Condition Template Pages Updated

Condition Template pages are based on bacteria shifts reported as significant in studies published on PubMed. These studies usually use the non-quantified expression “higher” or “lower”, often on the averages for a group. This makes application to individual microbiome samples a challenge — especially when averages very rarely are normal distributions.

With the improved quantiles (bins of samples) enhancement, we have a way to likely use these vague study results in a robustness safe way.

Consider a condition with 16 bacteria. If we select the 6% level, we would expect around 1 (16 x 6% =0.96) matches. With expected numbers and actual numbers, we can use a method called Pearson’s chi-squared test. Using expected numbers and actual numbers from a sample, we can compute chi-square probability of the matches being random.

For the example below, we pretend we have 16 bacteria. We expect 1 to be found(matches) and 15 not to be found. If we find 6 matches and thus 16-6 -= 10 non-matches, we can compute the odds of this happening at random. Around 0.032, 3.2% or 1/0.032 => 1 in 32 times.

  • With 0 matches on 16 items we have 0.309
  • With 1 matches on 16 items, we have 1.00
  • With 2 matches on 16 items, we have 0.54
  • With 3 matches on 16 items, we have 0.285
  • With 4 matches on 16 items, we have 0.144
  • With 5 matches on 16 items, we have 0.070
  • With 6 matches on 16 items, we have 0.032
Compute it yourself at: https://www.mathsisfun.com/data/chi-square-calculator.html

Traditionally, a value is 0.05 (5%) or less is needed to be significant (or, of concern). There is one change in 20 or this being random. If you check 20 conditions – you would expect one to be at 0.05 or below ( 5% x 20 = 1). So you need to use a bit of common (statistical) sense if some odd condition appears. On the other hand, if a suspect condition shows up very strong then you may wish to get definitive testing (if it is available). Not responsible for MD’s confused faces when you ask for testing because you microbiome results suggest it may be wise.

Bottom Line

Using these studies results have been an ongoing challenge for me. There is elegance in the approach implemented given that we are dealing with fuzzy data from studies that are all over the place in approach, default regional diets etc.

Again, this is not intended to predict, rather just inform people when their microbiome appears to be shifting to reported disease patterns.

Suggestions:

  • The smaller the probability and the smaller the percentile (i.e. 6%), the more likely that these is a real association.
  • The association should persist across multiple samples.
  • The all in one is likely the best measure of microbiome dysfunction

I will see about adding a timeline to these measures for those with multiple samples.