In the electronic health record, using clinical notes to identify entities such as disorders and their temporality (e.g. the order of an event relative to a time index) can inform many important analyses. However, creating training data for clinical entity tasks is time consuming and sharing labeled data is challenging due to privacy concerns. The information needs of the COVID-19 pandemic highlight the need for agile methods of training machine learning models for clinical notes. We present Trove, a framework for weakly supervised entity classification using medical ontologies and expert-generated rules. Our approach, unlike hand-labeled notes, is easy to share and modify, while offering performance comparable to learning from manually labeled training data. In this work, we validate our framework on six benchmark tasks and demonstrate Trove’s ability to analyze the records of patients visiting the emergency department at Stanford Health Care for COVID-19 presenting symptoms and risk factors.

Comorbid conditions appear to be common among individuals hospitalised with coronavirus disease 2019 (COVID-19) but estimates of prevalence vary and little is known about the prior medication use of patients. Here, we describe the characteristics of adults hospitalised with COVID-19 and compare them with influenza patients. We include 34,128 (US: 8362, South Korea: 7341, Spain: 18,425) COVID-19 patients, summarising between 4811 and 11,643 unique aggregate characteristics. COVID-19 patients have been majority male in the US and Spain, but predominantly female in South Korea. Age profiles vary across data sources. Compared to 84,585 individuals hospitalised with influenza in 2014-19, COVID-19 patients have more typically been male, younger, and with fewer comorbidities and lower medication use. While protecting groups vulnerable to influenza is likely a useful starting point in the response to COVID-19, strategies will likely need to be broadened to reflect the particular characteristics of individuals being hospitalised with COVID-19.

Foundation models are transforming artificial intelligence (AI) in healthcare by providing modular components adaptable for various downstream tasks, making AI development more scalable and cost-effective. Foundation models for structured electronic health records (EHR), trained on coded medical records from millions of patients, demonstrated benefits including increased performance with fewer training labels, and improved robustness to distribution shifts. However, questions remain on the feasibility of sharing these models across hospitals and their performance in local tasks. This multi-center study examined the adaptability of a publicly accessible structured EHR foundation model (FM_SM), trained on 2.57 M patient records from Stanford Medicine. Experiments used EHR data from The Hospital for Sick Children (SickKids) and Medical Information Mart for Intensive Care (MIMIC-IV). We assessed both adaptability via continued pretraining on local data, and task adaptability compared to baselines of locally training models from scratch, including a local foundation model. Evaluations on 8 clinical prediction tasks showed that adapting the off-the-shelf FM_SMmatched the performance of gradient boosting machines (GBM) locally trained on all data while providing a 13% improvement in settings with few task-specific training labels. Continued pretraining on local data showed FM_SMrequired fewer than 1% of training examples to match the fully trained GBM’s performance, and was 60 to 90% more sample-efficient than training local foundation models from scratch. Our findings demonstrate that adapting EHR foundation models across hospitals provides improved prediction performance at less cost, underscoring the utility of base foundation models as modular components to streamline the development of healthcare AI.

Neuronal damage secondary to traumatic brain injury (TBI) is a rapidly evolving condition, which requires therapeutic decisions based on the timely identification of clinical deterioration. Changes in S100B biomarker levels are associated with TBI severity and patient outcome. The S100B quantification is often difficult since standard immunoassays are time-consuming, costly, and require extensive expertise. A zero-length cross-linking approach on a cysteamine self-assembled monolayer (SAM) was performed to immobilize anti-S100B monoclonal antibodies onto both planar (AuEs) and interdigitated (AuIDEs) gold electrodes via carbonyl-bond. Surface characterization was performed by atomic force microscopy (AFM) and specular-reflectance FTIR for each functionalization step. Biosensor response was studied using the change in charge-transfer resistance (Rct) from electrochemical impedance spectroscopy (EIS) in potassium ferrocyanide, with [S100B] ranging 10–1000 pg/mL. A single-frequency analysis for capacitances was also performed in AuIDEs. Full factorial designs were applied to assess biosensor sensitivity, specificity, and limit-of-detection (LOD). Higher Rct values were found with increased S100B concentration in both platforms. LODs were 18 pg/mL(AuES) and 6 pg/mL(AuIDEs). AuIDEs provide a simpler manufacturing protocol, with reduced fabrication time and possibly costs, simpler electrochemical response analysis, and could be used for single-frequency analysis for monitoring capacitance changes related to S100B levels.

Incident depression has been reported to be associated with poor prognosis in patients with cardiovascular disease (CVD), which might be associated with beta-blocker therapy. Because early detection and intervention can alleviate the severity of depression, we aimed to develop a machine learning (ML) model predicting the onset of major depressive disorder (MDD). A model based on L1 regularized logistic regression was trained against the South Korean nationwide administrative claims database to identify risk factors for the incident MDD after beta-blocker therapy in patients with CVD. We identified 50,397 patients initiating beta-blockers for CVD, with 774 patients developing MDD within 365 days after initiating beta-blocker therapy. An area under the receiver operating characteristic curve (AUC) of 0.74 was achieved. A history of non-selective beta-blockers and factors related to anxiety disorder, sleeping problems, and other chronic diseases were the most strong predictors. AUCs of 0.62–0.71 were achieved in the external validation conducted on six independent electronic health records and claims databases in the USA and South Korea. In conclusion, an ML model that identifies patients at high-risk for incident MDD was developed. Application of ML to identify susceptible patients for adverse events of treatment may serve as an important approach for personalized medicine.