CLCarlo Lutz

Carlo Lutz

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Carlo Lutz has a total of 25 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Policy-Anchored Evaluation of Large Language Model Disposition Recommendations for ED Chest Pain: Single-Institution Feasibility Study (RAPID-ED Phase 0) (Preprint)

    by Carlo Lutz, Alan Teigman, Jennifer Zapata, Kyle Shibuya, Deborah White, Benjamin Friedman, Michael Jones, Shitij Arora

    BACKGROUND Large language models (LLMs) are increasingly deployed as clinical decision support systems (CDSS), but the accuracy of their emergency department (ED) disposition recommendations is unclear. We propose measuring not just accuracy but also direction of error and adherence to institutional protocols. OBJECTIVE To assess the accuracy of policy-anchored LLM disposition recommendations for chest pain presentations to the ED, and to characterize concordance, direction of discordance, and institutional disposition-vocabulary adherence across 5 artificial intelligence (AI) configurations. METHODS In a single-institution feasibility study, 31 ED chest pain cases (21 real, 10 synthetic borderline-risk vignettes) were scored on a 5-tier ordinal disposition scale (discharge, observation, floor, telemetry, intensive care/cath lab). An emergency physician (EP) adjudicating against the institutional chest pain policy served as the reference standard (Reviewer 1) and was compared to a second EP’s review to establish ceiling inter-rater reliability. Five AI configurations were asked to provide level of care and campus disposition: an ungrounded general-purpose model (ChatGPT-Edu, GPT-5.5); 3 retrieval-augmented generation (RAG) -like configurations grounded in the 2021 national chest pain guideline and institutional disposition policy, differing in knowledge-delivery prompting; and OpenEvidence, a literature-grounded RAG LLM. The primary metric was linear-weighted Gwet AC2, with 95% CIs by bootstrap. Discordant cases were classified as over- or under-triage; agreement with actual disposition was also measured. RESULTS Physician adjudication agreed on 30/31 cases; inter-rater reliability was AC2 0.98, 95% CI 0.95-1.00. Among LLM configurations, per-case in-chat delivery achieved the highest agreement (AC2 0.53, 95% CI 0.39-0.68) and single-batch delivery the lowest (0.45), with OpenEvidence at 0.49; the ungrounded baseline was lowest overall (0.31, 95% CI 0.10-0.52). Reviewer 1 agreed with actual disposition (n=21) at AC2 0.81 (0.64-0.96), exceeding every configuration. GPT base over-triaged (22 over vs 3 under; telemetry or intensive care in 77% of cases vs 26% for the reference), whereas RAG-batch and OpenEvidence tended to under-triage (6 over vs 14 under). CONCLUSIONS No LLM configuration approached the level of physician-physician agreement. Institutional grounding improved concordance but reversed the direction of residual error, from over-triage to under-triage. This error was driven predominantly by an observation pathway not actually present at the institution. Adherence to the institutional disposition vocabulary is a distinct evaluation axis from accuracy; deployment should constrain model output to the local disposition menu.

  • Open Access

    Respiratory-Responsive Vocal Biomarker for Asthma Exacerbation Monitoring: Prospective Cohort Study (Preprint)

    by Erik Larsen, Xinyu Song, Dale Joachim, Peter Ch'EN, Samuel Green, Emily Hunt, Savneet Kaur, Robin Nag, Olivia Pisani, Sherron Thomas, et al.

    Journal of Medical Internet Research - International Scientific Journal for Medical Research, Information and Communication on the Internet #Preprint #PeerReviewMe: Warning: This is a unreviewed preprint. Readers are warned that the document has not been peer-reviewed by expert/patient reviewers or an academic editor, may contain misleading claims, and is likely to undergo changes before final publication, if accepted, or may have been rejected/withdrawn. Readers with interest and expertise are encouraged to sign up as peer-reviewer, if the paper is within an open peer-review period. Please cite this preprint only for review purposes or for grant applications and CVs (if you are the author). Background: Asthma exacerbations are a significant issue due to poor perception and failure of traditional monitoring techniques to capture real-time changes in respiratory function. There is a need for strategies to enhance recognition and facilitate timely intervention. Objective: To evaluate the efficacy of a respiratory-responsive vocal biomarker (RRVB), accessible via smartphone application, as a convenient real-time tool for monitoring asthma exacerbations and respiratory states. Methods: In this three-month prospective cohort study, 84 asthma patients recorded daily 6-second vocal samples, generating RRVB scores. The RRVB's ability to predict asthma exacerbations and respiratory states (normal function, mild events, exacerbations) was assessed using risk ratios (RR) and confidence intervals (CIs). Respiratory states were defined using peak expiratory flow (PEF) and self-reported wellbeing, with exacerbations identified by a decline in both PEF and wellbeing, and mild events by a decline in one measure. Results: RRVB scores significantly predicted asthma exacerbations, with a risk ratio of 2.15 (95% CI: 1.62-2.85, p < .001), and 3.57 (95% CI: 2.70-4.73, p < .001) using normalized scores adjusted for individual characteristics. However, time-averaged RRVB scores did not significantly correlate with asthma control as measured by the Asthma Control Test (ACT), with a risk ratio of 1.17 (95% CI: 0.96-1.44, p = .12). A total of 58% of participants remained engaged throughout the study, reporting high satisfaction with the tool's usability and perceived benefits for asthma management. Conclusions: This study demonstrates potential enhancements in asthma management through more accessible and continuous monitoring of respiratory function offered by the RRVB tool, which was shown to provide an effective, real-time, and user-friendly approach for detecting potential exacerbations. Clinical Trial: clinicaltrials.gov: NCT05850390


  • Open Access

    Respiratory-Responsive Vocal Biomarker for Asthma Exacerbation Monitoring: Prospective Cohort Study (Preprint)

    by Erik Larsen, Xinyu Song, Dale Joachim, Peter Y Ch'EN, Samuel M Green, Emily Hunt, Savneet Kaur, Robin Nag, Olivia Pisani, Sherron Thomas, et al.

    BACKGROUND Asthma exacerbations remain a major challenge in asthma management, often due to delayed recognition and limitations of conventional monitoring tools such as peak flow meters and symptom questionnaires. These tools are typically effort dependent or retrospective, making them less suited for continuous, real-time monitoring. A novel, smartphone-based respiratory-responsive vocal biomarker (RRVB) may offer an accessible and noninvasive approach for dynamic assessment of respiratory health. This RRVB has previously demonstrated generalizable performance in cross-sectional cohorts across multiple respiratory conditions, including asthma, chronic obstructive pulmonary disease, and COVID-19, in populations spanning India and the United States. This study extended this work by evaluating the real-world, longitudinal performance of the same RRVB tool for daily asthma exacerbation monitoring via smartphones in home settings. OBJECTIVE This study aimed to evaluate the efficacy of the RRVB as a convenient real-time tool for monitoring asthma exacerbations and respiratory states in a real-world, longitudinal setting. METHODS In this prospective cohort study, 84 adult patients with asthma were enrolled from an academic medical center and followed for 90 days. Participants submitted daily 6-second voice samples and conducted peak expiratory flow measurements and surveys, including symptom reports and asthma control assessments. RRVB scores were generated in real time on the app. Asthma states (normal function, mild event, and exacerbation) were defined based on both peak expiratory flow and self-reported well-being. Risk ratios were calculated to assess the predictive validity of RRVB scores for identifying exacerbation events. Engagement was measured via frequency of completed sessions, and participant experience was evaluated through exit surveys. RESULTS RRVB scores significantly stratified asthma states. The risk of experiencing an exacerbation was 2.15 times higher (95% CI 1.62-2.85; P<.001) with elevated RRVB scores and 3.57 times higher (95% CI 2.70-4.73; P<.001) using normalized scores adjusted for individual characteristics. RRVB scores did not significantly correlate with the Asthma Control Test (risk ratio=1.17, 95% CI 0.96-1.44; P=.12), highlighting its role as a momentary signal rather than a proxy for longitudinal control. Engagement was moderate or higher (≥26 total app sessions) in 58% (49/84) of participants. Among survey respondents, 93% (43/46) found the app easy to use, 89% (41/46) reported a positive overall experience, and 87% (40/46) indicated that they would use a similar tool in the future. Fewer participants (32/46, 70%) reported understanding the RRVB scores, suggesting a need for improved score interpretability and user guidance in future implementations. CONCLUSIONS The RRVB tool demonstrated effective real-time detection of asthma exacerbations and dynamic respiratory states, supporting its potential as a noninvasive, user-friendly, and physiologically grounded digital biomarker for asthma monitoring. These findings provide foundational evidence for broader deployment and integration of voice-based tools to support proactive, real-world asthma management.