Predict Hormone Trends with Whuman App

If you have a wearable device such as, Apple watch, Oura ring, Fitbit, etc. then you can get daily hormone insights using Whuman app now live on app store. Can you trust it though?

Even though I built it the scientist in me questions this daily. So, I started testing it with my own hormone data using Mira plus at-home hormone testing. It was a good reminder how expensive and tedious at-home testing can be.

Here is the result (unbiased ran by claude with the data - best I can do with self-funding this project):

The production validity from single user, readings per hormone jumped from days 21–50, spanning a full post-ovulatory window.

What's real vs. noise here:

- Each AUC now rests on 6 positive × 12 negative = 72 comparisons — meaningfully stable but still one person, so the Hanley–McNeil CI is roughly ±0.15–0.20, not tight enough to call this "validated."

- The lag pattern is more physiologically coherent :
    - LH peaks sharply at lag0 (0.868) and falls off fast at lag±1/2
    - PdG peaks at lag0 (0.792), matching expectation for a wearable-driven proxy.
    - Estrogen's best is lag+2 (0.785) — plausible (E3G metabolite clearance can trail estradiol by more than a day), but with n=18 from one person this could also just be noise reshuffling the peak.
  - Compared to the mcPHASES LOOCV bounds (LH 0.668 / E 0.645 / PdG 0.724), all three numbers here beat the research ceiling — encouraging directionally, but expected to regress toward those numbers as more users are added, since a single user's tertile-labeling is the easiest case (their own baseline, no cross-user noise).

  Bottom line: The directional story — LH/PdG same-day, estrogen lagged — is a nice self-consistency check.
  It's still not a "plausible production validity" claim — it's one person's numbers, still short of the ~40–50 total pairs needed even at the single-user-cluster level for a defensible CI, let alone the between-user generalization gap (need ≥5 users to exclude chance at 95% CI).

As my friend Claude pointed out above, the directional story is pretty impressive (I am surprised too) but it is far far far from perfect. But I will keep at it asking friends/family to volunteer and help me with the data and hopefully one day women will trust this and find it useful to understand their PMS, PMOS, perimenopause symptoms, transitions from one stage of life to another, and be able to make improvements in their physical and mental health.