BACKGROUND:Reproducible quantification of pulmonary edema on chest radiographs is challenging for bedside clinicians. The Radiographic Assessment of Lung Edema (RALE) score offers a structured, semi-quantitative approach with prognostic implications in acute respiratory failure (ARF), yet its clinical use is limited by the absence of standardized training. OBJECTIVE:To determine whether a standardized RALE training program, combined with expert feedback, improves inter-rater reliability across clinician experience levels. METHODS:We analyzed 4487 chest radiographs from 864 critically ill patients with ARF enrolled in a prospective registry. Sixteen physicians (1 expert and 15 reviewers-including 5 interns, 4 residents, 3 fellows, and 3 attendings) completed a 3-phase RALE training program: (i) independent review of instructional materials, (ii) a live virtual session led by the expert, and (iii) iterative practice with expert feedback on discrepant scores. All scoring was performed using the Pulmo-Annotator platform. We assessed inter-rater reliability using intraclass correlation coefficients (ICC), Bland-Altman analysis, and Deming regression, while stratifying performance by experience level and self-reported confidence. RESULTS:Inter-rater reliability improved following the training program, with ICC increasing from 0.89 (95% CI: 0.85-0.92) to 0.93 (95% CI: 0.90-0.95) after expert feedback. Fellows demonstrated the largest gain of ICC from 0.89 to 0.97, while interns and attendings maintained high reliability. Post-feedback analyses indicated reduced systematic bias. Residents exhibited the greatest variability and revised fewer scores post-feedback, despite equivalent exposure to feedback. Score distributions also varied by self-reported confidence, with greater variability among less experienced reviewers. CONCLUSION:Following a structured training program, radiographic edema scoring using the RALE framework demonstrated excellent inter-rater reliability, which further improved with targeted expert feedback. This structured training program, which incorporated standardized instructional materials and iterative feedback, demonstrated that RALE scoring is teachable across different levels of clinical experience and can be integrated into educational curricula to enhance consistency in chest radiograph interpretation. These findings provide a foundation for future work exploring the role of RALE training in broader research and clinical applications.
BACKGROUND:Hypertension is common during and after pregnancy. Patterns of antihypertensive medication discontinuation (AMD) in the postpartum period are not well characterized. This study examined factors associated with AMD among postpartum women. METHODS:A retrospective claims analysis was conducted using the Komodo Health Healthcare Map. The study included 63,312 postpartum women aged 18-64 years who delivered between January 1, 2019, and December 31, 2022, and initiated an antihypertensive medication within 30 days after live delivery. AMD was defined as the absence of any anti-hypertensive medication from index medication (days' supply + 30-day). Multivariable Poisson regression models with a log link and robust variance estimators were used. RESULTS:Discontinuation occurred in 57.8% (36,576) of women. In adjusted analyses, older age was associated with lower risk of AMD compared with women aged 18-24 years (RR 0.93, 95% CI 0.93, 0.94 for ages 25-34; RR 0.89, 95% CI 0.88, 0.90 for ages 35-44; RR 0.85, 95% CI 0.82, 0.88 for ages ≥45). Hispanic (RR 1.03, 95% CI 1.02, 1.04) and Asian (RR 1.02, 95% CI 1.01, 1.04) women had higher risk of discontinuation compared with White women. Women with eclampsia (RR 0.97, 95% CI 0.97, 0.98), baseline hypertension (RR 0.95, 95% CI 0.95, 0.96), and postpartum depression (RR 0.95, 95% CI 0.94, 0.96) had lower risk of discontinuation. CONCLUSIONS:Postpartum AMD was common in this national claims-based cohort. Differences across demographic and clinical subgroups highlight patient populations that may benefit from directed postpartum blood pressure follow-up and medication management.
Background:Physician burnout, often driven by documentation burden in electronic health records, remains a major challenge in clinical care. Ambient artificial intelligence (AI) scribes generate draft notes from clinical encounters, offering a scalable approach to reduce administrative workload, yet population-level evaluations remain limited. Objectives:The objective of this study is to assess changes over time in clinician perceptions of workload, well-being, and care delivery following implementation of ambient AI scribes. Methods:We conducted a prospective repeated cross-sectional survey study of physicians and advanced practice providers at four time points: baseline and approximately 1, 3, and 6 months' postimplementation. Outcomes included cognitive workload (NASA Task Load Index), professional fulfillment and burnout (Professional Fulfillment Index and Mini-Z), and perceptions of documentation burden, communication, and clinical capacity. Responses were grouped into clinically meaningful categories, and chi-square tests compared distributions across time. Secondary analyses compared baseline characteristics of respondents with linkable data across time points to the full cohort and evaluated subgroup differences for specific items. Results:A total of 1,600 clinicians responded at baseline, with 172, 117, and 101 respondents at 1-, 3-, and 6-month follow-up, respectively. Over 95% of follow-up respondents reported generating ≥5 ambient AI-supported notes, with use increasing significantly over time (p < 0.05). Clinicians reported statistically significant reductions in mental demand, time pressure, and documentation effort at postimplementation intervals (p < 0.05). Time spent on documentation outside clinic declined, with more clinicians reporting <2 hours/day (p < 0.05). Burnout increased across follow-up time points (p < 0.05). Confidence in patient comprehension and willingness to increase patient volume improved. Subgroup analyses mirrored overall trends. Conclusion:Ambient AI scribes were associated with sustained reductions in cognitive workload and perceived documentation burden. Despite these improvements, burnout increased over time, likely reflecting multifactorial influences and limitations of unpaired survey data. These findings support continued implementation and further longitudinal evaluation of clinician well-being.
Rule-based natural language processing (NLP) tools can identify pulmonary embolism (PE) via radiology reports. However, their external validity remains uncertain.In this cross-sectional study, 1,712 hospitalized patients (with and without PE) at Mass General Brigham (MGB) hospitals (2016-2021) were analyzed. Two previously published NLP algorithms were applied to radiology reports to identify PE. Chart review by two physicians was the reference standard. We tested three approaches: (A) NLP applied to all patients; (B) NLP limited to radiology reports of patients with principal or secondary International Classification of Diseases 10th revision (ICD-10) PE discharge codes; and (C) NLP applied to patients with PE discharge codes or a Present-on-Admission (POA) indicator ("Y") for PE. All others were assumed PE-negative in Approaches B and C to minimize NLP false positives. Weighted estimates were derived from the MGB hospitalized cohort (n = 381,642) to calculate F1 scores (as the harmonic mean of sensitivity and positive predictive value [PPV]).In Approach A, both NLP tools showed high sensitivity (82.5%, 93.0%) and specificity (98.9%, 98.7%) but low PPV (60.3%, 59.6%). Approach B improved PPV (95.2%, 94.9%) but reduced sensitivity (74.1%, 76.2%), while Approach C preserved both high sensitivity (82.5%, 93.0%) and PPV (95.6%, 95.8%). Approach C demonstrated the best performance, yielding significantly higher F1 scores for both NLP tools (88.6%, 94.4%) compared with Approach A (69.7%, 72.6%) and Approach B (83.3%, 84.5%) (P < 0.001).The accuracy of PE detection improves when rule-based NLP algorithms are operationalized using administrative claims data in addition to radiology reports.
