Expert Physicians Research Articles

Conversational LLM Chatbot ChatGPT-4 for Colonoscopy Boston Bowel Preparation Scoring: An Artificial Intelligence-to-Head Concordance Analysis

Background/objectives:To date, no studies have evaluated Chat Generative Pre-Trained Transformer (ChatGPT) as a large language model chatbot in optical applications for digestive endoscopy images. This study aimed to weigh the performance of ChatGPT-4 in assessing bowel preparation (BP) quality for colonoscopy. Methods: ChatGPT-4 analysed 663 anonymised endoscopic images, scoring each according to the Boston BP scale (BBPS). Expert physicians scored the same images subsequently. Results: ChatGPT-4 deemed 369 frames (62.9%) to be adequately prepared (i.e., BBPS > 1) compared to 524 frames (89.3%) assessed by human assessors. The agreement was slight (κ: 0.099, p = 0.0001). The raw human BBPS score was higher at 3 (2–3) than that of ChatGPT-4 at 2 (1–3), demonstrating moderate concordance (W: 0.554, p = 0.036). Conclusions: ChatGPT-4 demonstrates some potential in assessing BP on colonoscopy images, but further refinement is still needed.

Abstract 4128519: Performance of a Popular Large Language Model in Answering Cardiovascular Related Queries: A Systematic Review and Pooled-Analysis

Background: The integration of large language models (LLMs) such as ChatGPT into healthcare can have significant implications for patient education and clinical decision-making. Aims: This systematic review and pooled analysis aim to evaluate the accuracy of ChatGPT 3.5 and 4 in answering simple queries across cardiovascular (CV) medicine disciplines. Methods: Literature searches were conducted in PubMed, Embase, and Cochrane Central in May 2024. Keywords included “ChatGPT”, “LLMs”, and “Chat-based artificial intelligence models”. Cross-sectional, peer-reviewed studies published in 2023 and 2024 investigating ChatGPT’s performance in CV medicine-related queries (Table/Figure) were extracted and included. All queries were evaluated by expert physicians in the corresponding fields within each study (and not by our readjudication), and a standardized grading system was employed for pooled analysis using an "accurate" and "inaccurate" grading scale for each answer. Results: Out of 127 identified and screened peer-reviewed studies, fourteen studies involving 542 CV-related queries were included. Pooled analysis revealed an overall accuracy of 84.5% (458/542) (95% CI [81.5, 87.6]). Stratification by model (ChatGPT-4 vs. ChatGPT-3.5) did not show a significant difference in accuracy (p=0.32). Furthermore, no significant differences in accuracies were seen between answers in 2023 and 2024 (p=0.07). The accuracies across the various topics were statistically comparable, except in the field of cardio-oncology, which showed significantly lower accuracy at 68% (p=0.02). Detailed performances per topic are included in the table and figure. Conclusion: ChatGPT demonstrated consistently high accuracy in answering CV-related queries with no significant differences across model versions or years. These results support the potential use of online-chat based LLMs as an informational tool in cardiology.

Abstract 4120657: A Personal Risk Assessment Device in Patients with Chest Pain

Introduction: Early prehospital chest pain assessment improves acute coronary syndrome (ACS) outcomes and can reduce unnecessary Emergency Room (ER) visits but requires medical personnel and electrocardiographic (ECG) equipment. Objective: To evaluate the feasibility of a personal credit-card sized chest pain self-assessment device (HeartBeam, Santa Clara, CA, USA) in diagnosing ACS. Methods: ACS risk was estimated by serial likelihood ratio analysis of 3 components: pre-existing coronary heart disease (CHD) risk (pre-calculated based on pooled ASCVD equation), chest pain characteristics (interactive proprietary questionnaire), and integrated near-orthogonal 3-lead self-recorded vectorcardiogram (VCG) with optional comparison to baseline using portable handheld HeartBeam recorder. Final ACS risk was graded High/Intermediate/Low by the cloud-based proprietary algorithm. High risk was considered positive for ACS. Results: 184 ER patients with chest pain (Age 57 +/- 7years, 90 Male, 43 (23%) ACS) comprised learning (n=96) and test (n=88) sets. ACS sensitivity was 28/28 (100%) in learning and 15/16 (94%) in test set with specificity of 43% and 42% respectively. In the subset of patients with pain-free portable VCG recorded 9 months later used as a “reverse baseline” (n=110) sensitivity was 17/17 (100%) and 11/12 (92%), specificity 54% and 58% in learning and test set, respectively (p>0.1 with no baseline). Single patient missed by the analysis had known coronary disease with usual anginal episode resulting in minor troponin leak. In “reverse baseline” analysis 41/110 (37%) patients were classified as Low risk compared to 18/184 (10%) without baseline (p<0.001), all negative for ACS. Based on the analysis of CHD risk factors, chest pain characteristics, and 12-lead ECG, a consensus of 3-member ER chest pain unit physicians panel had sensitivity/specificity of 94/54% for the whole and 93/48% for the “reverse baseline” groups (p>0.1 with HeartBeam device). Conclusion: HeartBeam chest pain self-assessment device provides accurate ACS detection comparable to expert ER physicians using 12 lead ECG. The presence of baseline recording reduces the number of false positive results. If confirmed in larger studies, it can be used to facilitate outpatient chest pain assessment and triage by remote physicians and patients themselves.

Barriers to opioid use in the Arab world.

The Arab world consists of 22 countries, representing 5.5% of the world's population. Morphine consumption accounts for less than 1% of the world's consumption. This is the first study to identify barriers to opioid use in the Arab world. This study aimed to investigate the barriers to opioid use in pain management in the Arab world. Online survey was developed to investigate barriers to governance, prescribing, distributing, dispensing and administering, as well as educational barriers in the Arab world. The questionnaire was sent via email and a mobile app to one expert physician in pain management and one licensed pharmacist from each Arab country. With the exception of Tunisia, Djibouti and Comoros, 34(77%) participants from 19 Arab countries answered the survey. Most countries lack local opioid production, necessitate special licenses for physicians, and restrict opioid prescribing to medical specialists. Special prescription forms are mandated, and pharmacists lack the authority to correct prescription errors or accept refill or verbal orders on opioids. Storage requirements for empty ampoules and prescriptions are enforced. Nurses are not allowed to carry opioids during home visits, and only first degree relatives can collect opioids for patients. Furthermore, the integration of palliative care and pain management curricula into pharmacies and medical schools is lacking. There is a wide range of regulatory and other barriers to opioid use in the Arab world. There is a substantial need for regulatory review and reform, as well as for educational initiatives, in most Arab countries.

Management of Adult Patients with Newly Diagnosed or Relapsed Primary Immune Thrombocytopenia in Eastern Austria.

Treatment sequence in primary immune thrombocytopenia (ITP) is based on national and international recommendations, treatment availability, and physician expertise. This article aimed to provide real-world data on treatment sequence and responses to first- and second-line treatments in newly diagnosed and relapsed adult ITP patients. We analyzed a cohort of 46 adult ITP patients from the Vienna ITP Biobank, who started first-line therapy within 1 week before their first study visit between February 2016 and March 2023. We investigated clinical patient characteristics and patient management in our specialized center and examined the impact of the international ASH guidelines on ITP treatment. Forty-six primary ITP patients, 27 (58.7%) with newly diagnosed ITP and 19 (41.3%) with relapsed ITP, were investigated. Most patients were female (65.2%) with a median platelet count of 9 × 109/L, and 31 patients (67.4%) had bleeding symptoms. All patients received first-line treatment with oral prednisolone; 15 patients received oral prednisolone combined with intravenous immunoglobulins (IVIGs), which were more commonly administered in newly diagnosed than in relapsed ITP patients. First-line therapy resulted an overall response in 82.6% of patients after a median (interquartile range [IQR]) time of 10 (5-25) days. There was no difference in treatment responses between newly diagnosed and relapsed ITP patients, but newly diagnosed patients had a shorter time to response (median [IQR]: 8 [5-14] and 14 [8-27], p = 0.02). Twenty-three (50%) of the patients (11/27 newly diagnosed [40.7%], 12/19 relapsed [63.2%]) required second-line ITP therapy. Thrombopoietin-receptor agonists (TPO-RAs) were the most commonly used second-line therapy with a response rate of 73.7%, and a median (IQR) time to treatment response of 15 (12-20) days. Overall response rates to TPO-RA treatment did not differ between newly diagnosed and relapsed ITP. Following the publication of novel guidelines in 2019, the median (IQR) duration of corticosteroid treatment shortened (100-52 days, p = 0.01), as did the time to second-line treatment (160-47 days, p = 0.01), and the median number of first-line therapies decreased from 2 (1-3) to 1 (1-2). Initial treatment with corticosteroids was effective in the majority of newly diagnosed and relapsed ITP. Response rates to initial corticosteroid treatment in ITP patients are consistent with previous data, but only 50% achieve sustained remission. TPO-RAs, which are well tolerated and effective, are the most commonly used second-line therapy in our study population. International guidelines have led to faster treatment transitions and reduced splenectomy rates. Integration of real-life experience, expert consensus, and guidelines optimizes ITP patient management.

Effectiveness of King's Theory of Goal Attainment in Blood Glucose Management for Newly Diagnosed Patients With Type 2 Diabetes: Randomized Controlled Trial.

Diabetes poses a significant public health challenge in China and globally, with the number of patients expected to reach 592 million by 2035, notably in Asia. In China alone, an estimated 140 million individuals are living with diabetes, and a significant portion is nonadherent to medications, underscoring the urgency of effective management strategies. Recognizing the necessity of early and comprehensive management for newly diagnosed patients with type 2 diabetes, this study leverages an online teach-back method and "Internet + Nursing" platform based on King's Theory of Goal Attainment. The approach aims to enhance glycemic control and reduce fear and misconceptions about the disease, addressing both the educational and emotional needs of the patients. The primary aim of this study was to assess the effectiveness of King's Goal Attainment Theory in the management of newly diagnosed patients with type 2 diabetes. This research sought to develop a collaborative model for blood glucose management, integrating the expertise and roles of physicians, nurses, and patients. The model is designed to enhance the synergy in health care provision, ensuring a comprehensive approach to diabetes management. In this study conducted at Changzhou Traditional Chinese Medicine Hospital between January 2022 and February 2023, eligible patients were randomized into a control group or an online feedback group. The control group received standard care, while the online feedback group participated in a King's Theory of Goal Attainment-based online teach-back program, enhanced by "Internet + Nursing" strategies. This included an interactive platform for goal planning, video content sharing, comprehension assessment, misconception correction, and patient-driven recaps of disease information. Health monitoring was facilitated through the "Internet + Nursing" platform. The study focused on comparing changes in glucose metabolism and emotional disorder symptoms between the groups to evaluate the intervention's effectiveness. Following a 24-week intervention, we observed significant differences in key metrics between the online feedback group and the control group, each comprising 60 participants. The online feedback group demonstrated significant reductions in fasting plasma glucose, 2-hour postprandial glucose, and hemoglobin A1c (P<.05). Additionally, there was a notable decrease in hypoglycemia-related anxiety and alexithymia within this group. Conversely, the control group maintained relatively higher values for these metrics at the same time point (P<.05). These findings underscore the efficacy of online feedback in managing glycemic control and reducing psychological distress associated with hypoglycemia. The online teaching-back method, guided by King's Theory of Goal Attainment, effectively enhances glycemic control, reducing fasting plasma glucose, 2-hour postprandial glucose, and hemoglobin A1c levels in newly diagnosed patients with type 2 diabetes. Simultaneously, it alleviates hypoglycemia-related anxiety and mitigates alexithymia. This approach merits widespread promotion and implementation in clinical settings. Chinese Clinical Trial Registry ChiCTR2400079547; https://www.chictr.org.cn/showproj.html?proj=208223.

Consensus Statement: Postoperative Management after Balloon Dilation of the Eustachian Tube.

Balloon dilation of Eustachian tube (BDET) has been widely used as a minimally invasive treatment for obstructive Eustachian tube dysfunction (ETD). We used a Delphi consensus methodology to explore recommendations for clinical management of BDET in obstructive ETD. A Delphi panel of 26 expert physicians of otology participated in two rounds of anonymous, iterative questionnaires. Consensus is defined as ≥ 70% of panelists agree with recommendation and disagreement as < 70% agree. The responses from the Delphi study were analyzed using both the content validity ratio and Kendall's coefficient of concordance. The panel finally evaluated 26 topics, agreed 9 and did not reach consensus on 17 topics after 2 rounds. Although no consensus was reached on the postoperative follow-up period, a period of 12 months was most adopted. The Valsalva maneuver and questionnaire responses showed the highest agreement as postoperative assessment tools after BDET. Consensus was reached on some of the recommendations for the management of BEDT in obstructive ETD. The resultant agreement will provide directions for future research to describe standard postoperative management of BDET.

VR-Aided Ankle Rehabilitation Decision-Making Based on Convolutional Gated Recurrent Neural Network.

Traditional rehabilitation training for stroke patients with ankle joint issues typically relies on the expertise of physicians. However, when confronted with complex challenges, such as online decision-making or assessing rehabilitation progress, even seasoned experts may not anticipate all potential hurdles. A novel approach is necessary-one that effectively addresses these complexities without solely leaning on expert experience. Previous studies have introduced a rehabilitation assessment method based on fuzzy neural networks. This paper proposes a novel approach, which is a VR-aided ankle rehabilitation decision-making model based on a convolutional gated recurrent neural network. This model takes various inputs, including ankle dorsiflexion range of motion, angular velocity, jerk, and motion performance scores, gathered from wearable motion inertial sensors during virtual reality rehabilitation. To overcome the challenge of limited data, data augmentation techniques are employed. This allows for the simulation of five stages of rehabilitation based on the Brunnstrom staging scale, providing tailored control parameters for virtual training scenarios suited to patients at different stages of recovery. Experiments comparing the classification performance of convolutional neural networks and long short-term memory networks were conducted. The results were compelling: the optimized convolutional gated recurrent neural network outperformed both alternatives, boasting an average accuracy of 99.16% and a Macro-F1 score of 0.9786. Importantly, it demonstrated a strong correlation (correlation coefficient r > 0.9) with the assessments made by clinical rehabilitation experts, showing its effectiveness in real-world applications.

The Dutch practice of euthanasia and assisted suicide in patients suffering from psychiatric disorders: a qualitative case review study.

Euthanasia or assisted suicide (EAS) in patients suffering from a psychiatric disorder (PD) is a controversial topic worldwide. In the Netherlands, this practice is regulated by law. All cases of EAS have to be reported and are assessed by the Regional Euthanasia Review Committees (RTEs), who publish a selection of all cases on their website. To provide insight into the Dutch practice of EAS in patients suffering from a psychiatric disorder. We performed a retrospective case review study in which all published cases of EAS in patients suffering from a PD between 2017 and 2022 were analyzed. Not applicable. Characteristics of patients who died by EAS because of suffering from a PD, characteristics of the reporting physician and consultant(s) and the RTEs assessment of published cases. Of the 72 cases studied, the majority of patients were female (n=48, 67%), suffered from 3 or more conditions (n=38, 53%) and died by euthanasia instead of assistance in suicide (n=56, 78%). In 63% of cases (n=45), the life termination was performed by a physician from the Euthanasia Expertise center (EE). The RTEs' judgement that the case did not meet the due care criteria (n=11) was in all cases related to issues regarding the (advice of the) independent physician or psychiatric expert. This qualitative study shows that the RTEs attach great importance to a careful evaluation procedure of physicians handling EAS requests and to the physician demonstrating ability to reflect on his views, especially when the independent consultant evaluates the case different than the physician. Training for physicians and more transparency in the assessment of EAS requests in patients with a PD may lower the threshold for physicians to handle requests of these patients themselves. Not applicable.

Empowering Healthcare: TinyML for Precise Lung Disease Classification

Respiratory diseases such as asthma pose significant global health challenges, necessitating efficient and accessible diagnostic methods. The traditional stethoscope is widely used as a non-invasive and patient-friendly tool for diagnosing respiratory conditions through lung auscultation. However, it has limitations, such as a lack of recording functionality, dependence on the expertise and judgment of physicians, and the absence of noise-filtering capabilities. To overcome these limitations, digital stethoscopes have been developed to digitize and record lung sounds. Recently, there has been growing interest in the automated analysis of lung sounds using Deep Learning (DL). Nevertheless, the execution of large DL models in the cloud often leads to latency, dependency on internet connectivity, and potential privacy issues due to the transmission of sensitive health data. To address these challenges, we developed Tiny Machine Learning (TinyML) models for the real-time detection of respiratory conditions by using lung sound recordings, deployable on low-power, cost-effective devices like digital stethoscopes. We trained three machine learning models—a custom CNN, an Edge Impulse CNN, and a custom LSTM—on a publicly available lung sound dataset. Our data preprocessing included bandpass filtering and feature extraction through Mel-Frequency Cepstral Coefficients (MFCCs). We applied quantization techniques to ensure model efficiency. The custom CNN model achieved the highest performance, with 96% accuracy and 97% precision, recall, and F1-scores, while maintaining moderate resource usage. These findings highlight the potential of TinyML to provide accessible, reliable, and real-time diagnostic tools, particularly in remote and underserved areas, demonstrating the transformative impact of integrating advanced AI algorithms into portable medical devices. This advancement facilitates the prospect of automated respiratory health screening using lung sounds.

The Digitized Memory Clinic.

Several major challenges, including an ageing population and declining workforce and the implementation of recent breakthrough therapies for Alzheimer disease, are prompting a necessary rethink of how people with neurodegenerative dementias are diagnosed and medically managed. Digital health technologies could play a pivotal part in this transformation, with new advances enabling the collection of millions of data points from a single individual. Possible applications include unobtrusive monitoring that aids early detection of disease and artificial intelligence-based health advice. To translate these advances to meaningful benefits for people living with a disease, technologies must be implemented within a system that retains the physician expert as a central figure in decision-making. This Perspective presents a new framework, termed the Digitized Memory Clinic, for the diagnostic pathway of neurodegenerative dementias that incorporates digital health technologies with currently available assessment tools, such as fluid and imaging biomarkers, in an interplay with the physician. The Digitized Memory Clinic will manage people across the entire disease spectrum, from the detection of risk factors for cognitive decline and the earliest symptoms to dementia, and will replace the present paradigm of a pure 'brick-and-mortar' memory clinic. Important ethical, legal and societal barriers associated with the implementation of digital health technologies in memory clinics need to be addressed. The envisioned Digitized Memory Clinic aims to improve diagnostics and enable precise disease-tracking prognostication for individuals with memory disorders and to open new possibilities, such as precision medicine for prevention and treatment.

A Quantitative Comparison Between Human and Artificial Intelligence in the Detection of Focal Cortical Dysplasia.

Artificial intelligence (AI) is thought to improve lesion detection. However, a lack of knowledge about human performance prevents a comparative evaluation of AI and an accurate assessment of its impact on clinical decision-making. The objective of this work is to quantitatively evaluate the ability of humans to detect focal cortical dysplasia (FCD), compare it to state-of-the-art AI, and determine how it may aid diagnostics. We prospectively recorded the performance of readers in detecting FCDs using single points and 3-dimensional bounding boxes. We acquired predictions of 3 AI models for the same dataset and compared these to readers. Finally, we analyzed pairwise combinations of readers and models. Twenty-eight readers, including 20 nonexpert and 5 expert physicians, reviewed 180 cases: 146 subjects with FCD (median age: 25, interquartile range: 18) and 34 healthy control subjects (median age: 43, interquartile range: 19). Nonexpert readers detected 47% (95% confidence interval [CI]: 46, 49) of FCDs, whereas experts detected 68% (95% CI: 65, 71). The 3 AI models detected 32%, 51%, and 72% of FCDs, respectively. The latter, however, also predicted more than 13 false-positive clusters per subject on average. Human performance was improved in the presence of a transmantle sign (P < 0.001) and cortical thickening (P < 0.001). In contrast, AI models were sensitive to abnormal gyration (P < 0.01) or gray-white matter blurring (P < 0.01). Compared with single experts, expert-expert pairs detected 13% (95% CI: 9, 18) more FCDs (P < 0.001). All AI models increased expert detection rates by up to 19% (95% CI: 15, 24) (P < 0.001). Nonexpert+AI pairs could still outperform single experts by up to 13% (95% CI: 10, 17). This study pioneers the comparative evaluation of humans and AI for FCD lesion detection. It shows that AI and human predictions differ, especially for certain MRI features of FCD, and, thus, how AI may complement the diagnostic workup.

Evaluation of a Physiologic-Driven Closed-Loop Resuscitation Algorithm in an Animal Model of Hemorrhagic Shock.

Appropriate resuscitation from hemorrhagic shock is critical to restore tissue perfusion and to avoid over-resuscitation. The objective of this study was to test the ability of a closed-loop diagnosis and resuscitation algorithm called resuscitation from shock using functional hemodynamic monitoring using invasive monitoring (ReFit1) and minimally invasive monitoring (ReFit2) to identify, treat, and stabilize a porcine model of severe hemorrhagic shock. We created a ReFit algorithm using dynamic hemodynamic parameters of pulse pressure variation (PPV), stroke volume variation (SVV), dynamic arterial elastance (Eadyn = PPV/SVV), driven by mean arterial pressure (MAP), mixed venous oxygen saturation, and heart rate targets to define the need for fluids, vasopressors, and inotropes. University-based animal laboratory. Twenty-seven female pigs. Anesthetized, intubated, and ventilated (8 mL/kg) pigs were bled at 10 mL/min until a MAP of less than 40 mm Hg, held for 30 minutes, then resuscitated. The ReFit algorithm used the above dynamic parameters to drive computer-controlled infusion pumps to deliver blood, lactated Ringer's solution, norepinephrine, and in ReFit1 dobutamine. In four animals, after initial resuscitation from hemorrhagic shock, the ability of the ReFit1 algorithm to treat acute air embolism-induced pulmonary hypertension and right heart failure was also tested. In 10 ReFit1 and 17 ReFit2 animals, the time to stabilization from shock was not dissimilar to open controlled resuscitation performed by an expert physician (52 ± 12, 50 ± 13, and 60 ± 15 min, respectively) with similar amounts of fluids and norepinephrine needed. In four ReFit1 animals after initial stabilization, the algorithm successfully resuscitated the animals after inducing an acute air embolism right heart failure, with all animals recovering stability within 30 minutes. Our physiologically based functional hemodynamic monitoring-centered closed-loop resuscitation system can effectively diagnose and treat cardiovascular shock due to hemorrhage and air embolism.

Malpractice litigation in multidisciplinary cancer care: Navigating medico-legal challenges and shared decision making in India

Healthcare professionals participating in multidisciplinary team (MDT) cancer meetings may not have a comprehensive understanding of their medicolegal responsibilities in current times. This article aims to delineate the principal medicolegal issues pertinent to multidisciplinary cancer care and offer recommendations for future implementation. Key concerns highlighted in this literature encompass patient consent and privacy during MDT meetings, professional liability, the formal expression of dissenting views and the duty of care. The analysis of existing literature prioritizes several recommendations for addressing these issues. Given the limited precedent available for formulating recommendations, this article identifies foundational evidence that can inform the practicing clinicians of these concerns in future MDT practices.Navigating decision-making processes in cancer care poses significant challenges for patients. This article seeks to elucidate the significance of shared decision making (SDM) within the Indian clinical context while pin pointing disparities compared to Western practices. Through a systematic search conducted in Medline and Google Scholar from 2000 to 2019, approximately 400 articles were screened, resulting in the selection of 43 relevant articles (5 from India and 38 from Western sources). The literature underscores a scarcity of information on shared decision making in India as compared to Western contexts, potentially leading to adverse physical, psychological and financial consequences reported by patients. According to one study, While Western data demonstrate extensive involvement of both patients and physicians in consensus-building for treatment decisions, such engagement in India is predominantly observed in tertiary care settings, academic institutions or cases with high therapy costs. Cultural beliefs and biases further influence patient engagement, while communication breakdowns correlate strongly with medicolegal malpractice litigations. Future research endeavours are warranted to explore strategies for integrating shared decision making into routine oncology practice in India. Physicians must actively involve patients or their immediate family members in decision-making processes to ensure a patient-centric approach, thereby mitigating the risks of un-informed decision-making or mistrust in the treating physician's expertise. Efforts such as physician and patient education, tool development, policy formulation, widespread implementation and periodic assessments hold promise for advancing the practice of shared decision making.

Physician perspectives about the diagnosis and management of acute chest syndrome.

Acute chest syndrome (ACS) is the leading cause of mortality, accounting for 25% of all deaths among individuals with sickle cell disease (SCD). There is a lack of evidence-based laboratory and clinical risk stratification guidelines for the diagnosis and management of ACS. To better understand physician practices for the management of ACS in the United States, we created an ACS Working Group including hematology and transfusion medicine physicians from four different SCD treatment centers in the United States. The working group created a physician survey that included physician demographics and ACS diagnostic criteria that they had to rate. The survey also included three case scenarios to assess physician attitudes about the management of ACS. Management options included supportive and preventive strategies in addition to transfusion therapy options. Out of 455 physicians who received the survey, 195 responded (response rate = 43%). The respondents were primarily hematology/oncology physicians. The responses showed wide variability among physicians in how diagnostic criteria for ACS are used and how physicians risk-stratify ACS patients in their practice. The responses also reflected variability in the use of transfusions for ACS. Based on our results, we conclude that ACS is diagnosed and managed inconsistently among expert physicians, especially in their transfusion practices due to a lack of consensus on risk stratification criteria. Our data suggest an urgent need for well-designed prospective studies to provide evidence-based guidelines and minimize management variability among physicians who care for individuals with SCD and ACS.

7870 Exploring The Role Of ChatGPT In Pediatric Endocrinology Education; Exam Preparation And Question Generation

Abstract Disclosure: J. Tarkoff: None. A.G. Martinez Sanchez: None. Large language models (LLMs) hold substantial promise for improving physician knowledge and expertise. Their role in medical education and the generation of diverse diagnoses can be crucial, potentially leading to positive effects on clinical outcomes. The New England Journal of Medicine pediatric case challenges have examined ChatGPT [1]. Yet, its performance in specialized areas like Pediatric Endocrinology remains unexplored. This study assesses the effectiveness of ChatGPT 4 in responding to questions from the Pediatric Endocrine Self-Assessment Program (PESAP). It also examines if the AI is suitable for creating an educational quiz for residents and fellows. Methods: ChatGPT 4 underwent testing with questions from the 2021-2022 version of PESAP, utilizing the prompt: “Can you assist from the perspective of a pediatric endocrinologist with the following patient case”. Responses were evaluated for initial correctness, and performance was analyzed across various competency categories corresponding to the 7 “umbrella sections” of the tool (Adrenal, Bone, Carbohydrate and Lipid Metabolism/Obesity, Growth, Pituitary, Reproductive System, and Thyroid). Subsequently, we personalized the model by incorporating the questions and detailed responses from the PESAP. This customization resulted in a model that was then utilized to create a 10-question proof-of-concept quiz for four board-certified pediatric endocrinologists. The quiz included a scoring system designed to measure the extent and depth of ChatGPT knowledge. Results: ChatGPT 4 accurately answered 52% of PESAP questions, demonstrating varying performance across specific categories, ranging from 30% (Adrenal) to 78% (Reproductive System). In 16 questions, ChatGPT 4 did not provide an initial answer, requiring a specific request for a response. For questions related to thyroid cancer, explicit prompts were necessary to instruct responses based on the American Thyroid Association 2015 guidelines. In the endocrinologist quiz, the average score was 80%, ranging from 60% to 100%. Discussion: Currently, using ChatGPT 4 as the final diagnostic tool, especially in pediatric endocrinology, should be approached with caution, based on our assessment. However, when incorporating distinct Pediatric Endocrinology case studies, ChatGPT 4 successfully generated valuable educational questions, reinforcing fundamental concepts in the field. We anticipate that as LLMs advance and receive direct medical training, the out-of-the-box diagnostic accuracy will improve, leading to a transformative impact on medical education through this technology.

Comparative Study to Evaluate the Accuracy of Differential Diagnosis Lists Generated by Gemini Advanced, Gemini, and Bard for a Case Report Series Analysis: Cross-Sectional Study.

Generative artificial intelligence (GAI) systems by Google have recently been updated from Bard to Gemini and Gemini Advanced as of December 2023. Gemini is a basic, free-to-use model after a user's login, while Gemini Advanced operates on a more advanced model requiring a fee-based subscription. These systems have the potential to enhance medical diagnostics. However, the impact of these updates on comprehensive diagnostic accuracy remains unknown. This study aimed to compare the accuracy of the differential diagnosis lists generated by Gemini Advanced, Gemini, and Bard across comprehensive medical fields using case report series. We identified a case report series with relevant final diagnoses published in the American Journal Case Reports from January 2022 to March 2023. After excluding nondiagnostic cases and patients aged 10 years and younger, we included the remaining case reports. After refining the case parts as case descriptions, we input the same case descriptions into Gemini Advanced, Gemini, and Bard to generate the top 10 differential diagnosis lists. In total, 2 expert physicians independently evaluated whether the final diagnosis was included in the lists and its ranking. Any discrepancies were resolved by another expert physician. Bonferroni correction was applied to adjust the P values for the number of comparisons among 3 GAI systems, setting the corrected significance level at P value <.02. In total, 392 case reports were included. The inclusion rates of the final diagnosis within the top 10 differential diagnosis lists were 73% (286/392) for Gemini Advanced, 76.5% (300/392) for Gemini, and 68.6% (269/392) for Bard. The top diagnoses matched the final diagnoses in 31.6% (124/392) for Gemini Advanced, 42.6% (167/392) for Gemini, and 31.4% (123/392) for Bard. Gemini demonstrated higher diagnostic accuracy than Bard both within the top 10 differential diagnosis lists (P=.02) and as the top diagnosis (P=.001). In addition, Gemini Advanced achieved significantly lower accuracy than Gemini in identifying the most probable diagnosis (P=.002). The results of this study suggest that Gemini outperformed Bard in diagnostic accuracy following the model update. However, Gemini Advanced requires further refinement to optimize its performance for future artificial intelligence-enhanced diagnostics. These findings should be interpreted cautiously and considered primarily for research purposes, as these GAI systems have not been adjusted for medical diagnostics nor approved for clinical use.

The Value of Liver Biopsy and Histology in Liver Disease Diagnosis and Patient Care-a Pragmatic Prospective Clinical Practice Study.

We aimed to examine the correlation of pre-biopsy clinical diagnosis with hepatic histopathology. Liver biopsy provides histologic information and informs physicians about the underlying clinical disease. We hypothesized that expert physicians' pre-biopsy clinical diagnoses may obviate the need for histopathological diagnosis. Patients undergoing liver biopsy to investigate a liver diagnosis were prospectively identified. In the 80 patients included, an anonymous validated questionnaire inquiring about the most likely clinical diagnosis and liver disease stage was completed prospectively by hepatologists before biopsy performance. The most common pre-biopsy diagnoses were alcoholic liver disease (19 diagnoses), followed by non-alcoholic steatohepatitis and autoimmune hepatitis (18 each). Overall, the predicted histologic diagnosis was the same as the histologic diagnosis in 51/80 patients (64%), and thus a new liver disease diagnosis was made in 36% of patients. The diagnosis with the greatest pre-biopsy and post-biopsy diagnosis discrepancy was autoimmune hepatitis, with the correct diagnosis being predicted in 6/18 (33%) of patients (other diagnoses included the following: non-alcoholic steatohepatitis/non-alcoholic fatty liver disease (4), alcoholic liver disease (3), drug-induced liver injury (3), others (2)). For fibrosis staging, when grouped as no fibrosis (F0), fibrosis (F1-F3), or cirrhosis (F4), the fibrosis stage was correctly predicted in 68% of patients (54/80). Notably, 7 patients (9%) who were initially thought to have no or early-stage fibrosis had F4 fibrosis, and 6/80 (8%) patients who were considered to have a liver disease diagnosis before their biopsy had normal histology. Although hepatology experts often predict the correct underlying liver disease diagnosis, histopathological diagnoses different from expected are common.

A quality assessment tool for focused abdominal sonography for trauma examinations using artificial intelligence.

Current tools to review focused abdominal sonography for trauma (FAST) images for quality have poorly defined grading criteria or are developed to grade the skills of the sonographer and not the examination. The purpose of this study is to establish a grading system with substantial agreement among coders, thereby enabling the development of an automated assessment tool for FAST examinations using artificial intelligence (AI). Five coders labeled a set of FAST clips. Each coder was responsible for a different subset of clips (10% of the clips were labeled in triplicate to evaluate intercoder reliability). The clips were labeled with a quality score from 1 (lowest quality) to 5 (highest quality). Clips of 3 or greater were considered passing. An AI training model was developed to score the quality of the FAST examination. The clips were split into a training set, a validation set, and a test set. The predicted scores were rounded to the nearest quality level to distinguish passing from failing clips. A total of 1,514 qualified clips (1,399 passing and 115 failing clips) were evaluated in the final data set. This final data set had a 94% agreement between pairs of coders on the pass/fail prediction, and the set had a Krippendorff α of 66%. The decision threshold can be tuned to achieve the desired tradeoff between precision and sensitivity. Without using the AI model, a reviewer would, on average, examine roughly 25 clips for every 1 failing clip identified. In contrast, using our model with a decision threshold of 0.015, a reviewer would examine roughly five clips for every one failing clip - a fivefold reduction in clips reviewed while still correctly identifying 85% of passing clips. Integration of AI holds significant promise in improving the accurate evaluation of FAST images while simultaneously alleviating the workload burden on expert physicians. Diagnostic Test/Criteria; Level II.

SMLS-YOLO: an extremely lightweight pathological myopia instance segmentation method.

Pathological myopia is a major cause of blindness among people under 50 years old and can result in severe vision loss in extreme cases. Currently, its detection primarily relies on manual methods, which are slow and heavily dependent on the expertise of physicians, making them impractical for large-scale screening. To tackle these challenges, we propose SMLS-YOLO, an instance segmentation method based on YOLOv8n-seg. Designed for efficiency in large-scale screenings, SMLS-YOLO employs an extremely lightweight model. First, StarNet is introduced as the backbone of SMLS-YOLO to extract image features. Subsequently, the StarBlock from StarNet is utilized to enhance the C2f, resulting in the creation of the C2f-Star feature extraction module. Furthermore, shared convolution and scale reduction strategies are employed to optimize the segmentation head for a more lightweight design. Lastly, the model incorporates the Multi-Head Self-Attention (MHSA) mechanism following the backbone to further refine the feature extraction process. Experimental results on the pathological myopia dataset demonstrate that SMLS-YOLO outperforms the baseline YOLOv8n-seg by reducing model parameters by 46.9%, increasing Box mAP@0.5 by 2.4%, and enhancing Mask mAP@0.5 by 4%. Furthermore, when compared to other advanced instance segmentation and semantic segmentation algorithms, SMLS-YOLO also maintains a leading position, suggesting that SMLS-YOLO has promising applications in the segmentation of pathological myopia images.