The data has to be collected under clinical conditions to control for confounding variables. People are notoriously bad at accurately recording their precise fo

The data has to be collected under clinical conditions to control for confounding variables. People are notoriously bad at accurately recording their precise food intake which is probably the most important variable here. They need to do smaller controlled studies with provided diets. Large scale population studies with sensor data in the wild is going to be a trickier proposition that requires linking the sensors to the rest of the patients medical records which is a privacy mine field.