================ PEARL: analysis set ================
randomised and analysed: 114 completers (placebo 39, 5 mg 40, 10 mg 35)
discontinued and excluded from analysis: 11 (placebo 6, 5 mg 2, 10 mg 3)
so 11 of 125 randomised participants (8.8%) are missing from the
analysis; the paper reports a completer analysis, not intention-to-treat.
women: 40 of 114 (35.1%); by arm placebo 15, 5 mg 17, 10 mg 8

--- The subgroup carrying the headline efficacy finding ---
Lean tissue mass improved in women on 10 mg. That arm contains 8 women.
Printed odds ratio for improvement: OR = 28, 95%% CI 2.42-323.7, p = 0.008.
The interval spans a factor of 134, which is what an OR estimated on
single-digit cell counts looks like. The all-participants OR for the same
measure is 2.29 (95% CI 0.81-4.49, p = 0.12), and in men 1.09 (p = 0.90).

================ Power for the primary outcome ================
Primary outcome (visceral adiposity) came out eta_p^2 = 0.001, p = 0.942.
Below: the smallest effect a one-way ANOVA with these group sizes would
detect with 80% and 90% power at alpha = 0.05.

  80% power: smallest detectable Cohen's f = 0.295  (eta_p^2 = 0.080)
  90% power: smallest detectable Cohen's f = 0.338  (eta_p^2 = 0.102)

  observed: eta_p^2 = 0.001 (Cohen's f = 0.032)
  power this trial had against its own observed effect: 5.8%
  power against a conventionally 'medium' effect (f = 0.25): 65.1%

  Reading: the trial can exclude a medium-or-larger effect on visceral
  adiposity. It cannot exclude a small one, and it was never able to.
  'No significant change' is therefore not evidence of equivalence.

--- Power in the female subgroup (the LTM finding) ---
  n = 40 women across three arms; 80% power needs f = 0.511 (eta_p^2 = 0.207)
  printed 48-week female LTM effect: eta^2 = 0.202, F(2,30) = 5.052, p = 0.013
  That is above the detectable threshold, but an effect only detectable
  when it is this large is also the kind most inflated by selection:
  it is the one significant result among the body-composition contrasts.

================ Multiplicity in the printed OR table ================
comparisons in Table 1: 24; nominally significant at 0.05: 2
expected false positives if nothing were real: 1.2
Holm-adjusted, surviving 0.05: 0
Benjamini-Hochberg, surviving 0.05: 0
The two nominally significant cells point in opposite directions: lean
tissue up in women on 10 mg (OR 28) and bone mineral density improvement
less likely on 5 mg (OR 0.24). The 10 mg BMD cell is OR 1.00. A real
dose effect would not normally appear at 5 mg and vanish at 10 mg.

================ What dose was actually delivered ================
The paper reports compounded rapamycin reached about 1/3 the 24-hour blood
concentration of commercial formulations, and states effective doses are
about 66% lower than the labelled dose.
  labelled  5 mg/week  ->  roughly 1.7 mg/week commercial-equivalent
  labelled 10 mg/week  ->  roughly 3.3 mg/week commercial-equivalent

This matters twice. It weakens 'no effect at 5-10 mg/week' as a statement
about rapamycin, because the exposure tested was lower than the label
implies. And it means the milligram numbers in this trial cannot be read
across to milligram numbers from any other source.

================ Safety events and their denominator ================
  placebo  n=39  adverse events 122 (3.1 per participant)  serious 3
  5 mg     n=40  adverse events 116 (2.9 per participant)  serious 2
  10 mg    n=35  adverse events 117 (3.3 per participant)  serious 1
  one case of anemia in the 5 mg arm, resolved after transfusion; that
  participant completed the trial.

  Total exposure behind these counts: 114 participants x 48 weeks =
  5472 participant-weeks (105.2 participant-years), and 11 randomised
  participants are not in it. Event counts being similar across arms at
  this size, this duration, and below-label exposure does not support a
  conclusion about long-term safety in healthy people.
  The one unfavourable direction worth keeping: BMD improvement was less
  likely on 5 mg (OR 0.24, p=0.04). It does not survive correction above,
  and the 10 mg cell is OR 1.00, so it is a signal to watch, not a finding.
