WHOOP measures over 100 parameters at a high frequency sampling compared to other wearables to come up with derived metrics such as Strain, sleep insights, Recovery and so on. These derived metrics are gathered from raw metrics (HR, HRV, Skin Temperature, Movement, SpO2) and User-Input Metrics (Lifestyle factors -Caffeine and alcohol consumption, Mood, Outlook to life, symptoms, and behaviours).

How do we use these metrics to further our understanding of our body and its health score?

Bayesian principle holds strong in the statistical realm of analysis which is a combination of “prior knowledge” supported with “observed data” driving to make “inferences about unknown parameters/probabilities”.

Given the explosive growth of “wearables”,  which helps in higher-sampling frequency of physiological parameters of an individual supporting the Nyquist criteria, it is now possible to track physiological parameters at smaller units of time. Supporting the tracking of physiological data, it is now possible to measure and record environmental data of an individual such as UV radiation exposure, exposure to sunlight, exposure to blue light, noise levels, AQI, travel conditions to work. Combining these metrics we can paint a 24 hour vivid picture of a person as he/she go through their daily routine with rich longitudinal context.

This capability of gathering a diverse set of health and environmental data is driving towards building a “Health Intelligence Engine, HIE” , empowering personal and preventive healthcare.

Answering a question such as “What is the probability of my HbA1c (a leading indicator or diabetes risk) reading to become above normal in the next 60 days?”, is possible to be answered with a certain confidence level.

Below is a high level thought process to address the above prediction question.

1.  Prior Knowledge: Bloodwork [Quarterly, structured data gathered from a pdf file] HbA1c, Ferritin, CRP, B12, Thyroid, Fasting Insulin

2. Observed Data: WHOOP [time series, at 1 minute intervals] RHR, HRV, Sleep metrics, Strain, Stress

3.  Observed Data: CGM (Continuous Glucose Monitor) data [time series, at 1 minute intervals] Time In Range, Avg glucose

4. Observed Data: Lifestyle [structured and/or unstructured data, captured daily] Supplements, Medication, Mood, Alcohol consumption

By analyzing the multimodal data, most of which is recorded at a high-sampling frequency and some derived metrics such as sleep quality, it is possible to predict an Estimated HbA1c for a near-time window.

Adaption of Bayesian Principles to HbA1c Estimation

HIE Prompt

“Based on your CGM and WHOOP trends from the last 30 days, your estimated HbA1c is 5.9% (±0.2). You have a 62% probability of crossing the 6.0% mark in the next 60 days. Recommend evaluating dietary consistency, increasing recovery, and retesting HbA1c at your next bloodwork.”

Such a prediction helps an individual work with their Health Manager to drive focussed, personalized and preventive intervention to drive optimal well-being.

This is the power that an individual wears on their wrists or fingers (smart watches, rings etc) that are propelling us towards Health 4.0.