Using Wearable Technology To Predict Cognitive Function in Patients With MCI

Mild cognitive impairment (MCI), a decline in an individual’s cognitive function that is greater than what would occur due to aging alone, impacts over 15% of older adults worldwide. A multi-institutional study, led by Dr. Yuri Rykov at Neuroglee Therapeutics and published in BMC Medicine, demonstrates the potential of using wearable technology to predict cognitive function in patients with MCI via physiological markers.

Lack of monitoring for patients with MCI

MCI is characterized by frequent lapses in thinking or memory during routine activities, which can manifest as consistently forgetting important people or events, struggling to maintain coherent thoughts and misplacing items on a regular basis. MCI often serves as a precursor for more severe conditions, such as Alzheimer’s disease and dementia. Traditional neuropsychologist tests for patients with MCI only capture clinical data during doctor visits, leaving gaps in data when the patients are not at the clinic. Wearable sensors, such as Fitbit, are already capable of collecting information including blood volume pulse, which can be used to calculate heart rate and heart rate variability (HRV). Previous research studies have shown a correlation between HRV metrics and cognitive function, specifically in executive function and episodic memory. Dr. Rykov’s team built on this research through this study in which they tested the ability of wearable sensors to predict cognitive function as a method to provide continuous monitoring of patients with MCI.

Using wearable technology to predict cognitive function

The 10-week clinical trial was designed for older adults, aged 50–70, including 30 individuals diagnosed with amnestic MCI and 10 age-matched cognitively normal individuals. During the trial, participants underwent regular digitally delivered multidomain therapeutic intervention sessions. They were asked to wear the Empatica E4 wrist-wearable device during these sessions, as well as during sleep and other routine activities. Cognitive performance was measured at baseline at the start of the trial and again at the end of the 10 weeks using the neuropsychological battery test (NTB), which includes assessments of executive function, processing speed, immediate memory, delayed memory and global cognition. The device recorded physiological data, such as blood volume pulse, electrodermal activity, acceleration and skin temperature. They utilized supervised machine learning to train models predicting NTB scores using digital physiological features and demographics. Out of the 30 individuals diagnosed with amnestic MCI, 27 completed the trial, and 7 out of the 10 controls completed the trial, with missing data being attributed to improper device usage or non-compliance.

The key findings from this study were that:

• Digital physiological features showed strong correlations with processing speed, executive function and global cognition composite scores.
• Measures of HRV, particularly cardiac sympathetic index (CSI) and the high frequency HRV (HRV-HF), correlated significantly with the cognitive composites.
• Predictive models that combined digital physiological features and demographics demonstrated the greatest predictability for executive function scores.

The potential of wearable technology for continuous cognitive monitoring

The research team found significant correlations between tracked physiological markers and key cognitive aspects such as processing speed, executive function and overall cognitive performance. These components of cognition are essential for everyday cognitive tasks, underscoring the potential of wearables for continuous monitoring of cognitive function in patients with MCI. HRV measures, in particular, showed promise in predicting cognitive performance.

Continuous monitoring offers clinicians a convenient way to gather patient data outside of the clinic. This can be especially advantageous after the implementation of a new treatment protocol or during periods of significant deterioration in patient health because it can fill in missing data gaps that would otherwise exist. If a patient were to come into the clinic with severe deterioration in cognitive function, physicians would have the ability to track changes specifically by looking at data from wearable sensors, which could help them point to a particular time or date at which these changes occurred.

Nevertheless, this study had limitations that could have impacted the findings. Firstly, the sample size is relatively small, with no control group that did not receive the intervention, and non-compliance led to the loss of potential data that could have contributed to the study. Specifically, only data from 17 individuals was viable to use after data cleanup. Secondly, the study period was relatively short, so long-term effects may have been missed. Additionally, given the setup of the study, factors such as the amount of physical activity or daily routine were not controlled for, which could have impacted the results, especially since many of the physiological markers, such as heart rate data, can easily be influenced by external factors. Furthermore, the study sample displayed a lack of ethnic variation, which makes the results less generalizable to a larger population.

Developing a more controlled study

Further research is needed to control for more external factors and ensure that these markers are reliable predictors of cognitive function. Moreover, a controlled study would allow for greater exploration of potential confounding variables, thereby enhancing the validity and generalizability of the findings. The authors acknowledge that the achieved accuracy of the predictive models is currently below desirable levels, preventing the replacement of standardized cognitive tests with wearable-based physiological measurements as they are. However, with further work, this approach shows promise for improving care for patients with MCI.

Reference: Rykov YG, Patterson MD, Gangwar BA, et al. Predicting cognitive scores from wearable-based digital physiological features using machine learning: data from a clinical trial in mild cognitive impairment. BMC Med. 2024;22(1):36. doi:10.1186/s12916-024-03252-y

Have you recently published research?

You can raise awareness of your work with an article like this too.

Find out more