The attached workflow does what you want by introducing an artificial ID and rejoining the relevant values. Maybe there is a simpler way. LAG column only seems to work for referring to the rows before.
Also if you have ‘messy’ CSV files you might want to consider using the R package readr to help you with that (cf. Error with File Reader node while running in a loop)
kn_example_lag_messy_csv.knwf (49.5 KB)