Very interesting. I am six months into a similar self-tracking project on this topic. My analysis covers a good one and half years of data quantifying health observations, vitals, medicines, stool scale, activity and many more areas for myself and my family.
I am trying to slowly chip away at the complexity in this area. When I started the project, I had a set of assumptions regarding predictors, which I now realize are either not correct or a very small fraction of what will ultimately prove to be the complete set.
I am confident that a project like this requires self-tracking data covering multiple dimensions of well-being - from physical health, to economic/financial, to social/environment, intellectual/psychological and emotional/spiritual.
When analyzing a recent episode of stress/anxiety, I concluded that the predictors (whatever they may be) accumulate day-over-day, until a tipping point that then catalyzes an episode of stress/anxiety. In this scenario, I agree with your findings that some predictors like social interactions, sleep etc. help to reduce this accumulation. An accurate model in this area would identify the balance required to avoid the tipping point.