Last Observation Before Exposure

1library(yamaa)23lb <- yamaa_domain("spec.yaml")$output4lb

1import yamaa23lb = yamaa.yamaa_domain("spec.yaml").output4lb

1# Reference solution for the yamaa benchmark sdtm-lb-last-observation-before-exposure (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(purrr)1011odm <- read_csv(12  "/app/input/odm.csv",13  col_types = cols(.default = col_character())14) |>15  mutate(Value = na_if(Value, ""))1617dm <- read_csv(18  "/app/input/dm.csv",19  col_types = cols(.default = col_character())20) |>21  mutate(RFSTDTC = na_if(RFSTDTC, ""))2223mapping <- read_csv(24  "/app/input/lb_mapping.csv",25  col_types = cols(.default = col_character())26)2728groups <- odm |>29  distinct(StudyOID, SubjectKey, ItemGroupRepeatKey)3031one_group <- function(study, subj, igrep) {32  sub <- odm |>33    filter(SubjectKey == subj, ItemGroupRepeatKey == igrep)34  get <- function(oid) {35    hit <- sub |> filter(ItemOID == oid)36    if (nrow(hit) == 0) return(NA_character_)37    hit$Value[[1]]38  }39  lbdtc <- get("IT.LB.LBDTC")40  found <- mapping |>41    filter(ItemOID %in% sub$ItemOID) |>42    slice_head(n = 1)43  if (nrow(found) == 0) return(tibble())44  raw <- get(found$ItemOID[[1]])45  has_result <- !is.na(raw)46  tibble(47    DOMAIN = "LB",48    STUDYID = study,49    USUBJID = subj,50    LBSEQ = as.integer(igrep),51    LBTESTCD = found$LBTESTCD[[1]],52    LBSPEC = found$LBSPEC[[1]],53    LBORRES = raw,54    LBDTC = lbdtc,55    LBSTAT = if_else(has_result, NA_character_, "NOT DONE"),56    HAS = has_result57  )58}5960lb <- pmap_dfr(61  list(groups$StudyOID, groups$SubjectKey, groups$ItemGroupRepeatKey),62  one_group63)6465flags <- lb |>66  filter(HAS, !is.na(LBDTC)) |>67  left_join(dm |> select(USUBJID, RFSTDTC), by = "USUBJID") |>68  mutate(DAYDATE = substr(LBDTC, 1, 10), REFDAY = substr(RFSTDTC, 1, 10)) |>69  filter(is.na(REFDAY) | DAYDATE <= REFDAY) |>70  arrange(USUBJID, LBTESTCD, LBSPEC, DAYDATE, LBSEQ) |>71  group_by(USUBJID, LBTESTCD, LBSPEC) |>72  slice_tail(n = 1) |>73  ungroup() |>74  transmute(USUBJID, LBTESTCD, LBSPEC, FLAGSEQ = LBSEQ)7576lb <- lb |>77  left_join(flags, by = c("USUBJID", "LBTESTCD", "LBSPEC")) |>78  mutate(79    LBLOBXFL = if_else(HAS & !is.na(FLAGSEQ) & LBSEQ == FLAGSEQ, "Y", NA_character_)80  ) |>81  arrange(STUDYID, USUBJID, LBSEQ) |>82  select(83    DOMAIN, STUDYID, USUBJID, LBSEQ, LBTESTCD, LBSPEC, LBORRES,84    LBDTC, LBSTAT, LBLOBXFL85  )8687dir.create("/app/output", showWarnings = FALSE)88write_csv(lb, "/app/output/lb.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-lb-last-observation-before-exposure (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 1.67from pathlib import Path89import polars as pl1011odm = pl.read_csv("/app/input/odm.csv", infer_schema=False)12dm = pl.read_csv("/app/input/dm.csv", infer_schema=False)13mapping = pl.read_csv("/app/input/lb_mapping.csv", infer_schema=False)1415ref_of = {16    r["USUBJID"]: (r["RFSTDTC"] if r["RFSTDTC"] not in (None, "") else None)17    for r in dm.to_dicts()18}19map_of = {20    r["ItemOID"]: (r["LBTESTCD"], r["LBSPEC"]) for r in mapping.to_dicts()21}22study_of = {}23groups = {}24for r in odm.to_dicts():25    study_of[r["SubjectKey"]] = r["StudyOID"]26    key = (r["SubjectKey"], r["ItemGroupRepeatKey"])27    groups.setdefault(key, {})[r["ItemOID"]] = r["Value"]2829rows = []30for (subj, repkey), items in groups.items():31    lbdtc = items.get("IT.LB.LBDTC")32    lbdtc = lbdtc if lbdtc not in (None, "") else None33    # The test item in this group, via the item dictionary.34    found = None35    for oid, (code, spec) in map_of.items():36        if oid in items:37            found = (oid, code, spec)38            break39    if found is None:40        continue41    oid, code, spec = found42    raw = items.get(oid)43    has_result = raw is not None and raw != ""44    lbstat = items.get("IT.LB.LBSTAT")45    rows.append(46        {47            "DOMAIN": "LB",48            "STUDYID": study_of[subj],49            "USUBJID": subj,50            "LBSEQ": int(repkey),51            "LBTESTCD": code,52            "LBSPEC": spec,53            "LBORRES": raw if has_result else None,54            "LBDTC": lbdtc,55            "LBSTAT": "NOT DONE" if not has_result else None,56            "_HAS": has_result,57        }58    )5960# Latest record with a result on or before the reference start date,61# per subject, test, and specimen; ties go to the higher sequence number.62flag_keys = {}63for row in sorted(rows, key=lambda r: ((r["LBDTC"] or "")[:10], r["LBSEQ"])):64    if not row["_HAS"] or row["LBDTC"] is None:65        continue66    ref = ref_of.get(row["USUBJID"])67    if ref is not None and row["LBDTC"][:10] > ref[:10]:68        continue69    flag_keys[(row["USUBJID"], row["LBTESTCD"], row["LBSPEC"])] = row["LBSEQ"]7071for row in rows:72    key = (row["USUBJID"], row["LBTESTCD"], row["LBSPEC"])73    row["LBLOBXFL"] = (74        "Y" if row["_HAS"] and flag_keys.get(key) == row["LBSEQ"] else None75    )76    del row["_HAS"]7778lb = (79    pl.DataFrame(80        rows,81        schema={82            "DOMAIN": pl.String,83            "STUDYID": pl.String,84            "USUBJID": pl.String,85            "LBSEQ": pl.Int64,86            "LBTESTCD": pl.String,87            "LBSPEC": pl.String,88            "LBORRES": pl.String,89            "LBDTC": pl.String,90            "LBSTAT": pl.String,91            "LBLOBXFL": pl.String,92        },93    )94    .sort(["STUDYID", "USUBJID", "LBSEQ"])95    .select(96        "DOMAIN",97        "STUDYID",98        "USUBJID",99        "LBSEQ",100        "LBTESTCD",101        "LBSPEC",102        "LBORRES",103        "LBDTC",104        "LBSTAT",105        "LBLOBXFL",106    )107)108109Path("/app/output").mkdir(parents=True, exist_ok=True)110lb.write_csv("/app/output/lb.csv")

1/* opensas reference solution for the yamaa benchmark sdtm-lb-last-observation-before-exposure.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data dm;6  length STUDYID USUBJID RFSTDTC $1024;7  infile "/app/input/dm.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. RFSTDTC : $char1024.;9run;1011data lb_mapping;12  length ItemOID LBTESTCD LBSPEC $1024;13  infile "/app/input/lb_mapping.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input ItemOID : $char1024. LBTESTCD : $char1024. LBSPEC : $char1024.;15run;1617data odm;18  length StudyOID MetaDataVersionOID SubjectKey StudyEventOID StudyEventRepeatKey FormOID FormRepeatKey ItemGroupOID ItemGroupRepeatKey ItemOID Value $1024;19  infile "/app/input/odm.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;20  input StudyOID : $char1024. MetaDataVersionOID : $char1024. SubjectKey : $char1024. StudyEventOID : $char1024. StudyEventRepeatKey : $char1024. FormOID : $char1024. FormRepeatKey : $char1024. ItemGroupOID : $char1024. ItemGroupRepeatKey : $char1024. ItemOID : $char1024. Value : $char1024.;21run;2223proc sort data=odm out=ordered; by StudyOID SubjectKey StudyEventOID StudyEventRepeatKey FormOID FormRepeatKey ItemGroupRepeatKey; run;24data dates;25  length LBDTC LBSTAT $1024;26  retain LBDTC LBSTAT;27  set ordered;2829  by StudyOID SubjectKey StudyEventOID StudyEventRepeatKey FormOID FormRepeatKey ItemGroupRepeatKey;30  if first.ItemGroupRepeatKey then do;31    LBDTC=''; LBSTAT='';32  end;33    if ItemOID='IT.LB.LBDTC' then LBDTC=Value;34  if ItemOID='IT.LB.LBSTAT' then LBSTAT=Value;35  if last.ItemGroupRepeatKey then output;36run;37proc sql;38create table joined as select a.*,b.LBDTC,b.LBSTAT,c.LBTESTCD,c.LBSPEC,d.RFSTDTC from odm a inner join dates b on a.StudyOID=b.StudyOID and a.SubjectKey=b.SubjectKey and a.StudyEventOID=b.StudyEventOID and a.StudyEventRepeatKey=b.StudyEventRepeatKey and a.FormOID=b.FormOID and a.FormRepeatKey=b.FormRepeatKey and a.ItemGroupRepeatKey=b.ItemGroupRepeatKey inner join lb_mapping c on a.ItemOID=c.ItemOID left join dm d on a.StudyOID=d.STUDYID and a.SubjectKey=d.USUBJID; quit;39data collected; set joined; length DOMAIN $2 STUDYID USUBJID LBORRES DAYDATE REFDAY $1024; DOMAIN='LB'; STUDYID=StudyOID; USUBJID=SubjectKey; LBSEQ=input(ItemGroupRepeatKey,best32.); LBORRES=Value;40if missing(Value) then LBSTAT='NOT DONE'; if LBSTAT='NOT DONE' then LBORRES=''; DAYDATE=''; REFDAY=''; if lengthn(LBDTC)>=10 then DAYDATE=substr(LBDTC,1,10); if lengthn(RFSTDTC)>=10 then REFDAY=substr(RFSTDTC,1,10); run;41data eligible; set collected; if not missing(LBORRES) and not missing(DAYDATE) and (missing(REFDAY) or DAYDATE<=REFDAY); run;42proc sort data=eligible; by STUDYID USUBJID LBTESTCD LBSPEC descending DAYDATE descending LBSEQ; run;43data flagged; set eligible; by STUDYID USUBJID LBTESTCD LBSPEC; if first.LBSPEC; length LBLOBXFL $1; LBLOBXFL='Y'; run;44proc sql; create table result as select a.*,b.LBLOBXFL from collected a left join flagged b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.LBSEQ=b.LBSEQ and a.LBTESTCD=b.LBTESTCD and a.LBSPEC=b.LBSPEC; quit;45proc sql;46  create table final as select DOMAIN, STUDYID, USUBJID, LBSEQ, LBTESTCD, LBSPEC, LBORRES, LBDTC, LBSTAT, LBLOBXFL from result;47quit;48proc export data=final outfile="/app/output/lb.csv" dbms=csv replace; run;