Abstract Objectives Early recognition of critical deterioration events (CDEs) in pediatric intensive care units (PICUs) is crucial to improving outcomes, but the real-time performance of machine-learning (ML)-based alert systems is rarely assessed. We prospectively evaluated the performance, operational utility, and deployment feasibility of the PICU Warning Index (P-WIN), a personalized ML alert system predicting CDE in the PICU. Materials and methods In this prospective cohort study at a quaternary-care children’s hospital, P-WIN estimated deterioration risk for PICU patients aged 0-24 years at 30-minute intervals between January 7 and April 7, 2026 using electronic health record data collected via Fast Healthcare Interoperability Resources and proprietary EPIC API endpoints. CDEs included cardiopulmonary resuscitation, initiation of extracorporeal membrane oxygenation, dilute (10 μg/mL) epinephrine bolus, or unplanned endotracheal intubation. P-WIN was compared to baseline workflows: an automated PICU Warning Tool and a clinician-huddle-based assessment. Primary measures were CDE-level sensitivity, admission-level positive predictive value (PPV), and alerts per patient-day. Results Among 1005 PICU patients, 99 CDEs were experienced by 62 patients. P-WIN detected 73% of these CDE 1-24 hours in advance, outperforming the PICU Warning Tool (39%, P < .05) and huddle workflow (23%, P < .05). P-WIN generated 0.21 (95% CI, 0.18-0.23) alerts per patient-day. The PICU Warning Tool and huddle workflow had similar PPV to P-WIN. Conclusion P-WIN detected significantly more CDEs than existing PICU Warning tool and huddle workflows while maintaining a similar PPV. Our findings underscore the feasibility of ML-based personalized warning systems to enhance shared situational awareness of PICU deterioration.
Abstract Purpose Digital mental health interventions could improve population health. However, our understanding of how to improve their effectiveness through personalization is limited. We describe the outcomes of a micro-randomized trial (MRT) of a 60-day text messaging program, delivered during the final stages of the COVID-19 pandemic. We analyze the effects of 6 different types of cognitive behavioral therapy (CBT) and dialectical behavioral therapy (DBT) messages, and examine whether socio-demographic, health-related, and contextual factors (eg, timing of sending a message) moderate effectiveness. Methods This secondary analysis included 373 adults aged 18 years or older (mean = 38.07), mostly English-speaking female, highly educated White individuals who reported being in good health, but having high stress and struggling to get their basic needs met. Participants were recruited between December 2021 and September 2022 through social media and a recruitment service (Health Match). We used the weighted and centered least squares method to examine the proximal effects of the text-message categories on daily mood, and whether sociodemographic, health-related factors, and the timing of message delivery moderated effectiveness. Results None of the 6 CBT/DBT-message categories significantly improved mood ratings marginally. However, their effectiveness varied based on the time of day and self-reported health. Overall, sending a message in the afternoon was more effective. However, certain message categories showed significant effects when sent in the morning among participants with poorer self-reported health, while some showed lower effectiveness when sent in the late afternoon for participants with better health. Conclusion Our analysis underscores the importance of considering the time of day and individual health status when developing digital mental health interventions. These insights can help in designing more personalized, adaptive, and effective digital interventions.
Abstract Objective To characterize patients’ perspectives on privacy on the disclosure, documentation, and visibility of social risk information collected during screening in healthcare settings, and to identify ways to design patient-centered, privacy-protecting screening and documentation systems. Materials and Methods Interviews were conducted with 18 patients from a Midwestern healthcare system who had experienced screening. Interviews were analyzed using a multi-cycle coding process. Results Participants expressed privacy concerns regarding the disclosure, storage, and visibility of social risk information. Patients judged the relevance of social risk information when choosing to disclose and when considering how long they would like these data to remain in the EHR. Disclosure and storage risks identified included embarrassment, the potential for these data to be shared with others or used against patients, and the possibility that patients would be unaware of stored social risk data in the electronic health record (EHR). Patients desired trusting relationships with physicians to facilitate disclosure and the ability to edit or delete stored data to ensure privacy. Discussion Findings highlight privacy-related concerns about social risk disclosure, as well as patients’ desire for control over these data. Trusting patient-clinician relationships and clear policies governing social risk data use may support disclosure. Patients advocated for screening to occur with trusted healthcare staff and control over EHR social risk data. Conclusion Social risk screening and documentation systems must address patients’ privacy concerns, facilitate safe disclosure, and clarify how data are used. Empowering patients with greater control over social risk data may affirm privacy and improve acceptability.
Abstract Objective Expert clinician chart review is the most accurate method to identify patients with infection but is laborious and costly. We aimed to develop and externally validate machine-learning models using structured and unstructured data to identify infection in hospitalized patients compared to chart review. Materials and Methods Adult inpatient encounters at 2 health systems (2006-2016) were manually annotated for suspected infection on admission. Data from one site were used to develop gradient-boosted machine-learning models. Structured features included labs, vital signs, and nursing assessments. Unstructured clinical notes were converted to concept unique identifiers (CUIs) using the clinical Text Analysis and Knowledge Extraction System. Unimodal and multimodal models were tuned to maximize the area under the receiver operating curve (AUROC) in the derivation site through cross-validation. Model discrimination and calibration were assessed in the external validation site. Results Of the 3014 included admissions, 677 (22.5%) were classified as having suspected infection on admission. A multimodal model combining n-gram TF-IDF and structured data demonstrated the best performance, significantly outperforming the structured data only model (AUROC 0.88 vs 0.82, P < .001, area under the precision-recall curve 0.71 vs 0.58, P < .001). The most important features came from unstructured data. Discussion and Conclusion Our results highlight the value of unstructured text data for developing models that accurately replicate chart reviews in identifying patients with infection. By expediting the annotation of infection cases, our models can facilitate epidemiological research and the curation of datasets necessary for developing clinical decision support systems for early sepsis detection.
Abstract Objectives This article summarizes the architecture, design principles, and implementation strategy of the ATRI Laboratory Information Management System (LIMS), a cloud‑native, open‑source platform supporting biomarker‑driven Alzheimer’s disease (AD) clinical research. The system addresses challenges associated with managing large‑scale, multimodal biomarker, and biobanking data across multisite studies. Methods The ATRI LIMS employs a Model–View–Controller architecture implemented in Django, with Single‑Page Application interfaces built in ReactJS. A hybrid database model integrates a normalized core schema with study‑specific extensions. Deployment on Amazon Web Services incorporates multifactor authentication, encryption, and a web application firewall. Regulatory support is provided through audit trails and electronic signatures. Development relied on collaboration between informatics engineers and laboratory personnel, with user acceptance testing guiding workflow and interface refinement. Results As of February 2026, the ATRI LIMS supports seven clinical trials and observational studies and manages more than 1.2 million biospecimens. The platform synchronizes biospecimen metadata and assay outputs with clinical datasets, improves operational efficiency through automated workflows and quality control, and facilitates compliant sample sharing for secondary research. AI‑assisted development accelerated feature delivery while maintaining code quality. Discussion The ATRI LIMS offers a scalable, adaptable informatics framework for biomarker‑intensive AD research. Its modular design, cloud‑native architecture, and comprehensive data management capabilities support accurate specimen tracking and coordinated data integration. An open-source release is planned to enable adoption and adaptation by other research programs. Conclusions The ATRI LIMS enhances informatics infrastructure for modern AD research by enabling scalable, efficient, and compliant biomarker and biospecimen workflow management.
