Assessing Biases in Medical Decisions via Clinician and AI Chatbot Responses to Patient Vignettes.
Ontology highlight
ABSTRACT:
SUBMITTER: Kim J
PROVIDER: S-EPMC10582782 | biostudies-literature | 2023 Oct
REPOSITORIES: biostudies-literature
Publications
Assessing Biases in Medical Decisions via Clinician and AI Chatbot Responses to Patient Vignettes.
Kim Jiyeong J Cai Zhuo Ran ZR Chen Michael L ML Simard Julia F JF Linos Eleni E
JAMA network open 20231002 10
Similar Datasets
Project description:BackgroundThe burden on healthcare systems is mounting continuously owing to population growth and aging, overuse of medical services, and the recent COVID-19 pandemic. This overload is also causing reduced healthcare quality and outcomes. One solution gaining momentum is the integration of intelligent self-assessment tools, known as symptom-checkers, into healthcare-providers' systems. To the best of our knowledge, no study so far has investigated the data-gathering capabilities of these tools, which represent a crucial resource for simulating doctors' skills in medical-interviews.ObjectivesThe goal of this study was to evaluate the data-gathering function of currently available chatbot symptom-checkers.MethodsWe evaluated 8 symptom-checkers using 28 clinical vignettes from the repository of MSD-Manual case studies. The mean number of predefined pertinent findings for each case was 31.8 ± 6.8. The vignettes were entered into the platforms by 3 medical students who simulated the role of the patient. For each conversation, we obtained the number of pertinent findings retrieved and the number of questions asked. We then calculated the recall-rates (pertinent-findings retrieved out of all predefined pertinent-findings), and efficiency-rates (pertinent-findings retrieved out of the number of questions asked) of data-gathering, and compared them between the platforms.ResultsThe overall recall rate for all symptom-checkers was 0.32(2,280/7,112;95 %CI 0.31-0.33) for all pertinent findings, 0.37(1,110/2,992;95 %CI 0.35-0.39) for present findings, and 0.28(1140/4120;95 %CI 0.26-0.29) for absent findings. Among the symptom-checkers, Kahun platform had the highest recall rate with 0.51(450/889;95 %CI 0.47-0.54). Out of 4,877 questions asked overall, 2,280 findings were gathered, yielding an efficiency rate of 0.46(95 %CI 0.45-0.48) across all platforms. Kahun was the most efficient tool 0.74 (95 %CI 0.70-0.77) without a statistically significant difference from Your.MD 0.69(95 %CI 0.65-0.73).ConclusionThe data-gathering performance of currently available symptom checkers is questionable. From among the tools available, Kahun demonstrated the best overall performance.
| S-EPMC9595333 | biostudies-literature
Project description:Chronic kidney disease (CKD) patients can benefit from personalized education on lifestyle and nutrition management strategies to enhance healthcare outcomes. The potential use of chatbots, introduced in 2022, as a tool for educating CKD patients has been explored. A set of 15 questions on lifestyle modification and nutrition, derived from a thorough review of three specific KDIGO guidelines, were developed and posed in various formats, including original, paraphrased with different adverbs, incomplete sentences, and misspellings. Four versions of AI were used to answer these questions: ChatGPT 3.5 (March and September 2023 versions), ChatGPT 4, and Bard AI. Additionally, 20 questions on lifestyle modification and nutrition were derived from the NKF KDOQI guidelines for nutrition in CKD (2020 Update) and answered by four versions of chatbots. Nephrologists reviewed all answers for accuracy. ChatGPT 3.5 produced largely accurate responses across the different question complexities, with occasional misleading statements from the March version. The September 2023 version frequently cited its last update as September 2021 and did not provide specific references, while the November 2023 version did not provide any misleading information. ChatGPT 4 presented answers similar to 3.5 but with improved reference citations, though not always directly relevant. Bard AI, while largely accurate with pictorial representation at times, occasionally produced misleading statements and had inconsistent reference quality, although an improvement was noted over time. Bing AI from November 2023 had short answers without detailed elaboration and sometimes just answered "YES". Chatbots demonstrate potential as personalized educational tools for CKD that utilize layman's terms, deliver timely and rapid responses in multiple languages, and offer a conversational pattern advantageous for patient engagement. Despite improvements observed from March to November 2023, some answers remained potentially misleading. ChatGPT 4 offers some advantages over 3.5, although the differences are limited. Collaboration between healthcare professionals and AI developers is essential to improve healthcare delivery and ensure the safe incorporation of chatbots into patient care.
| S-EPMC11353724 | biostudies-literature
Project description:Chatbot-based multimodal AI holds promise for collecting medical histories and diagnosing ophthalmic diseases using textual and imaging data. This study developed and evaluated the ChatGPT-powered Intelligent Ophthalmic Multimodal Interactive Diagnostic System (IOMIDS) to enable patient self-diagnosis and self-triage. IOMIDS included a text model and three multimodal models (text + slit-lamp, text + smartphone, text + slit-lamp + smartphone). The performance was evaluated through a two-stage cross-sectional study across three medical centers involving 10 subspecialties and 50 diseases. Using 15640 data entries, IOMIDS actively collected and analyzed medical history alongside slit-lamp and/or smartphone images. The text + smartphone model showed the highest diagnostic accuracy (internal: 79.6%, external: 81.1%), while other multimodal models underperformed or matched the text model (internal: 69.6%, external: 72.5%). Moreover, triage accuracy was consistent across models. Multimodal approaches enhanced response quality and reduced misinformation. This proof-of-concept study highlights the potential of chatbot-based multimodal AI for self-diagnosis and self-triage. (The clinical trial was registered on June 26, 2023, on ClinicalTrials.gov under the registration number NCT05930444.).
| S-EPMC11772878 | biostudies-literature
Project description: Not available
| S-EPMC9825143 | biostudies-literature
Project description: Not available
| S-EPMC10966080 | biostudies-literature
Project description:BackgroundChatGPT has gained global attention recently owing to its high performance in generating a wide range of information and retrieving any kind of data instantaneously. ChatGPT has also been tested for the United States Medical Licensing Examination (USMLE) and has successfully cleared it. Thus, its usability in medical education is now one of the key discussions worldwide.ObjectiveThe objective of this study is to evaluate the performance of ChatGPT in medical biochemistry using clinical case vignettes.MethodsThe performance of ChatGPT was evaluated in medical biochemistry using 10 clinical case vignettes. Clinical case vignettes were randomly selected and inputted in ChatGPT along with the response options. We tested the responses for each clinical case twice. The answers generated by ChatGPT were saved and checked using our reference material.ResultsChatGPT generated correct answers for 4 questions on the first attempt. For the other cases, there were differences in responses generated by ChatGPT in the first and second attempts. In the second attempt, ChatGPT provided correct answers for 6 questions and incorrect answers for 4 questions out of the 10 cases that were used. But, to our surprise, for case 3, different answers were obtained with multiple attempts. We believe this to have happened owing to the complexity of the case, which involved addressing various critical medical aspects related to amino acid metabolism in a balanced approach.ConclusionsAccording to the findings of our study, ChatGPT may not be considered an accurate information provider for application in medical education to improve learning and assessment. However, our study was limited by a small sample size (10 clinical case vignettes) and the use of the publicly available version of ChatGPT (version 3.5). Although artificial intelligence (AI) has the capability to transform medical education, we emphasize the validation of such data produced by such AI systems for correctness and dependability before it could be implemented in practice.
| S-EPMC10664016 | biostudies-literature
Project description:BackgroundRecent advancements in artificial intelligence (AI) have reshaped telehealth, with AI chatbots like Chat Generative Pretrained Transformer (ChatGPT) showing promise in various medical applications. ChatGPT is capable of offering basic patient education on procedures in plastic and reconstructive surgery (PRS), yet the preference between human AI VideoBots and traditional chatbots in plastic and reconstructive surgery remains unexplored.MethodsWe developed a VideoBot by integrating ChatGPT with Synthesia, a human AI avatar video platform. The VideoBot was then integrated into Tolstoy to create an interactive experience that answered four of the most asked questions related to breast reconstruction. We used Zapier to develop a ChatGPT-integrated chatbot. A 16-item survey adapted from the 2005 validated measurement of online trust by Corritore et al was distributed online to female participants via Amazon Mechanical Turk.ResultsA total of 396 responses were gathered. Participants were 18 to 64 years old. Perceptions of truthfulness, believability, content expertise, ease of use, and safety were similar between the VideoBot and chatbot. Most participants preferred the VideoBot compared with the traditional chatbot (63.5% versus 28.1%), as they found it more captivating than the text-based chatbot. Of the participants, 77% would have preferred to see someone who they identified with in terms of gender and race.ConclusionsBoth the VideoBot and text-based chatbot show comparable effectiveness, usability, and trust. Nonetheless, the VideoBot's human-like qualities enhance interactivity. Future research should explore the impact of race and gender concordance in telehealth to provide a more personalized experience for patients.
| S-EPMC11444614 | biostudies-literature
Project description:ObjectiveThis study aims to evaluate the impact of prompt construction on the quality of artificial intelligence (AI) chatbot responses in the context of head and neck surgery.Study designObservational and evaluative study.SettingAn international collaboration involving 16 researchers from 11 European centers specializing in head and neck surgery.MethodsA total of 24 questions, divided into clinical scenarios, theoretical questions, and patient inquiries, were developed. These questions were entered into ChatGPT-4o both with and without the use of a structured prompt format, known as SMART (Seeker, Mission, AI Role, Register, Targeted Question). The AI-generated responses were evaluated by experienced head and neck surgeons using the Quality Analysis of Medical Artificial Intelligence instrument (QAMAI), which assesses accuracy, clarity, relevance, completeness, source quality, and usefulness.ResultsThe responses generated using the SMART prompt scored significantly higher across all QAMAI dimensions compared to those without contextualized prompts. Median QAMAI scores for SMART prompts were 27.5 (interquartile range [IQR] 25-29) versus 24 (IQR 21.8-25) for unstructured prompts (P < .001). Clinical scenarios and patient inquiries showed the most significant improvements, while theoretical questions also benefited, but to a lesser extent. The AI's source quality improved notably with the SMART prompt, particularly in theoretical questions.ConclusionThis study suggests that the structured SMART prompt format significantly enhances the quality of AI chatbot responses in head and neck surgery. This approach improves the accuracy, relevance, and completeness of AI-generated information, underscoring the importance of well-constructed prompts in clinical applications. Further research is warranted to explore the applicability of SMART prompts across different medical specialties and AI platforms.
| S-EPMC11736147 | biostudies-literature
Project description:Decisions are often made by heterogeneous groups of individuals, each with distinct initial biases and access to information of different quality. We show that in large groups of independent agents who accumulate evidence the first to decide are those with the strongest initial biases. Their decisions align with their initial bias, regardless of the underlying truth. In contrast, agents who decide last make decisions as if they were initially unbiased, and hence make better choices. We obtain asymptotic expressions in the large population limit that quantify how agents' initial inclinations shape early decisions. Our analysis shows how bias, information quality, and decision order interact in non-trivial ways to determine the reliability of decisions in a group.
| S-EPMC10802676 | biostudies-literature
Project description:BackgroundEfforts to improve primary care depression treatment have assessed strategies across heterogeneous groups of patients, but few have examined clinician-level influences on depression treatment.ObjectiveTo examine clinician characteristics that affect depression treatment in primary care settings, using multilevel ordinal regression modeling to disentangle patient- from clinician-level effects.DesignSecondary analysis from the Quality Improvement in Depression Study dataset.ParticipantsThe participants were 1,023 primary care patients with depression who reported on treatment in the 6-month follow-up and whose clinicians (n = 158) had at least 4 patients in the study.MeasurementsPrimary outcome variable was depression treatment intensity, derived from assessment of concordance with AHCPR depression treatment guidelines based on patient-reported data on their treatment. Primary independent variable was clinical practice burden for treating depression, derived from patient- and clinician-reported composite measures tested for significant association with clinician-reported practice burden.ResultsClinicians who treat patients with more chronic medical comorbidities perceive less burden from treating depressed patients in their practice (Spearman's rho = -.30, p < .05). Clinicians who treat patients with more chronic medical comorbidities also provide greater intensity of depression treatment (adjusted OR = 1.44, p = .02), even after adjusting for the effects of patient-level chronic medical comorbidities (adjusted OR = 0.95, p = .45).ConclusionsClinicians who provide more chronic care also provide greater depression treatment intensity, suggesting that clinicians who care for complex patients can integrate depression care into their practice. Targeting interventions to these clinicians to enhance their ability to provide guideline-concordant depression care is a worthwhile endeavor and deserves further investigation.
| S-EPMC2585690 | biostudies-literature