=========== A. Pre-treatment balance (deaths before dosing began) ===========
Randomisation implies no group difference here. Tested for every rapamycin
monotherapy arm against its own cohort/sex control at that arm's start age.

  C2005 m Rapa                 start 20 mo: died before dosing  17.5% vs ctrl  29.9%  p=0.0018 *
  C2005 f Rapa                 start 20 mo: died before dosing   8.3% vs ctrl   7.1%  p=0.7005
  C2006 m Rapa                 start  9 mo: died before dosing   2.3% vs ctrl   2.0%  p=1.0000
  C2006 f Rapa                 start  9 mo: died before dosing   1.4% vs ctrl   0.4%  p=0.2760
  C2009 m Rapa_hi              start  9 mo: died before dosing   5.1% vs ctrl   4.7%  p=0.8209
  C2009 f Rapa_hi              start  9 mo: died before dosing   0.0% vs ctrl   0.4%  p=1.0000
  C2009 m Rapa_lo              start  9 mo: died before dosing   5.1% vs ctrl   4.7%  p=0.8209
  C2009 f Rapa_lo              start  9 mo: died before dosing   0.7% vs ctrl   0.4%  p=0.5475
  C2009 m Rapa_mid             start  9 mo: died before dosing   5.8% vs ctrl   4.7%  p=0.6542
  C2009 f Rapa_mid             start  9 mo: died before dosing   0.7% vs ctrl   0.4%  p=0.5475
  C2015 m rapa_hi_continuous   start 20 mo: died before dosing  25.0% vs ctrl  37.3%  p=0.0087 *
  C2015 f rapa_hi_continuous   start 20 mo: died before dosing   9.6% vs ctrl   6.1%  p=0.2262
  C2015 m rapa_hi_cycle        start 20 mo: died before dosing  33.3% vs ctrl  37.3%  p=0.4149
  C2015 f rapa_hi_cycle        start 20 mo: died before dosing   8.8% vs ctrl   6.1%  p=0.3100
  C2015 m rapa_hi_start_stop   start 20 mo: died before dosing  31.4% vs ctrl  37.3%  p=0.2146
  C2015 f rapa_hi_start_stop   start 20 mo: died before dosing   6.6% vs ctrl   6.1%  p=0.8312

  C2015 males, three rapamycin arms pooled against the shared control:
      pooled:  29.9% vs ctrl  37.3%  p=0.0338 *

  C2005 males by site (the paper names UT and possibly UM):
      TJL:  26.7% vs ctrl  28.6%  p=0.8622
      UM :  12.3% vs ctrl  23.8%  p=0.0774
      UT :  13.3% vs ctrl  37.3%  p=0.0010 *

=========== B. Dose-response, C2009 (one shared control) ===========
4.7 / 14 / 42 ppm all started at 9 months and share the same 580 controls,
so the three contrasts are correlated; the trend test uses control as 0 ppm.

  m: trend per log(ppm+1) HR=0.876 [0.831,0.922] p=5.72e-07
      departure from monotone log-linear trend: p=0.778
      4.7 ppm vs 0: HR=0.847 p=1.10e-01
      14 ppm vs 0: HR=0.706 p=1.15e-03
      42 ppm vs 0: HR=0.598 p=4.03e-06
  f: trend per log(ppm+1) HR=0.663 [0.622,0.706] p=1.09e-37
      departure from monotone log-linear trend: p=0.007 <- dose curve is not log-linear
      4.7 ppm vs 0: HR=0.414 p=1.01e-13
      14 ppm vs 0: HR=0.255 p=2.01e-27
      42 ppm vs 0: HR=0.240 p=1.36e-31

  Dose x sex interaction (is the dose curve itself different by sex?)
      p=6.32e-06 *

=========== C. Start age at the same dose (14 ppm continuous) ===========
C2006 starts at 9 months, C2005 at 20 months. Different cohorts, so
different control mice: this is a cross-cohort comparison, not randomised.

  m: C2006(start 9 mo): HR=0.59 [0.48,0.73], RMST +76 d  |  C2005(start 20 mo): HR=0.59 [0.47,0.73], RMST +81 d
  f: C2006(start 9 mo): HR=0.38 [0.30,0.48], RMST +146 d  |  C2005(start 20 mo): HR=0.45 [0.36,0.57], RMST +115 d

=========== D. Dosing regimen, C2015 (42 ppm from 20 mo, shared control) ===========
continuous / cycled (alternating) / stop-start (20 to 23 mo then stopped)

  m:
      rapa_hi_continuous     HR=0.73 [0.57,0.92] p=0.0081
      rapa_hi_cycle          HR=0.61 [0.48,0.79] p=0.0001
      rapa_hi_start_stop     HR=0.80 [0.62,1.03] p=0.0864
      stop-start vs continuous directly: HR=1.14 p=0.3596
  f:
      rapa_hi_continuous     HR=0.47 [0.38,0.59] p=0.0000
      rapa_hi_cycle          HR=0.61 [0.49,0.76] p=0.0000
      rapa_hi_start_stop     HR=0.85 [0.68,1.05] p=0.1299
      stop-start vs continuous directly: HR=1.86 p=0.0000

=========== H. Removal (censoring) by arm ===========
Miller 2014 reports more males removed for fighting at the highest dose
(11%, 18%, 22% of cages by site). Non-random censoring would bias survival.

  C2005 m Rapa                 removed: 5.1% vs ctrl 5.6%  p=1.0000
  C2005 f Rapa                 removed: 0.7% vs ctrl 2.4%  p=0.2827
  C2006 m Rapa                 removed: 15.2% vs ctrl 9.6%  p=0.0776
  C2006 f Rapa                 removed: 9.7% vs ctrl 12.9%  p=0.4246
  C2009 m Rapa_hi              removed: 23.1% vs ctrl 1.7%  p=0.0000 *
  C2009 m Rapa_lo              removed: 3.8% vs ctrl 1.7%  p=0.1977
  C2009 m Rapa_mid             removed: 9.0% vs ctrl 1.7%  p=0.0007 *
  C2009 f Rapa_hi              removed: 11.8% vs ctrl 13.6%  p=0.6445
  C2009 f Rapa_lo              removed: 15.4% vs ctrl 13.6%  p=0.6537
  C2009 f Rapa_mid             removed: 14.7% vs ctrl 13.6%  p=0.7642
  C2015 m rapa_hi_cycle        removed: 9.4% vs ctrl 12.0%  p=0.4393
  C2015 m rapa_hi_start_stop   removed: 17.0% vs ctrl 12.0%  p=0.1505
  C2015 m rapa_hi_continuous   removed: 6.4% vs ctrl 12.0%  p=0.0712
  C2015 f rapa_hi_continuous   removed: 0.0% vs ctrl 2.1%  p=0.1837
  C2015 f rapa_hi_cycle        removed: 2.2% vs ctrl 2.1%  p=1.0000
  C2015 f rapa_hi_start_stop   removed: 1.5% vs ctrl 2.1%  p=1.0000

  Worst-case bounds for the arms where removal differs, C2009 42 ppm males:
      as analysed:            HR=0.59 [0.48,0.74]
      worst case vs rapamycin: HR=0.79 [0.65,0.97]
      worst case for rapamycin: HR=0.37 [0.29,0.47]

=========== Truncation sensitivity (plan section 4) ===========
Main analysis conditions on surviving to the arm's start age. Below, the
same arms analysed over all mice in the file, as a reader would get by
ignoring start age.

  C2005 Rapa                 m: HR 0.59 (from start age) vs 0.57 (all mice)
  C2005 Rapa                 f: HR 0.45 (from start age) vs 0.49 (all mice)
  C2015 rapa_hi_continuous   m: HR 0.71 (from start age) vs 0.70 (all mice)
  C2015 rapa_hi_continuous   f: HR 0.48 (from start age) vs 0.53 (all mice)
  C2015 rapa_hi_cycle        m: HR 0.65 (from start age) vs 0.72 (all mice)
  C2015 rapa_hi_cycle        f: HR 0.61 (from start age) vs 0.65 (all mice)
  C2015 rapa_hi_start_stop   m: HR 0.79 (from start age) vs 0.78 (all mice)
  C2015 rapa_hi_start_stop   f: HR 0.83 (from start age) vs 0.86 (all mice)

=========== I. Rapamycin against every other ITP compound ===========
Same estimator for all arms: Cox vs own cohort control, own sex, from the
arm's start age, stratified by site. This is what 'most promising' has to
be judged against.


  males  : 93 arms tested; 34 with p<0.05 in the protective direction
     rapamycin arms rank 1,3,7,8,9,11,15,20,23,34,44 out of 93 by HR
     top 8 by HR:
        1. C2017  RaAc_9               HR=0.36 [0.28,0.45] p=7.1e-19   <- rapamycin
        2. C2009  ACA                  HR=0.49 [0.40,0.60] p=2.5e-11
        3. C2011  MetRapa              HR=0.50 [0.40,0.61] p=4.4e-11   <- rapamycin
        4. C2011  17aE2                HR=0.51 [0.42,0.64] p=1.2e-09
        5. C2013  ACA_mid              HR=0.55 [0.45,0.67] p=6.1e-09
        6. C2013  ACA_hi               HR=0.56 [0.46,0.68] p=1.5e-08
        7. C2017  RaAc_16              HR=0.58 [0.45,0.75] p=2.9e-05   <- rapamycin
        8. C2005  Rapa                 HR=0.59 [0.47,0.73] p=1.0e-06   <- rapamycin

  females: 89 arms tested; 17 with p<0.05 in the protective direction
     rapamycin arms rank 1,2,3,4,5,6,7,8,9,10,19 out of 89 by HR
     top 8 by HR:
        1. C2009  Rapa_hi              HR=0.22 [0.17,0.28] p=1.2e-29   <- rapamycin
        2. C2017  RaAc_9               HR=0.25 [0.19,0.32] p=4.3e-26   <- rapamycin
        3. C2009  Rapa_mid             HR=0.30 [0.23,0.39] p=5.3e-20   <- rapamycin
        4. C2011  MetRapa              HR=0.36 [0.29,0.45] p=5.4e-19   <- rapamycin
        5. C2006  Rapa                 HR=0.38 [0.30,0.48] p=2.4e-15   <- rapamycin
        6. C2017  RaAc_16              HR=0.39 [0.30,0.49] p=8.1e-15   <- rapamycin
        7. C2009  Rapa_lo              HR=0.42 [0.33,0.54] p=5.2e-12   <- rapamycin
        8. C2005  Rapa                 HR=0.45 [0.36,0.57] p=8.5e-12   <- rapamycin
