Treatment emergence by phase in a neoadjuvant, surgery and adjuvant study

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

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

1# Reference solution for the yamaa benchmark adam-adae-phase-emergence (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 1.67# Derive ADAE with phase-specific treatment-emergent flags from ADSL and AE.8suppressPackageStartupMessages({9  library(dplyr)10  library(readr)11})1213input_dir <- "/app/input"14output_dir <- "/app/output"15dir.create(output_dir, showWarnings = FALSE, recursive = TRUE)1617adsl <- read_csv(file.path(input_dir, "adsl.csv"),18                 col_types = cols(.default = col_character()),19                 na = c("", "NA"))20ae <- read_csv(file.path(input_dir, "ae.csv"),21               col_types = cols(.default = col_character()),22               na = c("", "NA"))2324to_date <- function(x) as.Date(x, format = "%Y-%m-%d")2526adsl <- adsl %>%27  mutate(across(matches("^TR0[123][SE]DT$"), to_date))2829# Determine the phase (1, 2, 3) an event started in, or NA when none.30phase_of_event <- function(astdt, aeser, s1, e1, s2, e2, s3, e3, eotstt) {31  if (is.na(astdt)) return(NA_integer_)32  starts <- c(s1, s2, s3)33  ends <- c(e1, e2, e3)34  entered <- which(!is.na(starts))35  if (length(entered) == 0) return(NA_integer_)36  if (astdt < starts[entered[1]]) return(NA_integer_)37  for (i in seq_along(entered)) {38    k <- entered[i]39    if (astdt < starts[k]) next40    if (i < length(entered)) {41      phase_end <- starts[entered[i + 1]] - 142      if (astdt <= phase_end) return(k)43    } else {44      if (!is.na(eotstt) && eotstt == "ONGOING") return(k)45      if (is.na(ends[k])) return(NA_integer_)46      window <- if (!is.na(aeser) && aeser == "Y") 90 else if (!is.na(aeser) && aeser == "N") 30 else NA47      if (is.na(window)) return(NA_integer_)48      if (astdt <= ends[k] + window) return(k)49    }50  }51  NA_integer_52}5354adae <- ae %>%55  left_join(adsl %>% select(-STUDYID), by = "USUBJID") %>%56  mutate(57    ASTDT = to_date(AESTDTC),58    AESEQ = as.integer(AESEQ)59  )6061adae$PHASE <- vapply(62  seq_len(nrow(adae)),63  function(r) {64    with(adae[r, ], phase_of_event(ASTDT, AESER, TR01SDT, TR01EDT, TR02SDT,65                                   TR02EDT, TR03SDT, TR03EDT, EOTSTT))66  },67  integer(1)68)6970adae <- adae %>%71  mutate(72    TREM01FL = if_else(!is.na(PHASE) & PHASE == 1L, "Y", NA_character_),73    TREM02FL = if_else(!is.na(PHASE) & PHASE == 2L, "Y", NA_character_),74    TREM03FL = if_else(!is.na(PHASE) & PHASE == 3L, "Y", NA_character_),75    TRTEMFL = if_else(!is.na(PHASE), "Y", NA_character_)76  ) %>%77  arrange(STUDYID, USUBJID, AESEQ) %>%78  select(STUDYID, USUBJID, AESEQ, AETERM, AESER, ASTDT,79         TREM01FL, TREM02FL, TREM03FL, TRTEMFL)8081write_csv(adae %>% mutate(ASTDT = format(ASTDT, "%Y-%m-%d")),82          file.path(output_dir, "adae.csv"), na = "")

1# Reference solution for the yamaa benchmark adam-adae-phase-emergence (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 1.67"""Derive ADAE with phase-specific treatment-emergent flags from ADSL and AE."""89from datetime import date, timedelta10from pathlib import Path1112import polars as pl1314INPUT = Path("/app/input")15OUTPUT = Path("/app/output")1617PHASES = [("TR01SDT", "TR01EDT"), ("TR02SDT", "TR02EDT"), ("TR03SDT", "TR03EDT")]18FLAGS = ["TREM01FL", "TREM02FL", "TREM03FL"]19COLUMNS = ["STUDYID", "USUBJID", "AESEQ", "AETERM", "AESER", "ASTDT",20           "TREM01FL", "TREM02FL", "TREM03FL", "TRTEMFL"]21WINDOW = {"N": 30, "Y": 90}222324def to_date(value):25    """Return a date for a complete ISO 8601 date, otherwise None."""26    if value is None:27        return None28    text = str(value).strip()29    if len(text) < 10:30        return None31    try:32        return date.fromisoformat(text[:10])33    except ValueError:34        return None353637def phase_windows(subject, aeser):38    """List (phase index, start, end) for the phases a subject entered; end None = open."""39    entered = []40    for index, (start_col, end_col) in enumerate(PHASES):41        start = to_date(subject.get(start_col))42        if start is not None:43            entered.append((index, start, to_date(subject.get(end_col))))44    windows = []45    for position, (index, start, last) in enumerate(entered):46        if position + 1 < len(entered):47            end = entered[position + 1][1] - timedelta(days=1)48        elif str(subject.get("EOTSTT") or "").strip().upper() == "ONGOING":49            end = None50        elif last is None:51            end = None52        else:53            end = last + timedelta(days=WINDOW.get(aeser, 30))54        windows.append((index, start, end))55    return windows565758def derive_flags(subject, astdt, aeser):59    flags = [None, None, None]60    if subject is None or astdt is None:61        return flags62    if to_date(subject.get("TR01SDT")) is None:63        return flags64    for index, start, end in phase_windows(subject, aeser):65        if astdt >= start and (end is None or astdt <= end):66            flags[index] = "Y"67            break68    return flags697071def main():72    # Every column is read as text; an empty field is null.73    adsl = pl.read_csv(INPUT / "adsl.csv", infer_schema=False)74    ae = pl.read_csv(INPUT / "ae.csv", infer_schema=False)75    subjects = {row["USUBJID"]: row for row in adsl.to_dicts()}7677    records = []78    for row in ae.to_dicts():79        astdt = to_date(row.get("AESTDTC"))80        aeser = (row.get("AESER") or "").strip().upper()81        flags = derive_flags(subjects.get(row["USUBJID"]), astdt, aeser)82        records.append({83            "STUDYID": row["STUDYID"],84            "USUBJID": row["USUBJID"],85            "AESEQ": int(row["AESEQ"]),86            "AETERM": row["AETERM"],87            "AESER": row["AESER"],88            "ASTDT": astdt.isoformat() if astdt else None,89            "TREM01FL": flags[0],90            "TREM02FL": flags[1],91            "TREM03FL": flags[2],92            "TRTEMFL": "Y" if "Y" in flags else None,93        })9495    schema = {name: pl.Utf8 for name in COLUMNS} | {"AESEQ": pl.Int64}96    adae = pl.DataFrame(records, schema=schema).sort(97        ["STUDYID", "USUBJID", "AESEQ"], maintain_order=True98    )99    OUTPUT.mkdir(parents=True, exist_ok=True)100    adae.write_csv(OUTPUT / "adae.csv")101102103if __name__ == "__main__":104    main()

1/* opensas reference solution for the yamaa benchmark adam-adae-phase-emergence.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 TR01SDT TR01EDT TR02SDT TR02EDT TR03SDT TR03EDT EOTSTT $1024;7  infile "/app/input/adsl.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. TR01SDT : $char1024. TR01EDT : $char1024.9        TR02SDT : $char1024. TR02EDT : $char1024. TR03SDT : $char1024. TR03EDT : $char1024.10        EOTSTT : $char1024.;11run;1213data ae;14  length STUDYID USUBJID AETERM AESER AESTDTC $1024;15  infile "/app/input/ae.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;16  input STUDYID : $char1024. USUBJID : $char1024. AESEQ AETERM : $char1024. AESER : $char1024.17        AESTDTC : $char1024.;18run;1920proc sql;21  create table joined as22  select a.*, b.TR01SDT, b.TR01EDT, b.TR02SDT, b.TR02EDT, b.TR03SDT, b.TR03EDT, b.EOTSTT23  from ae as a left join adsl as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID;24quit;2526data events;27  length ASTDT $10 TREM01FL TREM02FL TREM03FL TRTEMFL $1;28  set joined;29  ASTDT=AESTDTC; TREM01FL=''; TREM02FL=''; TREM03FL=''; TRTEMFL='';30  onset=.; s1=.; e1=.; s2=.; e2=.; s3=.; e3=.;31  if not missing(AESTDTC) then onset=input(AESTDTC, yymmdd10.);32  if not missing(TR01SDT) then s1=input(TR01SDT, yymmdd10.);33  if not missing(TR01EDT) then e1=input(TR01EDT, yymmdd10.);34  if not missing(TR02SDT) then s2=input(TR02SDT, yymmdd10.);35  if not missing(TR02EDT) then e2=input(TR02EDT, yymmdd10.);36  if not missing(TR03SDT) then s3=input(TR03SDT, yymmdd10.);37  if not missing(TR03EDT) then e3=input(TR03EDT, yymmdd10.);38  /* Phase 1 ends the day before the next phase the subject entered starts:39     surgery, or the first adjuvant dose for a subject who had no surgery. */40  next1=s2;41  if missing(next1) then next1=s3;42  /* The last phase a subject entered ends 30 days after its last date for a43     non-serious event and 90 days after it for a serious one, both days44     included, and has no end while treatment is ongoing. */45  lastend=e1;46  if not missing(s3) then lastend=e3;47  else if not missing(s2) then lastend=e2;48  tail=.;49  if AESER='Y' then tail=90;50  else if AESER='N' then tail=30;51  inlast=0;52  if EOTSTT='ONGOING' then inlast=1;53  else if not missing(onset) and not missing(lastend) and not missing(tail) then do;54    if onset<=lastend+tail then inlast=1;55  end;56  /* SAS missing dates sort below every date, so each phase start is tested57     for presence before the onset is compared with it. An untreated subject58     has no phase start and so no flag. */59  if not missing(onset) and not missing(s1) then do;60    if onset>=s1 then do;61      if not missing(next1) then do;62        if onset<next1 then TREM01FL='Y';63      end;64      else if inlast=1 then TREM01FL='Y';65    end;66  end;67  if not missing(onset) and not missing(s2) then do;68    if onset>=s2 then do;69      if not missing(s3) then do;70        if onset<s3 then TREM02FL='Y';71      end;72      else if inlast=1 then TREM02FL='Y';73    end;74  end;75  if not missing(onset) and not missing(s3) then do;76    if onset>=s3 and inlast=1 then TREM03FL='Y';77  end;78  if TREM01FL='Y' or TREM02FL='Y' or TREM03FL='Y' then TRTEMFL='Y';79run;8081proc sql;82  create table final as83  select STUDYID, USUBJID, AESEQ, AETERM, AESER, ASTDT, TREM01FL, TREM02FL, TREM03FL, TRTEMFL84  from events;85quit;86proc export data=final outfile="/app/output/adae.csv" dbms=csv replace; run;