ASAntoine Saab

Antoine Saab

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

Antoine Saab has a total of 43 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    AI-Enabled Early Detection of Epidemic Threats in a Resource-Constrained Health System: Implementation Research from Lebanon

    by Antoine Saab, Christo El Morr, Elie Salem Sokhn

    Background: Global South countries face heightened infectious disease risks compounded by fragmented surveillance systems and resource constraints. Lebanon’s health system crisis and recurrent outbreaks highlight the urgent need for new approaches to epidemic and pandemic preparedness. Objective: To describe the design, implementation process, and early implementation findings of an artificial intelligence (AI)–enabled early warning system (AI-EWS) that automates detection of epidemiologically relevant events using laboratory data in Lebanon, with emphasis on feasibility, workflow integration, stakeholder engagement, and implementation learnings. Methods: We applied an implementation research approach to co-design and validate an AI-EWS prototype integrated within a university hospital laboratory. The system incorporates a hybrid statistical and machine learning analytic pipeline and a real-time dashboard for detecting epidemic events across multiple pathogens using routine laboratory information systems (LIS) data. The system was developed in collaboration with the Ministry of Public Health (MoPH) to support future national adoption. Findings: The project produced a structured framework for pathogen selection, rule development, dataset creation, and platform co-design with clinical and epidemiological experts, medical informatics professionals, MoPH experts, and technical specialists. Early implementation observations indicate operational feasibility within existing infrastructure, strong stakeholder engagement, and trust-building around AI-enabled surveillance through participatory design and responsible-AI governance. Conclusions: AI-enabled laboratory surveillance is a viable and scalable approach to strengthening epidemic preparedness in low-resource health systems when grounded in implementation science and responsible AI, and when explicitly aligned with local workflows and national policy priorities.

  • Reclaiming Preparedness: AI, Justice, and the Futures We Choose

    by Christo El Morr, Antoine Saab, Elie Salem Sokhn

  • Open Access

    A protocol model for simulation trials evaluating clinical decision support systems

    by Jean-Baptiste Lamy, Cynthia Abi Khalil, Sophie Dubois, Hector Falcoff, Pierre Meneton, Akram Redjdal, Brigitte Séroussi, Antoine Saab

    Objective: Clinical Decision Support Systems (CDSSs) are computerized systems that help clinicians to make medical decisions. The evaluation of these systems through randomized clinical trials involving real clinicians and patients is very long and complex to set up, and the system should be tested for security first. On the contrary, simulation trials consist of evaluating the system with real clinicians but fictitious patients or simulated cases. While simulation trials cannot replace clinical trials, they are simpler to perform, permit a better control of the patient profile tested and can provide a higher statistical power.Methods: We analyzed lessons learned from two recent simulation studies, and the corresponding literature, in order to derive a model of protocols for simulation trials evaluating CDSS. Then, we tested our proposal in a new study.Results: We propose a model of protocols for evaluating a clinical decision support system during a simulation trial, based on a crossover design that balances interventions, fictitious patients, and orders of intervention and patients, and controls or measures several common biases including automation bias. Our proposal was subsequently successfully implemented in a third study. We also provide general recommendations for simulation trials and an R code template for statistical analysis using linear mixed models.Conclusions: we proposed a protocol model for simulation trial evaluating CDSS. This proposal should be adaptable to almost any clinical decision support system.

  • Open Access

    AI-based epidemic early warning using social media and search engine signals: A scoping review

    by Amir Bayat, Ayissha Pavakopethan, Nadine Kashmar, Iman Yousuf, Antoine Saab, Elie Salem Sokhn, Yahya El-Lahib, Christo El Morr

    Background: Early detection of infectious disease outbreaks is essential for timely public health response. Artificial intelligence (AI) combined with digital traces from social media and search engines may strengthen epidemic early warning systems (EWS). Objective: To map how AI-based EWS use social media and search engine signals for outbreak detection/forecasting, and to characterize whether studies integrate clinical and climate/environmental data. Methods: Following PRISMA-ScR, we reviewed 38 studies (past five years) across six databases. We extracted study characteristics, data sources, target diseases, prediction tasks, validation practices, and performance metrics. For cross-tabulated evidence maps, each study was coded once using its primary data-source category and primary modeling approach to avoid double-counting. Results: Deep learning and ensemble approaches were the most common methods. COVID-19 and influenza-like illness dominated the literature. Few studies combined social media and search engine signals. Clinical and climate/environmental data were rarely integrated. No study reported prospective, real-time evaluation. Geographic coverage was skewed toward high-income settings. Conclusion: AI-based epidemic early warning systems using digital signals show promise for outbreak detection and forecasting. However, the evidence remains largely retrospective, fragmented, and geographically uneven. Future research should prioritize multi-stream data integration, equity-aware design, and prospective validation to support real-world global applicability.

  • Détection, prédiction et prévention des évènements indésirables liés aux soins via l’intelligence artificielle, en associant approche basée sur les règles de décision et l’apprentissage automatique

    by Antoine Saab

    La sécurité des patients est considérée comme une grande priorité en santé publique. Les systèmes de santé au niveau mondial tentent de mobiliser des ressources pour la prévention des évènements indésirables cliniques chez les patients. Selon l'Organisation Mondiale de la Santé, des études ont montré qu'en moyenne un patient sur dix est sujet à un événement indésirable (EI) lors de son hospitalisation dans les pays à revenu élevé. L'estimation pour les pays à revenu faible et intermédiaire suggère même que jusqu'à un patient sur quatre sont concernés. Malgré une dynamique positive au cours des vingt dernières années, d'importantes opportunités d'amélioration subsistent, découlant souvent d'une mise en œuvre inefficace ou inadéquate des actions d'amélioration de la sécurité. Ainsi, les EI restent l'une des principales causes de morbidité et de mortalité en santé, entraînant des pressions économiques et sociales supplémentaires pour les systèmes de santé. L’analyse approfondie des raisons pour lesquelles l'incidence élevée des EI persiste semble identifier les facteurs suivants :- Le manque d'outils fiables et efficaces pour la mesure et la surveillance des EI cliniques. - L'incohérence dans l'utilisation de techniques de prévention recommandées - Le manque d'accessibilité aux techniques de prévention recommandées - Le manque d'utilisabilité de certaines solutions technologiques - L'existence de domaines cliniques pour lesquels il n'y a pas de solutions de prévention prouvées, dénommés les "angles morts" de la sécurité patient. L'intelligence artificielle (IA) détient un potentiel important qui peut être appliqué à l'identification, la prédiction et la prévention des EI pour les patients. En fait, elle peut être utilisée pour détecter les événements liés à la sécurité des patients, améliorer les performances des alarmes cliniques, prédire le risque de survenue d’EI et améliorer l’adhérence aux bonnes pratiques. D’autre part, quand bien même l’IA commence à générer des applications dans certains domaines cliniques, peu ont été développés pour le domaine de l’amélioration de la sécurité des patient et la grande majorité de ces applications sont encore au stade expérimental ou insuffisamment validées pour des scénarios d'utilisation en contexte réel.En fait, le développement d'applications basées sur l'IA pose de nombreux défis liés à la conception, à la validation et à l'acceptation par les utilisateurs qui doivent être relevés avant que ces applications puissent être transformées en systèmes d’aide à la décision clinique (SADC) ou d’amélioration. Dans le cadre de cette thèse, notre objectif est d'explorer le potentiel de l'IA pour le domaine d’application des événements indésirables cliniques. Après une présentation de l’état de l’art, notre point de départ sera un système d'outil de détection non basé sur l'IA pour la mesure automatisée des événements indésirables hospitaliers, conçu avant la thèse.Après avoir identifié et analysé les avantages et les inconvénients de ce genre de système, nous concevrons deux modèles basés sur l'apprentissage automatique, consacrés respectivement à la prédiction des réadmissions à l'hôpital et à la détérioration clinique des patients. Nous présenterons également un SADC basé sur l'IA qui associe le deuxième modèle de prédiction à des règles cliniques élaborées par des experts médicaux afin de gérer le risque de détérioration clinique.Finalement, nous analyserons les spécificités de l'utilisation de l'apprentissage automatique par rapport aux algorithmes basés sur des règles dans de telles applications, et comment ces deux techniques peuvent être utilisées en synergie.

  • Open Access

    Capabilities of computerized decision support systems supporting the nursing process in hospital settings: a scoping review

    by Cynthia Abi Khalil, Antoine Saab, Jihane Rahme, Jacques Bouaud, Brigitte Seroussi

    Background: The implementation of the Nursing process (NP) is a complex and cognitively demanding task that can be enhanced by using clinical decision support systems (CDSSs). While some CDSSs assist in the management of patients with specific patient conditions, nursing process clinical decision support systems (NP-CDSSs) are designed to support nursing decisions through all five sequential NP steps: assessment, nursing diagnosis (NDs), planning, intervention, and evaluation. Although an internationally agreed standard for NP-CDSS development (ICS4NP-CDSSs) was published in 2016, outlining recommended design and functional features for NP-CDSSs, current NP-CDSSs show limited compliance with these guidelines. Methods: A PubMed search was conducted to identify NP-CDSSs used in hospital settings and published between January 2013 and February 2024. The objective was to assess how well these systems adhered to ICS4NP-CDSSs including coverage of all NP steps, use of standardized nursing languages (SNLs), and compliance with recommended technical features. The search also examined whether NP-CDSSs had been implemented and evaluated and whether any outcome related to system impact or any nurses’ feedback on barriers and facilitators influencing NP-CDSS implementation was documented. Results: Of the 986 papers retrieved, 35 related to 21 systems met the eligibility criteria. Only seven systems addressed all NP steps, with the evaluation step remaining underdeveloped. Seventeen systems provided NDs based on entered characteristics or risk factors, 15 included automatic linkages between NDs and interventions, but only 10 included linkages to outcomes at the evaluation step. Most retrieved systems used NANDA-I/NANDA, NIC, NOC as SNL. Only four systems reached the post-implementation evaluation phase; none reported patient outcomes analysis, while four collected nursing outcomes. Barriers during implementation included a lack of trust and perceived usefulness to improve daily workflows, limited validation, and missing technical features. Conclusions: NP-CDSS development is still in its infancy. This review identifies a gap between ICS4NP-CDSS best-practice recommendations and their implementation in current NP-CDSSs, which still lack features like full coverage of NP steps and necessary linkages between them. To enhance NP-CDSSs, strategies should focus on reducing documentation burden, providing comprehensive education and support for nurses, and demonstrating the impact of NP-CDSSs through outcome studies.


  • Open Access

    Practical Approach for Evaluating Machine Learning Anomaly Detection Algorithms for Epidemic Early Warning Systems

    by Antoine Saab, Abdul Hamid Dabboussi, Cynthia Abi Khalil, Jihane Rahme, Elie Salem Sokhn, Christo El Morr

    Anomaly detection methods in time series data can play a pivotal role in epidemic surveillance Early Warning Systems (EWS). Statistical and rules-based methods have been traditionally employed in such systems, but are challenged by data dynamics and necessitate expert fine-tuning regularly. On the other hand, machine learning methods can handle complex and multidimensional data better, learn and adapt to changing patterns, and improve their performance. However, practical methodologies for their fitting and evaluation relative to gold standard data for infectious diseases epidemic surveillance are still lacking. In this study, a practical evaluation method was presented using an ensemble technique of four traditional statistical models to build the reference gold standard dataset, and results of validation of two machine learning (LSTM and Isolation Forest) relative to four pathogen data series (COVID19, Hepatitis C, Acinetobacter baumannii and Methicillin-resistant Staphylococcus aureus) was reported with promising results. Lessons learned can be useful in the perspective of adapting ML algorithms to epidemic surveillance EWS.


  • Open Access

    Optimizing ICU Care: Advancing Nursing Diagnoses with a Decision Support System

    by Cynthia Abi Khalil, Antoine Saab, Akram Redjdal, Jihane Rahme, Mouin Jammal, Brigitte Seroussi

    The Nursing Process (NP) is a multi-step, structured framework, that enables nurses to integrate scientific knowledge and optimize patient care. Establishing a Nursing Diagnosis (ND) is a fundamental step of this process, enabling the selection of appropriate interventions. This study presents the development of VIGIL Care, a clinical decision support system designed to enhance NP in intensive care units (ICUs). Using a Delphi method, 13 experts reviewed the 277 nursing diagnoses (NDs) from the latest NANDA-I taxonomy (2024–2026) and selected 32 as relevant for ICU care, and with high criticality or high prevalence. In a second step, a panel of three nursing experts and one internal medicine physician refined the diagnostic indicators associated to the 32 NDs as described in NANDA-I, selecting those deemed highly or moderately relevant. VIGIL Care integrates patient data and employs logical rules to propose patient-specific NDs, while offering an explainable user interface allowing for nursing decisional autonomy preservation. The assessment of 10 NDs across 20 randomly-selected patient cases demonstrated that VlGIL Care achieved 86% sensitivity and 78% specificity, outperforming traditional nursing assessments, especially under time constraints. However, the limited sample size highlights the need for further validation to confirm the system’s effectiveness.