Finding closest matching time for each patient

Question

I have two sets of data:

First set:

 patient<-c("A","A","B","B","C","C","C","C")
 arrival<-c("11:00","11:00","13:00","13:00","14:00","14:00","14:00","14:00")
 lastRow<-c("","Yes","","Yes","","","","Yes")

 data1<-data.frame(patient,arrival,lastRow)

Another set of data:

 patient<-c("A","A","A","A","B","B","B","C","C","C")
 availableSlot<-c("11:15","11:35","11:45","11:55","12:55","13:55","14:00","14:00","14:10","17:00")

 data2<-data.frame(patient, availableSlot)

I want to create add a column to the first dataset such that for each last row of each patient, it shows the available slot that is closest to the arrival time:

The results would be:

  patient arrival lastRow availableSlot
       A   11:00        
       A   11:00     Yes     11:15
       B   13:00        
       B   13:00     Yes     12:55
       C   14:00        
       C   14:00        
       C   14:00        
       C   14:00     Yes     14:00

Would appreciate if anyone can tell me how I can implement this in R.

Answer 1

I'd use data.table, first cleaning up by converting to ITime and ignoring redundant rows:

library(data.table)
setDT(data1)[, arrival := as.ITime(as.character(arrival))]
setDT(data2)[, availableSlot := as.ITime(as.character(availableSlot))]
DT1 = unique(data1, by="patient", fromLast=TRUE)

Then you can do a "rolling join":

res = data2[DT1, on=.(patient, availableSlot = arrival), roll="nearest", 
  .(patient, availableSlot = x.availableSlot)]

#    patient availableSlot
# 1:       A      11:15:00
# 2:       B      12:55:00
# 3:       C      14:00:00

How it works

The syntax is x[i, on=, roll=, j].

• on= are the merge-by columns.
• It's a join: for each row of i, we are looking for matches in x.
• With roll="nearest", the final column in the on= is "rolled" to its nearest match.
• The on= columns in the original tables can be referenced with x.* and i.* prefixes.
• The j argument should give a list of columns, and .() is an alias for list() here.

Check out the package's introductory materials at http://r-datatable.com/Getting-started and type ?data.table for the docs relevant to rolling joins.

---

I would stop at res, but if you really want it back in your original table...

# a very nonstandard step:
data1[lastRow == "Yes", availableSlot := res$availableSlot ]

#    patient  arrival lastRow availableSlot
# 1:       A 11:00:00                  <NA>
# 2:       A 11:00:00     Yes      11:15:00
# 3:       B 13:00:00                  <NA>
# 4:       B 13:00:00     Yes      12:55:00
# 5:       C 14:00:00                  <NA>
# 6:       C 14:00:00                  <NA>
# 7:       C 14:00:00                  <NA>
# 8:       C 14:00:00     Yes      14:00:00

Now, data1 has availableSlot in a new column, similar to when you do data1$col <- val.

Answer 2

Here is a solution (based on joel.wilson's answer to my question) that will work with base R

#Convert dates to POSIXct format
data1$arrival = as.POSIXct(data1$arrival, format = "%H:%M")
data2$availableSlot = as.POSIXct(data2$availableSlot, format = "%H:%M")

#Lookup times from data2$availableSlot closest to data1$arrival
data1$availableSlot = sapply(data1$arrival, function(x)
                    data2$availableSlot[which.min(abs(x - data2$availableSlot))])

#Keep just hour and minutes
data1$availableSlot = strftime(as.POSIXct(data1$availableSlot, 
                                origin = "1970-01-01"), format = "%H:%M")
data1$arrival = strftime(as.POSIXct(data1$arrival), format = "%H:%M")

#Remove times when lastrow is empty
data1$availableSlot[which(data1$lastRow != "Yes")] = ""