Influence of a Large Language Model on Diagnostic Reasoning: A Randomized Clinical Vignette Study.

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.

Bacterial DnaK reduces the activity of anti-cancer drugs cisplatin and 5FU.

BACKGROUND: Chemotherapy is a primary treatment for cancer, but its efficacy is often limited by cancer-associated bacteria (CAB) that impair tumor suppressor functions. Our previous research found that Mycoplasma fermentans DnaK, a chaperone protein, impairs p53 activities, which are essential for most anti-cancer chemotherapeutic responses. METHODS: To investigate the role of DnaK in chemotherapy, we treated cancer cell lines with M. fermentans DnaK and then with commonly used p53-dependent anti-cancer drugs (cisplatin and 5FU). We evaluated the cells' survival in the presence or absence of a DnaK-binding peptide (ARV-1502). We also validated our findings using primary tumor cells from a novel DnaK knock-in mouse model. To provide a broader context for the clinical significance of these findings, we investigated human primary cancer sequencing datasets from The Cancer Genome Atlas (TCGA). We identified F. nucleatum as a CAB carrying DnaK with an amino acid composition highly similar to M. fermentans DnaK. Therefore, we investigated the effect of F. nucleatum DnaK on the anti-cancer activity of cisplatin and 5FU. RESULTS: Our results show that both M. fermentans and F. nucleatum DnaKs reduce the effectiveness of cisplatin and 5FU. However, the use of ARV-1502 effectively restored the drugs' anti-cancer efficacy. CONCLUSIONS: Our findings offer a practical framework for designing and implementing novel personalized anti-cancer strategies by targeting specific bacterial DnaKs in patients with poor response to chemotherapy, underscoring the potential for microbiome-based personalized cancer therapies.

Effectiveness of an Artificial Intelligence-Enabled Intervention for Detecting Clinical Deterioration.

Importance: Inpatient clinical deterioration is associated with substantial morbidity and mortality but may be easily missed by clinicians. Early warning scores have been developed to alert clinicians to patients at high risk of clinical deterioration, but there is limited evidence for their effectiveness. Objective: To evaluate the effectiveness of an artificial intelligence deterioration model-enabled intervention to reduce the risk of escalations in care among hospitalized patients using a study design that facilitates stronger causal inference. Design, Setting, and Participants: This cohort study used a regression discontinuity design that controlled for confounding and was based on Epic Deterioration Index (EDI; Epic Systems Corporation) prediction model scores. Compared with other observational research, the regression discontinuity design facilitates causal analysis. Hospitalized adults were included from 4 general internal medicine units in 1 academic hospital from January 17, 2021, through November 16, 2022. Exposure: An artificial intelligence deterioration model-enabled intervention, consisting of alerts based on an EDI score threshold with an associated collaborative workflow among nurses and physicians. Main Outcomes and Measures: The primary outcome was escalations in care, including rapid response team activation, transfer to the intensive care unit, or cardiopulmonary arrest during hospitalization. Results: During the study, 9938 patients were admitted to 1 of the 4 units, with 963 patients (median [IQR] age, 76.1 [64.2-86.2] years; 498 males [52.3%]) included within the primary regression discontinuity analysis. The median (IQR) Elixhauser Comorbidity Index score in the primary analysis cohort was 10 (0-24). The intervention was associated with a -10.4-percentage point (95% CI, -20.1 to -0.8 percentage points; P = .03) absolute risk reduction in the primary outcome for patients at the EDI score threshold. There was no evidence of a discontinuity in measured confounders at the EDI score threshold. Conclusions and Relevance: Using a regression discontinuity design, this cohort study found that the implementation of an artificial intelligence deterioration model-enabled intervention was associated with a significantly decreased risk of escalations in care among inpatients. These results provide evidence for the effectiveness of this intervention and support its further expansion and testing in other care settings.

IFNα induces CCR5 in CD4+ T cells of HIV patients causing pathogenic elevation.

BACKGROUND: Among people living with HIV, elite controllers (ECs) maintain an undetectable viral load, even without receiving anti-HIV therapy. In non-EC patients, this therapy leads to marked improvement, including in immune parameters, but unlike ECs, non-EC patients still require ongoing treatment and experience co-morbidities. In-depth, comprehensive immune analyses comparing EC and treated non-EC patients may reveal subtle, consistent differences. This comparison could clarify whether elevated circulating interferon-alpha (IFNα) promotes widespread immune cell alterations and persists post-therapy, furthering understanding of why non-EC patients continue to need treatment. METHODS: Levels of IFNα in HIV-infected EC and treated non-EC patients were compared, along with blood immune cell subset distribution and phenotype, and functional capacities in some cases. In addition, we assessed mechanisms potentially associated with IFNα overload. RESULTS: Treatment of non-EC patients results in restoration of IFNα control, followed by marked improvement in distribution numbers, phenotypic profiles of blood immune cells, and functional capacity. These changes still do not lead to EC status, however, and IFNα can induce these changes in normal immune cell counterparts in vitro. Hypothesizing that persistent alterations could arise from inalterable effects of IFNα at infection onset, we verified an IFNα-related mechanism. The protein induces the HIV coreceptor CCR5, boosting HIV infection and reducing the effects of anti-HIV therapies. EC patients may avoid elevated IFNα following on infection with a lower inoculum of HIV or because of some unidentified genetic factor. CONCLUSIONS: Early control of IFNα is essential for better prognosis of HIV-infected patients.

The treatment for HIV, known as antiretroviral therapy (ART), does not cure HIV but enables individuals to live longer, healthier lives. In this study, we compared immune responses between elite controllers (ECs), who control their HIV infection without any treatment, and ART-treated and untreated patients. We demonstrate that IFNα, a small protein crucial in controlling immune system, is excessively produced at the onset of HIV infection and at levels that persist, resulting in poor HIV control without therapy. We show a mechanism for lack of control of HIV by IFNα. While inhibiting HIV, IFNα also simultaneously increases the HIV co-receptor, CCR5, thereby facilitating virus entry into the target cell. This is avoided by ECs which we hypothesize is associated with a lower infectious inoculum of HIV.

Early elevated IFNα is a key mediator of HIV pathogenesis.

BACKGROUND: A complete understanding of the different steps of HIV replication and an effective drug combination have led to modern antiretroviral regimens that block HIV replication for decades, but these therapies are not curative and must be taken for life. "Elite controllers" (ECs) is a term for the 0.5% of HIV-infected persons requiring no antiretroviral therapy, whose status may point the way toward a functional HIV cure. Defining the mechanisms of this control may be key to understanding how to replicate this functional cure in others. METHODS: In ECs and untreated non-EC patients, we compared IFNα serum concentration, distribution of immune cell subsets, and frequency of cell markers associated with immune dysfunction. We also investigated the effect of an elevated dose of IFNα on distinct subsets within dendritic cells, natural killer cells, and CD4+ and CD8 + T cells. RESULTS: Serum IFNα was undetectable in ECs, but all immune cell subsets from untreated non-EC patients were structurally and functionally impaired. We also show that the altered phenotype and function of these cell subsets in non-EC patients can be recapitulated when cells are stimulated in vitro with high-dose IFNα. CONCLUSIONS: Elevated IFNα is a key mediator of HIV pathogenesis.

Currently, HIV infection is not curable, but infected individuals can manage their condition by taking daily doses of antiretroviral therapy. Some individuals, known as elite controllers (ECs), control their infection without antiretroviral treatment, and studying how their immune system responds to HIV exposure could lead to a potential cure for others. Here, we compare immune cell responses between ECs and untreated non-ECs. We find that IFNα, a small protein with an important role in controlling white blood cell activity, is produced in excess in immune cells from non-ECs compared with ECs during early infection. This insight provides an important clue for the future development of a targeted cure for HIV.

Mycoplasma DnaK expression increases cancer development in vivo upon DNA damage.

Well-controlled repair mechanisms are involved in the maintenance of genomic stability, and their failure can precipitate DNA abnormalities and elevate tumor risk. In addition, the tumor microenvironment, enriched with factors inducing oxidative stress and affecting cell cycle checkpoints, intensifies DNA damage when repair pathways falter. Recent research has unveiled associations between certain bacteria, including Mycoplasmas, and various cancers, and the causative mechanism(s) are under active investigation. We previously showed that Mycoplasma fermentans DnaK, an HSP70 family chaperone protein, hampers the activity of proteins like PARP1 and p53, crucial for genomic integrity. Moreover, our analysis of its interactome in human cancer cell lines revealed DnaK's engagement with several components of DNA-repair machinery. Finally, in vivo experiments performed in our laboratory using a DnaK knock-in mouse model generated by our group demonstrated that DnaK exposure led to increased DNA copy number variants, indicative of genomic instability. We present here evidence that expression of DnaK is linked to increased i) incidence of tumors in vivo upon exposure to urethane, a DNA damaging agent; ii) spontaneous DNA damage ex vivo; and iii) expression of proinflammatory cytokines ex vivo, variations in reactive oxygen species levels, and increased ß-galactosidase activity across tissues. Moreover, DnaK was associated with increased centromeric instability. Overall, these findings highlight the significance of Mycoplasma DnaK in the etiology of cancer and other genetic disorders providing a promising target for prevention, diagnostics, and therapeutics.

Diagnostic reasoning prompts reveal the potential for large language model interpretability in medicine.

One of the major barriers to using large language models (LLMs) in medicine is the perception they use uninterpretable methods to make clinical decisions that are inherently different from the cognitive processes of clinicians. In this manuscript we develop diagnostic reasoning prompts to study whether LLMs can imitate clinical reasoning while accurately forming a diagnosis. We find that GPT-4 can be prompted to mimic the common clinical reasoning processes of clinicians without sacrificing diagnostic accuracy. This is significant because an LLM that can imitate clinical reasoning to provide an interpretable rationale offers physicians a means to evaluate whether an LLMs response is likely correct and can be trusted for patient care. Prompting methods that use diagnostic reasoning have the potential to mitigate the "black box" limitations of LLMs, bringing them one step closer to safe and effective use in medicine.

Preoperative Antidepressant Prescriptions Are Associated With Increased Opioid Prescriptions and Health Care Use but Similar Rates of Secondary Surgery Following Primary Anterior Cruciate Ligament Reconstruction in a Young Adult Population.

PURPOSE: To compare adverse events, postoperative opioid-prescribing patterns, health care use, and secondary anterior cruciate ligament reconstruction (ACLR) surgery rates of patients undergoing primary ACLR with a preoperative antidepressant prescription (ADP) against a propensity-matched group with no preoperative antidepressant prescription (NADP) using the TriNetX Diamond Network. METHODS: Patients undergoing primary ACLR between ages 18 and 35 years of age were queried from the database using International Classification of Diseases, Tenth Revision/Current Procedural Terminology codes. Patients with an ADP were propensity matched in a 1:1 ratio to patients with NADP based on 11 patient characteristics. Postoperative rates of adverse events, emergency department (ED) visits, in-patient hospitalizations, outpatient services, physical therapy evaluations, postoperative opioid prescriptions, and secondary ACLR were compared at various time points. RESULTS: In total, 3,736 patients with an ADP with an average age of 21.4 ± 4.5 years undergoing primary ACLR were propensity matched to patients with NADP. A significantly greater percentage of patients with an ADP received opioid prescriptions at 2 weeks (ADP 21%, NADP 11.3%, odds ratio [OR] 2.08), 6 weeks (ADP 25.5%, NADP 13.9%, OR 2.13), 3 months (ADP 27.6%, NADP 15.6%, OR 2.07), 6 months (ADP 30.5%, NADP 17.2%, OR 2.1), and 1 year (ADP 35.3%, NADP 20.2%, OR 2.16) postoperatively (P <.0001 for each time point). Patients with ADP had greater rates of ED visits (ADP 9.7%, NADP 7.1%, P < .0001, OR 1.39) and outpatient appointments (ADP 28.3%, NADP 21.8%) P < .0001, OR 1.42) at 3 months' postoperatively. Secondary surgery rates at 1 and 2 years were nonsignificant (P = .381 and P = .062, respectively). CONCLUSIONS: Following ACLR, patients with ADP had a significant increase in postoperative opioid prescriptions at all time points and used more ED resources and outpatient services compared with patients with NADP at 3 months' postoperatively. Thirty-day postoperative adverse events and both 1- and 2-year secondary ACL surgery rates demonstrated no significant differences between the groups. LEVEL OF EVIDENCE: Level III, retrospective comparative case series.

Training and Evaluation of Residents to Ensure Competent Physician Metrics to Recognize Struggling Residents.

Assessing competency across domains of knowledge, skills, and behavior is critical to ensure that graduating orthopaedic residents possess the requisite skills and attributes to enter independent orthopaedic practice. Of the domains, knowledge is most easily assessed. In addition to the AAOS Orthopaedic In-Training Examination®, which provides a yearly gauge of residents' orthopaedic knowledge relative to their peers, there are several online platforms such as Orthobullets, the American Academy of Orthopaedic Surgeons ResStudy program, and the Journal of Bone and Joint Surgery Clinical Classroom that offer online learning resources and question banks. Clinical skills are best assessed through a combination of observation tools, including live or video assessments, 360° evaluations, and objective structured clinical examinations. Surgical skills can be evaluated in two domains: live surgical cases or simulations. The American Board of Orthopaedic Surgery is attempting to standardize live surgical evaluations through the use of the O-P tool. Although most available models feature only arthroscopic procedures, surgical simulators provide for opportunity to objectively evaluate resident performance. Behavior and professionalism has traditionally been the most challenging domain to assess. The American Board of Orthopaedic Surgery's Behavior Assessment Tool has demonstrated success in pilot testing and is being introduced as the standard for measuring behavior and professionalism in orthopaedic training. Although no single assessment tool can accurately gauge a resident's overall performance, a combination of readily available tools should be used to assess competence across domains.

Telemedicine After COVID-19: Incorporating Virtual Visits Into Your Practice.

The use of telemedicine services within orthopaedics increased rapidly as a result of the COVID-19 pandemic. Telemedicine may improve access to care and save time and money for patients and clinicians; however, limitations such as technical issues and limited physical examination may reduce its widespread adoption. Virtual visits generally produce equivalent satisfaction and clinical outcomes compared with those performed in person. Although telemedicine has served many different roles within orthopaedic practices, its main utility is for patients who have to travel significant distances and for visits that do not require physical examination to determine a treatment plan. Several regulations govern the use of telemedicine. Most notably, clinicians must be licensed to practice medicine in the state in which the patient is located during the appointment. Although compliance issues remain a potential source of legal issues, experts cite misdiagnosis from limited physical examination as the most likely reason for medical liability. Clinicians should be familiar with techniques for virtual physical examination and should provide instruction to patients before the visit to optimize data obtained.

How to Recognize, Correct, and Avoid Intraoperative Complications During Anterior Cruciate Ligament Reconstructions.

Technical complications are a leading cause of graft failure following anterior cruciate ligament reconstructions. Complications can occur during any phase of the procedure, from graft harvesting to tunnel preparation to graft fixation. Predicting potential causes of technical difficulty and developing strategies to avoid potential pitfalls can limit the number of intraoperative complications. If adverse events do occur intraoperatively, prompt recognition and treatment can lead to favorable outcomes. It is important to discuss strategies to understand potential complications and develop tactics to avoid and correct adverse events that can occur during anterior cruciate ligament reconstruction.

ChatGPT Influence on Medical Decision-Making, Bias, and Equity: A Randomized Study of Clinicians Evaluating Clinical Vignettes.

In a randomized, pre-post intervention study, we evaluated the influence of a large language model (LLM) generative AI system on accuracy of physician decision-making and bias in healthcare. 50 US-licensed physicians reviewed a video clinical vignette, featuring actors representing different demographics (a White male or a Black female) with chest pain. Participants were asked to answer clinical questions around triage, risk, and treatment based on these vignettes, then asked to reconsider after receiving advice generated by ChatGPT+ (GPT4). The primary outcome was the accuracy of clinical decisions based on pre-established evidence-based guidelines. Results showed that physicians are willing to change their initial clinical impressions given AI assistance, and that this led to a significant improvement in clinical decision-making accuracy in a chest pain evaluation scenario without introducing or exacerbating existing race or gender biases. A survey of physician participants indicates that the majority expect LLM tools to play a significant role in clinical decision making.

Association of Systemic Hormonal Contraceptives With Reconstruction Procedures for Patellar Instability.

Background: Previous research suggests that estrogen plays a role in increased ligamentous laxity observed within the female population. Whereas many studies have sought to evaluate the impact of exogenous estrogen on anterior cruciate ligament injuries, research has not yet explored its impact on the medial patellofemoral ligament. Furthermore, less is known about the role of exogenous progesterone on ligamentous structures. Purpose: To determine whether women prescribed systemic estrogen (ethinyl estradiol) or progesterone (norgestimate or etonogestrel) hormonal contraceptives had an increased risk of undergoing reconstruction surgery for patellar instability compared with women without a prescription for systemic hormonal contraceptives. Study Design: Cross-sectional study; Level of evidence, 3. Methods: The TriNetX Research Network database was queried using International Classification of Disease and Common Procedural Terminology codes for women aged 15 to 26 years who underwent reconstruction procedures for patellar instability between 2012 and 2022. Women were grouped according to whether they had a coded prescription for a systemic hormonal contraceptive containing either ethinyl estradiol or etonogestrel; controls were matched by age, sex, race, and ethnicity. The relative risk (RR) of undergoing reconstruction for patellar instability was determined for each contraceptive. Results: After 1-to-1 propensity score matching, 0.054% (525/980,878) of women prescribed a systemic contraceptive containing ethinyl estradiol underwent reconstruction procedures for patellar instability compared with 0.043% (417/980,878) of controls (RR, 1.3; 95% CI, 1.1-1.4; P = .0004). Likewise, 0.058% (67/116,260) of women prescribed a form of systemic contraceptive containing only etonogestrel underwent reconstruction procedures for patellar instability compared with 0.026% (30/116,260) of controls (RR, 2.2; 95% CI, 1.5-3.4; P = .0002). Conclusion: Female patients prescribed systemic contraceptives containing estrogen or progesterone had an increased rate of reconstruction procedures for patellar instability.

Prophylactic Knee Bracing in Offensive Linemen of the National Football League: A Retrospective Analysis of Usage Trends, Player Performance, and Major Knee Injury.

Background: Offensive linemen in American football are prone to high-energy valgus forces to the knee, leading to associated injuries. Some offensive linemen in the National Football League (NFL) wear prophylactic knee braces (PKB) to prevent ligamentous injury. Purpose/Hypothesis: This purpose of the study was to compare injury rates and performance between NFL offensive linemen who wear PKB and those who do not. It was hypothesized that brace wear would be associated with fewer major knee injuries and no difference in gameplay performance. Study Design: Cohort study; Level of evidence, 3. Methods: For the 2014 through 2020 NFL seasons, offensive linemen with at least 200 game snaps per regular season were identified. Players were grouped by PKB status (bracers vs nonbracers) based on visualization of bilateral, dual-hinged metal knee braces as part of gameday uniforms on publicly available imaging databases and/or game videos. Major knee injuries, defined as those requiring the missing of games, were identified using publicly available data. Performance was assessed with Pro Football Focus grades for each season. Rates of major knee injury were compared between groups with the 2-sample Z test for proportions, and performance grades were compared with the unpaired t test. Results: For the cumulative study period, bracers demonstrated a significantly lower rate of major knee injuries than nonbracers (0.013 vs 0.049 injuries per player, respectively; P = .04). Isolated MCL injury was the most common injury for nonbracers. There was no group difference in performance for the cumulative study period or during most individual seasons. Yearly prevalence of PKB usage declined steadily from 16.3% in 2014 to 5.6% in 2020. A subgroup analysis of rookie players demonstrated an overall downtrend in usage during the study period as well. Conclusion: Results indicated that knee brace prophylaxis by NFL offensive linemen was associated with a reduced risk of major knee injury without a significant difference in performance when compared with nonbracers. Despite this, the prevalence of PKB declined over the study period.

Mycoplasma DnaK increases DNA copy number variants in vivo.

The human microbiota affects critical cellular functions, although the responsible mechanism(s) is still poorly understood. In this regard, we previously showed that Mycoplasma fermentans DnaK, an HSP70 chaperone protein, hampers the activity of important cellular proteins responsible for DNA integrity. Here, we describe a novel DnaK knock-in mouse model generated in our laboratory to study the effect of M. fermentans DnaK expression in vivo. By using an array-based comparative genomic hybridization assay, we demonstrate that exposure to DnaK was associated with a higher number of DNA copy number variants (CNVs) indicative of unbalanced chromosomal alterations, together with reduced fertility and a high rate of fetal abnormalities. Consistent with their implication in genetic disorders, one of these CNVs caused a homozygous Grid2 deletion, resulting in an aberrant ataxic phenotype that recapitulates the extensive biallelic deletion in the Grid2 gene classified in humans as autosomal recessive spinocerebellar ataxia 18. Our data highlight a connection between components of the human urogenital tract microbiota, namely Mycoplasmas, and genetic abnormalities in the form of DNA CNVs, with obvious relevant medical, diagnostic, and therapeutic implications.

The role of trained immunity in COVID-19: Lessons for the next pandemic.

Trained immunity is a long-term increase in responsiveness of innate immune cells, induced by certain infections and vaccines. During the last 3 years of the COVID-19 pandemic, vaccines that induce trained immunity, such as BCG, MMR, OPV, and others, have been investigated for their capacity to protect against COVID-19. Further, trained immunity-inducing vaccines have been shown to improve B and T cell responsiveness to both mRNA- and adenovirus-based anti-COVID-19 vaccines. Moreover, SARS-CoV-2 infection itself induces inappropriately strong programs of trained immunity in some individuals, which may contribute to the long-term inflammatory sequelae. In this review, we detail these and other aspects of the role of trained immunity in SARS-CoV-2 infection and COVID-19. We also examine the learnings from the trained immunity studies conducted in the context of this pandemic and discuss how they may help us in preparing for future infectious outbreaks.