上原 孝紀

BACKGROUND: An illness script is a specific script format geared to represent patient-oriented clinical knowledge organized around enabling conditions, faults (i.e., pathophysiological process), and consequences. Generative artificial intelligence (AI) stands out as an educational aid in continuing medical education. The effortless creation of a typical illness script by generative AI could help the comprehension of key features of diseases and increase diagnostic accuracy. No systematic summary of specific examples of illness scripts has been reported since illness scripts are unique to each physician. OBJECTIVE: This study investigated whether generative AI can generate illness scripts. METHODS: We utilized ChatGPT-4, a generative AI, to create illness scripts for 184 diseases based on the diseases and conditions integral to the National Model Core Curriculum in Japan for undergraduate medical education (2022 revised edition) and primary care specialist training in Japan. Three physicians applied a three-tier grading scale: "A" denotes that the content of each disease's illness script proves sufficient for training medical students, "B" denotes that it is partially lacking but acceptable, and "C" denotes that it is deficient in multiple respects. RESULTS: By leveraging ChatGPT-4, we successfully generated each component of the illness script for 184 diseases without any omission. The illness scripts received "A," "B," and "C" ratings of 56.0% (103/184), 28.3% (52/184), and 15.8% (29/184), respectively. CONCLUSION: Useful illness scripts were seamlessly and instantaneously created using ChatGPT-4 by employing prompts appropriate for medical students. The technology-driven illness script is a valuable tool for introducing medical students to key features of diseases.

Abstract Background Illness scripts, which are structured summaries of clinical knowledge concerning diseases, are crucial in disease prediction and problem representation during clinical reasoning. Clinicians iteratively enhance their illness scripts through their clinical practice. Because illness scripts are unique to each physician, no systematic summary of specific examples of illness scripts has been reported. Objective Generative artificial intelligence (AI) stands out as an educational aid in continuing medical education. The effortless creation of a typical illness script by generative AI could enhance the comprehension of disease concepts and increase diagnostic accuracy. This study investigated whether generative AI possesses the capability to generate illness scripts. Methods We used ChatGPT, a generative AI, to create illness scripts for 184 diseases based on the diseases and conditions integral to the National Model Core Curriculum for undergraduate medical education (2022 revised edition) and primary care specialist training in Japan. Three physicians applied a three-tier grading scale: “A” if the content of each disease’s illness script proves sufficient for training medical students, “B” if it is partially lacking but acceptable, and “C” if it is deficient in multiple respects. Moreover, any identified deficiencies in the illness scripts were discussed during the evaluation process. Results Leveraging ChatGPT, we successfully generated each component of the illness script for 184 diseases without any omission. The illness scripts received “A,” “B,” and “C” ratings of 56.0% (103/184), 28.3% (52/184), and 15.8% (29/184), respectively. Conclusion Useful illness scripts were seamlessly and instantaneously created by ChatGPT using prompts appropriate for medical students. The technology-driven illness script is a valuable tool for introducing medical students to disease conceptualization.

BACKGROUND<p>Medical history contributes approximately 80% to the diagnosis, although physical examinations and laboratory investigations increase a physician’s confidence in the medical diagnosis. The concept of artificial intelligence (AI] was first proposed more than 70 years ago. Recently, its role in various fields of medicine has grown remarkably. However, no studies have evaluated the importance of patient history in AI-assisted medical diagnosis.</p> OBJECTIVE<p>This study explored the contribution of patient history to AI-assisted medical diagnoses.</p> METHODS<p>Using 30 cases from clinical vignettes from the British Medical Journal, we evaluated the accuracy of diagnoses generated by the AI model ChatGPT. We compared the diagnoses made by ChatGPT based solely on the medical history with the correct diagnoses. We also compared the diagnoses made by ChatGPT after incorporating additional physical examination findings and laboratory data alongside the history with correct diagnoses.</p> RESULTS<p>ChatGPT accurately diagnosed 76.6% of the cases with the medical history alone, consistent with previous research targeting physicians. We also found that this rate was 93.3% when additional information was included.</p> CONCLUSIONS<p>Although adding additional information improves diagnostic accuracy, patient history remains a significant factor in AI-assisted medical diagnosis. Thus, when utilizing AI in medical diagnosis, it is crucial to include pertinent and correct patient histories for an accurate diagnosis. Our findings emphasize the continued significance of patient history in clinical diagnoses in this age and highlight the need for its integration into AI-assisted medical diagnosis systems.</p>

OBJECTIVE: The pathophysiological concepts of diseases are encapsulated in patients' medical histories. Whether information on the pathophysiology or anatomy of "infarction" can be preserved and objectively expressed in the distributed representation obtained from a corpus of scientific Japanese medical texts in the "infarction" domain is currently unknown. Word2Vec was used to obtain distributed representations, meanings, and word analogies of word vectors, and this process was verified mathematically. MATERIALS & METHODS: The texts were abstracts that were obtained by searching for "infarction," "abstract," and "case report" in the Japan Medical Journal Association's Ichushi Data Base. The abstracted text was morphologically analyzed to produce word sequences converted into their standard form. MeCab was used for morphological analysis and mecab-ipadic-NEologd and ComeJisyo were used as dictionaries. The accuracy of the known tasks for medical terms was evaluated using a word analogy task specific to the "infarction" domain. RESULTS: Only 33 % of the word analogy tasks for medical terminology were correct. However, 52 % of the new original tasks, which were specific to the "infarction" domain, were correct, especially those regarding anatomical differences. DISCUSSION: Documents related to "infarction" were collected from a corpus of Japanese medical documents and word-embedded expressions were obtained using Word2Vec. Terminology that had similar meanings to "infarction" included words such as "cavity" and "ischemia," which suggest the pathology of an infarction. CONCLUSION: The pathophysiological and anatomical features of an "infarction" may be retained in a distributed representation.

BACKGROUND: Physicians frequently experience patients as difficult. Our study explores whether more empathetic physicians experience fewer patient encounters as difficult. OBJECTIVE: To investigate the association between physician empathy and difficult patient encounters (DPEs). DESIGN: Cross-sectional study. PARTICIPANTS: Participants were 18 generalist physicians with 3-8 years of experience. The investigation was conducted from August-September 2018 and April-May 2019 at six healthcare facilities. MAIN MEASURES: Based on the Jefferson Scale of Empathy (JSE) scores, we classified physicians into low and high empathy groups. The physicians completed the Difficult Doctor-Patient Relationship Questionnaire-10 (DDPRQ-10) after each patient visit. Scores ≥ 31 on the DDPRQ-10 indicated DPEs. We implemented multilevel mixed-effects logistic regression models to examine the association between physicians' empathy and DPE, adjusting for patient-level covariates (age, sex, history of mental disorders) and with physician-level clustering. KEY RESULTS: The median JSE score was 114 (range: 96-126), and physicians with JSE scores 96-113 and 114-126 were assigned to low and high empathy groups, respectively (n = 8 and 10 each); 240 and 344 patients were examined by physicians in the low and high empathy groups, respectively. Among low empathy physicians, 23% of encounters were considered difficulty, compared to 11% among high empathy groups (OR: 0.37; 95% CI = 0.19-0.72, p = 0.004). JSE scores and DDPRQ-10 scores were negatively correlated (r = -0.22, p < 0.01). CONCLUSION: Empathetic physicians were less likely to experience encounters as difficult. Empathy appears to be an important component of physician perception of encounter difficulty.

BACKGROUND: A clinical diagnostic support system (CDSS) can support medical students and physicians in providing evidence-based care. In this study, we investigate diagnostic accuracy based on the history of present illness between groups of medical students using a CDSS, Google, and neither (control). Further, the degree of diagnostic accuracy of medical students using a CDSS is compared with that of residents using neither a CDSS nor Google. METHODS: This study is a randomized educational trial. The participants comprised 64 medical students and 13 residents who rotated in the Department of General Medicine at Chiba University Hospital from May to December 2020. The medical students were randomly divided into the CDSS group (n = 22), Google group (n = 22), and control group (n = 20). Participants were asked to provide the three most likely diagnoses for 20 cases, mainly a history of a present illness (10 common and 10 emergent diseases). Each correct diagnosis was awarded 1 point (maximum 20 points). The mean scores of the three medical student groups were compared using a one-way analysis of variance. Furthermore, the mean scores of the CDSS, Google, and residents' (without CDSS or Google) groups were compared. RESULTS: The mean scores of the CDSS (12.0 ± 1.3) and Google (11.9 ± 1.1) groups were significantly higher than those of the control group (9.5 ± 1.7; p = 0.02 and p = 0.03, respectively). The residents' group's mean score (14.7 ± 1.4) was higher than the mean scores of the CDSS and Google groups (p = 0.01). Regarding common disease cases, the mean scores were 7.4 ± 0.7, 7.1 ± 0.7, and 8.2 ± 0.7 for the CDSS, Google, and residents' groups, respectively. There were no significant differences in mean scores (p = 0.1). CONCLUSIONS: Medical students who used the CDSS and Google were able to list differential diagnoses more accurately than those using neither. Furthermore, they could make the same level of differential diagnoses as residents in the context of common diseases. TRIAL REGISTRATION: This study was retrospectively registered with the University Hospital Medical Information Network Clinical Trials Registry on 24/12/2020 (unique trial number: UMIN000042831).

BACKGROUND: To investigate whether speech recognition software for generating interview transcripts can provide more specific and precise feedback for evaluating medical interviews. METHODS: The effects of the two feedback methods on student performance in medical interviews were compared using a prospective observational trial. Seventy-nine medical students in a clinical clerkship were assigned to receive either speech-recognition feedback (n = 39; SRS feedback group) or voice-recording feedback (n = 40; IC recorder feedback group). All students' medical interviewing skills during mock patient encounters were assessed twice, first using a mini-clinical evaluation exercise (mini-CEX) and then a checklist. Medical students then made the most appropriate diagnoses based on medical interviews. The diagnostic accuracy, mini-CEX, and checklist scores of the two groups were compared. RESULTS: According to the study results, the mean diagnostic accuracy rate (SRS feedback group:1st mock 51.3%, 2nd mock 89.7%; IC recorder feedback group, 57.5%-67.5%; F(1, 77) = 4.0; p = 0.049), mini-CEX scores for overall clinical competence (SRS feedback group: 1st mock 5.2 ± 1.1, 2nd mock 7.4 ± 0.9; IC recorder feedback group: 1st mock 5.6 ± 1.4, 2nd mock 6.1 ± 1.2; F(1, 77) = 35.7; p < 0.001), and checklist scores for clinical performance (SRS feedback group: 1st mock 12.2 ± 2.4, 2nd mock 16.1 ± 1.7; IC recorder feedback group: 1st mock 13.1 ± 2.5, 2nd mock 13.8 ± 2.6; F(1, 77) = 26.1; p < 0.001) were higher with speech recognition-based feedback. CONCLUSIONS: Speech-recognition-based feedback leads to higher diagnostic accuracy rates and higher mini-CEX and checklist scores. TRIAL REGISTRATION: This study was registered in the Japan Registry of Clinical Trials on June 14, 2022. Due to our misunderstanding of the trial registration requirements, we registered the trial retrospectively. This study was registered in the Japan Registry of Clinical Trials on 7/7/2022 (Clinical trial registration number: jRCT1030220188).

This study aims to compare the effectiveness of Hybrid and Pure problem-based learning (PBL) in teaching clinical reasoning skills to medical students. The study sample consisted of 99 medical students participating in a clerkship rotation at the Department of General Medicine, Chiba University Hospital. They were randomly assigned to Hybrid PBL (intervention group, n = 52) or Pure PBL group (control group, n = 47). The quantitative outcomes were measured with the students' perceived competence in PBL, satisfaction with sessions, and self-evaluation of competency in clinical reasoning. The qualitative component consisted of a content analysis on the benefits of learning clinical reasoning using Hybrid PBL. There was no significant difference between intervention and control groups in the five students' perceived competence and satisfaction with sessions. In two-way repeated measure analysis of variance, self-evaluation of competency in clinical reasoning was significantly improved in the intervention group in "recalling appropriate differential diagnosis from patient's chief complaint" (F(1,97) = 5.295, p = 0.024) and "practicing the appropriate clinical reasoning process" (F(1,97) = 4.016, p = 0.038). According to multiple comparisons, the scores of "recalling appropriate history, physical examination, and tests on clinical hypothesis generation" (F(1,97) = 6.796, p = 0.011), "verbalizing and reflecting appropriately on own mistakes," (F(1,97) = 4.352, p = 0.040) "selecting keywords from the whole aspect of the patient," (F(1,97) = 5.607, p = 0.020) and "examining the patient while visualizing his/her daily life" (F(1,97) = 7.120, p = 0.009) were significantly higher in the control group. In the content analysis, 13 advantage categories of Hybrid PBL were extracted. In the subcategories, "acquisition of knowledge" was the most frequent subcategory, followed by "leading the discussion," "smooth discussion," "getting feedback," "timely feedback," and "supporting the clinical reasoning process." Hybrid PBL can help acquire practical knowledge and deepen understanding of clinical reasoning, whereas Pure PBL can improve several important skills such as verbalizing and reflecting on one's own errors and selecting appropriate keywords from the whole aspect of the patient.

総合診療医は領域横断的に、愁訴・疾患・プロブレムが多数併存した患者に対応することが多い。診療の基準となるガイドラインなどは単一の疾患や病態を扱うことが一般的であり、その複雑性は言語化しづらい。例えば、臨床推論のデータベースであれば、「仮説に合わない点の抽出」や「仮説同士の比較」など、推論の核を為すべきデータが十分に言語化されていない。これらの可視化は当面の臨床推論領域の喫緊の課題であり、本稿ではAIを用いて診断に到達した事例を用いて、総合診療医が考えるAIを用いた診断、治療の現状と今後の展望について述べたい。(著者抄録)

Group B Streptococcus (GBS) causes septic arthritis in healthy adults, and a significant number of GBS septic arthritis cases involve multiple joints. Nevertheless, septic arthritis is commonly monoarticular. Here, we report a case of a 45-year-old man who complained of subacute fever and right shoulder and right buttock pain for three weeks despite undergoing garenoxacin treatment for one week. Although synovitis, acne, pustulosis, hyperostosis, and osteitis (SAPHO) syndrome could be a possible differential diagnosis for this patient, the fever and subacute clinical course could not be explained. Blood cultures revealed the presence of GBS; therefore, he was diagnosed with septic arthritis. After antibiotic treatment for six weeks, his symptoms resolved.

The clinical manifestations of Takayasu arteritis (TA) greatly vary, and this ultimately leads to a delay in diagnosis. We describe a case of TA presenting with two coexisting rare symptoms of linear neck pain and prolonged cough. A 28-year-old Japanese female with a six-month history of ulcerative colitis presented with recurrent left neck pain, cough, and fever. The neck pain and fever started five months ago. Her symptoms briefly improved with nonsteroidal anti-inflammatory drug therapy, but eventually recurred one month prior to her latest presentation to the hospital, which was accompanied by a dry cough. Physical examination revealed a blood pressure discrepancy, with systolic blood pressure being >10 mmHg lower in her left arm than in her right arm, a bilateral carotid bruit, a weak left radial pulse and radio-radial delay without coolness in the upper extremities, and linear pulsatile tenderness in her left neck along the common carotid artery. No supraclavicular or infraclavicular bruit was noted. The erythrocyte sedimentation rate was elevated at 66 mm/hour. After obtaining the images from a contrast-enhanced computed tomography, she was diagnosed with TA. All her symptoms improved with prednisone therapy. Notably, neck pain and cough are both late-stage symptoms of TA, which are seen in 9.7% and 1.5% of patients, respectively. Although her unspecific symptoms could have been easily misdiagnosed, the recurring exacerbation of symptoms warranted careful attention to a focused physical examination. In conclusion, neck pain and cough are both uncommon presentations of TA, which may lead to physicians underdiagnosing it. It is important to recognize neck pain and cough as presenting complaints in patients with TA.

Behavioral science, the scientific study of human behavior and the elucidation of its laws, is also applied to medicine, and is included in pre-graduate education.Understanding patient behaviors that correspond to behavior-based medical diagnosis and interpreting the clinical information suggested by these patient behaviors can be useful in avoiding diagnostic errors in clinical practice.

Objective: To determine if inter-disease distances between word embedding vectors using the picot-and-cluster strategy (PCS) are a valid quantitative representation of similar disease groups in a limited domain.Materials and Methods: Abstracts were extracted from the Ichushi-Web database and subjected to morphological analysis and training using the Word2Vec. From this, word embedding vectors were obtained. For words including "infarction", we calculated the cophenetic correlation coefficient (CCC) as an internal validity measure and the adjusted rand index (ARI), normalized mutual information (NMI), and adjusted mutual information (AMI) with ICD-10 codes as the external validity measures. This was performed for each combination of metric and hierarchical clustering method.Results: Seventy-one words included "infarction", of which 38 diseases matched the ICD-10 standard with the appearance of 21 unique ICD-10 codes. The CCC was most significant at 0.8690 (metric and method: euclidean and centroid), while the AMI was maximal at 0.4109 (metric and method: cosine and correlation, and average and weighted). The NMI and ARI were maximal at 0.8463 and 0.3593, respectively (metric and method: cosine and complete).Discussion: The metric and method that maximized the internal validity measure were different from those that maximized the external validity measures; both produced different results. The Cosine distance should be used when considering ICD-10, and the Euclidean distance when considering the frequency of word occurrence.Conclusion: The distributed representation, when trained by Word2Vec on the "infarction" domain from a Japanese academic corpus, provides an objective inter-disease distance used in PCS.

Few studies consider socioeconomic status when assessing mortality risk in dyslipidemia cases. This study used cohort data from the 2010 Japan Gerontological Evaluation Study (JAGES), which contains data on older Japanese people, to associate socioeconomic status with mortality risk in patients treated for dyslipidemia. In this 6-year longitudinal study, we examined 47,275 older Japanese people aged ≥ 65 years who could independently perform activities of daily living. Patients' background characteristics were classified based on their dyslipidemia treatment status and were assessed using the chi-squared test. The mortality risk was assessed using the Cox proportional hazards model, wherein the objective and explanatory variables were total mortality and self-report of dyslipidemia treatment, respectively. The participants were stratified by sex and age into younger (aged 65-74 years) and older (aged ≥ 75 years) groups of men and women. The results were adjusted, with health condition, health behavior, and socioeconomic status as confounding factors. The adjusted hazard ratios of 5514 people who died during the follow-up who had self-reported dyslipidemia treatment were 0.49 [95% confidence interval (CI) 0.35-0.69] for younger men; 0.57 (95% CI 0.42-0.76) for older men; 0.52 (95% CI 0.34-0.80) for younger women; and 0.47 (95% CI 0.33-0.67) for older women. Older people undergoing treatment for dyslipidemia had factors beneficial for health, such as good socioeconomic status. Despite considering these factors, individuals undergoing dyslipidemia treatment had a negative association with mortality risk.

OBJECTIVE: To clarify the factors associated with prolonged hospital stays, focusing on the COMplexity PRediction Instrument (COMPRI) score's accuracy in predicting the length of stay of newly hospitalised patients in general internal medicine wards. DESIGN: A case-control study. SETTING: Three general internal medicine wards in Chiba Prefecture, Japan. PARTICIPANTS: Thirty-four newly hospitalised patients were recruited between November 2017 and December 2019, with a final analytic sample of 33 patients. We included hospitals in different cities with general medicine outpatient and ward facilities, who agreed to participate. We excluded any patients who were re-hospitalised within 2 weeks of a prior discharge. PRIMARY AND SECONDARY OUTCOME MEASURES: Patients' COMPRI scores and their consequent lengths of hospital stay. RESULTS: The 17 patients (52%) allocated to the long-term hospitalisation group (those hospitalised ≥14 days) had a significantly higher average age, COMPRI score and percentage of participants with comorbid chronic illnesses than the short-term hospitalisation group (<14 days). A logistic regression model (model A, comprising only the COMPRI score as the explanatory variable) and a multiple logistic regression model (model B, comprising variables other than the COMPRI score as explanatory variables) were created as prediction models for the long-term hospitalisation group. When age ≥75 years, a COMPRI score ≥6 and a physician with 10 years' experience were set as explanatory variables, model A showed better predictive accuracy compared with model B (fivefold cross-validation, area under curve of 0.87 vs 0.78). The OR of a patient with a COMPRI score of ≥6 joining the long-term hospitalisation group was 4.25 (95% CI=1.43 to 12.63). CONCLUSIONS: Clinicians can use the COMPRI score when screening for complexity assessment to identify hospitalised patients at high risk of prolonged hospitalisation. Providing such patients with multifaceted and intensive care may shorten hospital stays.