Quantified self for healthy aging and longevity

[LazaLong]

Hello forum,

Few years back I started a journey of extensive self tracking with the main focus on healthy aging
and longevity. At the moment I track more than 300 variables, many of them daily for more than
2 years.

Now, that I have collected sufficient data, and most crucially developed my analytical platform, I am starting to finally produce interesting insights.

I realised that with just a bit more extra effort, I could share my insights and tools with the rest of the community, and perhaps reap some rewards in the form of useful feedback. Hence the Quantified Longevity blog: https://www.quantifiedlongevity.org/

I want to document how I track biomarkers, how they are affected by interventions, ‘pre-register’ self-experiments and subsequently their results, discuss relevant hardware and software tools, and possibly other longevity topics.

I am posting here to spread the word and collect feedback.

I kicked off the site with the first 4 posts on:

1. Introducing the overall rationale of the project
2. Description of the different categories of tracked data
3. Description of the data collection, aggregation and analytical framework I have developed
4. First insights into sleep tracking (to be followed by 3 more posts in the coming weeks)

My intention is to post once a week, and I will take the liberty to add to this thread posts that might be of interest to the community.

Any feedback, either in this forum or directly on the website are most welcome!

[rain8dome9]

@LazaLong

This tab allows for semi-automated data analysis based on simple Pearson correlations.

Pearson correlation almost never works because:
https://wiki.openhumans.org/wiki/Finding_relations_between_variables_in_time_series#Reasons_time_series_analysis_especially_as_applied_to_QS_is_hard

@LazaLong

Note that in this case the time points look like ‘sitting on trajectories’ due to the Gaussian smoothing filter that is switched on for both variables. On the bottom right, the user can see the cross-correlation between the two variables, that can reveal a delayed relationship between the two variables.

nice but delay only fixes one of the problems

[LazaLong]

Hello rain8dome9. I appreciate your concern, but there is no need to be so dismissive. Indeed correlations have lot of problems and it is just the first step to start exploring data. I have also explained at length in the first post why I am not that concerned about accepting few spurious correlations as long as I also capture the real ones.

But most importantly, as I mentioned in the third post, I am developing advanced analysis techniques, including auto-regressive models as suggested by the very link you have posted.

Do you perhaps have some specific recommendations on techniques that you use for analyzing your data?

[rain8dome9]

I would suggest changepoint, granger causality, causal impact, and unit root / derivative though I am sure of only change point detection.

[LazaLong]

I have added a new post where I look into which interventions have the greatest impact on my sleep quality as measured by the Fitbit Sleep Score.

Apart from few obvious relationships that validate the data, such as alcohol negatively impacting sleep quality, even at smallest amounts (i.e. a small beer) most findings are quite surprising.

I find that both consumption of sour cherries (that I occasionally put in smoothies), cider vinegar (that I regularly put in my salads) and hyaluronic acid seem to have positive impact on my sleep.

On the other hand consumption of both tomatoes and tomato sauce worsen my sleep, and both do so independently.

Perhaps just as surprisingly I found no effect of several interventions of touted to improve sleep, such as Magnesium Threonate, or chamomile tea.

[Agaricus (Gary Isaac Wolf)]

Can you explain the graph. I’m not sure I understand. Is the left bar the average sleep score under the “no sour cherry” (0.0) condition? And the right is the average sleep score under the "sour cherry consumed (1.0) condition? Also: it appears the average sleep score goes from 78 to 80? Is that right? If all this is correct, some questions:

How many nights of data?
Over how long a period?
How much sour cherry is consumed?
Is there a dose effect?
(Same questions for the tomatoes.)

I may have misunderstood the analysis, so glad to have more explanation!!

[rain8dome9]

The correlation completely disappears if I shift the data just by one day

Granger ‘causality’.

The relationship is weak, but I noticed that it is stable over time - i.e. if I look at both the first and last year of my tracked data I find the same effect in both, which makes me believe it is real.

Good! What is the lag delay? How about build up?