PREDICT: The Science

This blog post is the fifth in a series on my participation in the PREDICT 2 study. Click the links below to read earlier parts of the series:

• Forging the Future of Personalized Nutrition: My Experience, My Contribution
• PREDICT 2 Study Experience: Set-up Day + The Study Pack
• PREDICT 2 Study Experience: Day 1 Reflection
• PREDICT 2 Study Experience: Day 2 Reflection

“Overview of the ZOE Scientific Project for Researchers & Clinicians”
A ScienceKara Summary

I’m technically neither a researcher nor a clinician, but let’s see if we can crack the code, eh?

What is PREDICT 1?

PREDICT 1 was the first of – potentially – many PREDICT studies to come in the future. Specifically, it was a multiple test-meal challenge study that 1,100 participants completed over 14 days. Scientists analyzed the participants’ metabolic response to standardized meals. With this data, researchers can drive the development of an algorithm for predicting metabolic responses. So in theory, the data from the PREDICT studies, with thousands of participants could help PREDICT scientists provide predictive insight on metabolic responses for millions of people interested in improving their health through diet changes.

There are five biological markers PREDICT 1 scientists analyzed as a response to standardized meals:

• Postprandial glucose (this marker most of all, at least for PREDICT 1)
• Triglycerides
• Insulin
• Inflammatory markers
• Self-reported hunger

Participants in the PREDICT 1 study were 1,100 healthy people mostly from King’s College in London, UK. 60% of these participants were made up of twins from the UK Twins Study, which is spearheaded by PREDICT scientist Tim Spector, PhD. A Harvard researcher team led a validation cohort at Massachusetts General Hospital, and ZOE researchers completed an additional three-week sub-study of 100 UK participants to investigate metabolic responses to different types of standardized meals.

Researchers finished gathering data in May 2019. PREDICT 2 launched in June 2019. This time, PREDICT would be a non-hospital-based study, but the two PREDICT studies’ sets of data would still be able to be “seamlessly combined.”

What is PREDICT’s end goal?

The ZOE report describes two doors that could be opened by the data and insights gained by the PREDICT studies.

1. “Data-scaling” – releasing a product outside of a clinical study in 2020
   • Customers undergo similar protocol as the one I’m following as a PREDICT 2 
   • Customers consent to sharing data anonymously
   • ZOE can make money AND gather data
2. “Traditional science”
   • Short- and long-term nutritional intervention studies to test the accuracy of any predictions made via experimental algorithms and (later) the efficacy of ZOE diets under controlled conditions
   • Additional PREDICT studies to continue to gather data, but in a lab setting; this door provides a chance to collect data unique to the lab setting versus at data collected at home (like what I’m doing with PREDICT 2)

The introduction to the report outlines five key hypotheses that provide a foundation for PREDICT. Here’s the gist.

1. Variability: People respond to food in unique ways and it can be predictable.
2. Postprandial: Scientists need to study how the body responds and changes after eating a meal (“postprandial”). Measuring “traditional fasting markers” is not enough.
3. Predict: Nutritional responses are directly linked to health conditions, and nutritional responses can be measured. Thus, health conditions can be predicted and addressed.
4. Algorithm: The more data scientists can analyze from individual participants in studies like PREDICT, the more accurate machine-learning will be to build predictive tests.
5. Complex: Science needs hands-on studies like PREDICT – not just traditional observational studies – to obtain insight needed for true predictive algorithm technology.

What’s taking so long?

Why haven’t scientists figured out how to predict individual responses to food yet? The variability and complexity of the factors involved is overwhelming; ZOE scientists list “background exposures” like:

• Time of day
• Recent exercise
• Previous meals 

It takes time, resources, and diligent volunteers to make a study like PREDICT function and produce useful results. We all have the same letters in our genetic code (AGTC); the words are just written in different ways. Millions of different ways.

POSTPRANDIAL

I started to count the number of times the ZOE scientists used the term “postprandial” and lost count. Needless to say, “postprandial” would be blowing up a word cloud made up of the text in this report. The necessity for postprandial study expressed by the scientists in this report is what led to both PREDICT 1 and PREDICT 2. These types of studies are a huge undertaking, needing to standardize meals for participants and control factors like fasting time without the participant coming to a clinic every day. Plus, the participants in PREDICT 2 apply and activate their own glucose monitors and collect, package, and mail their own saliva, stool, and blood samples! Not to toot my own horn, but I’m really doing science a favor with my dedicated and studious participation in this study (wink)!