detect if drug was given in interval - r

Question

can't figure this out, despite being rather close (supposedly). I want to check if a drug was given in a 4 hour window.
drug start stop
1 A 1 3
2 A 7 10
3 A 11 17
drug A was started on time 1 and was administered up to time 3; then started again at time 7 and given up to time 10 etc.
t1 t2
1 0 4
2 4 8
3 8 12
4 12 16
5 16 20
6 20 24
these are the windows in question
DATA:
t1 <- c(0,4,8,12,16,20)
t2 <- t1 + 4
chunks <- data.frame(t1=t1,t2=t2)
drug <- "A"
start <- c(1,7,11)
stop <- c(3,10,17)
times <- data.frame(drug,start,stop)
Expected solution
t1 t2 lsg
1 0 4 1
2 4 8 1
3 8 12 1
4 12 16 1
5 16 20 1
6 20 24 0
Attempt at solution
test <- function(){
n <- 1
for (row in times){
result <- (times$start[n] > chunks$t1 & times$stop[n] < chunks$t2) | ((times$start[n] > chunks$t1 & times$start[n] < chunks$t2) & (times$stop[n] > chunks$t2 | times$stop[n] < chunks$t2)) | (times$start[n] < chunks$t1 & times$stop[n] > chunks$t1)
n <- n + 1
print(result)
}
}
gives
[1] TRUE FALSE FALSE FALSE FALSE FALSE
[1] FALSE TRUE TRUE FALSE FALSE FALSE
[1] FALSE FALSE TRUE TRUE TRUE FALSE
which is correct! First administration fell into the first time window. 2nd and 3rd administration fell into the 2nd and 3rd windows etc. But how to get to the
expected solution?
As I said, I feel close but I don't know how to join the results to the chunks-df...

Answer 1

The first-half of this is #akrun's comment, but expanded to include the prerequisites. (If you come back and answer, I'll happily defer to you ... just giving more details here.) The second-half is new (and often over-looked).
data.table
data.table::foverlaps does joins based on overlaps/inequalities (as opposed to base merge and dplyr::*_join, which only operate on strict equalities). One prerequisite for using overlaps (in addition to being data.table class) is that the time fields be keyed correctly.
library(data.table)
setDT(times)
setDT(chunks)
# set the keys
setkey(times, start, stop)
setkey(chunks, t1, t2)
# the join
+(!is.na(foverlaps(chunks, times, which = TRUE, mult = 'first')))
# [1] 1 1 1 1 1 0
The function actually returns which row(s) each row in times corresponds to in chunks:
foverlaps(chunks, times, which = TRUE, mult = 'first')
# [1] 1 2 2 3 3 NA
sqldf
data.table is not the only R tool that lets this happen. This solution works on any variant of data.frame (base, data.table, or tbl_df).
Here's this:
library(sqldf)
sqldf("
select c.t1, c.t2,
(case when drug is null then 0 else 1 end) > 0 as n
from chunks c
left join times t on
(t.start between c.t1 and c.t2) or (t.stop between c.t1 and c.t2)
or (c.t1 between t.start and t.stop) or (c.t2 between t.start and t.stop)
group by c.t1, c.t2")
# t1 t2 n
# 1 0 4 1
# 2 4 8 1
# 3 8 12 1
# 4 12 16 1
# 5 16 20 1
# 6 20 24 0
(I don't know if it's possible to reduce the logic of that join, nor if it will mis-behave with other data.)
If you need the count of drugs that occur in each time frame, I think you can use sum(case when ... end) as n.