Carry the last post-baseline score to each missed scheduled visit

1library(yamaa)23yamaa_spec("spec.yaml")$run()$output

1import yamaa23yamaa.spec("spec.yaml").run().output

1# Reference solution for the yamaa benchmark adam-adqs-missed-visit-locf (R track).2#3# Written from the benchmark's full prompt and its inputs alone, as an agent4# would write it. Harbor's oracle agent runs it to check that the prompt can5# be solved and that the grader scores a correct answer 1.67library(dplyr, warn.conflicts = FALSE)8library(readr)9library(tidyr)1011adsl <- read_csv(12  "/app/input/adsl.csv",13  col_types = cols(.default = col_character())14)15qs <- read_csv(16  "/app/input/qs.csv",17  col_types = cols(18    QSSEQ = col_integer(),19    QSSTRESN = col_double(),20    VISITNUM = col_integer(),21    .default = col_character()22  )23)2425# Every QS record is kept, an empty score included.26collected <- qs |>27  left_join(28    select(adsl, STUDYID, USUBJID, EFFFL),29    by = c("STUDYID", "USUBJID")30  ) |>31  transmute(32    STUDYID, USUBJID,33    PARAMCD = QSTESTCD, AVISIT = VISIT, AVISITN = VISITNUM,34    AVAL = QSSTRESN, DTYPE = NA_character_, QSSEQ, EFFFL35  )3637# Scores by subject and visit; a score of zero counts.38scored <- qs |>39  filter(!is.na(QSSTRESN)) |>40  select(STUDYID, USUBJID, VISITNUM, QSSTRESN, QSSEQ)4142scheduled <- tibble(43  AVISIT = c("Week 8", "Week 16", "Week 24"),44  AVISITN = c(8L, 16L, 24L)45)4647# Each missed scheduled visit of an efficacy subject, carried from the48# closest earlier post-baseline score; no such score adds no record.49missed <- adsl |>50  filter(EFFFL %in% "Y") |>51  select(STUDYID, USUBJID, EFFFL) |>52  crossing(scheduled) |>53  anti_join(54    scored,55    by = c("STUDYID", "USUBJID", "AVISITN" = "VISITNUM")56  )5758carried <- missed |>59  inner_join(60    scored,61    by = c("STUDYID", "USUBJID"),62    relationship = "many-to-many"63  ) |>64  filter(VISITNUM > 0, VISITNUM < AVISITN) |>65  group_by(STUDYID, USUBJID, AVISITN) |>66  slice_max(VISITNUM, n = 1, with_ties = FALSE) |>67  ungroup() |>68  transmute(69    STUDYID, USUBJID, PARAMCD = "ACTOT", AVISIT, AVISITN,70    AVAL = QSSTRESN, DTYPE = "LOCF", QSSEQ, EFFFL71  )7273adqs <- bind_rows(collected, carried) |>74  arrange(USUBJID, AVISITN, !is.na(DTYPE))7576dir.create("/app/output", showWarnings = FALSE)77write_csv(adqs, "/app/output/adqs.csv", na = "")

1# Reference solution for the yamaa benchmark adam-adqs-missed-visit-locf (Python track).2#3# Written from the benchmark's full prompt and its inputs alone, as an agent4# would write it. Harbor's oracle agent runs it to check that the prompt can5# be solved and that the grader scores a correct answer 1.67import csv89INPUT_DIR = "/app/input"10OUTPUT_DIR = "/app/output"1112SCHEDULED = [("Week 8", 8), ("Week 16", 16), ("Week 24", 24)]131415def main():16    with open(f"{INPUT_DIR}/adsl.csv", newline="") as f:17        adsl = list(csv.DictReader(f))18    efffl = {(r["STUDYID"], r["USUBJID"]): r["EFFFL"] for r in adsl}1920    with open(f"{INPUT_DIR}/qs.csv", newline="") as f:21        qs = list(csv.DictReader(f))2223    # Every QS record is kept, an empty score included.24    rows = []25    for r in qs:26        rows.append(27            {28                "STUDYID": r["STUDYID"],29                "USUBJID": r["USUBJID"],30                "PARAMCD": r["QSTESTCD"],31                "AVISIT": r["VISIT"],32                "AVISITN": r["VISITNUM"],33                "AVAL": r["QSSTRESN"],34                "DTYPE": "",35                "QSSEQ": r["QSSEQ"],36                "EFFFL": efffl.get((r["STUDYID"], r["USUBJID"]), ""),37            }38        )3940    # Scores by subject and visit number. A score of zero counts.41    scored = {}42    for r in qs:43        if r["QSSTRESN"].strip() != "":44            key = (r["STUDYID"], r["USUBJID"])45            scored.setdefault(key, {})[int(r["VISITNUM"])] = r4647    # A carried-forward record for each missed scheduled visit of an48    # efficacy subject, from the closest earlier post-baseline score.49    for (studyid, usubjid), flag in efffl.items():50        if flag != "Y":51            continue52        visits = scored.get((studyid, usubjid), {})53        for avisit, avisitn in SCHEDULED:54            if avisitn in visits:55                continue56            earlier = [n for n in visits if 0 < n < avisitn]57            if not earlier:58                continue59            source = visits[max(earlier)]60            rows.append(61                {62                    "STUDYID": studyid,63                    "USUBJID": usubjid,64                    "PARAMCD": "ACTOT",65                    "AVISIT": avisit,66                    "AVISITN": str(avisitn),67                    "AVAL": source["QSSTRESN"],68                    "DTYPE": "LOCF",69                    "QSSEQ": source["QSSEQ"],70                    "EFFFL": flag,71                }72            )7374    rows.sort(key=lambda r: (r["USUBJID"], int(r["AVISITN"]), r["DTYPE"]))7576    columns = [77        "STUDYID",78        "USUBJID",79        "PARAMCD",80        "AVISIT",81        "AVISITN",82        "AVAL",83        "DTYPE",84        "QSSEQ",85        "EFFFL",86    ]87    with open(f"{OUTPUT_DIR}/adqs.csv", "w", newline="") as f:88        writer = csv.DictWriter(f, fieldnames=columns)89        writer.writeheader()90        writer.writerows(rows)919293if __name__ == "__main__":94    main()

1/* opensas reference solution for the yamaa benchmark adam-adqs-missed-visit-locf.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data adsl;6  length STUDYID USUBJID EFFFL $1024;7  infile "/app/input/adsl.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. EFFFL : $char1024.;9run;1011data qs;12  length STUDYID DOMAIN USUBJID QSTESTCD QSTEST QSSTAT VISIT $1024;13  infile "/app/input/qs.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. DOMAIN : $char1024. USUBJID : $char1024. QSSEQ QSTESTCD : $char1024. QSTEST : $char1024. QSSTRESN QSSTAT : $char1024. VISITNUM VISIT : $char1024.;15run;1617proc sql;18  create table collected as select a.STUDYID,a.USUBJID,a.QSTESTCD as PARAMCD,a.VISIT as AVISIT,a.VISITNUM as AVISITN,a.QSSTRESN as AVAL,'' as DTYPE,a.QSSEQ,b.EFFFL from qs as a left join adsl as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID;19  create table scored as select STUDYID,USUBJID,QSTESTCD,VISITNUM,QSSTRESN,QSSEQ from qs where not missing(QSSTRESN);20quit;21data schedule; length AVISIT PARAMCD $1024; set adsl; if EFFFL='Y'; PARAMCD='ACTOT'; do AVISITN=8,16,24; AVISIT=catx(' ','Week',strip(put(AVISITN,best32.))); output; end; run;22proc sql;23  create table missed as select a.* from schedule as a left join scored as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.PARAMCD=b.QSTESTCD and a.AVISITN=b.VISITNUM where missing(b.VISITNUM);24  create table candidates as select a.*,b.QSSTRESN as AVAL,b.QSSEQ,b.VISITNUM as sourcevisit from missed as a inner join scored as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.PARAMCD=b.QSTESTCD and b.VISITNUM>0 and b.VISITNUM<a.AVISITN;25quit;26proc sort data=candidates; by STUDYID USUBJID PARAMCD AVISITN descending sourcevisit; run;27data carried; length DTYPE $11; set candidates; by STUDYID USUBJID PARAMCD AVISITN; if first.AVISITN; DTYPE='LOCF'; run;28data result; length DTYPE $11; set collected carried; run;2930proc sql;31  create table final as select STUDYID, USUBJID, PARAMCD, AVISIT, AVISITN, AVAL, DTYPE, QSSEQ, EFFFL from result;32quit;33proc export data=final outfile="/app/output/adqs.csv" dbms=csv replace; run;