1
|
Ge J, Sun S, Owens J, Galvez V, Gologorskaya O, Lai JC, Pletcher MJ, Lai K. Development of a liver disease-specific large language model chat interface using retrieval-augmented generation. Hepatology 2024:01515467-990000000-00791. [PMID: 38451962 DOI: 10.1097/hep.0000000000000834] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Submit a Manuscript] [Subscribe] [Scholar Register] [Received: 12/13/2023] [Accepted: 02/24/2024] [Indexed: 03/09/2024]
Abstract
BACKGROUND AND AIMS Large language models (LLMs) have significant capabilities in clinical information processing tasks. Commercially available LLMs, however, are not optimized for clinical uses and are prone to generating hallucinatory information. Retrieval-augmented generation (RAG) is an enterprise architecture that allows the embedding of customized data into LLMs. This approach "specializes" the LLMs and is thought to reduce hallucinations. APPROACH AND RESULTS We developed "LiVersa," a liver disease-specific LLM, by using our institution's protected health information-complaint text embedding and LLM platform, "Versa." We conducted RAG on 30 publicly available American Association for the Study of Liver Diseases guidance documents to be incorporated into LiVersa. We evaluated LiVersa's performance by conducting 2 rounds of testing. First, we compared LiVersa's outputs versus those of trainees from a previously published knowledge assessment. LiVersa answered all 10 questions correctly. Second, we asked 15 hepatologists to evaluate the outputs of 10 hepatology topic questions generated by LiVersa, OpenAI's ChatGPT 4, and Meta's Large Language Model Meta AI 2. LiVersa's outputs were more accurate but were rated less comprehensive and safe compared to those of ChatGPT 4. RESULTS We evaluated LiVersa's performance by conducting 2 rounds of testing. First, we compared LiVersa's outputs versus those of trainees from a previously published knowledge assessment. LiVersa answered all 10 questions correctly. Second, we asked 15 hepatologists to evaluate the outputs of 10 hepatology topic questions generated by LiVersa, OpenAI's ChatGPT 4, and Meta's Large Language Model Meta AI 2. LiVersa's outputs were more accurate but were rated less comprehensive and safe compared to those of ChatGPT 4. CONCLUSIONS In this demonstration, we built disease-specific and protected health information-compliant LLMs using RAG. While LiVersa demonstrated higher accuracy in answering questions related to hepatology, there were some deficiencies due to limitations set by the number of documents used for RAG. LiVersa will likely require further refinement before potential live deployment. The LiVersa prototype, however, is a proof of concept for utilizing RAG to customize LLMs for clinical use cases.
Collapse
Affiliation(s)
- Jin Ge
- Department of Medicine, Division of Gastroenterology and Hepatology, University of California-San Francisco, San Francisco, California, USA
| | - Steve Sun
- UCSF Health Information Technology, University of California-San Francisco, San Francisco, California, USA
| | - Joseph Owens
- UCSF Health Information Technology, University of California-San Francisco, San Francisco, California, USA
| | - Victor Galvez
- UCSF Health Information Technology, University of California-San Francisco, San Francisco, California, USA
| | - Oksana Gologorskaya
- UCSF Health Information Technology, University of California-San Francisco, San Francisco, California, USA
- Bakar Computational Health Sciences Institute, University of California-San Francisco, San Francisco, California, USA
| | - Jennifer C Lai
- Department of Medicine, Division of Gastroenterology and Hepatology, University of California-San Francisco, San Francisco, California, USA
| | - Mark J Pletcher
- Department of Epidemiology and Biostatistics, University of California-San Francisco, San Francisco, California, USA
| | - Ki Lai
- UCSF Health Information Technology, University of California-San Francisco, San Francisco, California, USA
| |
Collapse
|
2
|
Beal EW, McNamara M, Owen M, McAlearney AS, Tsung A. Interventions to Improve Surveillance for Hepatocellular Carcinoma in High-Risk Patients: A Scoping Review. J Gastrointest Cancer 2024; 55:1-14. [PMID: 37328730 DOI: 10.1007/s12029-023-00944-1] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Accepted: 05/09/2023] [Indexed: 06/18/2023]
Abstract
PURPOSE Hepatocellular carcinoma (HCC) is most often a sequela of chronic liver disease or chronic hepatitis B infection. Among high-risk patients, surveillance for HCC every 6 months is recommended by international guidelines. However, rates of HCC surveillance are suboptimal (11-64%). Barriers at the patient, provider, and healthcare delivery system levels have been identified. METHODS We performed a systemic scoping review to identify and characterize interventions to improve HCC surveillance that has previously been evaluated. Searches using key terms in PubMed and Embase were performed to identify studies examining interventions designed to improve the surveillance rate for HCC in patients with cirrhosis or chronic liver disease that were published in English between January 1990 and September 2021. RESULTS Included studies (14) had the following study designs: (1) randomized clinical trials (3, 21.4%), (2) quasi-experimental (2, 14.3%), (3) prospective cohort (6, 42.8%), and (4) retrospective cohort (3, 21.4%). Interventions included mailed outreach invitations, nursing outreach, patient education with or without printed materials, provider education, patient navigation, chronic disease management programs, nursing-led protocols for image ordering, automated reminders to physicians and nurses, web-based clinical management tools, HCC surveillance databases, provider compliance reports, radiology-led surveillance programs, subsidized HCC surveillance, and the use of oral medications. It was found that HCC surveillance rates increased after intervention implementation in all studies. CONCLUSION Despite improvements in HCC surveillance rates with intervention, compliance remained suboptimal. Further analysis of which interventions yield the greatest increases in HCC surveillance, design of multi-pronged strategies, and improved implementation are needed.
Collapse
Affiliation(s)
- Eliza W Beal
- Departments of Surgery and Oncology, Barbara Ann Karmanos Cancer Institute, Wayne State University School of Medicine, 4100 John R, Mailcode: HW04HO, Detroit, MI, 48201, USA.
| | - Molly McNamara
- The Ohio State University College of Medicine, Columbus, OH, 43210, USA
| | - Mackenzie Owen
- The Ohio State University College of Medicine, Columbus, OH, 43210, USA
| | - Ann Scheck McAlearney
- The Center for the Advancement of Team Science, Analytics, and Systems Thinking in Health Services and Implementation Science Research (CATALYST), The Ohio State University College of Medicine, Columbus, OH, 43210, USA
- Department of Family and Community Medicine, The Ohio State University Wexner Medical Center, Columbus, OH, 43210, USA
| | - Allan Tsung
- Department of Surgery, Division of Surgical Oncology, University of Virginia, Charlottsville, VA, 22908, USA
| |
Collapse
|
3
|
Ge J, Sun S, Owens J, Galvez V, Gologorskaya O, Lai JC, Pletcher MJ, Lai K. Development of a Liver Disease-Specific Large Language Model Chat Interface using Retrieval Augmented Generation. MEDRXIV : THE PREPRINT SERVER FOR HEALTH SCIENCES 2023:2023.11.10.23298364. [PMID: 37986764 PMCID: PMC10659484 DOI: 10.1101/2023.11.10.23298364] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Grants] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 11/22/2023]
Abstract
Background Large language models (LLMs) have significant capabilities in clinical information processing tasks. Commercially available LLMs, however, are not optimized for clinical uses and are prone to generating incorrect or hallucinatory information. Retrieval-augmented generation (RAG) is an enterprise architecture that allows embedding of customized data into LLMs. This approach "specializes" the LLMs and is thought to reduce hallucinations. Methods We developed "LiVersa," a liver disease-specific LLM, by using our institution's protected health information (PHI)-complaint text embedding and LLM platform, "Versa." We conducted RAG on 30 publicly available American Association for the Study of Liver Diseases (AASLD) guidelines and guidance documents to be incorporated into LiVersa. We evaluated LiVersa's performance by comparing its responses versus those of trainees from a previously published knowledge assessment study regarding hepatitis B (HBV) treatment and hepatocellular carcinoma (HCC) surveillance. Results LiVersa answered all 10 questions correctly when forced to provide a "yes" or "no" answer. Full detailed responses with justifications and rationales, however, were not completely correct for three of the questions. Discussions In this study, we demonstrated the ability to build disease-specific and PHI-compliant LLMs using RAG. While our LLM, LiVersa, demonstrated more specificity in answering questions related to clinical hepatology - there were some knowledge deficiencies due to limitations set by the number and types of documents used for RAG. The LiVersa prototype, however, is a proof of concept for utilizing RAG to customize LLMs for clinical uses and a potential strategy to realize personalized medicine in the future.
Collapse
Affiliation(s)
- Jin Ge
- Division of Gastroenterology and Hepatology, Department of Medicine, University of California – San Francisco, San Francisco, CA
| | - Steve Sun
- UCSF Health Information Technology, University of California – San Francisco, San Francisco, CA
| | - Joseph Owens
- UCSF Health Information Technology, University of California – San Francisco, San Francisco, CA
| | - Victor Galvez
- UCSF Health Information Technology, University of California – San Francisco, San Francisco, CA
| | - Oksana Gologorskaya
- UCSF Health Information Technology, University of California – San Francisco, San Francisco, CA
- Bakar Computational Health Sciences Institute, University of California – San Francisco, San Francisco, CA
| | - Jennifer C. Lai
- Division of Gastroenterology and Hepatology, Department of Medicine, University of California – San Francisco, San Francisco, CA
| | - Mark J. Pletcher
- Department of Epidemiology and Biostatistics, University of California – San Francisco, San Francisco, CA
| | - Ki Lai
- UCSF Health Information Technology, University of California – San Francisco, San Francisco, CA
| |
Collapse
|
4
|
Beal EW, Owen M, McNamara M, McAlearney AS, Tsung A. Patient-, Provider-, and System-Level Barriers to Surveillance for Hepatocellular Carcinoma in High-Risk Patients in the USA: a Scoping Review. J Gastrointest Cancer 2023; 54:332-356. [PMID: 35879510 DOI: 10.1007/s12029-022-00851-x] [Citation(s) in RCA: 5] [Impact Index Per Article: 5.0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Accepted: 07/17/2022] [Indexed: 12/12/2022]
Abstract
PURPOSE Hepatocellular carcinoma has a dismal prognosis, except in patients diagnosed early who are candidates for potentially curative therapies. Most HCC cases develop in patients with chronic liver disease. Therefore, expert society guidelines recommend surveillance every 6 months with ultrasound with or without serum alpha-fetoprotein for high-risk patients. However, fewer than 20% of patients in the USA undergo appropriate surveillance. METHODS A systematic scoping review was performed with the objective of identifying barriers to screening among high-risk patients in the USA including mapping key concepts in the relevant literature, identifying the main sources and types of evidence available, and identifying gaps in the literature. A total of 43 studies published from 2007 to 2021 were included. Data were extracted and a conceptual framework was created. RESULTS Assessment of quantitative studies revealed poor surveillance rates, often below 50%. Three categories of barriers to surveillance were identified: patient-level, provider-level, and system-level barriers. Prevalent patient-level barriers included financial constraints, lack of awareness of surveillance recommendations, and scheduling difficulties. Common provider-level barriers were lack of provider awareness of guidelines for surveillance, difficulty accessing specialty resources, and time constraints in the clinic. System-level barriers included fewer clinic visits and rural/safety-net settings. Proposed interventions include improved patient/provider education, patient navigators, increased community/academic collaboration, and EMR-based reminders. CONCLUSION Based on these findings, there is a crucial need to implement and evaluate proposed interventions to improve HCC surveillance.
Collapse
Affiliation(s)
- Eliza W Beal
- The Ohio State University Comprehensive Cancer Center, Columbus, OH, 43210, USA.
- The Center for the Advancement of Team Science, Systems Thinking in Health Services and Implementation Science Research (CATALYST, The Ohio State University College of Medicine, AnalyticsColumbus, OH, 43210, USA.
| | - Mackenzie Owen
- The Ohio State University College of Medicine, Columbus, OH, 43210, USA
| | - Molly McNamara
- The Ohio State University College of Medicine, Columbus, OH, 43210, USA
| | - Ann Scheck McAlearney
- The Ohio State University Comprehensive Cancer Center, Columbus, OH, 43210, USA
- The Center for the Advancement of Team Science, Systems Thinking in Health Services and Implementation Science Research (CATALYST, The Ohio State University College of Medicine, AnalyticsColumbus, OH, 43210, USA
- The Department of Family and Community Medicine, The Ohio State University College of Medicine, Columbus, OH, 43210, USA
| | - Allan Tsung
- The Ohio State University Comprehensive Cancer Center, Columbus, OH, 43210, USA
- The Center for the Advancement of Team Science, Systems Thinking in Health Services and Implementation Science Research (CATALYST, The Ohio State University College of Medicine, AnalyticsColumbus, OH, 43210, USA
| |
Collapse
|
5
|
Wang AE, Terrault NA, Fong TL, Zhou K. Disparities in presentation and management of chronic hepatitis B among Hispanics in a diverse safety net system. Hepatol Commun 2023; 7:e0081. [PMID: 36972389 PMCID: PMC10043583 DOI: 10.1097/hc9.0000000000000081] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Submit a Manuscript] [Subscribe] [Scholar Register] [Received: 07/20/2022] [Accepted: 01/20/2023] [Indexed: 03/29/2023] Open
Abstract
BACKGROUND Chronic hepatitis B (HBV) prevalence is highest in foreign-born Asian and African individuals in the US, though Hispanics make up the largest proportion of the immigrant population. Differences in the diagnosis and management of chronic HBV in Hispanics might exist due to the lower awareness of risk. We aim to examine racial/ethnic disparities in the diagnosis, presentation, and immediate management of chronic HBV in a diverse safety net system enriched for Hispanics. METHODS In a large urban safety-net hospital system, we retrospectively identified patients with chronic HBV by serological data and categorized them into mutually exclusive self-identified racial/ethnic groups: Hispanics, Asians, Blacks, and Whites. We then examined differences in screening, disease phenotype and severity, follow-up testing, and referral by race/ethnicity. RESULTS Among 1063 patients, 302 (28%) were Hispanics, 569 (54%) Asians, 161 (15%) Blacks, and 31 (3%) Whites. More Hispanics (30%) were screened in the acute setting (defined as inpatient or emergency department encounters) than Asians (13%), Blacks (17%), or Whites (23%) (p<0.01). Hispanics also had lower rates of follow-up testing after HBV diagnosis than Asians including HBeAg status (43% vs. 60%, p<0.01) and HBV DNA levels (42% vs. 58%, p<0.01) and lower rates of linkage to specialty care (32% vs. 55%, p<0.01). Among those with available testing, however, the presence of immune-active chronic HBV was infrequent and similar across racial/ethnic groups. 25% of Hispanics had cirrhosis at initial presentation, proportionally higher than other groups (p<0.01). CONCLUSION Our results underscore the importance of raising chronic HBV awareness and increasing both screening and linkage to care among Hispanic immigrants in addition to the existing risk groups, with the goal of mitigating downstream liver-related complications.
Collapse
Affiliation(s)
- Allison E. Wang
- Department of Medicine, LAC+USC Medical Center, Los Angeles, California, USA
| | - Norah A. Terrault
- Division of Gastrointestinal and Liver Diseases, Department of Medicine, University of Southern California, Los Angeles, California, USA
| | - Tse-Ling Fong
- Division of Gastrointestinal and Liver Diseases, Department of Medicine, University of Southern California, Los Angeles, California, USA
- Liver Program, Hoag Memorial Hospital Presbyterian, Newport Beach, California, USA
| | - Kali Zhou
- Division of Gastrointestinal and Liver Diseases, Department of Medicine, University of Southern California, Los Angeles, California, USA
| |
Collapse
|
6
|
Maamor NH, Muhamad NA, Mohd Dali NS, Abdul Mutalip MH, Leman FN, Aris T, Lai NM, Abu Hassan MR. Seroprevalence of Hepatitis B Among Healthcare Workers in Asia and Africa and Its Association With Their Knowledge and Awareness: A Systematic Review and Meta-Analysis. Front Public Health 2022; 10:859350. [PMID: 35570890 PMCID: PMC9096243 DOI: 10.3389/fpubh.2022.859350] [Citation(s) in RCA: 3] [Impact Index Per Article: 1.5] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 01/21/2022] [Accepted: 04/04/2022] [Indexed: 12/03/2022] Open
Abstract
Introduction The hepatitis B virus (HBV) is a blood-borne virus that can be transmitted by percutaneous and mucocutaneous contact with infected bodily fluid. Healthcare workers (HCWs) are more exposed to HBV infection. They must have a thorough understanding of HBV infection since they can contract and spread the virus. In this study, we systematically reviewed all published evidence on the seroprevalence of Hepatitis B virus (HBV) infection among HCWs. and synthesize evidence on the association between knowledge and awareness with HBV infection. Methods We searched PubMed, EMBASE, Cochrane Library and Scopus for studies reporting on HBV seroprevalence from January 1997 to September 2021 among healthcare workers. We used random-effects meta-analyses to estimate the pool prevalence of HBV infection. Results We identified 25 studies that met our inclusion criteria, with data on 10,043 adults from 11 countries and two regions: Africa and Asia. The overall seroprevalence of HBV was 5.0% (95% confidence interval [CI] 3.6%), with Africa reporting higher estimates (5.0%, 95% CI 3.7%) than Asia population (4.0%, 95% CI 1.9%). The highest pooled prevalence estimate in African countries came from studies published in the Cameroon region (8.0%, 95% CI 5–10%) while the lowest came from Ethiopia (4.0%, 95% CI 2.6%). The overall seroprevalence estimates in the African population were significantly higher than those in the Asian group. Studies in Africa found that the average knowledge and seroprevalence were 1.4% and 11.0%, respectively where, eight studies (53.3%) reported good knowledge and seven studies (46.7%) reported average knowledge. In Asia, two studies (40.0%) reported good knowledge, one study (20.0%) reporting average knowledge, and two studies (40.0%) reporting poor knowledge. African studies demonstrated good knowledge despite the fact that their HBV infection rate was higher than 6.7%. Conclusion Africa and Asia have the highest seroprevalence of HBV infection. Improving the comparability of epidemiological and clinical studies constitutes an important step forward. More high-quality data is needed to improve the precision of burden estimates. Systematic Review Registration PROSPERO CRD42021279905.
Collapse
Affiliation(s)
- Nur Hasnah Maamor
- Sector for Evidence-Based Healthcare, National Institutes of Health, Ministry of Health, Setia Alam, Malaysia
| | - Nor Asiah Muhamad
- Sector for Evidence-Based Healthcare, National Institutes of Health, Ministry of Health, Setia Alam, Malaysia
| | - Nor Soleha Mohd Dali
- Institute for Medical Research, National Institutes of Health, Ministry of Health, Setia Alam, Malaysia
| | - Mohd Hatta Abdul Mutalip
- Institute for Public Health, National Institute of Health, Ministry of Health, Setia Alam, Malaysia
| | - Fatin Norhasny Leman
- Sector for Evidence-Based Healthcare, National Institutes of Health, Ministry of Health, Setia Alam, Malaysia
| | - Tahir Aris
- Institute for Medical Research, National Institutes of Health, Ministry of Health, Setia Alam, Malaysia
| | - Nai Ming Lai
- School of Medicine, Taylor's University, Subang Jaya, Malaysia
| | - Muhammad Radzi Abu Hassan
- Department of Medicine, Sultanah Bahiyah Hospital, Ministry of Health, Alor Setar, Malaysia.,Clinical Research Centre, Sultanah Bahiyah Hospital, Ministry of Health, Alor Setar, Malaysia
| |
Collapse
|
7
|
Jones PD, Gmunder K, Batrony S, Martin P, Kobetz E, Carrasquillo O. Acceptability and Feasibility of Home-Based Hepatitis B Screening Among Haitian Immigrants. J Immigr Minor Health 2021; 23:1170-1178. [PMID: 33686574 PMCID: PMC9166933 DOI: 10.1007/s10903-021-01165-z] [Citation(s) in RCA: 2] [Impact Index Per Article: 0.7] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Grants] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Accepted: 02/15/2021] [Indexed: 10/22/2022]
Abstract
Hepatitis B (HBV) is endemic in Haiti, therefore Haitian immigrants should be screened to identify and link affected individuals to care. Current screening approaches are ineffective. We assessed the acceptability and feasibility of home-based screening among Haitian immigrants using community health workers (CHWs). We recruited participants exiting a pragmatic trial evaluating strategies to improve care delivery (NCT02970136). Participants completed an acceptability questionnaire. Blood drawn by CHWs at participants' homes or community sites was tested for hepatitis B surface antigen, hepatitis B surface antibody and hepatitis B core antibody. Of 60 participants, 59 found screening acceptable; 53 had blood drawn. Of those, 45.3% had HBV previously, 49.1% remained susceptible and 5.7% were vaccinated. Respondents cited various reasons community members might find screening unacceptable. The high prior HBV rate highlights the need for effective outreach programs. Home-based HBV screening was both acceptable and feasible among Haitian immigrants.
Collapse
Affiliation(s)
- Patricia D Jones
- Sylvester Comprehensive Cancer Center, University of Miami Miller School of Medicine, Miami, FL, USA.
- Department of Medicine, Division of Digestive Health and Liver Diseases, University of Miami Miller School of Medicine, 1120 NW 14th Street, Suite 1112, Miami, FL, 33136, USA.
| | - Kristin Gmunder
- Sylvester Comprehensive Cancer Center, University of Miami Miller School of Medicine, Miami, FL, USA
| | - Saradjine Batrony
- Sylvester Comprehensive Cancer Center, University of Miami Miller School of Medicine, Miami, FL, USA
| | - Paul Martin
- Sylvester Comprehensive Cancer Center, University of Miami Miller School of Medicine, Miami, FL, USA
- Department of Medicine, Division of Digestive Health and Liver Diseases, University of Miami Miller School of Medicine, 1120 NW 14th Street, Suite 1112, Miami, FL, 33136, USA
| | - Erin Kobetz
- Sylvester Comprehensive Cancer Center, University of Miami Miller School of Medicine, Miami, FL, USA
- Department of Medicine, Division of Computational Medicine and Population Health, University of Miami Miller School of Medicine, Miami, FL, USA
| | - Olveen Carrasquillo
- Sylvester Comprehensive Cancer Center, University of Miami Miller School of Medicine, Miami, FL, USA
- Department of Medicine, Division of General Internal Medicine, University of Miami Miller School of Medicine, Miami, FL, USA
| |
Collapse
|