Progression-Free Survival

1library(yamaa)23adtte <- yamaa_domain("spec.yaml")$output4adtte

1import yamaa23adtte = yamaa.yamaa_domain("spec.yaml").output4adtte

1# Reference solution for the yamaa benchmark adam-adtte-pfs (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)910adsl <- read_csv(11  "/app/input/adsl.csv",12  col_types = cols(13    RANDDT = col_date(), NTXSTDT = col_date(), .default = col_character()14  )15)16rs <- read_csv(17  "/app/input/rs.csv",18  col_types = cols(RSSEQ = col_integer(), RSDTC = col_date(), .default = col_character())19)20ds <- read_csv(21  "/app/input/ds.csv",22  col_types = cols(DSSEQ = col_integer(), DSDTC = col_date(), .default = col_character())23)2425# The first adequate progression, the first death, and the last adequate26# assessment. Records sharing a date are ordered by sequence number.27progression <- rs |>28  filter(RSTESTCD %in% "OVRLRESP", RSSTRESC %in% "PD", ADEQFL %in% "Y", !is.na(RSDTC)) |>29  arrange(USUBJID, RSDTC, RSSEQ) |>30  distinct(USUBJID, .keep_all = TRUE) |>31  select(USUBJID, PDDT = RSDTC, PDSEQ = RSSEQ)3233death <- ds |>34  filter(DSDECOD %in% "DEATH", !is.na(DSDTC)) |>35  arrange(USUBJID, DSDTC, DSSEQ) |>36  distinct(USUBJID, .keep_all = TRUE) |>37  select(USUBJID, DTHDT = DSDTC, DTHSEQ = DSSEQ)3839last_assessment <- rs |>40  filter(RSTESTCD %in% "OVRLRESP", ADEQFL %in% "Y", !is.na(RSDTC)) |>41  arrange(USUBJID, desc(RSDTC), desc(RSSEQ)) |>42  distinct(USUBJID, .keep_all = TRUE) |>43  select(USUBJID, CENSORDT = RSDTC, LASTSEQ = RSSEQ)4445# The last adequate assessment dated on or before new-therapy start, for46# subjects who began one.47pre_therapy <- rs |>48  inner_join(select(adsl, USUBJID, NTXSTDT), by = "USUBJID") |>49  filter(50    RSTESTCD %in% "OVRLRESP", ADEQFL %in% "Y", !is.na(RSDTC),51    !is.na(NTXSTDT), RSDTC <= NTXSTDT52  ) |>53  arrange(USUBJID, desc(RSDTC), desc(RSSEQ)) |>54  distinct(USUBJID, .keep_all = TRUE) |>55  select(USUBJID, PRECENSORDT = RSDTC, PRESEQ = RSSEQ)5657adtte <- adsl |>58  left_join(progression, by = "USUBJID") |>59  left_join(death, by = "USUBJID") |>60  left_join(last_assessment, by = "USUBJID") |>61  left_join(pre_therapy, by = "USUBJID") |>62  mutate(63    PARAMCD = "PFS",64    PARAM = "Progression-Free Survival",65    STARTDT = RANDDT,66    # A progression or death dated after new-therapy start is not an event.67    PDDT = if_else(!is.na(NTXSTDT) & PDDT > NTXSTDT, as.Date(NA), PDDT),68    DTHDT = if_else(!is.na(NTXSTDT) & DTHDT > NTXSTDT, as.Date(NA), DTHDT),69    # The censoring assessment: the last adequate one overall, or the last70    # one dated on or before therapy start when therapy began.71    CENSORASSESSDT = if_else(is.na(NTXSTDT), CENSORDT, PRECENSORDT),72    CENSORSEQ = if_else(is.na(NTXSTDT), LASTSEQ, PRESEQ),73    # With no usable adequate assessment the subject is censored at74    # randomization, for a 1-day PFS.75    CENSORDT = coalesce(CENSORASSESSDT, STARTDT),76    EVENTDT = case_when(77      !is.na(PDDT) & !is.na(DTHDT) ~ pmin(PDDT, DTHDT),78      !is.na(PDDT) ~ PDDT,79      !is.na(DTHDT) ~ DTHDT80    ),81    # The event always wins, even after the last adequate assessment,82    # unless new anti-cancer therapy started first.83    CNSR = if_else(!is.na(EVENTDT), 0L, 1L),84    ADT = coalesce(EVENTDT, CENSORDT),85    AVAL = as.integer(ADT - STARTDT) + 1L,86    # A progression and a death on the same day count as progression.87    EVNTDESC = case_when(88      is.na(EVENTDT) ~ "CENSORED",89      !is.na(PDDT) & (is.na(DTHDT) | PDDT <= DTHDT) ~ "DISEASE PROGRESSION",90      .default = "DEATH"91    ),92    SRCDOM = case_when(93      EVNTDESC == "DEATH" ~ "DS",94      is.na(CENSORASSESSDT) ~ NA_character_,95      .default = "RS"96    ),97    SRCVAR = case_when(98      EVNTDESC == "DEATH" ~ "DSDTC",99      is.na(CENSORASSESSDT) ~ NA_character_,100      .default = "RSDTC"101    ),102    SRCSEQ = case_when(103      EVNTDESC == "DISEASE PROGRESSION" ~ PDSEQ,104      EVNTDESC == "DEATH" ~ DTHSEQ,105      is.na(CENSORASSESSDT) ~ NA_integer_,106      .default = CENSORSEQ107    )108  ) |>109  arrange(USUBJID) |>110  select(111    STUDYID, USUBJID, PARAMCD, PARAM, STARTDT, ADT, AVAL, CNSR, EVNTDESC,112    SRCDOM, SRCVAR, SRCSEQ113  )114115dir.create("/app/output", showWarnings = FALSE)116write_csv(adtte, "/app/output/adtte.csv", na = "")

1# Reference solution for the yamaa benchmark adam-adtte-pfs (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 pl101112def read(name: str) -> pl.DataFrame:13    """Read a benchmark input CSV with every column as text."""14    return pl.read_csv(f"/app/input/{name}", infer_schema=False)151617def first_per_subject(records: pl.DataFrame, by: list[str]) -> pl.DataFrame:18    """Keep the earliest record per subject, breaking ties by sequence."""19    return records.sort(by).unique("USUBJID", keep="first", maintain_order=True)202122def last_per_subject(records: pl.DataFrame, by: list[str]) -> pl.DataFrame:23    """Keep the latest record per subject, breaking ties by sequence."""24    return records.sort(by, descending=True).unique(25        "USUBJID", keep="first", maintain_order=True26    )272829adsl = read("adsl.csv").with_columns(30    pl.col("RANDDT").str.to_date(strict=False),31    pl.col("NTXSTDT").str.to_date(strict=False),32)33rs = read("rs.csv").with_columns(34    pl.col("RSDTC").str.to_date(strict=False),35    pl.col("RSSEQ").cast(pl.Int64, strict=False),36)37ds = read("ds.csv").with_columns(38    pl.col("DSDTC").str.to_date(strict=False),39    pl.col("DSSEQ").cast(pl.Int64, strict=False),40)4142# The first adequate progression, the first death, and the last adequate43# assessment. Records sharing a date are ordered by sequence number.44progression = first_per_subject(45    rs.filter(46        (pl.col("RSTESTCD") == "OVRLRESP")47        & (pl.col("RSSTRESC") == "PD")48        & (pl.col("ADEQFL") == "Y")49        & pl.col("RSDTC").is_not_null()50    ),51    ["RSDTC", "RSSEQ"],52).select("USUBJID", PDDT="RSDTC", PDSEQ="RSSEQ")5354death = first_per_subject(55    ds.filter(56        (pl.col("DSDECOD") == "DEATH") & pl.col("DSDTC").is_not_null()57    ),58    ["DSDTC", "DSSEQ"],59).select("USUBJID", DTHDT="DSDTC", DTHSEQ="DSSEQ")6061last_assessment = last_per_subject(62    rs.filter(63        (pl.col("RSTESTCD") == "OVRLRESP")64        & (pl.col("ADEQFL") == "Y")65        & pl.col("RSDTC").is_not_null()66    ),67    ["RSDTC", "RSSEQ"],68).select("USUBJID", CENSORDT="RSDTC", LASTSEQ="RSSEQ")6970# The last adequate assessment dated on or before new-therapy start, for71# subjects who began one.72pre_therapy = last_per_subject(73    rs.join(adsl.select("USUBJID", "NTXSTDT"), on="USUBJID", how="left").filter(74        (pl.col("RSTESTCD") == "OVRLRESP")75        & (pl.col("ADEQFL") == "Y")76        & pl.col("RSDTC").is_not_null()77        & pl.col("NTXSTDT").is_not_null()78        & (pl.col("RSDTC") <= pl.col("NTXSTDT"))79    ),80    ["RSDTC", "RSSEQ"],81).select("USUBJID", PRECENSORDT="RSDTC", PRESEQ="RSSEQ")8283PDDT, DTHDT = pl.col("PDDT"), pl.col("DTHDT")8485adtte = (86    adsl.join(progression, on="USUBJID", how="left")87    .join(death, on="USUBJID", how="left")88    .join(last_assessment, on="USUBJID", how="left")89    .join(pre_therapy, on="USUBJID", how="left")90    # A progression or death dated after new-therapy start is not an event.91    .with_columns(92        PDDT=pl.when(93            pl.col("NTXSTDT").is_not_null() & (pl.col("PDDT") > pl.col("NTXSTDT"))94        )95        .then(None)96        .otherwise("PDDT"),97        DTHDT=pl.when(98            pl.col("NTXSTDT").is_not_null() & (pl.col("DTHDT") > pl.col("NTXSTDT"))99        )100        .then(None)101        .otherwise("DTHDT"),102    )103    .with_columns(104        EVENTDT=pl.min_horizontal("PDDT", "DTHDT"),105        STARTDT=pl.col("RANDDT"),106        PARAMCD=pl.lit("PFS"),107        PARAM=pl.lit("Progression-Free Survival"),108        # The censoring assessment: the last adequate one overall, or the109        # last one dated on or before therapy start when therapy began.110        CENSORASSESSDT=pl.when(pl.col("NTXSTDT").is_null())111        .then(pl.col("CENSORDT"))112        .otherwise(pl.col("PRECENSORDT")),113        CENSORSEQ=pl.when(pl.col("NTXSTDT").is_null())114        .then(pl.col("LASTSEQ"))115        .otherwise(pl.col("PRESEQ")),116    )117    # With no usable adequate assessment the subject is censored at118    # randomization, for a 1-day PFS.119    .with_columns(120        CENSORDT=pl.coalesce("CENSORASSESSDT", "STARTDT"),121    )122    # The event always wins, even after the last adequate assessment,123    # unless new anti-cancer therapy started first.124    .with_columns(125        CNSR=pl.when(pl.col("EVENTDT").is_not_null()).then(0).otherwise(1),126        ADT=pl.coalesce("EVENTDT", "CENSORDT"),127    )128    .with_columns(129        AVAL=(pl.col("ADT") - pl.col("STARTDT")).dt.total_days() + 1,130        # A progression and a death on the same day count as progression.131        EVNTDESC=pl.when(pl.col("EVENTDT").is_null())132        .then(pl.lit("CENSORED"))133        .when(PDDT.is_not_null() & (DTHDT.is_null() | (PDDT <= DTHDT)))134        .then(pl.lit("DISEASE PROGRESSION"))135        .otherwise(pl.lit("DEATH")),136    )137    .with_columns(138        SRCDOM=pl.when(pl.col("EVNTDESC") == "DEATH")139        .then(pl.lit("DS"))140        .when(pl.col("CENSORASSESSDT").is_null())141        .then(None)142        .otherwise(pl.lit("RS")),143        SRCVAR=pl.when(pl.col("EVNTDESC") == "DEATH")144        .then(pl.lit("DSDTC"))145        .when(pl.col("CENSORASSESSDT").is_null())146        .then(None)147        .otherwise(pl.lit("RSDTC")),148        SRCSEQ=pl.when(pl.col("EVNTDESC") == "DISEASE PROGRESSION")149        .then(pl.col("PDSEQ"))150        .when(pl.col("EVNTDESC") == "DEATH")151        .then(pl.col("DTHSEQ"))152        .when(pl.col("CENSORASSESSDT").is_null())153        .then(None)154        .otherwise(pl.col("CENSORSEQ")),155    )156    .sort("USUBJID")157    .select(158        "STUDYID",159        "USUBJID",160        "PARAMCD",161        "PARAM",162        "STARTDT",163        "ADT",164        "AVAL",165        "CNSR",166        "EVNTDESC",167        "SRCDOM",168        "SRCVAR",169        "SRCSEQ",170    )171)172173Path("/app/output").mkdir(parents=True, exist_ok=True)174adtte.write_csv("/app/output/adtte.csv")

1/* opensas reference solution for the yamaa benchmark adam-adtte-pfs.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 RANDDT NTXSTDT $1024;7  infile "/app/input/adsl.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. RANDDT : $char1024. NTXSTDT : $char1024.;9run;1011data ds;12  length STUDYID USUBJID DSDTC DSDECOD $1024;13  infile "/app/input/ds.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. DSSEQ DSDTC : $char1024. DSDECOD : $char1024.;15run;1617data rs;18  length STUDYID USUBJID RSTESTCD RSDTC RSSTRESC ADEQFL $1024;19  infile "/app/input/rs.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;20  input STUDYID : $char1024. USUBJID : $char1024. RSSEQ RSTESTCD : $char1024. RSDTC : $char1024. RSSTRESC : $char1024. ADEQFL : $char1024.;21run;2223proc sql;24  create table usable as select a.* from rs as a inner join adsl as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID where a.RSDTC ne '' and a.ADEQFL='Y' and a.RSTESTCD='OVRLRESP' and (missing(b.NTXSTDT) or a.RSDTC<=b.NTXSTDT);25  create table progression as select a.* from rs as a inner join adsl as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID where a.RSDTC ne '' and a.RSSTRESC='PD' and a.ADEQFL='Y' and a.RSTESTCD='OVRLRESP' and (missing(b.NTXSTDT) or a.RSDTC<=b.NTXSTDT);26  create table deaths as select a.* from ds as a inner join adsl as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID where a.DSDECOD='DEATH' and a.DSDTC ne '' and (missing(b.NTXSTDT) or a.DSDTC<=b.NTXSTDT);27quit;28proc sort data=progression; by STUDYID USUBJID RSDTC RSSEQ; run;29data pd; set progression; by STUDYID USUBJID; if first.USUBJID; keep STUDYID USUBJID RSDTC RSSEQ; rename RSDTC=PDDT RSSEQ=PDSEQ; run;30proc sort data=deaths; by STUDYID USUBJID DSDTC DSSEQ; run;31data death; set deaths; by STUDYID USUBJID; if first.USUBJID; keep STUDYID USUBJID DSDTC DSSEQ; rename DSDTC=DEATHDT DSSEQ=DEATHSEQ; run;32proc sort data=usable; by STUDYID USUBJID descending RSDTC descending RSSEQ; run;33data lastassessment; set usable; by STUDYID USUBJID; if first.USUBJID; keep STUDYID USUBJID RSDTC RSSEQ; rename RSDTC=LASTDT RSSEQ=LASTSEQ; run;34proc sql; create table joined as select a.*,b.PDDT,b.PDSEQ,c.DEATHDT,c.DEATHSEQ,d.LASTDT,d.LASTSEQ from adsl as a35  left join pd as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID left join death as c on a.STUDYID=c.STUDYID and a.USUBJID=c.USUBJID left join lastassessment as d on a.STUDYID=d.STUDYID and a.USUBJID=d.USUBJID; quit;36data result;37  length PARAMCD $8 PARAM $40 STARTDT ADT $10 EVNTDESC CNSDTDSC $40 SRCDOM $4 SRCVAR $16;38  set joined;39  PARAMCD='PFS'; PARAM='Progression-Free Survival'; STARTDT=RANDDT;40  if not missing(PDDT) and (missing(DEATHDT) or PDDT<=DEATHDT) then do; ADT=PDDT; CNSR=0; EVNTDESC='DISEASE PROGRESSION'; SRCDOM='RS'; SRCVAR='RSDTC'; SRCSEQ=PDSEQ; end;41  else if not missing(DEATHDT) then do; ADT=DEATHDT; CNSR=0; EVNTDESC='DEATH'; SRCDOM='DS'; SRCVAR='DSDTC'; SRCSEQ=DEATHSEQ; end;42  else do;43    CNSR=1; EVNTDESC='CENSORED';44    if not missing(LASTDT) then do; ADT=LASTDT; SRCDOM='RS'; SRCVAR='RSDTC'; SRCSEQ=LASTSEQ; end; else ADT=RANDDT;45  end;46  if not missing(ADT) and not missing(STARTDT) then AVAL=input(ADT,yymmdd10.)-input(STARTDT,yymmdd10.)+1;47run;4849proc sql;50  create table final as select STUDYID, USUBJID, PARAMCD, PARAM, STARTDT, ADT, AVAL, CNSR, EVNTDESC, SRCDOM, SRCVAR, SRCSEQ from result;51quit;52proc export data=final outfile="/app/output/adtte.csv" dbms=csv replace; run;