Translating Science

Chatting with AI about Human Health

ChatGPT, HRV, and CGM

Photo by Andrea De Santis on Unsplash

Carter Williams, Managing Director and CEO iSelect Fund, recently posted on LinkedIn:

“HealthShift: Empowering People and Transforming Healthcare through Data

With a one-lead EKG and a CGM, can one interpolate that data (in conjunction with all the other data from watch sensors, food log, etc) in a time series to form a proxy for the degree of inflammation?

I think the path is through HRV and time series data on CGM in correlation to food log. And that Levels may have the first control arm dataset for this. Also, need more time series response data around the vagus nerve.”

LinkedIn post on HealthShift

I subsequently had a conversation with ChatGPT to write this reply:

It sounds like you’re suggesting to use a one-lead ECG to measure HRV to assess parasympathetic nervous system activity. Plus using the CGM for blood glucose levels. Plus getting user self-report on food consumption.

There’s good evidence from a recent meta-analysis that “HRV can be used to index the activity of inflammatory processes” (Williams et al., 2019). Approved as a strong index.

There’s a relationship between inflammation and glucose disorders (Brazilay et al., 2001, 2003). And conversely, elevated blood glucose levels can contribute to the development of chronic low-grade inflammation (Luc et al., 2019). There appears to be a bidirectional association.

ChatGPT suggests to use a multi-modal machine learning approach to integrate the time series response data from HRV and CGM to predict the level of inflammation.

Food log is going to give insight into degree of inflammatory foods consumed by a person.

ChatGPT acknowledges that adding data from a food log to the model for predicting inflammation can provide additional information about an individual’s diet, which has been shown to have an impact on inflammation levels. By incorporating food log data, the model can take into account the type and amount of food the individual has consumed, as well as the timing of meals. This can help improve the accuracy of inflammation predictions, as changes in diet can directly impact inflammation levels.

So, yes, these are known associations and you can train AI based on the autonomic (HRV), blood plasma (glucose), and behavioral (food intake) variables to predict degree of inflammation. Even better if you have repeated measurements as in a time series.

Photo by Jacopo Maia on Unsplash

But I have to agree with ChatGPT that some potential drawbacks of using a food log are:

1. Time-consuming: Recording every food item and its corresponding quantity can be time-consuming and may not fit into a busy lifestyle.
2. Inaccurate: People may forget to log their food, or may not know the exact portion sizes and calorie counts, leading to inaccurate data.
3. Demotivating: Constantly tracking and monitoring food intake can become demotivating, leading to burnout and decreased compliance.
4. Limited Information: A food log only provides information about food intake and does not take into account other factors such as exercise and medication that can impact glucose levels.
5. Tedious: The repetitive nature of logging food every day can become tedious and monotonous.
6. Dependence: Becoming overly reliant on a food log can lead to a lack of intuitive eating and a negative relationship with food.

These, in a nutshell, lead to limited reliability, validity, and generalizability.

Do you really need food log data for achieving high accuracy for predicting inflammation?

ChatGPT reasons that adding data from a food log to the model that uses HRV and CGM time series data can provide additional information about an individual’s diet and its impact on inflammation levels. However, incorporating food log data can also increase the complexity of the model and the need for pre-processing and feature engineering.

Ultimately, the best model for predicting inflammation levels will depend on the quality and availability of the data, the desired level of accuracy, and the computational resources available for building and evaluating models. It may be useful to build and compare models based on both sets of data and choose the one that provides the best results in terms of accuracy, precision, recall, and other relevant metrics.

When asked to recommend a set of physiological measures that can be continuously obtained in everyday life that are best for predicting level of inflammation in humans ChatGPT suggests the following:

1. Heart rate variability (HRV): HRV is a measure of the variation in time between successive heartbeats and has been associated with inflammation levels.
2. Activity levels: Wearable devices that track activity levels, such as accelerometers, can provide continuous data on physical activity, which has been linked to inflammation levels.
3. Sleep patterns: Wearable devices that track sleep patterns, such as wearable sleep monitors, can provide continuous data on sleep quality and quantity, which has been linked to inflammation levels.
4. Heart rate: Heart rate can be continuously monitored using wearable devices and has been associated with inflammation levels.
5. Body temperature: Continuous monitoring of body temperature using wearable devices has been associated with inflammation levels.

Photo by Solen Feyissa on Unsplash

Upon my probing why it did not list blood glucose from a CGM, ChatGPT politely replies: I apologize for the oversight. Continuous glucose monitoring (CGM) is a valid method for measuring blood glucose levels and can provide a continuous stream of data on an individual’s glucose levels. CGM is a non-invasive method and can provide valuable information for predicting inflammation levels.

CGM can be used in conjunction with other physiological measures such as heart rate variability, activity levels, sleep patterns, heart rate, and body temperature to provide a more comprehensive picture of a person’s physiological state and help predict inflammation levels.

It is important to note that while CGM provides continuous data on glucose levels, it may not reflect the systemic inflammation of the entire body, as other factors such as meal times, physical activity, and stress can also impact blood glucose levels. However, combined with other physiological measures, CGM data can provide valuable information for predicting inflammation levels.

I would be curious to learn more about what’s inflammation in a number and whether — one day — we will be comparing our level of inflammation in a gamified virtual environment, competing for the lowest score.

If you enjoyed this post, you might also like:

Systematic Guide to Author’s Note Writing

COVID-19 Feelings for Kids

Sylvia Kreibig, Ph.D., is emotion and sleep researcher at the Department of Psychology at Stanford University. She earned her Ph.D. in Psychology from the University of Geneva, Switzerland, completed her postdoctoral training at Stanford University, and has extensive training in psychophysiology — the mind–body connection. The overarching aim of her research is to better understand how emotions impact mental and physical health and how emotion regulation can be used toward health improvement. View my LinkedIn profile; connect via Twitter.