Recent studies indicate that Generative Pre-trained Transformer 4 with Vision (GPT-4V) outperforms human physicians in medical challenge tasks. However, these evaluations primarily focused on the accuracy of multi-choice questions alone. Our study extends the current scope by conducting a comprehensive analysis of GPT-4V’s rationales of image comprehension, recall of medical knowledge, and step-by-step multimodal reasoning when solving New England Journal of Medicine (NEJM) Image Challenges—an imaging quiz designed to test the knowledge and diagnostic capabilities of medical professionals. Evaluation results confirmed that GPT-4V performs comparatively to human physicians regarding multi-choice accuracy (81.6% vs. 77.8%). GPT-4V also performs well in cases where physicians incorrectly answer, with over 78% accuracy. However, we discovered that GPT-4V frequently presents flawed rationales in cases where it makes the correct final choices (35.5%), most prominent in image comprehension (27.2%). Regardless of GPT-4V’s high accuracy in multi-choice questions, our findings emphasize the necessity for further in-depth evaluations of its rationales before integrating such multimodal AI models into clinical workflows.

The B.1.1.529 (omicron) variant of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has a shorter incubation period and a higher transmission rate than previous variants.^1,2 Recently, the Centers for Disease Control and Prevention recommended shortening the strict isolation period for infected persons in non–health care settings from 10 days to 5 days after symptom onset or after the initial positive test, followed by 5 days of masking.³ However, the viral decay kinetics of the omicron variant and the duration of shedding of culturable virus have not been well characterized.

The accurate detection of SARS-CoV-2 through respiratory sampling is critical for the prevention of further transmission and the timely initiation of treatment for COVID-19. There is a diverse range of SARS-CoV-2 detection rates in reported studies, with uncertainty as to the optimal sampling strategy for COVID-19 diagnosis and monitoring. We performed a systematic review and meta-analysis of studies comparing respiratory sampling strategies for the detection of SARS-CoV-2 RNA. The inclusion criteria were studies that assessed at least two respiratory sampling sites (oropharyngeal swab, nasopharyngeal swab, and sputum) in participants with COVID-19. The percentage positive tests were compared between sampling modalities by constructing a Z-test assuming independence and using the standard errors obtained from the random effects meta-analysis. From 1039 total studies, we identified 11 studies that met our inclusion criteria, with SARS-CoV-2 testing results from a total of 3442 respiratory tract specimens. Compared to nasopharyngeal swab sampling, sputum testing resulted in significantly higher rates of SARS-CoV-2 RNA detection while oropharyngeal swab testing had lower rates of viral RNA detection. Earlier sampling after symptom onset was associated with improved detection rates, but the differences in SARS-CoV-2 RNA detection by sampling method was consistent regardless of the duration of symptoms. The results support sputum sampling as a valuable method of COVID-19 diagnosis and monitoring, and highlight the importance of early testing after symptom onset to increase the rates of COVID-19 diagnosis. This study was funded in part by the NIH grants U01AI106701 and by the Harvard University for AIDS Research (NIAID 5P30AI060354).

BACKGROUND. SARS-CoV-2 plasma viremia has been associated with severe disease and death in COVID-19 in small-scale cohort studies. The mechanisms behind this association remain elusive. METHODS. We evaluated the relationship between SARS-CoV-2 viremia, disease outcome, and inflammatory and proteomic profiles in a cohort of COVID-19 emergency department participants. SARS-CoV-2 viral load was measured using a quantitative reverse transcription PCR–based platform. Proteomic data were generated with Proximity Extension Assay using the Olink platform. RESULTS. This study included 300 participants with nucleic acid test–confirmed COVID-19. Plasma SARS-CoV-2 viremia levels at the time of presentation predicted adverse disease outcomes, with an adjusted OR of 10.6 (95% CI 4.4–25.5, P < 0.001) for severe disease (mechanical ventilation and/or 28-day mortality) and 3.9 (95% CI 1.5–10.1, P = 0.006) for 28-day mortality. Proteomic analyses revealed prominent proteomic pathways associated with SARS-CoV-2 viremia, including upregulation of SARS-CoV-2 entry factors (ACE2, CTSL, FURIN), heightened markers of tissue damage to the lungs, gastrointestinal tract, and endothelium/vasculature, and alterations in coagulation pathways. CONCLUSION. These results highlight the cascade of vascular and tissue damage associated with SARS-CoV-2 plasma viremia that underlies its ability to predict COVID-19 disease outcomes. FUNDING. Mark and Lisa Schwartz; the National Institutes of Health (U19AI082630); the American Lung Association; the Executive Committee on Research at Massachusetts General Hospital; the Chan Zuckerberg Initiative; Arthur, Sandra, and Sarah Irving for the David P. Ryan, MD, Endowed Chair in Cancer Research; an EMBO Long-Term Fellowship (ALTF 486-2018); a Cancer Research Institute/Bristol Myers Squibb Fellowship (CRI2993); the Harvard Catalyst/Harvard Clinical and Translational Science Center (National Center for Advancing Translational Sciences, NIH awards UL1TR001102 and UL1TR002541-01); and by the Harvard University Center for AIDS Research (National Institute of Allergy and Infectious Diseases, 5P30AI060354).

Despite vaccination and antiviral therapies, immunocompromised individuals are at risk for prolonged severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infection, but the immune defects that predispose an individual to persistent coronavirus disease 2019 (COVID-19) remain incompletely understood. In this study, we performed detailed viro-immunologic analyses of a prospective cohort of participants with COVID-19. The median times to nasal viral RNA and culture clearance in individuals with severe immunosuppression due to hematologic malignancy or transplant (S-HT) were 72 and 40 days, respectively, both of which were significantly longer than clearance rates in individuals with severe immunosuppression due to autoimmunity or B cell deficiency (S-A), individuals with nonsevere immunodeficiency, and nonimmunocompromised groups ( P < 0.01). Participants who were severely immunocompromised had greater SARS-CoV-2 evolution and a higher risk of developing resistance against therapeutic monoclonal antibodies. Both S-HT and S-A participants had diminished SARS-CoV-2–specific humoral responses, whereas only the S-HT group had reduced T cell–mediated responses. This highlights the varied risk of persistent COVID-19 across distinct immunosuppressive conditions and suggests that suppression of both B and T cell responses results in the highest contributing risk of persistent infection.

Data are conflicting regarding an association between treatment of acute COVID-19 with nirmatrelvir-ritonavir (N-R) and virologic rebound (VR). To compare the frequency of VR in patients with and without N-R treatment for acute COVID-19. Observational cohort study. Multicenter health care system in Boston, Massachusetts. Ambulatory adults with acute COVID-19 with and without use of N-R. Receipt of 5 days of N-R treatment versus no COVID-19 therapy. The primary outcome was VR, defined as either a positive SARS-CoV-2 viral culture result after a prior negative result or 2 consecutive viral loads above 4.0 log₁₀ copies/mL that were also at least 1.0 log₁₀ copies/mL higher than a prior viral load below 4.0 log₁₀ copies/mL. Compared with untreated persons (n = 55), those taking N-R (n = 72) were older, received more COVID-19 vaccinations, and more commonly had immunosuppression. Fifteen participants (20.8%) taking N-R had VR versus 1 (1.8%) who was untreated (absolute difference, 19.0 percentage points [95% CI, 9.0 to 29.0 percentage points]; P = 0.001). All persons with VR had a positive viral culture result after a prior negative result. In multivariable models, only N-R use was associated with VR (adjusted odds ratio, 10.02 [CI, 1.13 to 88.74]; P = 0.038). Virologic rebound was more common among those who started therapy within 2 days of symptom onset (26.3%) than among those who started 2 or more days after symptom onset (0%) (P = 0.030). Among participants receiving N-R, those who had VR had prolonged shedding of replication-competent virus compared with those who did not have VR (median, 14 vs. 3 days). Eight of 16 participants (50% [CI, 25% to 75%]) with VR also reported symptom rebound; 2 were completely asymptomatic. No post-VR resistance mutations were detected. Observational study design with differences between the treated and untreated groups; positive viral culture result was used as a surrogate marker for risk for ongoing viral transmission. Virologic rebound occurred in approximately 1 in 5 people taking N-R, often without symptom rebound, and was associated with shedding of replication-competent virus. National Institutes of Health.