Publications by Year: 2026

BACKGROUND: The use of generative large language models (LLMs) with electronic health record (EHR) data is rapidly expanding to support clinical and research tasks. This systematic review characterizes the clinical fields and use cases that have been studied and evaluated to date.

METHODS: We followed the Preferred Reporting Items for Systematic Review and Meta-Analyses guidelines to conduct a systematic review of articles from PubMed and Web of Science published between January 1, 2023, and November 9, 2024. Studies were included if they used generative LLMs to analyze real-world EHR data and reported quantitative performance evaluations. Through data extraction, we identified clinical specialties and tasks for each included article, and summarized evaluation methods.

RESULTS: Of the 18 735 articles retrieved, 196 met our criteria. Most studies focused on radiology (26.0%), oncology (10.7%), and emergency medicine (6.6%). Regarding clinical tasks, clinical decision support made up the largest proportion of studies (62.2%), while summarizations and patient communications made up the smallest, at 5.6% and 5.1%, respectively. In addition, GPT-4 and GPT-3.5 were the most commonly used generative LLMs, appearing in 60.2% and 57.7% of studies, respectively. Across these studies, we identified 22 unique non-NLP metrics and 35 unique NLP metrics. While NLP metrics offer greater scalability, none demonstrated a strong correlation with gold-standard human evaluations.

CONCLUSION: Our findings highlight the need to evaluate generative LLMs on EHR data across a broader range of clinical specialties and tasks, as well as the urgent need for standardized, scalable, and clinically meaningful evaluation frameworks.

Using 2023-2024 U.S. National Health Interview Survey data, we found that digital health literacy (dHL) mediated nearly half of the difference in telehealth use between Latino adults with non-English and English language preference. These findings identify dHL as a modifiable mechanism linking linguistic and digital access barriers, underscoring the need for multilingual, inclusive, and equitable telehealth design.

The Psychiatric Consultation Service at Massachusetts General Hospital sees medical and surgical inpatients with comorbid psychiatric symptoms and conditions. During their twice-weekly rounds, Dr Stern and other members of the Consultation Service discuss diagnosis and management of hospitalized patients with complex medical or surgical problems who also demonstrate psychiatric symptoms or conditions. These discussions have given rise to rounds reports that will prove useful for clinicians practicing at the interface of medicine and psychiatry. Prim Care Companion CNS Disord 2026;28(1):25f04057. Author affiliations are listed at the end of this article.

Throughout the COVID-19 pandemic, the Centers for Medicare and Medicaid Services and state health agencies implemented numerous regulations for nursing homes addressing infection control and data reporting. This study explores interviews (n = 156) with administrators of 40 nursing homes across the U.S. between July 2020 and December 2021 to better understand their experiences with governmental agencies amid shifting regulatory standards, frequent inspections, and possible enforcement actions. Administrators highlighted confusion due to the evolving and sometimes conflicting guidance between state and federal agencies, although some states offered valuable COVID-19-specific assistance. They also described challenges in understanding and implementing new, frequently changing requirements, resulting in potential inspection deficiencies. Although enforcement actions, including financial penalties, are intended to deter noncompliance, administrators expressed concerns about added resource strain. Recommendations included increasing collaboration and data collection between regulatory agencies; reducing administrative burden during outbreaks and incorporating feedback from centers during regulatory changes; and increasing reimbursement to support compliance. Continued changes to oversight, including increased penalization and risk-based survey prioritization, should be evaluated to determine differential impacts on nursing home operations and resident care.

OBJECTIVE: Immunothrombosis contributes to ischemic stroke pathophysiology through neutrophil extracellular trap (NET) formation, which promotes thrombus stabilization and microvascular dysfunction. DNase1 is the principal endonuclease responsible for NET degradation. The rs1053874 polymorphism in DNase1 gene influences enzymatic activity and protein stability in vitro, but its clinical relevance in ischemic stroke remains unexplored. We investigated whether this variant modulates systemic NET burden and impacts stroke-related outcomes.

METHODS: We conducted a prospective observational cohort study including 492 patients with acute ischemic stroke. Genotyping of rs1053874 was performed via Sanger sequencing and categorized into AA versus GG + GA genotypes (dominant model). Clinical variables, NET biomarkers (elastase, myeloperoxidase [MPO], and dsDNA), DNAse1 activity, infarct volume, thrombectomy metrics, and survival were assessed. Multivariable regression and Cox proportional hazards models were used to explore associations between genotype and outcomes.

RESULTS: AA genotype carriers (7.9%) had a significantly lower burden of prior vascular events compared to GG + GA carriers. At admission, they exhibited higher DNAse1 activity, reduced levels of circulating NET markers (elastase, MPO, and dsDNA), and lower neutrophil and monocyte counts. Despite similar initial stroke severity, AA carriers required fewer thrombectomy passes and had significantly better early neurological recovery and smaller infarcts. In adjusted models, both the AA genotype and dyslipidemia were independently associated with improved long-term survival. However, stratified analyses revealed the most robust survival benefit among AA carriers without dyslipidemia. No significant interaction was observed.

INTERPRETATION: DNase1 rs1053874 polymorphism influences NET-related inflammation and is associated with improved vascular profile, procedural efficiency, and long-term outcomes in ischemic stroke. These findings support the potential of DNase1 as a therapeutic and prognostic target in personalized stroke care. ANN NEUROL 2026.

BACKGROUND: Technetium-99 m-dimercaptosuccinic acid (DMSA) renal cortical scintigraphy is commonly used in the evaluation of children with urinary tract infections. Pyelonephritis and post-pyelonephritic scarring manifest as renal cortical defects on DMSA renal scintigraphy, including DMSA Single Photon Emission Computed Tomography (SPECT) imaging. SPECT image quality can be degraded by blurring related to respiratory motion. Thus we hypothesize that image quality will be improved with estimation and correction of respiratory motion.

PURPOSE: The purpose of this study is to develop and evaluate a data-driven methodology that estimates surrogate respiratory signals and then employs these surrogate signals in correction of respiratory-motion in pediatric DMSA renal SPECT imaging.

METHODS: The XCAT digital anthropomorphic phantom was used with SPECT Monte Carlo simulation to form a population of 100 ms projections of DMSA renal SPECT imaging acquired with clinically relevant count-levels. These 100 ms projections emulated the framing of list-mode acquisitions at Boston Children's Hospital (BCH). The axial (superior/inferior) center-of-count-mass (aCOM) approach was utilized to estimate a surrogate respiratory signal for combining the 100 ms projections into seven respiratory-motion states with each having different extents of motion. The motion-states were then reconstructed and rigid-body respiratory-motion of the kidneys between the three motion-states on either side of the center state versus the center state was estimated by rigid-body registration. This estimated motion was then used to correct respiratory motion as part of a second pass through reconstruction of the projections of the motion states. To evaluate the surrogate signal, Pearson's correlation coefficient was calculated between the true respiratory signals used in creating the XCAT projection data and the surrogate respiratory signals. The respiratory motion corrected reconstructions and the images reconstructed without respiratory motion compensation were quantitatively compared to the ground truth images (where no respiratory motion was simulated) using the Normalized Root Mean Square Error (NRMSE) as a measure of fidelity.

RESULTS: The average over our entire population of XCAT phantoms of Pearson's correlation coefficient (r) between the aCOM estimated surrogate respiratory and the actual average motion simulated for each of the 100 ms time intervals was 0.76. The average standard error of the estimate (SEE) for this r-value was 3.06 mm. For the group of XCAT phantoms with a simulated average amplitude of motion between 6-10 mm, there were no significant differences in the NRMSE versus ground truth reconstructions for the reconstructions with either the estimated or true motion correction methods compared against reconstructions without motion correction. For the groups with simulated average amplitudes of motion between 10-14 mm, 14-18 mm, and >18 mm, there were significant differences in the NRMSE for the reconstructions with motion with either the estimated and true motion correction applied in comparison to reconstructions without motion correction.

CONCLUSION: Respiratory motion correction in pediatric renal SPECT imaging using a data-driven approach can improve image quality, with potential for improved diagnostic accuracy for studies with a moderate amount of motion.

This CME/CE integrates real patient stories, current evidence, evolving guideline recommendations, and expert clinical experience to equip hematology/oncology clinicians with practical strategies for successful asparaginase-based therapy in young adults with acute lymphoblastic leukemia (ALL). The overarching goal is to improve outcomes for young adults with ALL through more consistent application of pediatric-inspired regimens, optimized asparaginase use, and comprehensive, patient-centered care. Leukemia experts synthesize the latest evidence on the efficacy and safety of asparaginase-based ALL treatment for young adults. Using a case-based approach, the curriculum provides structured guidance on mitigation, monitoring, and management of key asparaginase-related toxicities. Practical recommendations include therapeutic drug monitoring of asparaginase activity, detection of clinical and silent hypersensitivity reactions, and timely substitution of Escherichia coli-derived asparaginase with Erwinia-derived asparaginase to preserve therapeutic activity and efficacy after immune-mediated inactivation. Beyond treatment selection and toxicity management, the activity addresses system-level and psychosocial barriers that uniquely affect young adults with ALL, such as distance from specialty centers, employment and family responsibilities, lower rates of clinical trial participation, and survivorship concerns. To view this activity, and obtain CME/CE credit, click here.

Wildfire-induced fine particulate matter (PM2.5) poses a growing public health risk. To examine public responses to wildfire smoke, we analyzed smoke PM2.5 and Google Trends data across 11 Designated Market Areas (DMAs) in California from 2016 to 2020. Scaled Relative Search Interest (RSI) for 'air pollution' and 'air purifier' was used as a proxy for public awareness and preparedness. Applying a Bayesian spatiotemporal distributed lag model, we found that a one-unit increase in smoke PM2.5 (6.6 μg/m3) was associated with 149% (95% Credible Interval (CrI): 107%, 486%) higher 'air pollution' searches and 34% (95% CrI: 3%, 76%) and 68% (95% CrI: 26%, 123%) higher 'air purifier' searches over the first two weeks, respectively. Among all DMAs, Palm Springs showed the strongest responses. Public search responses were amplified under higher temperature and pressure but decreased with higher wind speed and humidity. Communities with greater socioeconomic vulnerability exhibited lower search interest in both awareness- and preparedness-related terms. These findings highlight the need for timely, targeted, and equitable public communication strategies that align with real-time environmental conditions and improve access to protective resources among vulnerable populations.

BACKGROUND: Suboptimal blood pressure (BP) control remains a major cardiovascular disease risk factor. Whether genetically predicted BP independently predicts long-term BP control is unknown. We examined the associations of BP polygenic scores (PGSs) with long-term BP control and treatment-resistant hypertension.

METHODS: We identified 22 456 Mass General Brigham Biobank participants with hypertension. Longitudinal BP control was defined as the percentage of time above-target systolic BP (SBP) ≥130 mm Hg or diastolic BP (DBP) ≥80 mm Hg over 5 years. Using multivariable regression, we assessed the associations of BP PGS with duration above-target BP and lifetime treatment-resistant hypertension incidence. Incremental prognostic utility of BP PGSs was assessed based on the discrimination C-index, Brier score, and net reclassification index. Validation was performed in the population-based UK Biobank cohort using the SBP/DBP ≥140/90 mm Hg threshold.

RESULTS: Among 10 853 (48.3%) were female, the mean SBP/DBP (SD) at index date was 132 (18)/75 (11) mm Hg, and 4126 (18.4%) developed treatment-resistant hypertension over lifetime. In reference to the low (<20th percentile) PGS group, the high (≥80th percentile) BP PGS was associated with 8.01 (95% CI, 6.68%-9.34%) longer duration with above-target SBP and 6.19 (95% CI, 5.05%-7.33%) with high DBP. Each high SBP and DBP PGS conferred 2.36 (95% CI, 2.07-2.68) and 1.75 (95% CI, 1.55-1.99)-fold higher odds of treatment-resistant hypertension. Adding BP PGSs to traditional risk factors improved treatment-resistant hypertension prediction from C-index (95% CI), 0.74 (0.73-0.75) to 0.78 (0.77-0.79). BP PGSs consistently predicted longitudinal BP management to a comparable extent in the UK Biobank.

CONCLUSIONS: Harnessing BP PGSs may inform anticipated trends in BP control to warrant vigilant monitoring and augment prioritization of intensive therapy.

BACKGROUND: Higher relative risk for cardiovascular disease (CVD) events at lower blood pressure (BP) thresholds in female versus male adults suggest that hypertension thresholds should be sex-specific.

METHODS: We used the ARIC study (Atherosclerosis Risk in Communities) visit 1 (1987-1989) to compare the BP distribution, estimated risk (via the 10-year Predicting Risk of Cardiovascular Disease Events score), absolute risk, and relative risk of CVD according to BP thresholds, stratified by sex and hypertension treatment status, in participants without prior CVD.

RESULTS: Of 13 418 participants (56% women, mean age [54±5.7 years]), 25% were treated for hypertension. Males had higher average 10-year CVD risk scores regardless of treatment. The distribution of BP and prevalence of CVD risk factors was similar for male and female adults. Incidence rates (per 10 000 person-years) comparing a systolic BP threshold of ≥140 versus <140 mm Hg for coronary heart disease were 30.9 and 12.0 among untreated male and female adults (P=0.07) and 27.4 versus 16.5 among treated male and female adults (P=0.63). HRs comparing a systolic BP threshold of ≥140 versus <140 mm Hg for coronary heart disease were 1.49 and 1.72 among untreated male and female adults (P=0.16) and 1.30 versus 1.40 among treated male and female adults (P=0.93).

CONCLUSIONS: In this middle-aged population, there were no consistent differences in BP distribution, risk factor burden, absolute risk, or relative risk of CVD between male and female adults. These findings do not support a sex-specific threshold for hypertension.