对某些列的跨行求和，如果全部为 NA，则保留 NA

Question

I have clinical data that looks something like this... I have a bunch of different binary outcomes but I only want to sum a few of the outcomes to create a total outcome/composite score. My data looks something like this

``patientid <- c(100,101,102,103,104,105,106)
outcome1 <- c(0,NA,1,0,1,NA,1)
outcome2 <- c(0,1,1,0,0,NA,1) 
outcome3 <- c(0,NA,NA,0,1,NA,0)
outcome4 <- c(NA,NA,NA,0,1,NA,0)
Data<-data.frame(patientid=patientid,outcome1=outcome1,outcome2=outcome2,outcome3=outcome3,outcome4=outcome4)
Data''

Now I want to create a composite score for just three of the outcomes. NA should count as a zero UNLESS it is NA in every outcome chosen to sum in which case it will stay NA. I assume this is done with rowsums? Here is what my desire database should like (summing just outcome 1, 2, 4)

``patientid <- c(100,101,102,103,104,105,106)
  outcome1 <- c(0,NA,1,0,1,NA,1)
  outcome2 <- c(0,1,1,0,1,NA,1) 
  outcome3 <- c(0,NA,NA,0,1,NA,0)
  outcome4 <- c(NA,NA,NA,0,1,NA,0)
  composite <- c(0,1,2,0,3,NA,2)
 data.frame(patientid=patientid,outcome1=outcome1,outcome2=outcome2,outcome3=outcome3,outcome4=outcome4, composite= composite)
    Data''

Answer 1

Try this approach using c_across() . I am a bit confused on why the final output has some columns different from the original output. You can use c_across() and rowwise() to sum certain rows and after that flag those with all ``NA`. Here the code:

library(tidyverse)
#Code
NewData <- Data %>% rowwise(patientid) %>% 
  mutate(Composite=sum(c_across(c(outcome1,outcome2,outcome4)),na.rm=T)) %>%
  mutate(Flag=ifelse(sum(!is.na(c_across(c(outcome1,outcome2,outcome4))))==0,1,0),
         Composite=ifelse(Flag==1,NA,Composite)) %>% select(-Flag)

Output:

# A tibble: 7 x 6
# Rowwise:  patientid
  patientid outcome1 outcome2 outcome3 outcome4 Composite
      <dbl>    <dbl>    <dbl>    <dbl>    <dbl>     <dbl>
1       100        0        0        0       NA         0
2       101       NA        1       NA       NA         1
3       102        1        1       NA       NA         2
4       103        0        0        0        0         0
5       104        1        0        1        1         2
6       105       NA       NA       NA       NA        NA
7       106        1        1        0        0         2

Answer 2

In base R, you can use rowSums :

#select the columns that we want to count
cols <- paste0('outcome', c(1:2, 4))
#sum them rowwise
Data$composite <- rowSums(Data[cols], na.rm  =TRUE)
#turn all NA rows to NA.
Data$composite[rowSums(!is.na(Data[cols])) == 0] <- NA
Data

#  patientid outcome1 outcome2 outcome3 outcome4 composite
#1       100        0        0        0       NA         0
#2       101       NA        1       NA       NA         1
#3       102        1        1       NA       NA         2
#4       103        0        0        0        0         0
#5       104        1        0        1        1         2
#6       105       NA       NA       NA       NA        NA
#7       106        1        1        0        0         2

Answer 3

library(tidyverse)

Data %>%
  rowwise() %>%
  mutate(
    Composite = if_else(
      c(outcome1, outcome2, outcome4) %>% is.na() %>% mean() %>% `==`(1), # looking for cases where all columns are NA
      NA_real_, # all NA columns produce NA
      c(outcome1, outcome2, outcome4) %>% sum(na.rm = T) # for other columns, NAs are treated as 0s
      )
  )

#  patientid outcome1 outcome2 outcome3 outcome4 composite
#1       100        0        0        0       NA         0
#2       101       NA        1       NA       NA         1
#3       102        1        1       NA       NA         2
#4       103        0        0        0        0         0
#5       104        1        0        1        1         2
#6       105       NA       NA       NA       NA        NA
#7       106        1        1        0        0         2