forked from ruthkeogh/causal_sim
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathanalysis.R
152 lines (117 loc) · 6.77 KB
/
analysis.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
#------------------------------------------------------
#------------------------------------------------------
# analysis.R
# - Fits a correctly specified MSM (using IPTW) to the simulated data. The MSM is an additive hazard model. See Section 6.1 of the paper.
# - Obtains estimates of cumulative coefficients from the MSM.
# - Obtains estimates of survival probabilities under the treatment regimes 'always treated' and 'never treated'.
# - Repeats over 1000 simulated data sets.
# Created by Ruth Keogh
#------------------------------------------------------
#------------------------------------------------------
#----
#set seed
set.seed(443792)
#--------------------------
#times at which we obtain estimated cumulative coefficients of the MSM and estimated survival probabilities
#--------------------------
t.hor=seq(0,5,0.01)
#--------------------------
#analysis in n.sim simulated data sets
#--------------------------
n.sim=1000
#----
#store results
coef_0=matrix(nrow=n.sim,ncol=length(t.hor))
coef_A=matrix(nrow=n.sim,ncol=length(t.hor))
coef_Alag1=matrix(nrow=n.sim,ncol=length(t.hor))
coef_Alag2=matrix(nrow=n.sim,ncol=length(t.hor))
coef_Alag3=matrix(nrow=n.sim,ncol=length(t.hor))
coef_Alag4=matrix(nrow=n.sim,ncol=length(t.hor))
surv1=matrix(nrow=n.sim,ncol=length(t.hor))
surv0=matrix(nrow=n.sim,ncol=length(t.hor))
#----
#start of simulation loop
for(i in 1:n.sim){
print(i)
#-----
#simulate data using dat_sim
#could be replaced with dat_sim_GEN.R
source("dat_sim.R")
#------------------
#stabilized weights
#------------------
wt.mod=glm(A~L+Alag1,family="binomial",data=dat.long)
pred.wt=predict(wt.mod,type = "response")
dat.long$wt=ifelse(dat.long$A==1,pred.wt,1-pred.wt)
dat.long$wt.cum=ave(dat.long$wt,dat.long$id,FUN=cumprod)
wt.mod.num=glm(A~Alag1,family="binomial",data=dat.long)
pred.wt.num=predict(wt.mod.num,type = "response")
dat.long$wt.num=ifelse(dat.long$A==1,pred.wt.num,1-pred.wt.num)
dat.long$wt.cum.num=ave(dat.long$wt.num,dat.long$id,FUN=cumprod)
dat.long$ipw.s=dat.long$wt.cum.num/dat.long$wt.cum
#-----
#Fit the correctly specified MSM using IPTW
#See notes in analysis_TRUE_VALUES.R which explain why we fit the model separately by time period
ah.p0=aalen(Surv(time,time.stop,event)~A,data=dat.long[dat.long$time==0,],n.sim=0,weights = dat.long[dat.long$time==0,]$ipw.s)
ah.p1=aalen(Surv(time,time.stop,event)~A+Alag1,data=dat.long[dat.long$time==1,],n.sim=0,weights = dat.long[dat.long$time==1,]$ipw.s)
ah.p2=aalen(Surv(time,time.stop,event)~A+Alag1+Alag2,data=dat.long[dat.long$time==2,],n.sim=0,weights = dat.long[dat.long$time==2,]$ipw.s)
ah.p3=aalen(Surv(time,time.stop,event)~A+Alag1+Alag2+Alag3,data=dat.long[dat.long$time==3,],n.sim=0,weights = dat.long[dat.long$time==3,]$ipw.s)
ah.p4=aalen(Surv(time,time.stop,event)~A+Alag1+Alag2+Alag3+Alag4,data=dat.long[dat.long$time==4,],n.sim=0,weights = dat.long[dat.long$time==4,]$ipw.s)
#-----
#Create a step function for each estimated cumulative coefficient, enabling us to obtain the estimated cumulative coefficient at any time
maxrow.p0=dim(ah.p0$cum)[1]
maxrow.p1=dim(ah.p1$cum)[1]
maxrow.p2=dim(ah.p2$cum)[1]
maxrow.p3=dim(ah.p3$cum)[1]
maxrow.p4=dim(ah.p4$cum)[1]
ah.stepfun.0=stepfun(c(ah.p0$cum[,1], ah.p1$cum[-1,1],ah.p2$cum[-1,1],ah.p3$cum[-1,1],ah.p4$cum[-1,1]),
c(0,ah.p0$cum[,2],
ah.p0$cum[maxrow.p0,2]+ah.p1$cum[-1,2],
ah.p0$cum[maxrow.p0,2]+ah.p1$cum[maxrow.p1,2]+ah.p2$cum[-1,2],
ah.p0$cum[maxrow.p0,2]+ah.p1$cum[maxrow.p1,2]+ah.p2$cum[maxrow.p2,2]+ah.p3$cum[-1,2],
ah.p0$cum[maxrow.p0,2]+ah.p1$cum[maxrow.p1,2]+ah.p2$cum[maxrow.p2,2]+ah.p3$cum[maxrow.p3,2]+ah.p4$cum[-1,2]))
ah.stepfun.A=stepfun(c(ah.p0$cum[,1], ah.p1$cum[-1,1],ah.p2$cum[-1,1],ah.p3$cum[-1,1],ah.p4$cum[-1,1]),
c(0,ah.p0$cum[,3],
ah.p0$cum[maxrow.p0,3]+ah.p1$cum[-1,3],
ah.p0$cum[maxrow.p0,3]+ah.p1$cum[maxrow.p1,3]+ah.p2$cum[-1,3],
ah.p0$cum[maxrow.p0,3]+ah.p1$cum[maxrow.p1,3]+ah.p2$cum[maxrow.p2,3]+ah.p3$cum[-1,3],
ah.p0$cum[maxrow.p0,3]+ah.p1$cum[maxrow.p1,3]+ah.p2$cum[maxrow.p2,3]+ah.p3$cum[maxrow.p3,3]+ah.p4$cum[-1,3]))
ah.stepfun.Alag1=stepfun(c(ah.p0$cum[,1], ah.p1$cum[-1,1],ah.p2$cum[-1,1],ah.p3$cum[-1,1],ah.p4$cum[-1,1]),
c(0,rep(0,dim(ah.p0$cum)[1]),
ah.p1$cum[-1,4],
ah.p1$cum[maxrow.p1,4]+ah.p2$cum[-1,4],
ah.p1$cum[maxrow.p1,4]+ah.p2$cum[maxrow.p2,4]+ah.p3$cum[-1,4],
ah.p1$cum[maxrow.p1,4]+ah.p2$cum[maxrow.p2,4]+ah.p3$cum[maxrow.p3,4]+ah.p4$cum[-1,4]))
ah.stepfun.Alag2=stepfun(c(ah.p0$cum[,1], ah.p1$cum[-1,1],ah.p2$cum[-1,1],ah.p3$cum[-1,1],ah.p4$cum[-1,1]),
c(0,rep(0,dim(ah.p0$cum)[1]),
rep(0,dim(ah.p1$cum)[1]-1),
ah.p2$cum[-1,5],
ah.p2$cum[maxrow.p2,5]+ah.p3$cum[-1,5],
ah.p2$cum[maxrow.p2,5]+ah.p3$cum[maxrow.p3,5]+ah.p4$cum[-1,5]))
ah.stepfun.Alag3=stepfun(c(ah.p0$cum[,1], ah.p1$cum[-1,1],ah.p2$cum[-1,1],ah.p3$cum[-1,1],ah.p4$cum[-1,1]),
c(0,rep(0,dim(ah.p0$cum)[1]),
rep(0,dim(ah.p1$cum)[1]-1),
rep(0,dim(ah.p2$cum)[1]-1),
ah.p3$cum[-1,6],
ah.p3$cum[maxrow.p3,6]+ah.p4$cum[-1,6]))
ah.stepfun.Alag4=stepfun(c(ah.p0$cum[,1], ah.p1$cum[-1,1],ah.p2$cum[-1,1],ah.p3$cum[-1,1],ah.p4$cum[-1,1]),
c(0,rep(0,dim(ah.p0$cum)[1]),
rep(0,dim(ah.p1$cum)[1]-1),
rep(0,dim(ah.p2$cum)[1]-1),
rep(0,dim(ah.p3$cum)[1]-1),
ah.p4$cum[-1,7]))
#----
#estimated cumulative coefficients from the MSM at times t.hor
coef_0[i,]=ah.stepfun.0(t.hor)
coef_A[i,]=ah.stepfun.A(t.hor)
coef_Alag1[i,]=ah.stepfun.Alag1(t.hor)
coef_Alag2[i,]=ah.stepfun.Alag2(t.hor)
coef_Alag3[i,]=ah.stepfun.Alag3(t.hor)
coef_Alag4[i,]=ah.stepfun.Alag4(t.hor)
#----
#estimated survival probabilities at times t.hor in the 'always treated' (surv1) and in the 'never treated' (surv0)
surv1[i,]=exp(-(ah.stepfun.0(t.hor)+ah.stepfun.A(t.hor)+ah.stepfun.Alag1(t.hor)+ah.stepfun.Alag2(t.hor)+ah.stepfun.Alag3(t.hor)+ah.stepfun.Alag4(t.hor)))
surv0[i,]=exp(-(ah.stepfun.0(t.hor)))
}
#end of simulation loop
#----