EFElgar Fleisch

Elgar Fleisch

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Elgar Fleisch has a total of 267 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Protocol for the REVELIO test-track pilot study: A randomised, controlled, single-centre trial in healthy recreational cannabis users investigating real-time in-vehicle detection of cannabis-impaired driving

    by Robin Deuber, Michal Bechny, Christoph Heck, Joel Brügger, Matthias Pfäffli, Mia Jovanova, Elgar Fleisch, Felix Wortmann, Wolfgang Weinmann

    Background: Driving under the influence of cannabis is associated with impaired cognitive and psychomotor performance and an increased risk of traffic accidents. Reliable real-time in-vehicle systems for detecting cannabis-related driving impairment are currently lacking; but hold great potential for improving road safety. Methods: This protocol describes the REVELIO test-track pilot study: a randomised, controlled, open-label, interventional, single-centre trial. The study assesses the feasibility and methodological requirements for developing and evaluating a multimodal in-vehicle detection approach using vehicle and driver state data. A total of 45 healthy recreational cannabis users will be enrolled and randomly allocated to an intervention or a reference control group. During the main study day, all participants will undergo biological sampling for tetrahydrocannabinol (THC) and related metabolites, as well as pre-driving assessments, followed by a sober baseline driving session on a closed test track using a dual-pedal vehicle with a certified driving instructor onboard. Participants in the intervention group will then receive a single controlled inhalative cannabis dose (target 0.67 mg THC per kg body weight), while the reference group will receive no cannabis. All participants will subsequently complete three additional standardized 50-minute driving sessions at predefined time points up to approximately six hours after administration, following identical schedules to enable within- and between-group comparisons. Between driving sessions, structured breaks will include recovery periods, repeated biological sampling, and traffic-medical, traffic-psychological, and pre-driving performance assessments, to characterise the temporal dynamics of cannabis-related impairment. Discussion: Multimodal data will be collected, including vehicle controller area network (CAN) data, driver monitoring camera (DMC) data, physiological signals using wearables, and biological samples (capillary blood, breath, oral fluid. Machine-learning–based models will be developed and evaluated to distinguish sober from cannabis-influenced driving states under controlled conditions. Secondary analyses will examine changes in driving performance over time and associations between functional measures and biological THC concentrations. As an exploratory pilot study conducted on a secured test track, the protocol aims to generate standardized reference data and quantitative performance metrics to inform both feasibility and system design considerations. Trial registration: The study is registered at ClinicalTrials.gov ( NCT07401628 ).

  • Open Access

    Effect of food swap recommendations on the healthfulness and carbon footprint of grocery purchases in a naturalistic online store: A randomized parallel trial

    by Anna H. Grummon, Kevin O’SUllivan, Amanda B. Zeitlin, Lindsey Smith Taillie, Cristina J. Y. Lee, Ashe J. Walker, Verena Tiefenbeck, Elgar Fleisch, Clayton Ulm, Thomas N. Robinson

    Background: Modeling studies suggest that simple dietary “swaps” in which just one food is replaced with another could yield both health and environmental benefits, but it remains unclear whether consumers make such changes when prompted. We tested whether recommending swaps improved the healthfulness and reduced the carbon footprint of simulated online grocery selections. Methods and findings: In May–June 2025, we recruited 1,201 US adults ages 18+ for a randomized parallel trial. Participants completed 3 weekly study visits. In the first visit, participants shopped in a naturalistic online grocery store with no intervention. They were then randomized in parallel to 1 of 4 trial arms: (1) climate swaps ( n = 301), (2) health swaps ( n = 300), (3) climate + health swaps ( n = 300), or (4) control ( n = 300). In the second and third visits, participants completed the shopping task in their randomly assigned trial arm. Participants were not masked to their trial arm. Using intent-to-treat analyses, we assessed 2 primary endpoints: the healthfulness (operationalized as the average Ofcom Nutrient Profiling Model score across the selected items [range: 1–100]) and carbon footprint (operationalized as the average kg of carbon dioxide equivalents [CO 2 -eq] per kg associated with producing the selected items) of participants’ grocery selections. We examined whether outcomes changed from the first (no-intervention) visit to the second and third (intervention) visits in the swaps arms compared to the control arm. The climate ( B = 2.4 on the 1–100 scale, 95% CI [1.1, 3.7]), health ( B = 4.7, 95% CI [3.3, 6.0]), and climate + health ( B = 4.0, 95% CI [2.7, 5.4]) swaps each improved the healthfulness of participants’ selections compared to control (all p s < 0.001). The health and climate + health swaps both increased healthfulness more than the climate swaps ( p s ≤ 0.02) but did not differ from one another ( p = 0.34). The climate ( B = −3.3 kg CO 2 -eq per kg, 95% CI [−4.6, −2.1]; p < 0.001) and climate + health swaps ( B = −2.8, 95% CI [−4.1, −1.6]; p < 0.001), but not the health swaps ( B = −0.4, 95% CI [−1.7, 0.8]; p = 0.51), reduced the carbon footprint of participants’ selections compared to control. The climate and climate + health swaps led to larger reductions in carbon footprint than the health swaps ( p s < 0.001) but did not differ from one another ( p = 0.43). Swaps’ effects on healthfulness and carbon footprint did not differ by age, health consciousness, or environmental consciousness ( p s for interactions ≥0.10). The main limitations were that the sample differed somewhat from the US population and study results may not generalize to brick-and-mortar stores. Conclusions: Encouraging swaps to more healthful and lower-carbon foods improved the healthfulness and reduced the carbon footprint of online grocery selections. Clinical trials registration: ClinicalTrials.gov (NCT#06648226), first submitted October 16, 2024.

  • Open Access

    Development, Deployment, and Evaluation of DyMand: Wearable AI for In-Situ Detection and Capture of Couples’ Dyadic Interactions in Chronic Disease Management

    by George Boateng, Prabhakaran Santhanam, Elgar Fleisch, Janina Lüscher, Theresa Pauly, Urte Scholz, Guy Bodenmann, Tobias Kowatsch

    Dyadic interactions of couples are of interest as they provide insight into relationship quality and chronic disease management. Existing sensing systems primarily infer social structure within groups retrospectively, or capture couples’ interactions at random or scheduled times, which could miss meaningful interpersonal interactions between partners. In this work, we developed a smartwatch-based algorithm that detects interaction moments between partners and enables interaction-aware data capture in everyday settings. It uses the Bluetooth signal strength between two smartwatches, each worn by one partner, and a voice activity detection machine-learning model that we trained, to infer that the partners are interacting and then trigger data collection. We operationalized this sensing approach by developing, deploying, and evaluating DyMand, an open-source smartwatch and smartphone system that integrates multimodal sensor and self-report data collection in situ. We deployed the DyMand system in a 7-day field study and collected and analyzed data about social support and diabetes discussions from 13 (N=26) Swiss-based heterosexual couples managing diabetes mellitus type 2 of one partner. Our system evaluation showed that 77.6% of algorithm-triggered recordings contained partners’ conversation moments compared to 43.8% for scheduled triggers, and that DyMand was easy to use. Preliminary insights into disease management behavior revealed that 5.2% of couples’ conversation moments contained social support, and partners discussed diabetes management 2.4% of the time based on the audio data analyses. In contrast, the self-report data showed that patients received support in 67.7% of couples’ interactions and caregivers provided support in 71.8% of couples’ interactions. DyMand enables insights into dyadic behavior in-situ, and could be used by social, clinical, or health psychology researchers to understand the social dynamics of couples that are managing chronic diseases in everyday life to inform the development and delivery of behavioral interventions.

  • Open Access

    Structured Large Language Model Workflows for Motivational Interviewing in Health Behavior Change: Proof-of-Concept Study

    by Akshaye Shenoi, Tianze Li, Ahmad Ishqi Jabir, Amanda Pitkethly, Elgar Fleisch, Tobias Kowatsch, Jacqueline Louise Mair

    Background: Motivational interviewing (MI) is an effective approach for supporting health behaviorchange, but face-to-face delivery is resource-intensive and difficult to scale. Rule-based conversational agents (CAs) can improve access; however, their scripted interactions and limited language flexibility constrain MI delivery. While large language models (LLMs) are increasingly being used for MI coaching, their conversational fidelity and quality compared with human coaches and rule-based CAs remain understudied. Objective: This study aimed to describe the development of an LLM-based CA, Artificially Intelligent Motivational Interviewing (Aimi), orchestrated with structured workflows, and to evaluate its feasibility, conversational fidelity, and user perceptions during MI coaching interactions. Methods: We developed Aimi using structured LLM workflows designed to enhance MI fidelity. We conducted a within-participants study, where 18 adults interacted with (1) Aimi, (2) a novice MI-trained human coach, and (3) a rule-based CA during live text-based role-play coaching sessions. Transcripts were independently evaluated by an MI expert using the Motivational Interviewing Skill Code, Version 2.0 (MISC-2), to assess MI competency and fidelity. Participants completed a user experience questionnaire to provide general feedback and to assess session alliance, dialogue relevance, empathy, engagement, linguistic quality, and perceived motivation to change. Feedback from users was thematically summarized and categorized under strengths and weaknesses for each approach. Results: Aimi achieved fidelity scores comparable to those of the novice human coach and higher than those of the rule-based CA on summary metrics, including higher reflection-to-question ratios (median 0.84, IQR 0.62-0.92 vs 0.62, IQR 0.42-0.74 vs 0.25, IQR 0.17-0.38), more complex reflections (median 66.67%, IQR 46.97%-76.92% vs 50%, IQR 34.38%-61.88% vs 0.00%, IQR 0%-50%), and greater elicitation of client change talk (median 90.83%, IQR 85.89%-100% vs 73.21%, IQR 63.10%-83.19% vs 66.67%, IQR 57.86%-81.94%). User experience ratings showed no significant differences across conditions. User feedback revealed distinct strengths and limitations across the coaching interactions. Participants described Aimi’s interactions as personalized, fluid, and adaptive, though sometimes overly reflective and lengthy. The novice human coach was viewed as empathetic and supportive but slow to respond, whereas the rule-based coach was viewed as efficient and structured yet limited in depth and personalization. Conclusions: This study demonstrates the technical feasibility of structured LLM-workflows for MI coaching and their capacity to maintain conversational fidelity comparable to that of a novice MI-trained human coach. Given the role-play paradigm, single-rater coding, and small convenience sample, these comparative findings should be interpreted as exploratory. Our findings serve as a foundational baseline for the development of scalable behavior change interventions in clinical settings.

  • Open Access

    Showcasing volunteer selection bias: Behavior of opt-in and not-opt-in participants in a time-of-use tariff study

    by Markus Kreft, Ludger Leenders, Marina González Vayá, Elgar Fleisch, Thorsten Staake

  • Open Access

    “Are You Okay, Honey?”: Recognizing Emotions Among Couples Managing Diabetes in Daily Life Using Multimodal Real-World Smartwatch Data

    by George Boateng, Xiangyu Zhao, Malgorzata Speichert, Elgar Fleisch, Janina Lüscher, Theresa Pauly, Urte Scholz, Guy Bodenmann, Tobias Kowatsch

    Couples generally manage chronic diseases together and the management takes an emotional toll on both patients and their romantic partners. Consequently, recognizing the emotions of each partner in daily life could provide insight into their emotional well-being in chronic disease management. Currently, the process of assessing each partner’s emotions is manual, time-intensive, and costly. Despite the existence of works on emotion recognition among couples, none of these works have used data collected from couples’ interactions in daily life. In this work, we collected 85 h (1021 5-min samples) of real-world multimodal smartwatch sensor data (speech, heart rate, accelerometer, and gyroscope) and self-reported emotion data (n = 612) from 26 partners (13 couples) managing diabetes mellitus type 2 in daily life. We extracted physiological, movement, acoustic, and linguistic features, and trained machine learning models (support vector machine and random forest) to recognize each partner’s self-reported emotions (valence and arousal). Our results from the best models—balanced accuracies of 63.8% and 78.1% for arousal and valence respectively—are better than the results from (1) chance, (2) prior work that also used data from German-speaking, Swiss-based couples, and (3) partners’ perceptions of each other’s emotions. This work contributes toward building automated emotion recognition systems that would eventually enable partners to monitor their emotions in daily life and enable the delivery of interventions to improve their emotional well-being.


  • Open Access

    Digital Quality Monitoring for Type 2 Diabetes in Swiss Primary Care: Qualitative Interview Study

    by Odile-Florence Giger, Lola Jo Ackermann, Marinja Principe, Simon Meier, Elgar Fleisch, Susanna Gallani, Michael Brändle, Tobias Kowatsch, Mia Jovanova

    Background: General practitioners (GPs) manage most type 2 diabetes (T2D) cases worldwide but face increasing workloads. To support guideline-based T2D care, the Swiss Society of Endocrinology and Diabetology (SSED) developed a quality monitoring tool for GPs to track and document patient clinical outcomes, such as glycated hemoglobin testing, blood pressure measurement, and diabetes-related complications in primary care. By aggregating these elements, the SSED score provides a structured way to assess adherence to recommended standards of diabetes care. Although this approach shows promise for improving health outcomes, the score is currently available in paper format and creates an additional administrative burden for physicians. Objective: This study examined challenges in T2D management in Swiss primary care, assessed limitations of the current SSED quality score, and identified strategies to enable the potential digital implementation and wider adoption of a digital SSED quality score in routine practice. Methods: A total of 38 qualitative interviews were conducted with 39 participants, representing 4 stakeholder groups relevant to digital T2D care: health care professionals (n=12), individuals with T2D (n=12), software developers (n=10), and health insurer representatives (n=5). Data were analyzed using thematic analysis. Results: Participants highlighted persistent challenges in T2D management, including time pressure for health care professionals, fragmented care, and a lack of personalized support for individuals with T2D. Barriers to digital quality monitoring included poor integration with GP systems, misaligned incentives, and limited relevance to multimorbidity. Suggested facilitators included embedding digital tools, such as a digital SSED score, into existing workflows, optimizing task shifting for GPs, integrating patient-centered features, and improving data sharing. Conclusions: A digital SSED score could support GPs by integrating patient-entered data, generating automated alerts, and enabling pharmacy-based follow-up. Successful adoption is likely to depend on GP workflow integration, system interoperability, and aligned incentives. Involving stakeholders early is essential to ensure that new digital tools meet the needs of both individuals with T2D and GPs.


  • Open Access

    Protocol for the REVELIO test-track pilot study: a randomised, controlled, single-centre trial in healthy recreational cannabis users investigating real-time in-vehicle detection of cannabis-impaired driving

    by Robin Deuber, Michal Bechny, Christoph Heck, Joel Brügger, Matthias Pfäffli, Mia Jovanova, Elgar Fleisch, Felix Wortmann, Wolfgang Weinmann

    Background: Driving under the influence of cannabis is associated with impaired cognitive and psychomotor performance and an increased risk of traffic accidents. Reliable real-time in-vehicle systems for detecting cannabis-related driving impairment are currently lacking; but hold great potential for improving road safety. Methods: This protocol describes the REVELIO test-track pilot study: a randomised, controlled, open-label, interventional, single-centre trial. The study assesses the feasibility and methodological requirements for developing and evaluating a multimodal in-vehicle detection approach using vehicle and driver state data. A total of 45 healthy recreational cannabis users will be enrolled and randomly allocated to an intervention or a reference control group. During the main study day, all participants will undergo biological sampling for tetrahydrocannabinol (THC) and related metabolites, as well as pre-driving assessments, followed by a sober baseline driving session on a closed test track using a dual-pedal vehicle with a certified driving instructor onboard. Participants in the intervention group will then receive a single controlled inhalative cannabis dose (target 0.67 mg THC per kg body weight), while the reference group will receive no cannabis. All participants will subsequently complete three additional standardized 50-minute driving sessions at predefined time points up to approximately six hours after administration, following identical schedules to enable within- and between-group comparisons. Between driving sessions, structured breaks will include recovery periods, repeated biological sampling, and traffic-medical, traffic-psychological, and pre-driving performance assessments, to characterise the temporal dynamics of cannabis-related impairment. Discussion: Multimodal data will be collected, including vehicle controller area network (CAN) data, driver monitoring camera (DMC) data, physiological signals using wearables, and biological samples (capillary blood, breath, oral fluid. Machine-learning–based models will be developed and evaluated to distinguish sober from cannabis-influenced driving states under controlled conditions. Secondary analyses will examine changes in driving performance over time and associations between functional measures and biological THC concentrations. As an exploratory pilot study conducted on a secured test track, the protocol aims to generate standardized reference data and quantitative performance metrics to inform both feasibility and system design considerations. Ethics and trial registration: The study was approved by the Cantonal Ethics Committee Bern, Switzerland (BASEC ID: 2025-01590) and is registered at ClinicalTrials.gov ( NCT07401628 ).

  • Open Access

    Integrating Prevention Into Health Systems: A Prevention-first Framework for Noncommunicable Diseases

    by Wasu Mekniran, Jakob Robert, Jacqueline Mair, Kai Gand, Hannes Schlieter, Oscar Castro, Marcia Nißen, Panitda Huynh, Odile-Florence Giger, Anja Yvonne Bischof, et al.

    The global healthcare system confronts a growing crisis due to the rising prevalence and escalating costs of noncommunicable diseases (NCDs). Although lifestyle intervention and targeted public health initiatives can considerably prevent NCDs, prevention efforts consistently receive inadequate funding, as healthcare budgets predominantly prioritize treatment. This article argues that healthcare systems can systematically address integration friction. We propose a multidisciplinary "Reach, Retain, Report, and Reimburse" framework, developed from existing prevention models and focus group discussions, to integrate NCD prevention effectively into current healthcare systems. This analytical lens recommends customizing the reach of prevention programs for at-risk populations, sustaining their engagement, developing relevant outcome reports, and reforming reimbursement structures to incentivize prevention. Ultimately, this multi-sector, interdisciplinary framework offers decision-makers a structure to expand access, sustain engagement, measure outcomes, and align reimbursement.

  • Open Access

    An LLM-based Motivational Interviewing Conversational Agent for Health Behaviour Change: Comparative Evaluation Study (Preprint)

    by Akshaye Shenoi, Tianze Li, Ahmad I. Jabir, Amanda Pitkethly, Elgar Fleisch, Tobias Kowatsch, Jacqueline L. Mair

    BACKGROUND Motivational interviewing (MI) is an effective approach for supporting health behaviour change, but face-to-face delivery is resource-intensive and difficult to scale. Rule-based Conversational Agents (CA) can improve access, yet their scripted interactions and limited language flexibility constrain MI delivery. This study evaluates the feasibility of using Large Language Models (LLMs) to deliver text-based MI coaching sessions. OBJECTIVE To describe the development of an LLM-based MI CA “Aimi”, and to evaluate its performance alongside a human coach delivering text-based MI and a rule-based MI-inspired CA in a physical activity behaviour change context. METHODS We developed Aimi using structured LLM-workflows that aim to enhance MI fidelity. We conducted a within-subjects study, where 18 adults interacted with (i) Aimi, (ii) an MI-trained human coach and (iii) a rule-based CA during live text-based role-play coaching sessions, in a randomised order. Transcripts were independently evaluated by an MI expert using the Manual for the Motivational Interviewing Skill Code Version 2.0 (MISC-2) to assess MI competency and fidelity. Participants completed a user experience questionnaire to provide general feedback and to assess session alliance, dialogue relevance, empathy, engagement, linguistic quality, and perceived motivation to change. Qualitative feedback was thematically summarized and categorized under strengths and weaknesses for each approach. RESULTS Across MISC-2 summary metrics, Aimi achieved higher fidelity scores than the trained human coach and rule-based CA, showing higher reflection-to-question ratios, more complex reflections, and greater elicitation of client change talk (91% vs. 71%). User experience ratings showed no significant differences across conditions. Qualitative analysis revealed distinct strengths and limitations across the coaching interactions: participants describing Aimi’s interactions as personalized, fluid, and adaptive, though sometimes overly reflective and lengthy; the human coach was viewed as empathetic and supportive but slow to respond; and the rule-based coach was viewed as efficient and structured, yet limited in depth and personalisation. CONCLUSIONS This study demonstrates that LLM-based CAs can deliver MI with fidelity comparable to trained coaches while more reliably eliciting client change talk than trained humans or rule-based systems. The perceived robotic quality and response style represent important areas for future refinement. LLM-based CAs orchestrated through structured workflows offer a scalable pathway to deliver MI for health behaviour change.