Predictive Analytics: A crystal ball for personal health insights

Data: The Big, the Diverse, and the Real-Time

I was curious on the of the term Big Data, and found something very interesting when I was trying to trace it. Turns out, every decade, since the rise of the microprocessor, there’s been a niche term coined with data and analytics. Going all the way back to 1965 to the first Main frames storing troves of tax data on Magnetic tape. In 1986, our nomenclature changed from bucketing it under Statistics to calling it “Data Science”. Data informs decision making with the highest probability for profit - probably Decision Science never sounded as glamorous as Data Science.

Irrespective of how it’s marketed, in Healthcare, 2000-2010, we sought Big Data to predict disease outbreaks. From 2010-2020 we sought Diverse data sets so that we could build generalisable models for disease detection and drug discovery. In this decade, we’re building systems with RLHF (Reinforcement learning from human feedback) with near Real-time data flow into Production servers. We’re not there yet in Healthcare, but atleast we’ve got our TikTok and Instagram reels personalized with our likes and shares.

Algorithms: The potential of unlocked models over locked

Regulatory bodies have been exploring Guidance with Dynamic models and Manufacturer upgrades, even though most of the current algorithms are “locked”, and we should see significant changes in the SaMD (Software as a Medical Device) space over the next 2 years. When there’s chaos amidst order, there’s bound to be evolution, and as a developer, I’d much rather be asked, “what is your Annual Model Upgrade and Maintenance Plan” rather than “what is the Model accuracy”.

The RLHF model at a personal and patient level would always be guiding on which new risks we may be exposed to. But if we zoom out, the indirect benefits at a health system level would be phenomenal. The patients’ altruism of clicking on the “I Agree” – for data sharing - could translate to epidemic prediction at an incipient stage, outbreak prediction models that could do away the need for lockdowns with targeted isolation. In the context of chronic diseases, I see tremendous potential of data from an army of glucose sensors and wearables informing drug dose titration on the population scale. Type 1 diabetics learn or are instructed to do this real-time based on what they eat, but imagine what we could do for Type II Diabetics, and other NCDs, when real-time data starts informing Lifestyle management. Algorithms adapting on the fly based on trends seen due to minor adjustments in therapy. Imagine what we could achieve if all participants actively subscribed to this paradigm of data sharing and adhering to algorithmic advice. We wouldn’t be operating in healthcare as interventional and therapeutic specialists. We would be operating in healthcare as preventive specialists.

DTx: What's next?

And as we gaze into the crystal ball for this future, we’re not very far. At Fitterfly, we’ve researched and developed Personalised Glycemic scores (PGR), where every meal gets rated based on the Continuous Glucose Monitor readings. We overlay meal logs, movement indices, and other data streamed from wearables along with CGM data to predict future glucose trends. We’re calling it vCGM and we’re showcasing our results at the ATTD 2024 conference in Florence, do stop by our booth! ?