Background While artificial intelligence (AI) offers possibilities of advanced clinical prediction and decision-making in healthcare, models trained on relatively homogeneous datasets, and populations poorly-representative of underlying diversity, limits generalisability and risks biased AI-based decisions. Here, we describe the landscape of AI in clinical medicine to delineate population and data-source disparities. Methods We performed a scoping review of clinical papers published in PubMed in 2019 using AI techniques. We assessed differences in dataset country source, clinical specialty, and author nationality, sex, and expertise. A manually tagged subsample of PubMed articles was used to train a model, leveraging transfer-learning techniques (building upon an existing BioBERT model) to predict eligibility for inclusion (original, human, clinical AI literature). Of all eligible articles, database country source and clinical specialty were manually labelled. A BioBERT-based model predicted first/last author expertise. Author nationality was determined using corresponding affiliated institution information using Entrez Direct. And first/last author sex was evaluated using the Gendarize.io API. Results Our search yielded 30,576 articles, of which 7,314 (23.9%) were eligible for further analysis. Most databases came from the US (40.8%) and China (13.7%). Radiology was the most represented clinical specialty (40.4%), followed by pathology (9.1%). Authors were primarily from either China (24.0%) or the US (18.4%). First and last authors were predominately data experts (i.e., statisticians) (59.6% and 53.9% respectively) rather than clinicians. And the majority of first/last authors were male (74.1%). Interpretation U.S. and Chinese datasets and authors were disproportionately overrepresented in clinical AI, and almost all of the top 10 databases and author nationalities were from high income countries (HICs). AI techniques were most commonly employed for image-rich specialties, and authors were predominantly male, with non-clinical backgrounds. Development of technological infrastructure in data-poor regions, and diligence in external validation and model re-calibration prior to clinical implementation in the short-term, are crucial in ensuring clinical AI is meaningful for broader populations, and to avoid perpetuating global health inequity.

No abstract

The reason for the striking differences in clinical outcomes of SARS-CoV-2 infected patients is still poorly understood. While most recover, a subset of people become critically ill and succumb to the disease. Thus, identification of biomarkers that can predict the clinical outcomes of COVID-19 disease is key to help prioritize patients needing urgent treatment. Given that an unbalanced gut microbiome is a reflection of poor health, we aim to identify indicator species that could predict COVID-19 disease clinical outcomes. Here, for the first time and with the largest COVID-19 patient cohort reported for microbiome studies, we demonstrated that the intestinal and oral microbiome make-up predicts respectively with 92% and 84% accuracy (Area Under the Curve or AUC) severe COVID-19 respiratory symptoms that lead to death. The accuracy of the microbiome prediction of COVID-19 severity was found to be far superior to that from training similar models using information from comorbidities often adopted to triage patients in the clinic (77% AUC). Additionally, by combining symptoms, comorbidities, and the intestinal microbiota the model reached the highest AUC at 96%. Remarkably the model training on the stool microbiome found enrichment of Enterococcus faecalis, a known pathobiont, as the top predictor of COVID-19 disease severity. Enterococcus faecalis is already easily cultivable in clinical laboratories, as such we urge the medical community to include this bacterium as a robust predictor of COVID-19 severity when assessing risk stratification of patients in the clinic.

The Diagnostic Rating Scale (DRS) was completed by the parents and teachers of 82 children referred for clinical evaluations, 73 referred children seen twice, and 218 non-referred children from the community. The DRS, which uses a categorical rather than a dimensional rating approach, was 70% to 90% sensitive to diagnoses of Attention Deficit/Hyperactivity Disorder (ADHD) made by blind clinical teams. In research and clinical applications, the DRS could improve screening efficiency, especially in situations where it would be desirable to exclude all children who might have ADHD or identify all children with Hyperactive-Impulsive symptoms. Because of its objectivity and consistency with the Diagnostic and Statistical Manual (DSM)-IV criteria, the DRS could facilitate comparison of participant samples across studies.

Background Recurrent hypoglycemia blunts counter-regulatory responses to subsequent hypoglycemic episodes, a syndrome known as hypoglycemia-associated autonomic failure (HAAF). Since adrenergic receptor blockade has been reported to prevent HAAF, we investigated whether the hypoglycemia-associated rise in plasma epinephrine contributes to pathophysiology and reported interindividual differences in susceptibility to HAAF. Methods To assess the role of hypoglycemia-associated epinephrine responses in the susceptibility to HAAF, 24 adult nondiabetic subjects underwent two 2-hour hyperinsulinemic hypoglycemic clamp studies (nadir 54 mg/dL; 0-2 hours and 4-6 hours) on Day 1, followed by a third identical clamp on Day 2. We challenged an additional 7 subjects with two 2-hour infusions of epinephrine (0.03 μg/kg/min; 0-2 hours and 4-6 hours) vs saline on Day 1 followed by a 200-minute stepped hypoglycemic clamp (90, 80, 70, and 60 mg/dL) on Day 2. Results Thirteen out of 24 subjects developed HAAF, defined by ≥20% reduction in average epinephrine levels during the final 30 minutes of the third compared with the first hypoglycemic episode (P < 0.001). Average epinephrine levels during the final 30 minutes of the first hypoglycemic episode were 2.3 times higher in subjects who developed HAAF compared with those who did not (P = 0.006). Compared to saline, epinephrine infusion on Day 1 reduced the epinephrine responses by 27% at the 70 and 60 mg/dL glucose steps combined (P = 0.04), with a parallel reduction in hypoglycemic symptoms (P = 0.03) on Day 2. Conclusions Increases in plasma epinephrine reproduce key features of HAAF in nondiabetic subjects. Marked interindividual variability in epinephrine responses to hypoglycemia may explain an individual’s susceptibility to developing HAAF.