No description available.

<p dir="ltr">The use of connected health technology is becoming increasingly significant in the field of healthcare. Artificial Intelligence- augmented workflows connected to treatment guidelines promise more inclusive care delivery. The AI/ML-based connected health informatics plays an integral role in every stage of medical product development, from initial discovery to providing guidance to healthcare providers, and finally to delivering patient care. The exponential growth of meta data and the rapid advancement of connected health technologies provide greater opportunities for novel healthcare solutions, delivery mechanisms, and clinical trial designs.</p><p dir="ltr">However, it poses complexity of the AI/ML-specific challenges besides all the challenges SaMD products face. The regulations for AI/ML-based connected solutions have yet to mature. The AI/ML SaMD development process requires additional considerations such as data quality and management, continuous deployment, and validation.</p><p><br></p><p dir="ltr">This study delves into the integration of Machine Learning (ML) with medical software devices and how the current lifecycle models fit the needs of the AI industry. AI/ML-based SaMD development process artifacts are identified through the theory and AI/ML SaMD regulations and standards requirements. Moreover, this study analyzes collected data from interviews, surveys, and an experimental case study to identify success factors in building quality and agility for AI/ML-based SaMD development projects.</p><p dir="ltr">Incorporating of Artificial Intelligence (AI) in healthcare requires continuous deployment and validation processes, which may not be in line with the current workflow, capability, or authority of regulators. This research also highlights that model governance and technology access can be key challenges in implementing AI/ML development process artifacts, especially when integrated into connected health solutions.</p><p dir="ltr">This work sets the foundation for future research to reduce bottlenecks in the machine-learning process. The focus should be on aiding model governance to streamline development and ensure machine reliability. A suitable software toolchain is necessary for exploratory data analysis, data integration, documentation, model governance, monitoring, version control, and integration with other software and services within a connected health solution. Additionally, conducting more focused research on security and privacy in the context of connected health would be valuable.</p>

Background: Functional Assessment (FA) in older adults is an important measure of their health status. FA using Activities of Daily Living (ADL) is a strong predictor of health outcomes, especially as we age. With the development of increasingly-connected health, we have a new opportunity for more robust and improved FA. Objective: The objective of this thesis is to collate and discuss published evidence on FA predictors and how the FA predictors can be collected using the paradigm of Connected Health (CH) architectures through an industrial case study in CHAPTER 5: INDUSTRIAL CASE STUDY. Methods: The method is to do two Systematic Literature Reviews (SLRs). The two SLRs were undertaken with Preferred Reporting Items for Systematic Reviews and Meta-Analyses statement (PRISMA) and Parsifal, an online tool for SLR. This thesis catalogs various FA and state-of-the-art Software Engineering Architectural Tactics and Styles (SEATS) used within Connected Health (CH) that focus on ADL. The results of the cataloged information were used in the industrial case study where some of the FA predictors were automated. Articles obtained from the data source during the SLRs were filtered based on the titles, abstracts, full-text provision, English language literature, including age, which must be sixty-five years and above. Another reviewer was also included in this study, while all the defined inclusion and exclusion criteria detailed in this thesis were applied. Information about FA via ADL were extracted from the articles with further extraction on the SEATS used for computer-supported FA during the industrial case study. Data Source: During the SLRs processes, database searched included PubMed, EBSCOhost, Engineering Village, IEEE Xplore Digital Library, and ScienceDirect. The conducted search contains both controlled terms called Medical Subject Headings(MeSH) such as activities of daily living and search strings such as functional assessment, older adults, geriatrics, seniors, elderly care, and aging. Results: From four hundred and ninety-five initial abstracts and titles, nineteen full-text journal articles were included in the final review for the SLR on FA predictors. Six full-text journal articles were obtained from the SLR on CH architectures after reading its 449 titles and abstracts. In the SLR on FA predictors, predictor metrics for FA via ADL were extracted from each of the articles. Gait speed, sleep quality, and movement activities were assessed as ADL predictor metrics for FA in older adults. Other FA predictors published involved self-reported metric scale measurement using Barthel-20 scale and performance-based scale through Timed-UP and Go test. This thesis reviewed each metric for sleep quality and movement activities. In the SLR on CH architectures, quick response of ADL and resource efficiency such as sensors were some of the major tactics related to performance in Software Engineering (SE) quality in CH, while confidentiality and integrity of FA measures related to security in SE quality in CH was another major concern. Conclusion: Having conducted the two SLRs, a wide range of measures were used for FA in older adults, including consideration on the SEATS used for computer-supported FA. Overall, these FA measures and SEATS provide inexpensive and easy-to-implement FA. The diversity of the FA measures and SEATS contributes towards the development of computer-supported FA. However, future work is needed to consider the result of this study as an open-source computer-supported FA tool, and such tool should also be evaluated and verified through direct examination with older adults. / Graduate

Face à la multiplication exponentielle du nombre de personnes souffrant d’une maladie chronique d’origine comportementale (e.g., tabagisme, alcoolisme, mauvaise alimentation, sédentarité), des interventions non médicamenteuses (INM) agissant sur ces comportements modifiables sont devenues incontournables en prévention et en complément des traitements. Parmi ces INM, les interventions numériques santé (INS) ouvrent un champ prometteur de changement durable de comportement de santé (e.g., objet connecté santé, application pour le téléphone, jeu vidéo). La thèse s’intéresse, au-delà de leur ergonomie et de leur fonctionnalité, à leur évaluation santé, de leur validation à leur surveillance. La première étude recense les modèles proposés dans le monde pour évaluer ces INS et les catégorise selon leurs paradigmes épistémologiques sous-jacents. Les résultats montrent une augmentation exponentielle de ces modèles et une absence de consensus ou de convergence vers un modèle comme cela a été le cas dans le médicament à la fin du XXème siècle. La deuxième étude s’appuie sur une revue systématique ayant identifié 90 essais interventionnels publiés testant les bénéfices et les risques de solutions numériques visant à lutter contre le tabagisme. Les résultats montrent une efficacité de certaines INS sur le tabagisme mais issue d’un corpus méthodologique très hétérogène limitant la portée des conclusions. Cette hétérogénéité est liée aux caractéristiques intrinsèques des INS (e.g., technologies utilisées et combinaison entre elles, multiplicité des théories du changement de comportement utilisées), aux méthodes d’évaluation utilisées (e.g., type de groupe contrôle, durée de suivi) et aux critères de jugement choisis (e.g., réduction du tabagisme ou arrêt). La discussion porte sur les limites actuelles dans la mise en évidence de l’efficacité et des risques des INS à cause d’approches paradigmatiques parallèles, le paradigme biomédical, le paradigme d’ingénierie et le paradigme comportemental. Le manque de consensus limite la comparabilité et la reproductibilité des résultats des études testant ces solutions numériques de santé. Elles restent pour la plupart à ce jour des solutions gadgets malgré un potentiel prometteur et promis par les industriels. La thèse encourage à la convergence vers un modèle consensuel de détermination du bénéfice/risque santé fondé sur la science de chaque INS et fait des propositions en ce sens. / To deal with the exponential increase of chronic diseases caused by health behavior (e.g., smoking, alcoholism, unhealthy eating, physical inactivity), non-pharmacological interventions (NPI) have become essential as a prevention tool and as a complement to treatments. Among these NPIs, behavioral intervention technologies (BIT) open up a promising field to a sustainable change in health behaviour (e.g., connected health devices, smartphone health apps, serious games). Beyond their ergonomics and their features, this thesis focuses on their evaluation in health, from their validation to their surveillance. The first study identifies the existing frameworks proposed around the world to evaluate these BITs and categorizes them, based on their underlying epistemological paradigm. The results show an exponential increase of these frameworks and a lack of consensus or convergence towards a common framework, as it had been the case for the drugs, by the end of the twentieth century. The second study is based on a systematic review used to identify 90 published interventional studies evaluating the benefits and the risks of digital solutions to fight against smoking. The results show that some BITs are effective against smoking but their effectiveness is based on a heterogeneous methodological corpus limiting the significance of the results produced. This heterogeneity is related to the inherent characteristics of the BITs (e.g., employed technologies and combination of technologies, multiplicity of the theories to change health behavior), to the chosen assessment methods (e.g., kind of control group, follow-up time) and to the chosen outcome measures (e.g., smoking reduction, smoking cessation). The discussion is focused on the current limitations to demonstrate the effectiveness and the risks of the BITs., due to parallel paradigmatic approaches, the biomedical paradigm, the engineering paradigm and the behavioral paradigm. The lack of consensus limits the comparability and the reproducibility of the results of the studies evaluating these BITs. Most of them are still gadgets, despite a promising potential, as predicted by the manufacturers. This thesis promotes the convergence to a consensual framework to determine the evidence-based benefits and risks of each BITs and introduces proposals to this effect.