← SASnRRetain maximum score upto the current observation

Introduction

The mirror of the running-minimum lesson: carry forward the largest score seen so far within each subject.
Running maximums show up in worst-observed-value summaries on the high side — highest lab value observed to date, peak pain score, and similar clinical derivations.
As with the minimum, the running maximum resets at the start of each subject so no values leak across subjects.
SAS uses RETAIN with max(score, maxscore), resetting on first.usubjid.
Tidyverse uses group_by() + cummax(), with -Inf as the NA placeholder so missing values never win the maximum.
Base R uses ave() with a helper that masks NAs as -Inf and calls cummax() within each subject.

Retain maximum score upto the current observation

SAS (Base SAS)+

*==============================================================================;
* Create the input dataset;
*==============================================================================;

data pmax;
    input usubjid visitnum score;
    datalines;
101 1 20
101 2 .
101 3 30
102 1 90
102 2 10
102 3 .
102 4 91
102 5 .
;
run;

*==============================================================================;
* Sort the dataset by usubjid and visitnum;
*==============================================================================;

proc sort data=pmax;
    by usubjid visitnum;
run;

*==============================================================================;
* Retain previous non-missing low score;
*==============================================================================;

data pmax01;
    set pmax;
    by usubjid visitnum;

    retain maxscore;

    * Store the original score;
    orig_score = score;

    if first.usubjid then maxscore = .;

    *Store maximum score and assign maxscore when score is missing;
    if not missing(score) then maxscore = max(score,maxscore);
    else score=maxscore;
run;

This is the mirror image of the running-minimum pattern — we carry forward the largest score seen so far within each subject.
After sorting by subject and visit, we reset maxscore at the start of each subject with if first.usubjid then maxscore = .;.
On every non-missing row, we update maxscore = max(score, maxscore). SAS treats missing maxscore as smaller than any real number, so the first non-missing score becomes the starting maximum naturally.
On missing-score rows we substitute score = maxscore, which fills the NA with the running high.
We keep orig_score so the original NAs remain visible after the fill.
After running, we verify maxscore never decreases within a subject and missing rows now carry the running maximum.

R (tidyverse)+

#==============================================================================;
#Retain previous non-missing high score;
#==============================================================================;

pmax <- tribble(
  ~usubjid,~visitnum,~score,
  101,1,20,
  101,2,NA,
  101,3,30,
  102,1,90,
  102,2,10,
  102,3,NA,
  102,4,91,
  102,5,NA
)


pmax01 <- pmax %>%
  arrange(usubjid,visitnum) %>% 
  group_by(usubjid) %>% 
  mutate(
    orig_score = score,
    temp_score = if_else(is.na(score),-Inf,score),
    maxscore=cummax(temp_score),
    score = if_else(is.na(score),maxscore,score)
  )

The tidyverse version mirrors the minimum case — same structure, opposite comparison.
We use cummax() to get the running maximum within each group.
cummax() also does not handle NAs the way SAS does, so we substitute them with -Inf in a temporary column first. -Inf is smaller than every real score, so it never wins the maximum.
group_by(usubjid) resets the cumulative maximum at each subject boundary — our analogue of if first.usubjid then maxscore = .;.
Then we assign maxscore into score only where the original value was NA, preserving real observations.
After running, we confirm that within each subject maxscore is monotonically non-decreasing and the filled rows show the running high.

Happy learning!

Search lessons

Introduction

Retain maximum score upto the current observation