Response to: Refining the clinical interpretation of activity variability in cognitive impairment: The need for phenotypic specificity

Abstract

Dear Guo et al.,

We thank Guo et al.¹ for their interest in our article² and for providing innovative thoughts regarding clinical translation of our activity variability metric.

We would first like to emphasize that we positioned our work as an exploratory analysis of a conceptually novel metric developed using available accelerometry data. It is fit for application in future longitudinal and clinical studies with accelerometry data, which can replicate our findings and establish clinically meaningful changes relevant to activity variability and cognition. Our cross-sectional study suggests that low activity variability is strongly associated with cognitive impairment, yet we acknowledge that this metric requires further validation before implementation as a digital biomarker of cognitive decline and impairment. We intend to extend this work to other studies and encourage other researchers and clinicians to pursue validation studies, which were beyond the scope of our analysis.

We agree with Guo et al.¹ that cognitive impairment is heterogeneous and that it would be useful to distinguish subtypes of dementia, such as vascular versus Alzheimer's disease pathologies. The data we used have strengths and limitations, as noted in our article. As Guo et al. mention, and we discussed in our article, the National Health and Aging Trends Study (NHATS) dementia classification criteria are not equivalent to a dementia diagnosis, nor do they distinguish subtypes of cognitive decline and impairment.² As Guo et al. compellingly state, it is important to further extend the conceptual link between activity variability and cognitive risk to clinical neurological outcomes, not limited to dementia (e.g., traumatic brain injury). Guo et al.’s proposal to link activity variability to specific brain regions via neuroimaging and associated cognitive domains via neuropsychological testing presents exciting additional mechanistic areas of research, which would greatly complement our findings. We agree that identifying underlying brain regions and cognitive domains that may be specifically related to activity variability would enhance this metric's clinical utility for greater sensitivity in detection of cognitive and functional impairment. In addition, if activity variability can be linked to specific brain regions, it may serve as a potential intermediate outcome for cognitive interventions.

Guo et al. commented that, because activity variability and gait speed were correlated, we cannot disentangle “the cognitive versus biomechanical determinants of variability”. Importantly, our study was not intended to examine the cognitive versus biological determinants of activity variability. Although activity variability and gait speed were correlated, activity variability remained strongly associated with cognitive impairment even after controlling for gait speed in our analysis (Model 4 in Table 2), suggesting that activity variability is independently associated with cognitive impairment. Moreover, we do not advocate that activity variability reflects a specific neurological or causal pathway. Section 4.3 of our article (“Possible mechanisms”) discusses potential hypotheses not specified a priori, which may explain the observed associations in our article. The goal of our work was to examine whether activity variability may be a non-invasive and valuable functional marker of cognitive impairment. As noted above, important future work by our team and others will need to apply this metric to longitudinal data to examine causal pathways and potential mediating mechanisms, which were beyond the scope of our cross-sectional analysis.

The multidimensional framework that Guo et al. mentioned appears interesting and posits a composition of several risk factors for cognitive decline, including activity variability. However, it is unclear whether they are proposing a biological framework for cognitive impairment, or are interested in designing a prediction model to identify patients at high risk for cognitive impairment. Either of these objectives may be useful, and we encourage Guo et al. and others to consider assessing activity variability in other cohorts with accelerometry data. In future work, it will be important to include detailed neuroimaging, biological risk factors (e.g., amyloid, tau, apolipoprotein E), and longitudinal follow-up for clinically adjudicated dementia outcomes. We hope that other researchers and clinicians see the promise of digital biomarkers–like activity variability–and will consider examining these biomarkers as informative metrics in dementia prevention, treatment, and care.

Jennifer A. Schrack is a consultant for Edwards Lifesciences. Michelle C. Carlson serves as a scientific advisory board member of AARP Staying Sharp. Patrick T. Donahue and Johannes Thru have no conflicts to disclose. Author disclosures are available in the supporting information.