Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
3.8. Health data
Health data can be collected from measurement loggers or manual entries into a symptom log.
For collecting health data the most attractive way of collecting data is from measurement loggers, often wearable devices. I own the Garmin Forerunner 305 which can provide GPS, heart rate, and shoe pedometer. I did not find the bundled software to answer my questions, but it at least would dump my data to an text file.The silliest shortcoming of the bundled software was that it claim I was burning a huge number of calories when driving. Clearly software should correlate the GPS and the heart rate monitor and conclude that if I speed up but my heart rate doesn't increase, I am not running 30 miles per hour. Individual measurements are fairly noisy, the heart rate monitor drops out when your skin under it is too dry, and the foot pedometer lasted only a few months. A pedometer type device could help measure activity such as:
- how many steps do I take around the house (even with my shoes off)
- how much tossing and turning do I do at night?
- how much time do I spend in bed?
- how much time am I sleeping? (this seems particularly difficult to measure accurately)
A symptom log records a series of symptoms and challenges, which can be statistically analyzed to uncover low-signal to noise relationships. An example might be to track food allergies and sensitivities. You would record everything you eat and record every symptom. For food allergies that are not fatal or too severe, you can enhance the signal to-noise ratio by following an elimination and challenge diet. In an elimination diet all suspect foods are eliminated until symptoms are relieved, then suspect foods are reintroduced one at a time until symptoms return.
The value of statistical analysis is due to the following:
- encourages careful and specific record keeping,
- symptom time delays increase the number of confounding variables,
- most purchased foods do not have precise ingredient amounts listed,
- symptoms may result from unrelated causes (is it what I ate or the flu?)
- Examining all pairs of possible challenges and possible symptoms. For the positive response, select only those with a plausible time relationship, say less than three days from challenge to symptom. Examine the time delay from challenge to symptom. If it shows a peak response around a particular delay, you may have found a true response. If it is evenly distributed, it could be a chance relationship. For the negative response, examine all the time intervals for which the challenge was present but the symptom was absent. Then plot a point for each candidate challenge symptom pair, the number of positive responses as the y coordinate and the number of negative responses as the x coordinate. Hopefully the plot will separate into at least two regions:
- probable chance associations with more negative responses than positive ones,
- probable positive associations with many more positive associations than negative ones,
- candidates for further investigation with some positive responses and few or no negative responses, but not a convincing sample size.