mirror of
https://github.com/agdamsbo/daDoctoR.git
synced 2024-11-22 11:50:23 +01:00
142 lines
3.8 KiB
R
142 lines
3.8 KiB
R
#' OBSOLETE - use 'print_log'
|
|
#'
|
|
#' Print regression results according to STROBE
|
|
#'
|
|
#' Printable table of logistic regression analysis according to STROBE.
|
|
#' @param meas outcome meassure variable name in data-data.frame as a string. Can be numeric or factor. Result is calculated accordingly.
|
|
#' @param var exposure variable to compare against (active vs placebo). As string.
|
|
#' @param adj variables to adjust for, as string.
|
|
#' @param data dataframe of data.
|
|
#' @param dec decimals for results, standard is set to 2. Mean and sd is dec-1.
|
|
#' @keywords logistic
|
|
#' @export
|
|
|
|
strobe_log<-function(meas,var,adj,data,dec=2){
|
|
## Ønskeliste:
|
|
##
|
|
## - Sum af alle, der indgår (Overall N)
|
|
## - Ryd op i kode, der der er overflødig %-regning, alternativt, så fiks at NA'er ikke skal regnes med.
|
|
##
|
|
|
|
require(dplyr)
|
|
|
|
d<-data
|
|
m<-d[,c(meas)]
|
|
v<-d[,c(var)]
|
|
|
|
ads<-d[,c(adj)]
|
|
dat<-data.frame(m,v)
|
|
df<-data.frame(matrix(ncol=4))
|
|
|
|
mn <- glm(m ~ .,family = binomial(), data = dat)
|
|
|
|
dat<-data.frame(dat,ads)
|
|
ma <- glm(m ~ .,family = binomial(), data = dat)
|
|
|
|
ctable <- coef(summary(mn))
|
|
pa <- ctable[, 4]
|
|
pa<-ifelse(pa<0.001,"<0.001",round(pa,3))
|
|
pa <- ifelse(pa<=0.05|pa=="<0.001",paste0("*",pa),
|
|
ifelse(pa>0.05&pa<=0.1,paste0(".",pa),pa))
|
|
pv<-c("REF",pa[2:length(coef(mn))])
|
|
|
|
co<-round(exp(coef(mn)),dec)[-1]
|
|
ci<-round(exp(confint(mn)),dec)[-1,]
|
|
lo<-ci[,1]
|
|
up<-ci[,2]
|
|
|
|
or_ci<-c("REF",paste0(co," (",lo," to ",up,")"))
|
|
|
|
nr<-c()
|
|
|
|
for (r in 1:length(levels(dat[,2]))){
|
|
vr<-levels(dat[,2])[r]
|
|
dr<-dat[dat[,2]==vr,]
|
|
n<-as.numeric(nrow(dr))
|
|
|
|
## Af en eller anden grund bliver der talt for mange med.
|
|
# nall<-as.numeric(nrow(dat[!is.na(dat[,2]),]))
|
|
nl<-levels(m)[r]
|
|
# pro<-round(n/nall*100,0)
|
|
# rt<-paste0(n," (",pro,"%)")
|
|
nr<-rbind(nr,cbind(nl,n))
|
|
}
|
|
|
|
mms<-data.frame(cbind(nr,or_ci,pv))
|
|
header<-data.frame(matrix(var,ncol = ncol(mms)))
|
|
names(header)<-names(mms)
|
|
|
|
ls<-list(unadjusted=data.frame(rbind(header,mms)))
|
|
|
|
actable <- coef(summary(ma))
|
|
pa <- actable[,4]
|
|
pa<-ifelse(pa<0.001,"<0.001",round(pa,3))
|
|
pa <- ifelse(pa<=0.05|pa=="<0.001",paste0("*",pa),
|
|
ifelse(pa>0.05&pa<=0.1,paste0(".",pa),pa))
|
|
|
|
apv<-pa[1:length(coef(ma))]
|
|
|
|
aco<-round(exp(coef(ma)),dec)
|
|
aci<-round(exp(confint(ma)),dec)
|
|
alo<-aci[,1]
|
|
aup<-aci[,2]
|
|
aor_ci<-paste0(aco," (",alo," to ",aup,")")
|
|
|
|
dat2<-dat[,-1]
|
|
# names(dat2)<-c(var,names(ads))
|
|
nq<-c()
|
|
|
|
for (i in 1:ncol(dat2)){
|
|
if (is.factor(dat2[,i])){
|
|
vec<-dat2[,i]
|
|
ns<-names(dat2)[i]
|
|
for (r in 1:length(levels(vec))){
|
|
vr<-levels(vec)[r]
|
|
dr<-vec[vec==vr]
|
|
n<-as.numeric(length(dr))
|
|
# nall<-as.numeric(nrow(dat[!is.na(dat2[,c(ns)]),]))
|
|
nl<-paste0(ns,levels(vec)[r])
|
|
# pro<-round(n/nall*100,0)
|
|
# rt<-paste0(n," (",pro,"%)")
|
|
nq<-rbind(nq,cbind(nl,n))
|
|
}
|
|
}
|
|
if (!is.factor(dat2[,i])){
|
|
num<-dat2[,i]
|
|
ns<-names(dat2)[i]
|
|
nall<-as.numeric(nrow(dat[!is.na(dat2[,c(ns)]),]))
|
|
nq<-rbind(nq,cbind(ns,nall))
|
|
}
|
|
}
|
|
|
|
rnames<-c()
|
|
|
|
for (i in 1:ncol(dat2)){
|
|
if (is.factor(dat2[,i])){
|
|
rnames<-c(rnames,names(dat2)[i],paste0(names(dat2)[i],levels(dat2[,i])))
|
|
}
|
|
if (!is.factor(dat2[,i])){
|
|
rnames<-c(rnames,paste0(names(dat2)[i],".all"),names(dat2)[i])
|
|
}
|
|
}
|
|
res<-cbind(aor_ci,apv)
|
|
rest<-data.frame(names=row.names(res),res,stringsAsFactors = F)
|
|
|
|
numb<-data.frame(names=nq[,c("nl")],N=nq[,c("n")],stringsAsFactors = F)
|
|
namt<-data.frame(names=rnames,stringsAsFactors = F)
|
|
|
|
coll<-left_join(left_join(namt,numb,by="names"),rest,by="names")
|
|
|
|
header<-data.frame(matrix("Adjusted",ncol = ncol(coll)))
|
|
names(header)<-names(coll)
|
|
|
|
ls$adjusted<-data.frame(rbind(header,coll))
|
|
|
|
fnames<-c("Variable","N","OR (95 % CI)","p value")
|
|
|
|
names(ls$unadjusted)<-fnames
|
|
names(ls$adjusted)<-fnames
|
|
|
|
return(ls)
|
|
}
|