Nov 2026: Igor Matias PhD Defence
6 November, 9h, PhD Defence of Igor Matias titled: “Providemus alz: Ubiquitous Preclinical Alzheimer’s Disease Screening”.
PhD Candidate Igor Matias
Thesis Title “Providemus alz: Ubiquitous Preclinical Alzheimer’s Disease Screening”
Abstract
Brain health encompasses multiple dimensions of human functioning and evolves throughout adulthood under the influence of aging, health, lifestyle, and environmental conditions. Cognitive performance and affective functioning represent two central and complementary dimensions of brain health, whose longitudinal characterization is increasingly relevant in the context of population aging and the growing burden of dementia. Alzheimer’s disease, the most common cause of dementia, presents a particularly important challenge because its underlying pathological processes may begin many years before clinical symptoms become apparent, during a prolonged and heterogeneous preclinical phase in which individuals can remain cognitively healthy and function independently.
Understanding brain health before clinically apparent impairment requires moving beyond isolated assessments toward the longitudinal characterization of cognitive and affective functioning in everyday life. Conventional clinical and laboratory evaluations provide standardized measurements of these constructs, but they are generally conducted infrequently and capture only specific moments. Consequently, they may overlook gradual, fluctuating, or individual-specific changes emerging between assessments. Consumer-grade smartphones and wearable devices offer an opportunity to complement these evaluations by continuously and unobtrusively capturing behavioral, physiological, and environmental signals in naturalistic settings. Such observations may support the development of digital biomarkers, defined as objective, quantifiable measures derived from digital data that reflect or indicate biological, physiological, behavioral, or health-related processes. However, despite evidence linking everyday-life digital signals to selected psychological outcomes, it remains unclear whether passive sensing can support a broad, integrated, and longitudinal characterization of cognitive and affective functioning, whether its predictive informativeness varies across individuals, and whether it can provide information about future brain health.
This thesis addresses these questions through the \textit{Providemus alz} study, a remote longitudinal observational study in which community-dwelling adults aged 45 years and older were monitored for brain health outcomes for up to 24 months. The study required the development and deployment of a dedicated research infrastructure integrating a cross-platform mobile application, consumer-grade wearable sensing, automated data acquisition and quality monitoring, and procedures for sustained participant engagement. Participants continuously wore a consumer-grade smartwatch and completed repeated cognitive and affective assessments through a custom smartphone application approximately every three months. The resulting dataset integrated physical activity, sleep, heart rate, weather, air quality, participant characteristics, and study-engagement information with a broad range of validated patient-reported and performance-reported outcomes. A systematic review and a sequence of empirical and methodological studies examined the feasibility, informativeness, individual-level model reliability, temporal interpretation, individualization, and prospective value of these data.
Taken together, the findings show that consumer-grade technologies can support sustained ecological observation and that everyday behavioral, physiological, and environmental signals contain information associated with multiple dimensions of cognitive and affective functioning. Their informativeness was nevertheless neither uniform nor interchangeable. It differed across psychological constructs, individuals, temporal windows, and modeling objectives. Affective functioning was generally more closely reflected in contemporaneous daily-life signals (e.g., heart rate patterns and physical activity duration), whereas cognitive functioning was comparatively more difficult to model and exhibited different temporal patterns (with, for example, sleep characteristics and environmental exposures). Furthermore, passive observations supported the joint forecasting of cognitive and affective outcomes at multiple future assessment horizons up to six months, extending the potential of ecological sensing from estimating contemporaneous functioning to anticipating individual patterns of future psychological functioning.
A central implication is that population-based average model performance is insufficient for individualized brain-health assessment. Participants differed substantially in how reliably their outcomes could be modeled, and higher- and lower-reliability profiles were associated with individual characteristics, health status, behavior patterns, and study engagement. Predictability should therefore be regarded as an individual- and outcome-dependent property to be evaluated explicitly, rather than assumed from population-level performance. A model that performs adequately on average may remain unreliable for particular individuals, precisely those for whom its outputs would need to be interpreted cautiously.
The thesis also clarifies what potential passive digital biomarkers can and cannot represent. Wearable and contextual signals do not directly measure cognition or affective functioning. Rather, they provide indirect observations of the behavioral, physiological, and environmental conditions through which these constructs may be expressed. The validity of a potential digital biomarker is not an intrinsic property of a sensor or feature, but depends on its demonstrated relationship with a defined psychological construct within a specific population, temporal context, and analytical framework. Potential passive digital biomarkers should consequently be understood as construct-specific, context-specific, individual-specific, and temporally bounded indicators.
This interpretation positions passive sensing as complementary to, rather than a replacement for, validated cognitive, affective, biological, and clinical assessments. Repeated active assessments provide psychological anchors, while passive technologies characterize the periods between them at a temporal density that conventional visits cannot achieve. Their integration may support, in the future, a means of distinguishing stable individual functioning from temporary fluctuation, recovery, gradual change, or sustained departure from an individual baseline. It also provides the foundation for interpretable approaches that identify when changes in everyday behavior precede or accompany later changes in brain-health outcomes, without treating predictive associations as causal mechanisms or clinical prescriptions.
Overall, this thesis advances a shift from isolated assessment scores and population-level thresholds toward continuous, multidimensional, and individualized brain-health trajectories. It does not demonstrate the detection of preclinical Alzheimer’s disease, establish causal mechanisms, or provide a clinically deployable screening system. Instead, this thesis establishes several prerequisites for future research seeking to identify early neurodegenerative deviation from everyday-life data: feasibility of the sustained ecological observation, repeated anchoring to validated outcomes, multidimensional modeling, participant-aware evaluation, explicit assessment of individual-level model reliability, uncertainty assessment, temporally structured interpretation, and prospective analysis. Future validation in larger, more diverse, clinically heterogeneous, and biomarker-enriched populations will be necessary to determine whether these individual brain health trajectories can reveal meaningful deviations from expected functioning and complement established approaches to preclinical Alzheimer’s disease research.
PhD project: providemus.unige.ch/
The Thesis Jury
Chairperson: Prof. Dr. Marcel Paulssen, University of Geneva, Geneva, Switzerland
Prof. Dr. Jakob Eyvind Bardram Technical, University of Denmark, Kongens Lyngby, Denmark
Prof. Dr. Nuno M. Garcia, University of Lisbon, Lisbon, Portugal
Prof. Dr. Dimitri Konstantas, University of Geneva, Geneva, Switzerland
Academic Supervisors
Prof. Dr. Katarzyna Wac, University of Geneva, Geneva, Switzerland
Prof. Dr. Matthias Kliegel, University of Geneva, Geneva, Switzerland
Thesis Defense Location/Time, Battelle D, Auditorium, 9h
ZOOM link https://unige.zoom.us/s/68384586026
Passcode 915817