Chapter 14 Prendre en compte l’interaction

14.1 Avec le sport

exp1 <- cmest(
  data = base3,                    # base
  model = "gformula",             # approach, defaut is rb (regression-based)
                                  # if postc is not empty only gformula or msm
  estimation = "imputation",      # method of estimation. "imputation" is conterfactual estimation
  inference = "bootstrap",        # method for se and CI
  nboot = 100,                    # defaut is 200
  EMint = TRUE,                  # interaction exposure mediator
  
  multimp = TRUE,                 # imputation multiple des DM 
  m = 10,

  outcome = "t8_SBP",                 
  exposure = "t0_baby_sex", 
  mediator = c("t4_act_sport"), 
  basec = c("t0_mother_scol_crt",
            "t0_fathers_csp_defav"),# confusion baseline
  postc = c("t4_litteraciepbm",
                     "t4_numeraciepbm",
                     "t4_act_read",
                     "t4_O_level",
                      "t4_hadchildren",
                     "t4_married",
                     "t4_act_friend",
                     "t4_religious"),    # confusion intermédiaire
  
  yreg = "linear",              # outcome regression model

  a = "Homme",                    # "active" value of exposure
  astar = "Femme",                # "control" value of exposure
  
  mreg = list("logistic"),          # regression model for each mediator
  mval = list("Non"),   # ref for M
  postcreg = list("logistic",
              "logistic",
              "logistic",
              "logistic",
              "logistic",
              "logistic",
              "logistic",
              "logistic")                 # ref for M
  ) ;    set.seed(28062022)

type	effect	ci.low	ci.high
cde	14.78	11.34	19.56
rpnde	13.49	10.97	17.48
rtnde	11.90	9.94	15.58
rpnie	-1.25	-3.53	-0.55
rtnie	-2.84	-5.36	-1.58
te	10.66	8.23	13.67
rintref	-1.28	-2.80	0.09
rintmed	-1.59	-3.51	0.13
cde(prop)	1.39	1.16	1.73
rintref(prop)	-0.12	-0.24	0.01
rintmed(prop)	-0.15	-0.33	0.01
rpnie(prop)	-0.12	-0.35	-0.05
rpm	-0.27	-0.55	-0.14
rint	-0.27	-0.58	0.03
rpe	-0.39	-0.73	-0.16

On peut décrire plusieurs effets :

l’effet direct (qui ne passe pas par l’activité physique) “moyen” d’être né homme est CDE = +14.78mmHg (95CI = [11.34 to 19.56]
il y a interaction avec le sport de : Int_Ref = -1.28mmHg (95CI = [-2.8 to 0.09]. Le fait de faire du sport change l’effet direct (chemins non en lien avec le sport) chez les hommes par rapport aux femmes.
la proportion médiée “moyenne” est PM = -26.78% (95CI = [-55.15% to -14.29%`]
L’effet du n’est pas le meme entre les hommes et les femmes : chez les hommes le fait de faire du sport diminue la TAS de Int_MEd = -1.59mmHg (95CI = [-3.51 to 0.13] de plus que chez les hommes.

14.2 Avec la lecture

exp1 <- cmest(
  data = base3,                    # base
  model = "gformula",             # approach, defaut is rb (regression-based)
                                  # if postc is not empty only gformula or msm
  estimation = "imputation",      # method of estimation. "imputation" is conterfactual estimation
  inference = "bootstrap",        # method for se and CI
  nboot = 100,                    # defaut is 200
  EMint = TRUE,                  # interaction exposure mediator
  
  multimp = TRUE,                 # imputation multiple des DM 
  m = 10,

  outcome = "t8_SBP",                 
  exposure = "t0_baby_sex", 
  mediator = c("t4_act_read"), 
  basec = c("t0_mother_scol_crt",
            "t0_fathers_csp_defav"),# confusion baseline
  postc = c("t4_litteraciepbm",
                     "t4_O_level",
                      "t4_hadchildren",
                     "t4_married"),    # confusion intermédiaire
  
  yreg = "linear",              # outcome regression model

  a = "Homme",                    # "active" value of exposure
  astar = "Femme",                # "control" value of exposure
  
  mreg = list("logistic"),          # regression model for each mediator
  mval = list("Oui"),   # ref for M
  postcreg = list("logistic",
              "logistic",
              "logistic",
              "logistic")                 # ref for M
  ) ;    set.seed(28062022)

type	effect	ci.low	ci.high
cde	9.49	6.70	12.84
rpnde	9.97	7.78	12.91
rtnde	10.36	8.07	13.26
rpnie	0.34	-0.32	1.04
rtnie	0.73	-0.05	1.80
te	10.70	8.50	13.77
rintref	0.48	-0.38	1.86
rintmed	0.39	-0.25	1.25
cde(prop)	0.89	0.68	1.04
rintref(prop)	0.05	-0.05	0.18
rintmed(prop)	0.04	-0.03	0.12
rpnie(prop)	0.03	-0.04	0.10
rpm	0.07	-0.00	0.16
rint	0.08	-0.07	0.29
rpe	0.11	-0.04	0.32

On peut décrire plusieurs effets :

l’effet direct (qui ne passe pas par l’activité de lecture) “moyen” d’être né homme est CDE = +9.49mmHg (95CI = [6.7 to 12.84]
il ne semble pas y avoir d’interaction avec la lecture de : Int_Ref = +0.48mmHg (95CI = [-0.38 to 1.86].
la proportion médiée “moyenne” est PM = 6.84% (95CI = [-0.31 to 15.87]
l’effet de la lecture semble a peu près similaire chez les hommes et les femmes Int_Med = +0.39mmHg (95CI = [-0.25 to 1.25]