Derive the Last Known Alive Date and Death Date from Multiple Domains

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

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

1# Reference solution for the yamaa benchmark adam-adsl-alive-date (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 1.67suppressPackageStartupMessages({8  library(dplyr)9  library(readr)10  library(stringr)11  library(lubridate)12})1314input_dir <- "/app/input"15output_dir <- "/app/output"1617read_domain <- function(name) {18  read_csv(19    file.path(input_dir, paste0(name, ".csv")),20    col_types = cols(.default = col_character()),21    na = character(),22    trim_ws = TRUE,23    progress = FALSE24  ) |>25    mutate(across(everything(), \(x) if_else(is.na(x) | x == "", NA_character_, str_trim(x))))26}2728# Split an ISO 8601 date or date-time into year, month and day parts.29# Anything after the date (a time) is dropped.30parse_parts <- function(x) {31  m <- str_match(x, "^(\\d{4})(?:-(\\d{2})(?:-(\\d{2}))?)?(?:$|[T ])")32  tibble(33    year = as.integer(m[, 2]),34    month = as.integer(m[, 3]),35    day = as.integer(m[, 4])36  )37}3839# Date a value counts on: a date-time on its date, a year and month on the40# first of that month, a year alone on 1 January.41to_date <- function(x) {42  p <- parse_parts(x)43  make_date(p$year, coalesce(p$month, 1L), coalesce(p$day, 1L))44}4546dm <- read_domain("dm")47ae <- read_domain("ae")48ds <- read_domain("ds")49ex <- read_domain("ex")50ss <- read_domain("ss")51sv <- read_domain("sv")52vs <- read_domain("vs")5354alive_dates <- bind_rows(55  ex |> transmute(STUDYID, USUBJID, DT = to_date(EXSTDTC)),56  ex |> transmute(STUDYID, USUBJID, DT = to_date(EXENDTC)),57  ae |> transmute(STUDYID, USUBJID, DT = to_date(AESTDTC)),58  ae |> transmute(STUDYID, USUBJID, DT = to_date(AEENDTC)),59  ds |>60    filter(!(str_to_upper(coalesce(DSDECOD, "")) %in% c("DEATH", "LOST TO FOLLOW-UP"))) |>61    transmute(STUDYID, USUBJID, DT = to_date(DSSTDTC)),62  sv |> transmute(STUDYID, USUBJID, DT = to_date(SVSTDTC)),63  vs |> transmute(STUDYID, USUBJID, DT = to_date(VSDTC)),64  ss |>65    filter(str_to_upper(coalesce(SSSTRESC, "")) == "ALIVE") |>66    transmute(STUDYID, USUBJID, DT = to_date(SSDTC))67) |>68  filter(!is.na(DT)) |>69  group_by(STUDYID, USUBJID) |>70  summarise(MAXALIVE = max(DT), .groups = "drop")7172fmt_date <- function(d) if_else(is.na(d), NA_character_, format(d, "%Y-%m-%d"))7374adsl <- dm |>75  distinct(STUDYID, USUBJID, .keep_all = TRUE) |>76  left_join(alive_dates, by = c("STUDYID", "USUBJID")) |>77  mutate(78    parts = parse_parts(DTHDTC),79    dth_year = parts$year,80    dth_month = parts$month,81    dth_day = parts$day,82    dth_base = make_date(dth_year, coalesce(dth_month, 1L), coalesce(dth_day, 1L)),83    dth_partial = !is.na(dth_year) & (is.na(dth_month) | is.na(dth_day)),84    # the latest alive date replaces the imputed day when it falls later85    # inside the collected month (or year, when only the year was collected)86    inside_collected = !is.na(MAXALIVE) & !is.na(dth_year) &87      year(MAXALIVE) == dth_year &88      (is.na(dth_month) | month(MAXALIVE) == dth_month),89    take_alive = dth_partial & inside_collected & !is.na(MAXALIVE) &90      !is.na(dth_base) & MAXALIVE > dth_base,91    DTHDT_D = if_else(take_alive, MAXALIVE, dth_base),92    DTHDTF = case_when(93      dth_partial & is.na(dth_month) ~ "M",94      dth_partial ~ "D",95      TRUE ~ NA_character_96    ),97    DTHFL = if_else(98      str_to_upper(coalesce(DTHFL, "")) == "Y" | !is.na(dth_year),99      "Y",100      NA_character_101    ),102    LSTALVDT_D = if_else(!is.na(DTHDT_D) & !is.na(MAXALIVE) & MAXALIVE > DTHDT_D, DTHDT_D, MAXALIVE),103    LSTALVDT = fmt_date(LSTALVDT_D),104    DTHDT = fmt_date(DTHDT_D)105  ) |>106  select(STUDYID, USUBJID, LSTALVDT, DTHFL, DTHDT, DTHDTF)107108dir.create(output_dir, showWarnings = FALSE, recursive = TRUE)109write_csv(adsl, file.path(output_dir, "adsl.csv"), na = "")

1# Reference solution for the yamaa benchmark adam-adsl-alive-date (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 ADSL (STUDYID, USUBJID, LSTALVDT, DTHFL, DTHDT, DTHDTF) from SDTM.89Reads the SDTM datasets from /app/input and writes /app/output/adsl.csv.10"""1112import csv13import re14from datetime import date1516INPUT = "/app/input"17OUTPUT = "/app/output"1819DATE_RE = re.compile(r"^(\d{4})(?:-(\d{2})(?:-(\d{2}))?)?(?:[T ].*)?$")202122def read(name):23    with open(f"{INPUT}/{name}.csv", newline="", encoding="utf-8-sig") as handle:24        return list(csv.DictReader(handle))252627def parse_iso(text):28    """Return (year, month, day) with None for components not collected."""29    match = DATE_RE.match((text or "").strip())30    if not match:31        return None32    year, month, day = match.groups()33    return int(year), (int(month) if month else None), (int(day) if day else None)343536def complete(text):37    """Complete a date to a day: a year and month to the 1st, a year to 1 Jan."""38    parts = parse_iso(text)39    if parts is None:40        return None41    year, month, day = parts42    try:43        return date(year, month or 1, day or 1)44    except ValueError:45        return None464748def norm(value):49    return (value or "").strip().upper()505152def main():53    dm = read("dm")54    alive = {}5556    def add(usubjid, text):57        day = complete(text)58        if day is None:59            return60        current = alive.get(usubjid)61        if current is None or day > current:62            alive[usubjid] = day6364    for row in read("ex"):65        add(row["USUBJID"], row.get("EXSTDTC"))66        add(row["USUBJID"], row.get("EXENDTC"))67    for row in read("ae"):68        add(row["USUBJID"], row.get("AESTDTC"))69        add(row["USUBJID"], row.get("AEENDTC"))70    for row in read("ds"):71        if norm(row.get("DSDECOD")) in {"DEATH", "LOST TO FOLLOW-UP"}:72            continue73        add(row["USUBJID"], row.get("DSSTDTC"))74    for row in read("sv"):75        add(row["USUBJID"], row.get("SVSTDTC"))76    for row in read("vs"):77        add(row["USUBJID"], row.get("VSDTC"))78    for row in read("ss"):79        if norm(row.get("SSSTRESC")) == "ALIVE":80            add(row["USUBJID"], row.get("SSDTC"))8182    out = []83    seen = set()84    for row in dm:85        usubjid = row["USUBJID"]86        if usubjid in seen:87            continue88        seen.add(usubjid)8990        last_alive = alive.get(usubjid)91        dthdtc = (row.get("DTHDTC") or "").strip()92        parts = parse_iso(dthdtc) if dthdtc else None93        dead = norm(row.get("DTHFL")) == "Y" or parts is not None9495        dthdt = None96        dthdtf = ""97        if parts is not None:98            year, month, day = parts99            base = complete(dthdtc)100            if day is not None:101                dthdt = base102            else:103                dthdt = base104                dthdtf = "D" if month is not None else "M"105                if last_alive is not None and last_alive > base:106                    same_period = last_alive.year == year and (107                        month is None or last_alive.month == month108                    )109                    if same_period:110                        dthdt = last_alive111112        lstalvdt = last_alive113        if lstalvdt is not None and dthdt is not None and lstalvdt > dthdt:114            lstalvdt = dthdt115116        out.append(117            {118                "STUDYID": row["STUDYID"],119                "USUBJID": usubjid,120                "LSTALVDT": lstalvdt.isoformat() if lstalvdt else "",121                "DTHFL": "Y" if dead else "",122                "DTHDT": dthdt.isoformat() if dthdt else "",123                "DTHDTF": dthdtf,124            }125        )126127    columns = ["STUDYID", "USUBJID", "LSTALVDT", "DTHFL", "DTHDT", "DTHDTF"]128    with open(f"{OUTPUT}/adsl.csv", "w", newline="", encoding="utf-8") as handle:129        writer = csv.DictWriter(handle, fieldnames=columns, lineterminator="\n")130        writer.writeheader()131        writer.writerows(out)132133134if __name__ == "__main__":135    main()

1/* opensas reference solution for the yamaa benchmark adam-adsl-alive-date.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 DTHDTC DTHFL $1024;7  infile "/app/input/dm.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. DTHDTC : $char1024. DTHFL : $char1024.;9run;1011data ex;12  length STUDYID USUBJID EXSEQ EXTRT EXSTDTC EXENDTC $1024;13  infile "/app/input/ex.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input STUDYID : $char1024. USUBJID : $char1024. EXSEQ : $char1024. EXTRT : $char1024. EXSTDTC : $char1024. EXENDTC : $char1024.;15run;1617data ae;18  length STUDYID USUBJID AESEQ AEDECOD AESTDTC AEENDTC AEOUT $1024;19  infile "/app/input/ae.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;20  input STUDYID : $char1024. USUBJID : $char1024. AESEQ : $char1024. AEDECOD : $char1024. AESTDTC : $char1024. AEENDTC : $char1024. AEOUT : $char1024.;21run;2223data ds;24  length STUDYID USUBJID DSSEQ DSCAT DSDECOD DSTERM DSSTDTC $1024;25  infile "/app/input/ds.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;26  input STUDYID : $char1024. USUBJID : $char1024. DSSEQ : $char1024. DSCAT : $char1024. DSDECOD : $char1024. DSTERM : $char1024. DSSTDTC : $char1024.;27run;2829data sv;30  length STUDYID USUBJID VISITNUM VISIT SVSTDTC $1024;31  infile "/app/input/sv.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;32  input STUDYID : $char1024. USUBJID : $char1024. VISITNUM : $char1024. VISIT : $char1024. SVSTDTC : $char1024.;33run;3435data vs;36  length STUDYID USUBJID VSSEQ VSTESTCD VSDTC $1024;37  infile "/app/input/vs.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;38  input STUDYID : $char1024. USUBJID : $char1024. VSSEQ : $char1024. VSTESTCD : $char1024. VSDTC : $char1024.;39run;4041data ss;42  length STUDYID USUBJID SSSEQ SSTESTCD SSTEST SSSTRESC SSDTC $1024;43  infile "/app/input/ss.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;44  input STUDYID : $char1024. USUBJID : $char1024. SSSEQ : $char1024. SSTESTCD : $char1024. SSTEST : $char1024. SSSTRESC : $char1024. SSDTC : $char1024.;45run;4647/* Every collected date that shows the subject alive. */48proc sql;49  create table alive_text as50  select STUDYID,USUBJID,EXSTDTC as DTC from ex51  union all select STUDYID,USUBJID,EXENDTC from ex52  union all select STUDYID,USUBJID,AESTDTC from ae53  union all select STUDYID,USUBJID,AEENDTC from ae54  union all select STUDYID,USUBJID,DSSTDTC from ds where DSDECOD not in ('DEATH','LOST TO FOLLOW-UP')55  union all select STUDYID,USUBJID,SVSTDTC from sv56  union all select STUDYID,USUBJID,VSDTC from vs57  union all select STUDYID,USUBJID,SSDTC from ss where SSSTRESC='ALIVE';58quit;5960/* A date counts on the day it names: a time is dropped, a year and month61   take the first of the month, and a year alone takes 1 January. */62data alive_days;63  length day $10;64  set alive_text;65  txt=strip(DTC);66  if prxmatch('/^[0-9]{4}-[0-9]{2}-[0-9]{2}/',txt) then day=substr(txt,1,10);67  else if prxmatch('/^[0-9]{4}-[0-9]{2}$/',txt) then day=cats(txt,'-01');68  else if prxmatch('/^[0-9]{4}$/',txt) then day=cats(txt,'-01-01');69  alive=input(day,yymmdd10.);70  if not missing(alive);71  keep STUDYID USUBJID alive;72run;7374proc sql;75  create table latest as select STUDYID,USUBJID,max(alive) as alive from alive_days group by STUDYID,USUBJID;76  create table joined as select a.*,b.alive from dm as a left join latest as b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID;77quit;7879data result;80  length LSTALVDT DTHDT $10 DTHDTF $1;81  set joined;82  t=strip(DTHDTC);83  death=.;84  if prxmatch('/^[0-9]{4}-[0-9]{2}-[0-9]{2}/',t) then death=input(substr(t,1,10),yymmdd10.);85  else if prxmatch('/^[0-9]{4}-[0-9]{2}$/',t) then do;86    death=input(cats(t,'-01'),yymmdd10.);87    DTHDTF='D';88    if not missing(alive) then do;89      if alive>death and substr(put(alive,yymmdd10.),1,7)=t then death=alive;90    end;91  end;92  else if prxmatch('/^[0-9]{4}$/',t) then do;93    death=input(cats(t,'-01-01'),yymmdd10.);94    DTHDTF='M';95    if not missing(alive) then do;96      if alive>death and substr(put(alive,yymmdd10.),1,4)=t then death=alive;97    end;98  end;99  last=alive;100  if not missing(last) and not missing(death) then do;101    if last>death then last=death;102  end;103  if not missing(last) then LSTALVDT=put(last,yymmdd10.);104  if not missing(death) then DTHDT=put(death,yymmdd10.);105  if strip(DTHFL)='Y' or not missing(death) then DTHFL='Y';106  else DTHFL='';107run;108109proc sql;110  create table final as select STUDYID, USUBJID, LSTALVDT, DTHFL, DTHDT, DTHDTF from result;111quit;112proc export data=final outfile="/app/output/adsl.csv" dbms=csv replace; run;