Jake Hayward

Background: Clinical documentation is a major contributor to physician workload and burnout, particularly in high-acuity settings such as emergency departments. Ambient AI scribes have emerged as a response, but commercial products cost US $99 to more than US $600 per physician per month, operate as opaque systems, and often process data on vendor-controlled infrastructure. These characteristics limit organizational control over data sovereignty, quality improvement, and clinical workflows and create barriers to adoption in resource-constrained settings. Objective: We aimed to develop an open-source, modular AI scribe platform that could be deployed entirely within institutional infrastructure, and to evaluate its real-world use across a provincial health system. Methods: We developed Berta, an open-source, modular platform that pairs a Next.js front end and a FastAPI back end with configurable automatic speech recognition and large language model components. A closed-source implementation developed in parallel with Berta was deployed at Alberta Health Services (AHS), Canada’s largest provincially integrated health system at the time of implementation, and operated within AHS’s existing Snowflake AI Data Cloud infrastructure. All clinical audio and text remained within the secure AHS environment. Use and session metadata were automatically recorded during routine clinical use by emergency physicians across AHS between November 2024 and July 2025, and physicians were emailed an optional survey. Operating costs were estimated from Snowflake service consumption over the same period. Implementation (Results): During the 9-month study period, 197 emergency physicians registered at 105 urban and rural facilities used the system, generating 21,548 clinical sessions and more than 2700 hours of audio. Monthly session volume grew from 675 sessions in November 2024 to 5463 in July 2025. Mean recorded session length was 7.6 minutes, and 42% (82/197) of users customized at least 1 documentation template. Operating costs averaged less than US $30 per physician per month, representing a 70% to 95% reduction in cost compared with commercial subscription prices; this comparison excludes personnel costs for development and maintenance. Survey respondents (69-73 per item) rated user-friendliness as good (median 5, IQR 4-6), reported a median time saving of 16 to 30 (IQR 0-15 to 46-60) minutes per shift, and rated documentation quality the lowest. After the pilot, access was opened to other departments, and by September 30, 2026, approximately 2000 users had registered. Conclusions: An in-house AI scribe can be deployed at provincial scale within existing health system infrastructure at a fraction of the operating cost of commercial alternatives while preserving data sovereignty and institutional control. The full cost of ownership and documentation quality require further evaluation. By releasing Berta publicly, we provide a reproducible foundation that health systems can adapt to evaluate and deploy AI documentation technology within their own secure environments.

Background and Purpose: Walking capacity is a common focus of stroke rehabilitation; however, the extent to which walking capacity is associated with daily movement behaviours (primary lying, sedentary time, standing and stepping) during different phases of stroke recovery remains unclear. This study examined the relationship between walking capacity and daily movement behaviours among individuals in the acute and subacute phases of ischemic stroke recovery. Methods: In this cross‐sectional study, adults with ischemic stroke who were ambulatory (able to walk ≥ 5 m with or without a gait aid) were recruited during the acute (≤ 7 days post‐stroke; n = 19) and subacute (2 weeks‐3 months post‐stroke; n = 25) phases of stroke recovery. Movement behaviours were assessed using 24‐h activPAL accelerometry, including primary lying time, sedentary time, standing time, stepping time and daily step count. Walking capacity was assessed using the 6‐Minute Walk Test, 10‐m Walk Test (gait speed) and Timed Up and Go. Between‐group differences were analysed using Welch’s t ‐tests or Mann‐Whitney U tests, as appropriate, and associations were examined using Spearman correlations and exploratory partial correlations adjusted for stroke severity (NIHSS). Results: The subacute group demonstrated faster gait speed than the acute group (0.89 vs. 0.71 m/s; p = 0.03), and shorter Timed Up and Go times (median 15.9 vs. 19.0 seconds; p = 0.03), whereas the 6‐Minute Walk Test and daily movement behaviours were similar. Both groups accumulated high sedentary time (∼12 hours/day). Gait speed was the only walking capacity measure associated with daily movement behaviours. In exploratory analyses adjusted for stroke severity, gait speed remained associated with sedentary time in both groups, whereas associations with stepping and standing time were attenuated in the acute group. Conclusions: In this ambulatory sample of adults in the acute and subacute phases of ischemic stroke recovery, participants accumulated approximately 12 hours of sedentary time per day despite walking speeds approaching thresholds associated with community ambulation. These findings are consistent with a potential capacity‐performance gap and highlight the importance of assessing daily movement behaviours alongside standardized walking capacity measures.

Workplace mental health issues are a prevalent concern. A potential solution is remote health monitoring. In this scoping review, we map evidence on artificial intelligence (AI) for workplace mental health monitoring. MEDLINE, PsycINFO, Scopus, Embase, and Web of Science were searched until March 2024. Studies that evaluated AI for monitoring of a common workplace mental health concern in an actual workplace setting were included. Thirteen studies were included; most (8/13, 62%) were conducted in an office environment, and 4/13 (31%) were in healthcare. Most studies monitored stress (10/13, 77%). The most common method was wearable sensors (8/13, 62%). Studies monitored a wide variety of parameters; the most common were physical activity (7/13, 54%), heart rate or heart rate variability (7/13, 54%). A wide variety of participant-report surveys, including validated (7/13, 54%) and non-validated (6/13, 46%) surveys, were used to measure mental health outcomes. Studies employed diverse AI methods, including supervised, semi-supervised and other (Markov chain and Monte Carlo). The most common AI models used were support vector machines (5/13, 38%), decision trees (5/13, 38%), naive Bayes (4/13, 31%), and random forest (4/13, 38%). Model accuracy ranged from 62%-99%. Future work should identify standard metrics to enable cross-comparability and use larger, high-quality datasets for enriched training, advanced optimization techniques, and careful model tuning.

BACKGROUND Clinical documentation is a major contributor to physician workload and burnout, particularly in high-acuity settings such as emergency departments. Ambient artificial intelligence (AI) scribes have emerged as a response, but commercial products cost US $99 to $600 per physician per month, operate as opaque systems, and do not return data to institutional infrastructure. These characteristics limit organizational control over data governance, quality improvement, and clinical workflows, and create barriers to adoption in resource-constrained settings OBJECTIVE To develop an open-source, modular AI scribe platform deployable entirely within institutional infrastructure, and to evaluate its real-world use across a provincial health system. METHODS We developed Berta, an open-source modular platform that pairs a Next.js front-end and a FastAPI backend with configurable automatic speech recognition and large language model components. A customized implementation was deployed at Alberta Health Services (AHS), Canada’s largest provincially integrated health system at the time of implementation, integrated with AHS’s existing Snowflake AI Data Cloud infrastructure. All clinical audio and text remained within the secure AHS environment. Usage and session metadata were automatically recorded during routine clinical use by emergency physicians across AHS between November 2024 and July 2025. Operating costs were estimated from Snowflake service consumption over the same period. RESULTS During 8 months, 198 emergency physicians used the system in 105 urban and rural facilities, generating 22,148 clinical sessions and more than 2800 hours of audio. Monthly volume grew from 680 sessions in November 2024 to 5530 in July 2025. Mean recorded session length was 7.6 minutes, and 42% of users customized at least one documentation template. Operating costs averaged less than US $30 per physician per month, a 70% to 95% reduction compared with commercial alternatives. Based on pilot outcomes, AHS has approved expansion to all provincial acute care physicians CONCLUSIONS An open-source AI scribe can be deployed at provincial scale within existing health system infrastructure at a fraction of the cost of commercial alternatives, while preserving data sovereignty and institutional control. By releasing Berta publicly, we provide a reproducible foundation deploy AI documentation technology within their own secure environments.

Emergency department (ED) opioid prescribing has been implicated in the opioid crisis, yet Canadian ED prescribing patterns and related harms remain poorly

Background: CT scans are the gold-standard diagnostic test for pulmonary embolisms (PE). Despite stable PE prevalence, CT use is rising in emergency departments (EDs), suggesting test overuse. Current methods for measuring test yield are error-prone or not scalable, thus we tested the accuracy of an open-source, foundational large language model (LLM) for identifying PEs from free-text radiology reports. Methods: Our retrospective diagnostic accuracy study used 10,173 CT-PE reports from 216 radiologists at 38 EDs across Alberta, Canada from April 2021- April 2023. Reports were classified as ‘PE present’, ‘PE absent’, or ‘Indeterminate’ by human labelers. An LLM (LLAMA-2-70B) was then prompt-engineered to label the reports. Label accuracy was compared against ICD-10-CA codes and a rule-based natural language processing (NLP) algorithm (ChartExtract; University of Toronto). Descriptive statistics were performed to analyze results. Results: 1070 (11.8%) reports were PE positive. The LLM achieved an Area Under the Curve (AUC) of 99.1%, outperforming both ICD-10-CA (AUC=90.6%) and ChartExtract (AUC=86.5%), while demonstrating a 16-25% higher sensitivity (LLM: sensitivity=98.8%, specificity=99.1%; ICD-10-CA: sensitivity=82.0%, specificity=99.1%; ChartExtract: sensitivity=73.6%, specificity=99.4%). The LLM took an average of 143 milliseconds to label each report and produce a paragraph justifying the classification. Conclusions: Open-source, foundational LLMs are an accurate and scalable method for interpreting radiology reports and identifying PEs from ED data. If IT resources are available, this is a cost-effective approach to quality metric derivation for diagnostic processes in large health systems.

Background: The exponential growth of publications regarding the application of machine learning (ML) tools in medicine highlights the significant potential for ML to revolutionize the field. Despite the multitude of literature surrounding this topic, there are limited publications addressing the implementation and feasibility of ML models in clinical practice. Currently, Machine Learning Operations (MLOps), a set of practices designed to deploy and maintain ML models in production, is used in various information technology and industrial settings. However, the MLOps pipeline is not well researched in medical settings, where multiple barriers exist to implementing ML pipelines into practice. Objective: This study aims to detail how MLOps is implemented in health care and propose a maturity framework for the health care implementations. Methods: A scoping review search was conducted according to the Joanna Briggs Institute Manual for Evidence Synthesis. Results were synthesized using the 3-stage basic qualitative content analysis. We searched 4 databases (eg, MEDLINE, Embase, Web of Science, and Scopus) to include any studies that involved proof of concept or real-world implementation of MLOps in health care. Studies not reported in English were excluded. Results: A total of 19 studies were included in this scoping review. The MLOps workflow identified within the studies included (1) data extraction (19/19 studies), (2) data preparation and engineering (18/19 studies), (3) model training (19/19 studies), (4) measured ML metrics and model evaluation (17/19 studies), (5) model validation and test in production (14/19 studies), (6) model serving and deployment (15/19 studies), (7) continuous monitoring (14/19 studies), and (8) continual learning (13/19 studies). We proposed a 3-stage MLOps maturity framework for health care based on existing studies in the field, that is, low (5/19 studies), partial (1/19 studies), and full maturity (13/19 studies). There were 8/19 studies that discussed ethical, legislative, and stakeholder considerations for MLOps implementations in health care settings. Conclusions: We investigated the implementation of MLOps in health care with a corresponding maturity framework. It is evident that only a limited number of studies reported the implementation of ML in health care contexts. Hence, it is imperative that we shift our focus toward creating an environment that supports the development of ML health care applications, such as improving existing data infrastructure, and engaging partners to support the development of MLOps applications. Specifically, we can include patients, policymakers, and health care professionals in the creation and implementation of ML applications. One of the main limitations includes the varying quality of each extracted study in terms of how the MLOps implementation was presented. Hence, it was difficult to verify the presence and discuss in depth all steps of the MLOps workflow for each study. Furthermore, due to the inherent nature of a scoping review protocol, there may be a compromise on an in-depth discussion of each step within the MLOps workflow.

Background: The use of wearable devices for remote health monitoring is a rapidly expanding field. These devices might benefit patients and providers; however, they are not yet widely used in healthcare. This scoping review assesses the current state of the literature on wearable devices for remote health monitoring in non-hospital settings. Methods: CINAHL, Scopus, Embase and MEDLINE were searched until August 5, 2024. We performed citation searching and searched Google Scholar. Studies on wearable devices in an outpatient setting with a clinically relevant, measurable outcome were included and were categorized according to intended use of data: monitoring of existing disease vs. diagnosis of new disease. Results: Eighty studies met eligibility criteria. Most studies used device data to monitor a chronic disease (68/80, 85%), most often neurodegenerative (22/68, 32%). Twelve studies (12/80, 15%) used device data to diagnose new disease, majority being cardiovascular (9/12, 75%). A range of wearable devices were studied with watches and bracelets being most common (50/80, 63%). Only six studies (8%) were randomized controlled trials, four of which (67%) showed evidence of positive clinical impact. Feasibility determinants were inconsistently reported, including compliance (51/80, 64%), patient-reported useability (13/80, 16%), and participant technology literacy (1/80, 1%). Conclusions: Evidence for clinical effectiveness of wearable devices remains scant. Heterogeneity across studies in terms of devices, disease targets and monitoring protocols makes data synthesis challenging, especially given the rapid pace of technical innovation. These findings provide direction for future research and implementation of wearable devices in healthcare.

The digital delivery of mental health services became increasingly common following the onset of the COVID-19 pandemic. There is still much to learn regarding tailoring interventions for trauma-affected populations (military members, Veterans, public safety personnel). Through the current pilot study, we explored the perceptions of digitally delivered psychotherapies utilized for trauma-affected populations, as reported by Canadian military members, Veterans, and public safety personnel who completed such interventions. Quantitative data were collected from 11 Canadian clients (military members, Veterans, and public safety personnel with posttraumatic stress injury). Survey questions were based on the Alberta Quality Matrix of Health and the Unified Theory of Acceptance and Use of Technology model. As a follow-up, clients were invited to partake in a semi-structured interview to further explore their perspectives on digitally delivered trauma-focused and adjunct therapies. Four clients participated in an interview. The client participants reported that digitally delivered trauma and adjunct therapies offered similar treatment effectiveness to in-person delivery while also improving treatment access. The participants indicated several unique advantages of digital delivery, including the increased accessibility of treatment, cost-effectiveness, and more efficient use of resources, although the small sample size limits the generalizability of our findings. Further research with a larger, more diverse population is required to corroborate our results and identify other avenues in which psychotherapies utilized for trauma-affected populations can be engaged with and improved upon.