Manisha Desai

Background: Accelerometers produce high-resolution physical activity data, but commonly used summary measures often reduce these data to total volume, intensity, or variability and may not retain interpretable temporal structure across the day. Cluster-based summaries may provide a way to characterize daily physical activity profiles while preserving information about when activity occurs. Objective: This study evaluated whether cluster-derived accelerometer summary measures could represent daily physical activity patterns and explain variation in pediatric cardiometabolic outcomes comparably to traditional accelerometer summary metrics. Methods: This baseline cross-sectional methodological analysis used data from 268 Latino children with overweight or obesity from low-income families participating in the Stanford GOALS trial. Participants were aged 7 to 11 years old and wore accelerometers for a minimum of 1 week according to study protocol. Valid daily activity profiles were summarized within a 7:00 AM-11:00 PM analytic window using consecutive, nonoverlapping 10-minute intervals. Daily profiles were clustered using unsupervised learning, and participant-level cluster-derived measures were created from the distribution of valid days assigned to each cluster. We compared these measures with traditional accelerometer summaries, including time spent in activity intensity states, Time Active Mean, Time Active Variability, Activity Intensity Mean, and Activity Intensity Variability. Linear regression models were used to evaluate associations with waist circumference, fasting insulin, and fasting triglycerides, adjusting for age and sex. Model performance was compared using R ² and the Akaike information criterion. Results: Cluster-derived measures explained a comparable proportion of variation in the 3 cardiometabolic outcomes to traditional accelerometer summary metrics. For example, the highest R ² values among the cluster-derived measures were 25%, 11%, and 6% for waist circumference, fasting insulin, and fasting triglycerides, respectively, compared with 25%, 10%, and 6% for Time Active Mean. No single summary-measure approach consistently yielded the highest R ² across all 3 outcomes. Conclusions: Cluster-derived accelerometer measures provide a regression-ready approach for summarizing daily physical activity patterns while preserving interpretable clock-time structure. In this baseline analysis, these measures performed comparably to traditional accelerometer summaries while capturing temporal features not represented by conventional volume- or intensity-based metrics. Future work should evaluate external validation across diverse populations and settings and assess the use of these measures in longitudinal and intervention analyses.

Artificial intelligence (AI) tools in health care often demonstrate technical validity yet fail to achieve sustained integration within academic health centers. This persistent gap between pilot implementation and routine use suggests that translational barriers extend beyond model performance and regulation to structural differences between industry and health systems. We conducted an exploratory thematic synthesis of a multi-sector panel convened by the Clinical and Translational Science Awards (CTSA) Principles for AI Translation in Healthcare (PATH) Working Group, including representatives from industry and academic leadership. Recurrent tensions emerged across economic sustainability, liability allocation, data governance, infrastructure readiness, and lifecycle monitoring. From these themes, we propose a five-domain Industry–Academic Alignment Conceptual Model for clinical AI translation. The model conceptualizes translation as dependent on alignment across economic, risk and liability, data governance, infrastructure, and lifecycle monitoring domains, each reflecting differences in incentives, governance, and operational capacity. Misalignment across domains may contribute to stagnation at the pilot stage despite technical validity. This work frames clinical AI translation as an inter-organizational alignment problem and identifies multidomain alignment as a potentially important condition for sustained clinical integration.

Doctors increasingly rely on AI in the clinic, yet which report features make AI-generated responses useful and trustworthy remains unclear. In this randomized mixed-methods study, 34 oncology physicians provided 294 ratings of four blinded AI systems across five vignettes, alongside 20 semi-structured interviews analyzed with a prespecified LLM-assisted qualitative pipeline. Despite similar references, an evidence-graded report adapted from OpenEvidence was rated significantly lower in overall utility than standard OpenEvidence (mean difference, −0.96; 95% CI, −1.26 to −0.66; P<.001). Qualitative analysis identified six themes and seven design requirements. Oncologists valued rapid orientation, evidence retrieval, and verification, preferring concise, scannable reports with quantitative outcomes, recognizable bolded guidelines, explicit uncertainty, and verifiable citations. Trust deteriorated with citation mismatch, buried provenance, evidence misclassification, overconfident recommendations, and poor organization. Evidence presented differently can alter perceptions of clinical utility and trust; accuracy alone is insufficient, and report design must also be empirically evaluated.

Background: In the COVID-19 pandemic, several phase II and III randomized trials were launched to evaluate the effectiveness of camostat, an orally administered TMPRSS2 inhibitor previously approved for other indications, for treating SARS-CoV-2 infections. Owing to the rapidly changing landscape during the pandemic, many of these trials were unable to reach completion. Further, methods for synthesizing trials that were launched and not completed were critical. Methods: This systematic review aimed to consolidate global evidence by identifying placebo controlled, randomized trials of camostat and analyzing their collective clinical and virologic impact on SARS-CoV-2 through an individual patient data meta-analysis (IPDMA). We harmonized data from the studies and utilized Bayesian statistical models to assess virologic outcomes (measured by the rate of change in viral shedding) and clinical outcomes (based on the time to the first of two consecutive symptom-free days), adjusting for age and sex. Results: The IPDMA incorporated data from six countries, totaling 431 patients across the studies; 118 patients contributed data for the primary virologic outcome and 240 for the clinical symptom outcome. Camostat did not improve the rate of change in viral load (difference in rate of change = 0.11 Ct value/day higher, 95% credible interval 2.04 lower to 2.23 higher) or time to symptom resolution (hazard ratio = 0.87, 95% credible interval 0.51, 1.55) when compared to placebo. Conclusions: Despite its theoretically promising mode of action, camostat did not demonstrate a statistically significant virologic or clinical benefit in treating COVID-19, highlighting the complexity of drug repurposing in emergency health situations.

Numerous symposia and conferences have been held to discuss the promise of Artificial Intelligence (AI). Many center on its potential to transform fields like health and medicine, law, education, business, and more. Further, while many AI-focused events include those data scientists involved in developing foundational models, to our knowledge, there has been little attention on AI’s role for data science and the data scientist. In a new symposium series with its inaugural debut in December 2024 titled AI for Data Science , thought leaders convened to discuss both the promises and challenges of integrating AI into the workflows of data scientists. A keynote address by Michael Pencina from Duke University together with contributions from three panels covered a wide range of topics including rigor, reproducibility, the training of current and future data scientists, and the potential of AI’s integration in public health.

Introduction: Biostatisticians increasingly use large language models (LLMs) to enhance efficiency, yet practical guidance on responsible integration is limited. This study explores current LLM usage, challenges, and training needs to support biostatisticians. Methods: A cross-sectional survey was conducted across three biostatistics units at two academic medical centers. The survey assessed LLM usage across three key professional activities: communication and leadership, clinical and domain knowledge, and quantitative expertise. Responses were analyzed using descriptive statistics, while free-text responses underwent thematic analysis. Results: Of 208 eligible biostatisticians (162 staff and 46 faculty), 69 (33.2%) responded. Among them, 44 (63.8%) reported using LLMs; of the 43 who answered the frequency question, 20 (46.5%) used them daily and 16 (37.2%) weekly. LLMs improved productivity in coding, writing, and literature review; however, 29 of 41 respondents (70.7%) reported significant errors, including incorrect code, statistical misinterpretations, and hallucinated functions. Key verification strategies included expertise, external validation, debugging, and manual inspection. Among 58 respondents providing training feedback, 44 (75.9%) requested case studies, 40 (69.0%) sought interactive tutorials, and 37 (63.8%) desired structured training. Conclusions: LLM usage is notable among respondents at two academic medical centers, though response patterns likely reflect early adopters. While LLMs enhance productivity, challenges like errors and reliability concerns highlight the need for verification strategies and systematic validation. The strong interest in training underscores the need for structured guidance. As an initial step, we propose eight core principles for responsible LLM integration, offering a preliminary framework for structured usage, validation, and ethical considerations.