Toward AI-Driven Precision Measurement of Cognition, Behavior, and Psychological Function in Aging and Alzheimer’s Disease and Alzheimer’s Disease-Related Dementias

Abstract

The National Institute on Aging (NIA) is at the forefront of leveraging advances in artificial intelligence (AI) to better understanding of aging and the diagnosis and treatment of Alzheimer’s Disease (AD) and Alzheimer’s disease-related dementias (ADRD). Recent NIA-supported projects have highlighted the transformative potential of AI, digital health, and computational approaches in improving the modeling, detection, and monitoring of changes in healthy aging and AD/ADRD. This perspective is forward looking, emphasizing key areas and efforts in AI-driven precision measurement in cognition, behavior, and psychological function.

Precision measurement is fundamental to the development of personalized diagnostics and the subsequent tailoring, and delivery of interventions that address what matters most to people as they age and in the context of Alzheimer’s disease and Alzheimer’s disease-related dementias (AD/ADRD). This includes predicting early neuropsychological and behavioral changes before clinical diagnosis of AD/ADRD and consideration of both benefits and harms in disclosure of early risk/resilience and future change. There are limitations in the tools currently used for early prediction of cognitive, behavioral, and functional changes. For example, traditional neuropsychological assessments can take hours, often require travel to a research site for in-person administration by trained staff, have floor and ceiling effects that reduce precision across ages and stages of disease, and in the case of clinical evaluations, there are often long wait lists to be seen by a professional. Precision measurement is the diagnostic paradigm, which can enable early detection to help us identify sensitive and critical periods for timely interventions that may improve quality of life, delay the onset of AD/ADRD, and support adaptive aging by tailoring lifestyle, enhancing neuroplasticity, modifying disease risk factors like amyloid deposition in the brain, and delaying or preventing functional decline. Digital assessments and health technologies combined with advances in artificial intelligence (AI) and computational modeling could revolutionize our ability to detect, predict, and track intraindividual change patterns in neuropsychological function and behavior to improve adaptive aging, disease diagnosis, personalization of intervention, and response to intervention.

This special issue was developed, in part, as a response to the Landscape of Early Neuropsychological Changes in AD/ADRD Project (https://www.nia.nih.gov/research/dbsr/workshops/landscape-early-neuropsychological-changes-ad-adrd-webinar-series), led by the National Institutes on Aging. The project was launched in September 2022 where experts convened for a series of webinars to discuss gaps in measuring early changes in AD/ADRD. These webinars identified constructs and measures that characterize early changes in AD/ADRD beyond cognition alone and discussed the need to develop precision measurement techniques that yield new meaningful endpoints for AD/ADRD prevention trials. One high-priority need was defining a roadmap for digital measurements and biomarkers using AI-driven measurement approaches. Although there is tremendous enthusiasm for the potential AI brings to realizing the goal of precision measurement, there are also challenges to achieving these goals, such as limitations in AI model access; bias and explainability of AI models; lack of an AI regulatory framework; unclear benchmarks to compare newer methodologies with traditional research methods; lack of validation (e.g., predictive, clinical); gaps in AI-ready data, especially in representative samples and longitudinally collected datasets; unclear utility for use in real-world settings; and need for collaborative efforts between researchers, clinicians, policy-makers, and industry partners. We hope future research efforts can focus, in part, on addressing these difficult challenges in addition to leveraging exciting AI advances.

Key Research Gaps, Opportunities, and Measurement Needs

• 1. Assessment of neuropsychological change across the lifespan.

  • a. Incorporating codesign or participatory design principles by involving the people who will be impacted by the solutions throughout the development process.

  • b. Development of new measures and methods that account for diverse cultural, linguistic, and sociodemographic backgrounds and lived experiences and inclusion of diverse longitudinal samples that begin earlier in the lifespan, capture comorbidities, and integrate neurobiological, psychological, and social and behavioral data types.

  • c. Expand native measurement beyond “cold cognitive” functioning to capture social cognition, emotional processing, reward/motivation, effort mobilization, and decision-making.

  • d. Identify strategies for integration of multimodal data, including curation and harmonization of large, well-annotated datasets, creating standards, and multimodal fusion techniques, including the use of innovative psychometric and computational approaches.

  • e. Validate innovative approaches to track intraindividual change across a range of timescales.

  • f. Measuring what matters most to participants and their care team.

  • g. Ethical, social, and regulatory considerations, especially for newer measurement approaches such as predictive models that raise questions about consent, communication of risk, psychological impact, and insurance or employment discrimination.

• 2. Subjective cognitive change/decline (SCC/SCD)

  • a. Compare measures of SCC/SCD (e.g., everyday, dyadic, brief) for association with biomarkers, longitudinal cognitive and behavioral assessments, and functional and clinical outcome measures.

  • b. Identify factors that influence SCC/SCD measurement across communities.

  • c. Develop best practices for participant engagement.

• 3. Digital phenotyping and measures

  • a. Conduct community-partnered participatory research, including user interface and user experience projects.

  • b. Adopt privacy and safety standards, as well as clear data-use agreements.

  • c. Develop standards for validation, including outcome definitions and probabilistic thresholds.

• 4. Neuropsychiatric symptoms (NPS), behaviors, and socioemotional function

  • a. Apply a multidisciplinary approach combining psychometric expertise, computational methods, and repeated measurement of affect/behavior in longitudinal preclinical cohorts.

  • b. Develop new methods and digital measures for automation, remote assessment, innovation in data processing, and comparability over time.

  • c. Plan a workshop focused on objective physiologic and behavioral biomarkers of Mild Behavioral Impairment and NPS.

• 5. Clinical and real-world measures

  • a. Compare different novel measures related to AD/ADRD onset and biomarkers to determine what matters most to participants and their care teams, including multidimensional assessment (e.g., patient report, biomarker, cognition, motor, functional, social, emotional).

  • b. Create a new architecture for health data aggregation.

  • c. Develop regulations, guidelines, and shared policies.

Future Directions: A Roadmap for AI-Driven Precision Measurement in Cognition, Behavior, and Psychological Function Development

To guide future efforts, this special issue will continue the conversation on building a roadmap for AI-driven measurement in cognition, behavior, and psychological function to advance precision aging and AD/ADRD research. By leveraging advances in AI, computational modeling, and digital phenotyping and remote assessments, researchers can identify subtle cognitive and behavioral changes due to aging and AD/ADRD. Combining cognitive, behavioral, genetic, biomarker, clinical, and real-world data enables a multidimensional approach to understanding change in the context of aging and AD/ADRD. Recently released tools like the Mobile Toolbox (https://mobiletoolbox.org) already provide high-frequency, ecologically valid cognitive assessments, which could be extended to other measurement types. These tools can be combined with other innovative methods like passive data collection and sensor-based technologies to provide rich, multidimensional data, but require evaluation for reliability, scalability, predictive, and clinical validity for use in proactive detection of meaningful change. AI moves us closer to the goal of precision diagnostics by extracting features from complex data (i.e., multidimensional, multimodal, multilingual) to help identify patterns in moment-to-moment changes in cognition, behavior, and psychological function in the real-world context. To realize this goal, we also need representative data. However, there are concerns about bias and limited generalizability in existing data and a need for large, harmonized data sets, rigorous validation, and prospective data collection to establish both diagnostic and predictive utility. Once these measurement approaches are validated, we can more accurately measure and model cognitive, behavioral, psychological, and functional trajectories from early adulthood to late life to help identify critical periods for intervention. We can then integrate these AI solutions into clinical workflows, aiding early screening, informing intervention design and decisions, and potentially guiding treatment trials for disease-modifying therapies. However, if these tools are validated in people and contexts that do not generalize, it could lead to consequential errors in clinical decision-making, especially for individuals disproportionately impacted by age- and disease-related change.

As the adoption of digital measures and biomarkers, AI, and computational modeling increases, it is critical to engage individuals who are underrepresented in conducting aging and AD research. NIA initiatives like the Butler–Williams Scholars Program promote diversity in the scientific workforce and ensure research benefits underserved populations. Using programs to train the next generation and upskill the current scientific workforce and creating new data challenges like PREPARE (National Institute on Aging [NIA], 2023) can better prepare and engage communities in aging and AD/ADRD research efforts. This includes engaging participants, care partners and teams, and diverse communities to ensure measures reflect real-world needs. Collaborative events like the NIA Computational Approaches to Aging and AD/ADRD Workshop (https://www.nia.nih.gov/research/dn/workshops/workshop-harnessing-computational-approaches-advance-aging-and-ad-adrd) can connect AI and computational scientists with the aging and AD/ADRD research community.

Open-source models and open- science principles are particularly important for advancing AI research in aging and AD/ADRD. Unlike traditional single-provider and proprietary models, open principles and practices promote equity and innovation through an open-source model that emphasizes collaboration and transparency. Single-provider refers to a centralized system where all services and assessments are exclusively offered and controlled by one organization, entity, or resource. Open software tools improve the quality and sustainability of research software in the face of ongoing technical enhancements and allow community development responsive to rapidly evolving needs. Open practices promote the sharing of data, methods, and results, enabling other researchers to replicate and build upon existing work. It includes open measures and scoring algorithms, as well as open, but protected data. Transparent practices for data privacy, informed consent, and algorithmic fairness are prioritized to mitigate biases and protect patient information. Expanding efforts to harmonize data across international cohorts and enable cross-cultural research.

Conclusion

NIA has been a leader in the use of technology to advance research in adaptive aging and AD/ADRD. We now have an opportunity to continue to lead during this period of tremendous growth in AI that promises to transform the landscape of aging and AD/ADRD research. Integrating cutting-edge AI, computational, and digital measurement technologies into the research pipeline, teaming across scientific and clinical disciples (e.g., genetics, neuropsychology, psychology, sociology, anthropology, kinesiology, neurology, psychiatry, engineering, data science, bioinformatics), and fostering an open, inclusive scientific practice can create opportunities for innovation in precision measurement. By addressing key methodological gaps, leveraging AI and computational tools, and prioritizing ethical considerations, NIA is paving the way for next-generation measurement strategies for aging and AD/ADRD. The focus remains on ensuring that these advancements benefit the global community, especially those historically underserved.

Funding

None.

Conflict of Interest

None.

Disclaimer

The content is solely the responsibility of the authors and does not represent the official views of the National Institutes of Health or federal government.