGPT (OpenAI) in November 2022 drastically reduced the barrier to using artificial intelligence by allowing a simple web-based text interface to a large language model (LLM). One use case where ChatGPT could be useful is in triaging patients at the site of a disaster using the Simple Triage and Rapid Treatment (START) protocol. However, LLMs experience several common errors including hallucinations (also called confabulations) and prompt dependency. OBJECTIVE This study addresses the research problem: “Can ChatGPT adequately triage simulated disaster patients using the START protocol?” by measuring three outcomes: repeatability, reproducibility, and accuracy. METHODS Nine prompts were developed by 5 disaster medicine physicians. A Python script queried ChatGPT Version 4 for each prompt combined with 391 validated simulated patient vignettes. Ten repetitions of each combination were performed for a total of 35,190 simulated triages. A reference standard START triage code for each simulated case was assigned by 2 disaster medicine specialists (JMF and MV), with a third specialist (LC) added if the first two did not agree. Results were evaluated using a gage repeatability and reproducibility study (gage R and R). Repeatability was defined as variation due to repeated use of the same prompt. Reproducibility was defined as variation due to the use of different prompts on the same patient vignette. Accuracy was defined as agreement with the reference standard. RESULTS Although 35,102 (99.7%) queries returned a valid START score, there was considerable variability. Repeatability (use of the same prompt repeatedly) was 14% of the overall variation. Reproducibility (use of different prompts) was 4.1% of the overall variation. The accuracy of ChatGPT for START was 63.9% with a 32.9% overtriage rate and a 3.1% undertriage rate. Accuracy varied by prompt with a maximum of 71.8% and a minimum of 46.7%. CONCLUSIONS This study indicates that ChatGPT version 4 is insufficient to triage simulated disaster patients via the START protocol. It demonstrated suboptimal repeatability and reproducibility. The overall accuracy of triage was only 63.9%. Health care professionals are advised to exercise caution while using commercial LLMs for vital medical determinations, given that these tools may commonly produce inaccurate data, colloquially referred to as hallucinations or confabulations. Artificial intelligence–guided tools should undergo rigorous statistical evaluation—using methods such as gage R and R—before implementation into clinical settings.
- 19 December 2023
- Published by JMIR Publications Inc. in JMIR Preprints
Open AccessAccuracy of a Commercial Large Language Model (ChatGPT) to Perform Disaster Triage of Simulated Patients Using the Simple Triage and Rapid Treatment (START) Protocol: Gage Repeatability and Reproducibility Study (Preprint)
BACKGROUND The release of ChatGPT (OpenAI) in November 2022 drastically reduced the barrier to using artificial intelligence by allowing a simple web-based text interface to a large language model (LLM). One use case where ChatGPT could be useful is in triaging patients at the site of a disaster using the Simple Triage and Rapid Treatment (START) protocol. However, LLMs experience several common errors including hallucinations (also called confabulations) and prompt dependency. OBJECTIVE This study addresses the research problem: “Can ChatGPT adequately triage simulated disaster patients using the START protocol?” by measuring three outcomes: repeatability, reproducibility, and accuracy. METHODS Nine prompts were developed by 5 disaster medicine physicians. A Python script queried ChatGPT Version 4 for each prompt combined with 391 validated simulated patient vignettes. Ten repetitions of each combination were performed for a total of 35,190 simulated triages. A reference standard START triage code for each simulated case was assigned by 2 disaster medicine specialists (JMF and MV), with a third specialist (LC) added if the first two did not agree. Results were evaluated using a gage repeatability and reproducibility study (gage R and R). Repeatability was defined as variation due to repeated use of the same prompt. Reproducibility was defined as variation due to the use of different prompts on the same patient vignette. Accuracy was defined as agreement with the reference standard. RESULTS Although 35,102 (99.7%) queries returned a valid START score, there was considerable variability. Repeatability (use of the same prompt repeatedly) was 14% of the overall variation. Reproducibility (use of different prompts) was 4.1% of the overall variation. The accuracy of ChatGPT for START was 63.9% with a 32.9% overtriage rate and a 3.1% undertriage rate. Accuracy varied by prompt with a maximum of 71.8% and a minimum of 46.7%. CONCLUSIONS This study indicates that ChatGPT version 4 is insufficient to triage simulated disaster patients via the START protocol. It demonstrated suboptimal repeatability and reproducibility. The overall accuracy of triage was only 63.9%. Health care professionals are advised to exercise caution while using commercial LLMs for vital medical determinations, given that these tools may commonly produce inaccurate data, colloquially referred to as hallucinations or confabulations. Artificial intelligence–guided tools should undergo rigorous statistical evaluation—using methods such as gage R and R—before implementation into clinical settings.
- 17 July 2023
- Published by Elsevier in Public Health
- Vol. 221, 166-169
Clashes and crowds: protests, riots, and other mass gathering events in North America 2021–2022
- 17 July 2023
- Published by Elsevier in Public Health
- Vol. 221, 166-169
Clashes and crowds: protests, riots, and other mass gathering events in North America 2021–2022
- Open Access
- 1 June 2023
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 18 (6) , e0286472
Factors associated with international humanitarian aid appeal for disasters from 1995 to 2015: A retrospective database study
Introduction: International humanitarian aid during disasters should be needs-based and coordinated in response to appeals from affected governments. We identify disaster and population factors associated with international aid appeal during disasters and hence guide preparation by international humanitarian aid providers. Methods: In this retrospective database analysis, we searched the Emergency Events Database for all disasters from 1995 to 2015. Disasters with and without international aid appeals were compared by location, duration, type of disaster, deaths, number of people affected, and total estimated damage. Logistic regression was used to examine the association of each factor with international aid appeal. Results: Of 13,961 disasters recorded from 1995 to 2015, 168 (1.2%) involved international aid appeals. Aid appeals were more likely to be triggered by disasters which killed more people (OR 1.29 [95% confidence interval (CI) 1.02–1.64] log10 persons), affected more people (OR 1.85 [95%CI 1.57–2.18] / log10 persons), and occurred in Africa (OR 1.67 [95%CI 1.06–2.62). Earthquakes (OR 4.07 [95%CI 2.16–7.67]), volcanic activity (OR 6.23 [95%CI 2.50–15.53]), and insect infestations (OR 12.14 [95%CI 3.05–48.35]) were more likely to trigger international aid appeals. International aid appeals were less likely to be triggered by disasters which occurred in Asia (OR 0.46 [95%CI 0.29–0.73]) and which were transport accidents (OR 0.12 [95%CI 0.02–0.89]). Conclusion: International aid appeal during disasters was associated with greater magnitude of damage, disasters in Africa, and specific types of disasters such as earthquakes, volcanic activity, and insect infestations. Humanitarian aid providers can focus preparation on these identified factors.
- 1 June 2023
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 18 (6) , e0286472
Open AccessFactors associated with international humanitarian aid appeal for disasters from 1995 to 2015: A retrospective database study
Introduction: International humanitarian aid during disasters should be needs-based and coordinated in response to appeals from affected governments. We identify disaster and population factors associated with international aid appeal during disasters and hence guide preparation by international humanitarian aid providers. Methods: In this retrospective database analysis, we searched the Emergency Events Database for all disasters from 1995 to 2015. Disasters with and without international aid appeals were compared by location, duration, type of disaster, deaths, number of people affected, and total estimated damage. Logistic regression was used to examine the association of each factor with international aid appeal. Results: Of 13,961 disasters recorded from 1995 to 2015, 168 (1.2%) involved international aid appeals. Aid appeals were more likely to be triggered by disasters which killed more people (OR 1.29 [95% confidence interval (CI) 1.02–1.64] log10 persons), affected more people (OR 1.85 [95%CI 1.57–2.18] / log10 persons), and occurred in Africa (OR 1.67 [95%CI 1.06–2.62). Earthquakes (OR 4.07 [95%CI 2.16–7.67]), volcanic activity (OR 6.23 [95%CI 2.50–15.53]), and insect infestations (OR 12.14 [95%CI 3.05–48.35]) were more likely to trigger international aid appeals. International aid appeals were less likely to be triggered by disasters which occurred in Asia (OR 0.46 [95%CI 0.29–0.73]) and which were transport accidents (OR 0.12 [95%CI 0.02–0.89]). Conclusion: International aid appeal during disasters was associated with greater magnitude of damage, disasters in Africa, and specific types of disasters such as earthquakes, volcanic activity, and insect infestations. Humanitarian aid providers can focus preparation on these identified factors.