RESUMO
BACKGROUND: Severe coronavirus disease 2019 (Covid-19) is associated with dysregulated inflammation. The effects of combination treatment with baricitinib, a Janus kinase inhibitor, plus remdesivir are not known. METHODS: We conducted a double-blind, randomized, placebo-controlled trial evaluating baricitinib plus remdesivir in hospitalized adults with Covid-19. All the patients received remdesivir (≤10 days) and either baricitinib (≤14 days) or placebo (control). The primary outcome was the time to recovery. The key secondary outcome was clinical status at day 15. RESULTS: A total of 1033 patients underwent randomization (with 515 assigned to combination treatment and 518 to control). Patients receiving baricitinib had a median time to recovery of 7 days (95% confidence interval [CI], 6 to 8), as compared with 8 days (95% CI, 7 to 9) with control (rate ratio for recovery, 1.16; 95% CI, 1.01 to 1.32; P = 0.03), and a 30% higher odds of improvement in clinical status at day 15 (odds ratio, 1.3; 95% CI, 1.0 to 1.6). Patients receiving high-flow oxygen or noninvasive ventilation at enrollment had a time to recovery of 10 days with combination treatment and 18 days with control (rate ratio for recovery, 1.51; 95% CI, 1.10 to 2.08). The 28-day mortality was 5.1% in the combination group and 7.8% in the control group (hazard ratio for death, 0.65; 95% CI, 0.39 to 1.09). Serious adverse events were less frequent in the combination group than in the control group (16.0% vs. 21.0%; difference, -5.0 percentage points; 95% CI, -9.8 to -0.3; P = 0.03), as were new infections (5.9% vs. 11.2%; difference, -5.3 percentage points; 95% CI, -8.7 to -1.9; P = 0.003). CONCLUSIONS: Baricitinib plus remdesivir was superior to remdesivir alone in reducing recovery time and accelerating improvement in clinical status among patients with Covid-19, notably among those receiving high-flow oxygen or noninvasive ventilation. The combination was associated with fewer serious adverse events. (Funded by the National Institute of Allergy and Infectious Diseases; ClinicalTrials.gov number, NCT04401579.).
Assuntos
Monofosfato de Adenosina/análogos & derivados , Alanina/análogos & derivados , Antivirais/uso terapêutico , Azetidinas/uso terapêutico , Tratamento Farmacológico da COVID-19 , Purinas/uso terapêutico , Pirazóis/uso terapêutico , Sulfonamidas/uso terapêutico , Monofosfato de Adenosina/efeitos adversos , Monofosfato de Adenosina/uso terapêutico , Adulto , Idoso , Alanina/efeitos adversos , Alanina/uso terapêutico , Antivirais/efeitos adversos , Azetidinas/efeitos adversos , COVID-19/mortalidade , COVID-19/terapia , Método Duplo-Cego , Quimioterapia Combinada , Feminino , Mortalidade Hospitalar , Hospitalização , Humanos , Inibidores de Janus Quinases/efeitos adversos , Inibidores de Janus Quinases/uso terapêutico , Masculino , Pessoa de Meia-Idade , Oxigenoterapia , Purinas/efeitos adversos , Pirazóis/efeitos adversos , Respiração Artificial , Sulfonamidas/efeitos adversos , Resultado do TratamentoRESUMO
BACKGROUND: The COVID-19 standard of care (SOC) evolved rapidly during 2020 and 2021, but its cumulative effect over time is unclear. OBJECTIVE: To evaluate whether recovery and mortality improved as SOC evolved, using data from ACTT (Adaptive COVID-19 Treatment Trial). DESIGN: ACTT is a series of phase 3, randomized, double-blind, placebo-controlled trials that evaluated COVID-19 therapeutics from February 2020 through May 2021. ACTT-1 compared remdesivir plus SOC to placebo plus SOC, and in ACTT-2 and ACTT-3, remdesivir plus SOC was the control group. This post hoc analysis compared recovery and mortality between these comparable sequential cohorts of patients who received remdesivir plus SOC, adjusting for baseline characteristics with propensity score weighting. The analysis was repeated for participants in ACTT-3 and ACTT-4 who received remdesivir plus dexamethasone plus SOC. Trends in SOC that could explain outcome improvements were analyzed. (ClinicalTrials.gov: NCT04280705 [ACTT-1], NCT04401579 [ACTT-2], NCT04492475 [ACTT-3], and NCT04640168 [ACTT-4]). SETTING: 94 hospitals in 10 countries (86% U.S. participants). PARTICIPANTS: Adults hospitalized with COVID-19. INTERVENTION: SOC. MEASUREMENTS: 28-day mortality and recovery. RESULTS: Although outcomes were better in ACTT-2 than in ACTT-1, adjusted hazard ratios (HRs) were close to 1 (HR for recovery, 1.04 [95% CI, 0.92 to 1.17]; HR for mortality, 0.90 [CI, 0.56 to 1.40]). Comparable patients were less likely to be intubated in ACTT-2 than in ACTT-1 (odds ratio, 0.75 [CI, 0.53 to 0.97]), and hydroxychloroquine use decreased. Outcomes improved from ACTT-2 to ACTT-3 (HR for recovery, 1.43 [CI, 1.24 to 1.64]; HR for mortality, 0.45 [CI, 0.21 to 0.97]). Potential explanatory factors (SOC trends, case surges, and variant trends) were similar between ACTT-2 and ACTT-3, except for increased dexamethasone use (11% to 77%). Outcomes were similar in ACTT-3 and ACTT-4. Antibiotic use decreased gradually across all stages. LIMITATION: Unmeasured confounding. CONCLUSION: Changes in patient composition explained improved outcomes from ACTT-1 to ACTT-2 but not from ACTT-2 to ACTT-3, suggesting improved SOC. These results support excluding nonconcurrent controls from analysis of platform trials in rapidly changing therapeutic areas. PRIMARY FUNDING SOURCE: National Institute of Allergy and Infectious Diseases.
Assuntos
Antivirais , Tratamento Farmacológico da COVID-19 , Adulto , Humanos , Antivirais/uso terapêutico , Ensaios Clínicos Fase III como Assunto , Dexametasona , Método Duplo-Cego , Ensaios Clínicos Controlados Aleatórios como Assunto , Resultado do TratamentoRESUMO
BACKGROUND: It is unclear whether asthma and its allergic phenotype are risk factors for hospitalization or severe disease from SARS-CoV-2. METHODS: All patients over 28 days old testing positive for SARS-CoV-2 between March 1 and September 30, 2020, were retrospectively identified and characterized through electronic analysis at Stanford. A sub-cohort was followed prospectively to evaluate long-term COVID-19 symptoms. RESULTS: 168,190 patients underwent SARS-CoV-2 testing, and 6,976 (4.15%) tested positive. In a multivariate analysis, asthma was not an independent risk factor for hospitalization (OR 1.12 [95% CI 0.86, 1.45], p = .40). Among SARS-CoV-2-positive asthmatics, allergic asthma lowered the risk of hospitalization and had a protective effect compared with non-allergic asthma (OR 0.52 [0.28, 0.91], p = .026); there was no association between baseline medication use as characterized by GINA and hospitalization risk. Patients with severe COVID-19 disease had lower eosinophil levels during hospitalization compared with patients with mild or asymptomatic disease, independent of asthma status (p = .0014). In a patient sub-cohort followed longitudinally, asthmatics and non-asthmatics had similar time to resolution of COVID-19 symptoms, particularly lower respiratory symptoms. CONCLUSIONS: Asthma is not a risk factor for more severe COVID-19 disease. Allergic asthmatics were half as likely to be hospitalized with COVID-19 compared with non-allergic asthmatics. Lower levels of eosinophil counts (allergic biomarkers) were associated with a more severe COVID-19 disease trajectory. Recovery was similar among asthmatics and non-asthmatics with over 50% of patients reporting ongoing lower respiratory symptoms 3 months post-infection.
Assuntos
Asma , COVID-19 , Asma/diagnóstico , Asma/epidemiologia , Teste para COVID-19 , Humanos , Fenótipo , Estudos Retrospectivos , SARS-CoV-2RESUMO
It is well recognised that medical training globally and at all levels lacks sufficient incorporation of genetics and genomics education to keep up with the rapid advances and growing application of genomics to clinical care. However, the best strategy to implement these desired changes into postgraduate medical training and engage learners is still unclear. We developed a novel elective rotation in 'Genomic Medicine and Undiagnosed Diseases' for categorical Internal Medicine Residents to address this educational gap and serve as an adaptable model for training that can be applied broadly across different specialties and at other institutions. Key curriculum goals achieved include increased understanding about genetic testing modalities and tools available for diagnosis and risk analysis, the role of genetics-trained allied health professionals, and indications and limitations of genetic and genomic testing in both rare and common conditions.
Assuntos
Genômica/educação , Internato e Residência , Atitude do Pessoal de Saúde , California , Currículo , Humanos , Medicina Interna , Inquéritos e QuestionáriosRESUMO
BACKGROUND: Thrombophilia testing is frequently ordered in the inpatient setting despite its limited impact on clinical decision-making and unreliable results in the setting of acute thrombosis or ongoing anticoagulation. We sought to determine the effect of an educational intervention in reducing inappropriate thrombophilia testing for hospitalized patients. METHODS: During the 2014 academic year, we implemented an educational intervention with a phase implementation design for Internal Medicine interns at Stanford University Hospital. The educational session covering epidemiology, appropriate thrombophilia evaluation and clinical rationale behind these recommendations. Their ordering behavior was compared with a contemporaneous control (non-medicine and private services) and a historical control (interns from prior academic year). From the analyzed data, we determined the proportion of inappropriate thrombophilia testing of each group. Logistic generalized estimating equations were used to estimate odds ratios for inappropriate thrombophilia testing associated with the intervention. RESULTS: Of 2151 orders included, 934 were deemed inappropriate (43.4%). The two intervention groups placed 147 orders. A pooled analysis of ordering practices by intervention groups revealed a trend toward reduction of inappropriate ordering (p = 0.053). By the end of the study, the intervention groups had significantly lower rates of inappropriate testing compared to historical or contemporaneous controls. CONCLUSION: A brief educational intervention was associated with a trend toward reduction in inappropriate thrombophilia testing. These findings suggest that focused education on thrombophilia testing can positively impact inpatient ordering practices.
Assuntos
Hospitalização , Medicina Interna/educação , Internato e Residência , Trombofilia/diagnóstico , Adulto , Feminino , Hospitais Universitários , Humanos , Masculino , Seleção de PacientesRESUMO
OBJECTIVE: The aim of this study was to characterize the relationship between computed tomography angiography imaging characteristics of coronary artery and atherosclerotic cardiovascular disease (ASCVD) score. METHODS: We retrospectively identified all patients who underwent a coronary computed tomography angiography at our institution from December 2013 to July 2016, then we calculated the 10-year ASCVD score. We characterized the relationship between coronary artery imaging findings and ASCVD risk score. RESULTS: One hundred fifty-one patients met our inclusion criteria. Patients with a 10-year ASCVD score of 7.5% or greater had significantly more arterial segments showing stenosis (46.4%, P = 0.008) and significantly higher maximal plaque thickness (1.25 vs 0.53, P = 0.001). However, among 56 patients with a 10-year ASCVD score of 7.5% or greater, 30 (53.6%) had no arterial stenosis. Furthermore, among the patients with a 10-year ASCVD score of less than 7.5%, 24 (25.3%) had some arterial stenosis. CONCLUSIONS: There is some concordance but not a perfect overlap between 10-year ASCVD risk scores and coronary artery imaging findings.
Assuntos
Angiografia por Tomografia Computadorizada/métodos , Angiografia Coronária/métodos , Doença da Artéria Coronariana/diagnóstico por imagem , Adulto , Idoso , American Heart Association , Vasos Coronários/diagnóstico por imagem , Feminino , Humanos , Masculino , Pessoa de Meia-Idade , Valor Preditivo dos Testes , Estudos Retrospectivos , Medição de Risco/métodos , Fatores de Risco , Índice de Gravidade de Doença , Estados UnidosAssuntos
COVID-19 , SARS-CoV-2 , COVID-19/epidemiologia , COVID-19/virologia , Hospitalização , Humanos , SARS-CoV-2/genéticaRESUMO
PURPOSE: Most residency programmes do not have a formal high value care curriculum. Our goal was to design and implement a multidisciplinary high value care curriculum specifically targeted at interns. DESIGN: Our curriculum was designed with multidisciplinary input from attendings, fellows and residents at Stanford. Curricular topics were inspired by the American Board of Internal Medicine's Choosing Wisely campaign, Alliance for Academic Internal Medicine, American College of Physicians and Society of Hospital Medicine. Our topics were as follows: introduction to value-based care; telemetry utilisation; lab ordering; optimal approach to thrombophilia work-ups and fresh frozen plasma use; optimal approach to palliative care referrals; antibiotic stewardship; and optimal approach to imaging for low back pain. Our curriculum was implemented at the Stanford Internal Medicine residency programme over the course of two academic years (2014 and 2015), during which 100 interns participated in our high value care curriculum. After each high value care session, interns were offered the opportunity to complete surveys regarding feedback on the curriculum, self-reported improvements in knowledge, skills and attitudinal module objectives, and quiz-based knowledge assessments. RESULTS: The overall survey response rate was 67.1%. Overall, the material was rated as highly useful on a 5-point Likert scale (mean 4.4, SD 0.6). On average, interns reported a significant improvement in their self-rated knowledge, skills and attitudes after the six seminars (mean improvement 1.6 points, SD 0.4 (95% CI 1.5 to 1.7), p<0.001). CONCLUSIONS: We successfully implemented a novel high value care curriculum that specifically targets intern physicians.
Assuntos
Currículo , Educação de Pós-Graduação em Medicina/organização & administração , Medicina Interna/educação , Internato e Residência , Adulto , Competência Clínica , Avaliação Educacional , Retroalimentação , Feminino , Humanos , MasculinoRESUMO
OBJECTIVE: The aim of the study was to examine the impact of a surgical comanagement (SCM) hospitalist program on patient outcomes at an academic institution. BACKGROUND: Prior studies may have underestimated the impact of SCM due to methodological shortcomings. METHODS: This is a retrospective study utilizing a propensity score-weighted intervention (n = 16,930) and control group (n = 3695). Patients were admitted between January 2009 to July 2012 (pre-SCM) and September 2012 to September 2013 (post-SCM) to Orthopedic or Neurosurgery at our institution. Using propensity score methods, linear regression, and a difference-in-difference approach, we estimated changes in outcomes between pre and post periods, while adjusting for confounding patient characteristics. RESULTS: The SCM intervention was associated with a significant differential decrease in the proportion of patients with at least 1 medical complication [odds ratio (OR) 0.86; 95% confidence interval (CI), 0.74-0.96; P = 0.008), the proportion of patients with length of stay at least 5 days (OR 0.75; 95% CI, 0.67-0.84; P < 0.001), 30-day readmission rate for medical cause (OR 0.67; 95% CI, 0.52-0.81; P < 0.001), and the proportion of patients with at least 2 medical consultants (OR 0.55; 95% CI, 0.49-0.63; P < 0.001). There was no significant change in patient satisfaction (OR 1.08; 95% CI, 0.87-1.33; P = 0.507). We estimated average savings of $2642 to $4303 per patient in the post-SCM group. The overall provider satisfaction with SCM was 88.3%. CONCLUSIONS: The SCM intervention reduces medical complications, length of stay, 30-day readmissions, number of consultants, and cost of care.
Assuntos
Hospitalização , Procedimentos Neurocirúrgicos/efeitos adversos , Procedimentos Ortopédicos/efeitos adversos , Equipe de Assistência ao Paciente/organização & administração , Assistência Perioperatória , Complicações Pós-Operatórias/prevenção & controle , Idoso , Idoso de 80 Anos ou mais , Estudos Controlados Antes e Depois , Feminino , Humanos , Masculino , Pessoa de Meia-Idade , Satisfação do Paciente , Pontuação de Propensão , Encaminhamento e Consulta , Estudos RetrospectivosAssuntos
Delírio , Idoso , Delírio/diagnóstico , Delírio/epidemiologia , Avaliação Geriátrica , Hospitais , HumanosAssuntos
Delírio , Adolescente , Adulto , Idoso , Delírio/diagnóstico , Delírio/epidemiologia , Hospitalização , Hospitais , Humanos , Pessoa de Meia-Idade , Fatores de Risco , Adulto JovemRESUMO
OBJECTIVE: The aim of this study is to evaluate the effect of 2 hospital-wide interventions on achieving a discharge-before-noon rate of 40%. BACKGROUND: A multidisciplinary team led by administrative and physician leadership developed a plan to diminish capacity constraints by minimizing late afternoon hospital discharges using 2 patient flow management techniques. METHODS: The study was a preintervention/postintervention retrospective analysis observing all inpatients discharged across 19 inpatient units in a 484-bed, academic teaching hospital measuring calendar month discharge-before-noon percentage, patient satisfaction, and readmission rates. Patient satisfaction and readmission rates were used as baseline metrics. RESULTS: The discharge-before-noon percentage increased from 14% in the 11-month preintervention period to an average of 24% over the 11-month postintervention period, whereas patient satisfaction scores and readmission rates remained stable. CONCLUSIONS: Implementation of the 2 interventions successfully increased the percentage of discharges before noon yet did not achieve the goal of 40%. Patient satisfaction and readmission rates were not negatively impacted by the program.
Assuntos
Fortalecimento Institucional/normas , Equipes de Administração Institucional/organização & administração , Alta do Paciente/normas , Fortalecimento Institucional/métodos , Fortalecimento Institucional/organização & administração , Eficiência Organizacional , Hospitais de Ensino/organização & administração , Hospitais de Ensino/normas , Humanos , Equipes de Administração Institucional/normas , Comunicação Interdisciplinar , Estudos de Casos Organizacionais , Alta do Paciente/estatística & dados numéricos , Readmissão do Paciente/estatística & dados numéricos , Satisfação do Paciente/estatística & dados numéricos , Fatores de Tempo , Gestão da Qualidade Total/métodos , Gestão da Qualidade Total/organização & administração , Gestão da Qualidade Total/normasRESUMO
Catastrophic antiphospholipid syndrome (CAPS) is fatal in approximately 44% of patients in whom the diagnosis is made, thus demonstrating the inadequacy of current medical therapy. In this report, we discuss a 47-year-old man with a known history of primary antiphospholipid syndrome, who presented with CAPS after undergoing cholecystectomy and a treatment-refractory early relapse after development of colitis. Given the potential therapeutic efficacy of complement inhibition in antiphospholipid syndrome, the patient was administered eculizumab, a terminal complement inhibitor. Progressive clinical improvement and laboratory improvement were observed upon initiation of eculizumab. He has remained in remission for over 16 months of follow-up while on eculizumab. In conclusion, this case represents successful use of eculizumab for the treatment of primary CAPS.
Assuntos
Anticorpos Monoclonais Humanizados/administração & dosagem , Síndrome Antifosfolipídica , Complicações Pós-Operatórias , Síndrome Antifosfolipídica/diagnóstico , Síndrome Antifosfolipídica/tratamento farmacológico , Síndrome Antifosfolipídica/fisiopatologia , Doença Catastrófica , Colecistectomia/efeitos adversos , Inativadores do Complemento/administração & dosagem , Monitoramento de Medicamentos/métodos , Humanos , Masculino , Pessoa de Meia-Idade , Complicações Pós-Operatórias/diagnóstico , Complicações Pós-Operatórias/tratamento farmacológico , Complicações Pós-Operatórias/fisiopatologia , Recidiva , Resultado do TratamentoRESUMO
Importance: Diagnostic errors are common and cause significant morbidity. Large language models (LLMs) have shown promise in their performance on both multiple-choice and open-ended medical reasoning examinations, but it remains unknown whether the use of such tools improves diagnostic reasoning. Objective: To assess the impact of the GPT-4 LLM on physicians' diagnostic reasoning compared to conventional resources. Design: Multi-center, randomized clinical vignette study. Setting: The study was conducted using remote video conferencing with physicians across the country and in-person participation across multiple academic medical institutions. Participants: Resident and attending physicians with training in family medicine, internal medicine, or emergency medicine. Interventions: Participants were randomized to access GPT-4 in addition to conventional diagnostic resources or to just conventional resources. They were allocated 60 minutes to review up to six clinical vignettes adapted from established diagnostic reasoning exams. Main Outcomes and Measures: The primary outcome was diagnostic performance based on differential diagnosis accuracy, appropriateness of supporting and opposing factors, and next diagnostic evaluation steps. Secondary outcomes included time spent per case and final diagnosis. Results: 50 physicians (26 attendings, 24 residents) participated, with an average of 5.2 cases completed per participant. The median diagnostic reasoning score per case was 76.3 percent (IQR 65.8 to 86.8) for the GPT-4 group and 73.7 percent (IQR 63.2 to 84.2) for the conventional resources group, with an adjusted difference of 1.6 percentage points (95% CI -4.4 to 7.6; p=0.60). The median time spent on cases for the GPT-4 group was 519 seconds (IQR 371 to 668 seconds), compared to 565 seconds (IQR 456 to 788 seconds) for the conventional resources group, with a time difference of -82 seconds (95% CI -195 to 31; p=0.20). GPT-4 alone scored 15.5 percentage points (95% CI 1.5 to 29, p=0.03) higher than the conventional resources group. Conclusions and Relevance: In a clinical vignette-based study, the availability of GPT-4 to physicians as a diagnostic aid did not significantly improve clinical reasoning compared to conventional resources, although it may improve components of clinical reasoning such as efficiency. GPT-4 alone demonstrated higher performance than both physician groups, suggesting opportunities for further improvement in physician-AI collaboration in clinical practice.
RESUMO
Importance: Large language model (LLM) artificial intelligence (AI) systems have shown promise in diagnostic reasoning, but their utility in management reasoning with no clear right answers is unknown. Objective: To determine whether LLM assistance improves physician performance on open-ended management reasoning tasks compared to conventional resources. Design: Prospective, randomized controlled trial conducted from 30 November 2023 to 21 April 2024. Setting: Multi-institutional study from Stanford University, Beth Israel Deaconess Medical Center, and the University of Virginia involving physicians from across the United States. Participants: 92 practicing attending physicians and residents with training in internal medicine, family medicine, or emergency medicine. Intervention: Five expert-developed clinical case vignettes were presented with multiple open-ended management questions and scoring rubrics created through a Delphi process. Physicians were randomized to use either GPT-4 via ChatGPT Plus in addition to conventional resources (e.g., UpToDate, Google), or conventional resources alone. Main Outcomes and Measures: The primary outcome was difference in total score between groups on expert-developed scoring rubrics. Secondary outcomes included domain-specific scores and time spent per case. Results: Physicians using the LLM scored higher compared to those using conventional resources (mean difference 6.5 %, 95% CI 2.7-10.2, p<0.001). Significant improvements were seen in management decisions (6.1%, 95% CI 2.5-9.7, p=0.001), diagnostic decisions (12.1%, 95% CI 3.1-21.0, p=0.009), and case-specific (6.2%, 95% CI 2.4-9.9, p=0.002) domains. GPT-4 users spent more time per case (mean difference 119.3 seconds, 95% CI 17.4-221.2, p=0.02). There was no significant difference between GPT-4-augmented physicians and GPT-4 alone (-0.9%, 95% CI -9.0 to 7.2, p=0.8). Conclusions and Relevance: LLM assistance improved physician management reasoning compared to conventional resources, with particular gains in contextual and patient-specific decision-making. These findings indicate that LLMs can augment management decision-making in complex cases. Trial Registration: ClinicalTrials.gov Identifier: NCT06208423 ; https://classic.clinicaltrials.gov/ct2/show/NCT06208423. Key Points: Question: Does large language model (LLM) assistance improve physician performance on complex management reasoning tasks compared to conventional resources?Findings: In this randomized controlled trial of 92 physicians, participants using GPT-4 achieved higher scores on management reasoning compared to those using conventional resources (e.g., UpToDate).Meaning: LLM assistance enhances physician management reasoning performance in complex cases with no clear right answers.
RESUMO
Analyzing vast textual data and summarizing key information from electronic health records imposes a substantial burden on how clinicians allocate their time. Although large language models (LLMs) have shown promise in natural language processing (NLP) tasks, their effectiveness on a diverse range of clinical summarization tasks remains unproven. Here we applied adaptation methods to eight LLMs, spanning four distinct clinical summarization tasks: radiology reports, patient questions, progress notes and doctor-patient dialogue. Quantitative assessments with syntactic, semantic and conceptual NLP metrics reveal trade-offs between models and adaptation methods. A clinical reader study with 10 physicians evaluated summary completeness, correctness and conciseness; in most cases, summaries from our best-adapted LLMs were deemed either equivalent (45%) or superior (36%) compared with summaries from medical experts. The ensuing safety analysis highlights challenges faced by both LLMs and medical experts, as we connect errors to potential medical harm and categorize types of fabricated information. Our research provides evidence of LLMs outperforming medical experts in clinical text summarization across multiple tasks. This suggests that integrating LLMs into clinical workflows could alleviate documentation burden, allowing clinicians to focus more on patient care.
Assuntos
Documentação , Semântica , Humanos , Registros Eletrônicos de Saúde , Processamento de Linguagem Natural , Relações Médico-PacienteRESUMO
Over 85 million computed tomography (CT) scans are performed annually in the US, of which approximately one quarter focus on the abdomen. Given the current shortage of both general and specialized radiologists, there is a large impetus to use artificial intelligence to alleviate the burden of interpreting these complex imaging studies while simultaneously using the images to extract novel physiological insights. Prior state-of-the-art approaches for automated medical image interpretation leverage vision language models (VLMs) that utilize both the image and the corresponding textual radiology reports. However, current medical VLMs are generally limited to 2D images and short reports. To overcome these shortcomings for abdominal CT interpretation, we introduce Merlin - a 3D VLM that leverages both structured electronic health records (EHR) and unstructured radiology reports for pretraining without requiring additional manual annotations. We train Merlin using a high-quality clinical dataset of paired CT scans (6+ million images from 15,331 CTs), EHR diagnosis codes (1.8+ million codes), and radiology reports (6+ million tokens) for training. We comprehensively evaluate Merlin on 6 task types and 752 individual tasks. The non-adapted (off-the-shelf) tasks include zero-shot findings classification (31 findings), phenotype classification (692 phenotypes), and zero-shot cross-modal retrieval (image to findings and image to impressions), while model adapted tasks include 5-year chronic disease prediction (6 diseases), radiology report generation, and 3D semantic segmentation (20 organs). We perform internal validation on a test set of 5,137 CTs, and external validation on 7,000 clinical CTs and on two public CT datasets (VerSe, TotalSegmentator). Beyond these clinically-relevant evaluations, we assess the efficacy of various network architectures and training strategies to depict that Merlin has favorable performance to existing task-specific baselines. We derive data scaling laws to empirically assess training data needs for requisite downstream task performance. Furthermore, unlike conventional VLMs that require hundreds of GPUs for training, we perform all training on a single GPU. This computationally efficient design can help democratize foundation model training, especially for health systems with compute constraints. We plan to release our trained models, code, and dataset, pending manual removal of all protected health information.
RESUMO
BACKGROUND: Different patient characteristics influence the decision to order diagnostic imaging for deep venous thrombosis (DVT) and pulmonary embolism (PE) in different settings (emergency department (ED), hospital, and office). Diagnostic yield is defined as the proportion of tests that report positive results. We hypothesize different patient characteristics are associated with higher or lower diagnostic yield of imaging for DVT and PE in different settings. METHODS: We used Optum Clinformatics™ national claims database (2015-2019) to assess the diagnostic yield of imaging for DVT and PE in three settings: (a) ED discharge, (b) Hospitalized, and (c) Office. We studied the patient characteristics associated with diagnostic yield using logistic regression. RESULTS: Diagnostic imaging for DVT and PE was performed in 1,502,417 and 710,263 visits, respectively. Diagnostic yield for DVT and PE was 9.8 ± 0.1 % and 12.7 ± 0.1 %, respectively in the overall cohort. In the ED discharge, hospitalized, and office settings, diagnostic yield for DVT was 10.4 ± 0.1 %, 16.9 ± 0.1 %, and 6.5 ± 0.1 %, respectively, and that for PE 6.4 ± 0.1 %, 18.7 ± 0.1 %, and 8.8 ± 0.2 %, respectively. Of the patients who underwent imaging for DVT, higher diagnostic yield was more likely with thrombophilia, central venous access, and cancer. Of the patients who underwent imaging for PE, higher diagnostic yield was most likely with thrombophilia, respiratory failure, and heart failure or acute myocardial infarction. CONCLUSIONS: In each setting, different patient characteristics influence the diagnostic yield of imaging for DVT and PE and can inform clinical practice. Judicious use of imaging for DVT and PE could reduce costs and avoid exposure to radiation and contrast.
Assuntos
Embolia Pulmonar , Trombofilia , Trombose Venosa , Humanos , Trombose Venosa/diagnóstico por imagem , Trombose Venosa/complicações , Embolia Pulmonar/diagnóstico por imagem , Embolia Pulmonar/complicações , Diagnóstico por Imagem , Hospitais , Trombofilia/complicações , Fatores de RiscoRESUMO
Sifting through vast textual data and summarizing key information from electronic health records (EHR) imposes a substantial burden on how clinicians allocate their time. Although large language models (LLMs) have shown immense promise in natural language processing (NLP) tasks, their efficacy on a diverse range of clinical summarization tasks has not yet been rigorously demonstrated. In this work, we apply domain adaptation methods to eight LLMs, spanning six datasets and four distinct clinical summarization tasks: radiology reports, patient questions, progress notes, and doctor-patient dialogue. Our thorough quantitative assessment reveals trade-offs between models and adaptation methods in addition to instances where recent advances in LLMs may not improve results. Further, in a clinical reader study with ten physicians, we show that summaries from our best-adapted LLMs are preferable to human summaries in terms of completeness and correctness. Our ensuing qualitative analysis highlights challenges faced by both LLMs and human experts. Lastly, we correlate traditional quantitative NLP metrics with reader study scores to enhance our understanding of how these metrics align with physician preferences. Our research marks the first evidence of LLMs outperforming human experts in clinical text summarization across multiple tasks. This implies that integrating LLMs into clinical workflows could alleviate documentation burden, empowering clinicians to focus more on personalized patient care and the inherently human aspects of medicine.