UNASSIGNED: Given the strikingly high diagnostic error rate in hospitals, and the recent development of Large Language Models (LLMs), we set out to measure the diagnostic sensitivity of two popular LLMs: GPT-4 and PaLM2. Small-scale studies to evaluate the diagnostic ability of LLMs have shown promising results, with GPT-4 demonstrating high accuracy in diagnosing test cases. However, larger evaluations on real electronic patient data are needed to provide more reliable estimates.
UNASSIGNED: To fill this gap in the literature, we used a deidentified Electronic Health Record (EHR) data set of about 300,000 patients admitted to the Beth Israel Deaconess Medical Center in Boston. This data set contained blood, imaging, microbiology and vital sign information as well as the patients\' medical diagnostic codes. Based on the available EHR data, doctors curated a set of diagnoses for each patient, which we will refer to as ground truth diagnoses. We then designed carefully-written prompts to get patient diagnostic predictions from the LLMs and compared this to the ground truth diagnoses in a random sample of 1000 patients.
UNASSIGNED: Based on the proportion of correctly predicted ground truth diagnoses, we estimated the diagnostic hit rate of GPT-4 to be 93.9%. PaLM2 achieved 84.7% on the same data set. On these 1000 randomly selected EHRs, GPT-4 correctly identified 1116 unique diagnoses.
UNASSIGNED: The results suggest that artificial intelligence (AI) has the potential when working alongside clinicians to reduce cognitive errors which lead to hundreds of thousands of misdiagnoses every year. However, human oversight of AI remains essential: LLMs cannot replace clinicians, especially when it comes to human understanding and empathy. Furthermore, a significant number of challenges in incorporating AI into health care exist, including ethical, liability and regulatory barriers.

OBJECTIVE: Acute myocardial infarction (AMI) is a cardiovascular disease with high morbidity and mortality. We collected patients with AMI from the Medical Information Mart for Intensive Care IV (v2.0) database and explored the association between serum albumin-corrected anion gap (ACAG) level and mortality in patients with AMI.
RESULTS: Data of adult patients with AMI were collected. According to the 360 day prognosis, patients were divided into survival and non-survival groups. Based on the ACAG level, patients were then divided into normal and high ACAG groups. Cox hazard proportional models and restricted cubic splines (RCSs) were used to investigate the correlation between ACAG and mortality. Kaplan-Meier curves were created to compare the cumulative survival rates between the high and normal ACAG groups. The receiver operating characteristic (ROC) curve was used to analyse the predictive value of ACAG for the prognosis of patients with AMI. Sensitivity and subgroup analyses were conducted to revalidate the results. Finally, 1783 patients were included. Elevated ACAG (>20 mmol/L) was significantly associated with 30 and 360 day mortality (P < 0.001). Adjusted for multiple confounding factors, the Cox proportional hazard analysis showed that elevated ACAG (>20 mmol/L) was an independent risk factor of increased all-cause mortality in patients with AMI (hazard ratio 1.423, 95% confidence interval 1.206-1.678, P < 0.001). RCS analysis further showed that there was a non-linear trend relationship between ACAG and the risk of all-cause mortality at 30 and 360 days (χ2 = 10.750, P = 0.013; χ2 = 13.960, P = 0.003). Kaplan-Meier survival curves showed that the 30 and 360 day cumulative survival rates of patients with AMI were significantly lower (log-rank test, χ2 = 98.880, P < 0.001; χ2 = 105.440, P < 0.001) in the high ACAG group. ROC curve analysis showed that the area under the curve (AUC) of ACAG was 0.651, while the AUC of anion gap (AG) was 0.609, indicating that ACAG had a higher predictive value for 360 day mortality than AG. When combined with Sequential Organ Failure Assessment score, the predictive performance of ACAG for 360 day mortality was better, with an AUC of 0.699. Sensitivity and subgroup analyses were conducted suggesting the stability of our results.
CONCLUSIONS: Elevated serum ACAG (≥20 mmol/L) is an independent risk factor for short-term and long-term mortality in critically ill patients with AMI, and it may assist clinicians and nurses identifying high-risk patients.

OBJECTIVE: Hypocalcemia occurs commonly among patients with acute pancreatitis (AP) in the intensive care unit (ICU). Calcium therapy could be used to correct hypocalcemia and maintain calcium levels, but its impact on the prognosis has not been demonstrated. Our study aimed to determine whether calcium therapy could benefit the multiple outcomes of AP patients with hypocalcemia.
METHODS: We extracted 807 AP patients with hypocalcemia from the Beth Israel Deaconess Medical Center (MIMIC-IV) database and performed retrospective analyses. The outcomes were in-hospital, 28 days, ICU mortality, and the length of stay (LOS) in the hospital and ICU. We performed propensity matching (PSM) and inverse probability weighting (IPTW) to balance the baseline differences and conducted multivariate regression to investigate the impact of calcium therapy.
RESULTS: A total of 620 patients (76.8%) received calcium treatment (calcium group) during hospitalization, while 187 patients (non-calcium group) did not. Patients in the calcium group did not present significant survival differences between groups before and after matching. After including covariates, calcium administration had no association with patients\' in-hospital (HR: 1.03, 95% Cl: 0.47-2.27, p = .942), 28 days and ICU mortality and was significantly associated with prolonged length of stay in the hospital (effect estimate: 6.18, 95% Cl: 3.27-9.09, p < .001) and ICU (effect estimate: 1.72, 95% Cl: 0.24-3.20, p < .001). Calcium therapy could not benefit patients in subgroups with exclusive parenteral infusion, early calcium therapy (<48 h), or various degrees of hypocalcemia.
CONCLUSIONS: AP patients with hypocalcemia could not benefit from calcium administration, which has no association with multiple mortality and is significantly associated with prolonged LOS in the hospital and ICU.