Background: Whether GPT-4, the conversational artificial intelligence, can accurately diagnose and triage health conditions and whether it presents racial and ethnic biases in its decisions remain unclear. Objective: We aim to assess the accuracy of GPT-4 in the diagnosis and triage of health conditions and whether its performance varies by patient race and ethnicity. Methods: We compared the performance of GPT-4 and physicians, using 45 typical clinical vignettes, each with a correct diagnosis and triage level, in February and March 2023. For each of the 45 clinical vignettes, GPT-4 and 3 board-certified physicians provided the most likely primary diagnosis and triage level (emergency, nonemergency, or self-care). Independent reviewers evaluated the diagnoses as “correct” or “incorrect.” Physician diagnosis was defined as the consensus of the 3 physicians. We evaluated whether the performance of GPT-4 varies by patient race and ethnicity, by adding the information on patient race and ethnicity to the clinical vignettes. Results: The accuracy of diagnosis was comparable between GPT-4 and physicians (the percentage of correct diagnosis was 97.8% (44/45; 95% CI 88.2%-99.9%) for GPT-4 and 91.1% (41/45; 95% CI 78.8%-97.5%) for physicians; P=.38). GPT-4 provided appropriate reasoning for 97.8% (44/45) of the vignettes. The appropriateness of triage was comparable between GPT-4 and physicians (GPT-4: 30/45, 66.7%; 95% CI 51.0%-80.0%; physicians: 30/45, 66.7%; 95% CI 51.0%-80.0%; P=.99). The performance of GPT-4 in diagnosing health conditions did not vary among different races and ethnicities (Black, White, Asian, and Hispanic), with an accuracy of 100% (95% CI 78.2%-100%). P values, compared to the GPT-4 output without incorporating race and ethnicity information, were all .99. The accuracy of triage was not significantly different even if patients’ race and ethnicity information was added. The accuracy of triage was 62.2% (95% CI 46.5%-76.2%; P=.50) for Black patients; 66.7% (95% CI 51.0%-80.0%; P=.99) for White patients; 66.7% (95% CI 51.0%-80.0%; P=.99) for Asian patients, and 62.2% (95% CI 46.5%-76.2%; P=.69) for Hispanic patients. P values were calculated by comparing the outputs with and without conditioning on race and ethnicity. Conclusions: GPT-4’s ability to diagnose and triage typical clinical vignettes was comparable to that of board-certified physicians. The performance of GPT-4 did not vary by patient race and ethnicity. These findings should be informative for health systems looking to introduce conversational artificial intelligence to improve the efficiency of patient diagnosis and triage.

Background: The Japan Coma Scale (JCS) is the most frequently adopted method for evaluating level of consciousness in Japan. However, no validated method for converting the JCS to Glasgow Coma Scale (GCS) exists. The aims of the present study were to develop and validate a method to convert the JCS to GCS.Methods: This is a multicenter retrospective cohort study involving three emergency departments (EDs) in Japan. We included all adult patients who visited the ED between 2017 and 2020. The participating facilities were divided into two cohorts—one cohort to develop a table to convert the JCS to GCS (development cohort), and the other cohort to validate the conversion table (validation cohort). The conversion table of the JCS to GCS was developed based on the median values of the GCS. The outcome was the concordance rate between the JCS and GCS.Results: We identified 8,194 eligible patients. The development cohort included 7,373 patients and the validation cohort included 821 patients. In the validation cohort, the absolute and relative concordance rates were 80.3% (95% confidence interval, 77.4–82.9%) and 93.2% (95% confidence interval, 91.2–94.8%), respectively.Conclusions: This study developed and validated a novel method for converting the JCS to GCS. Assuming the offset by a single category between the JCS and GCS is acceptable, the concordance rate was over 90% in the general adult patient population visiting the ED. The conversion method may assist researchers to convert JCS into GCS, which is more commonly recognized among global audiences.

Background: Vitamin C is a well-documented antioxidant that reduces oxidative stress and fluid infusion in high doses; however, the association between high-dose vitamin C and reduced mortality remains unclear. This study evaluates the effect of high-dose vitamin C in severe burn patients under two varying thresholds. Methods: We enrolled adult patients with severe burns (burn index ≥ 15) who were registered in the Japanese Diagnosis Procedure Combination national inpatient database from 2010 to 2016. Propensity score matching was performed between patients who received high-dose vitamin C within 1 day of admission (vitamin C group) and those who did not (control group). High-dose vitamin C was defined as a dosage in excess of 10 g or 24 g within 2 days of admission. The primary outcome was in-hospital mortality. Results: Eligible patients (n = 2713) were categorized into the vitamin C group (n = 157) or control group (n = 2556). After 1:4 propensity score matching, we compared 157 and 628 patients who were administered high-dose vitamin C (> 10-g threshold) and controls, respectively. Under this particular threshold, high-dose vitamin C therapy was associated with reduced in-hospital mortality (risk ratio, 0.79; 95% confidence interval, 0.66–0.95; p = 0.006). In contrast, in-hospital mortality did not differ between the control and high-dose vitamin C group under the > 24-g threshold (risk ratio, 0.83; 95% confidence interval, 0.68–1.02; p = 0.068). Conclusions: High-dose vitamin C therapy was associated with reduced mortality in patients with severe burns when used under a minimum threshold of 10 g within the first 2 days of admission. While “high-dose” vitamin C therapy lacks a universal definition, the present study reveals that different “high-dose” regimens may yield improved outcomes.

Background The lactate albumin ratio (LAR), a simple prognostic marker used in intensive care units (ICUs), combines lactate and serum albumin levels to predict patient outcomes. Despite its potential, the predictive accuracy of the LAR remains insufficiently explored. This study aimed to assess the usefulness of the LAR in predicting mortality among patients in the ICU. Methods This retrospective study conducted a secondary analysis of prospectively obtained clinical data from the Japanese Intensive Care Patient Database. We included all patients admitted to ICUs between 2015 and 2021, excluding those under the age of 16 years. The main outcome was in-hospital mortality. The LAR predictive value for this outcome was assessed by examining the area under the receiver operating characteristic curve and comparing it against prognostic indicators such as age, lactate, albumin and Sequential Organ Failure Assessment score. LAR shape was assessed using unrestricted spline curves, and the optimal cut-off value was identified from sensitivity and negative likelihood ratio. Subgroup analysis was used to evaluate the predictive accuracy of the LAR across different patient attributes and clinical scenarios. Results Of 2 34 774 cases analysed, in-hospital mortality was 8.8% (20 723 deaths). The LAR had an area under the curve of 0.761 (95%CI 0.757 to 0.765), indicating a fair predictive performance for in-hospital mortality. Unrestricted spline curves demonstrated that LAR can predict mortality through a monotonic positive dose–response relationship with 0.4 as the optimal cut-off value. In subgroup analysis, areas under the curve were significantly higher in subgroups defined by younger age, female sex, unplanned ICU admission, non-surgical patients, non-infectious patients, non-heart failure patients and lack of end-stage renal disease. Conclusions The LAR might be a useful predictor for screening mortality in ICU patients. However, further research to establish appropriate cut-off values for the LAR and identify the optimal target population is warranted.