Build SDTM subject visits from ODM extracts

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

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

1# Reference solution for the yamaa benchmark sdtm-sv-subject-visits (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(lubridate)9library(readr)10library(tidyr)1112study_day <- function(ref, coll) {13  if (is.na(ref) || is.na(coll)) {14    return(NA_integer_)15  }16  delta <- as.integer(coll - ref)17  if (delta >= 0L) delta + 1L else delta18}1920dm <- read_csv(21  "/app/input/dm.csv",22  col_types = cols(.default = col_character())23)24odm <- read_csv(25  "/app/input/odm.csv",26  col_types = cols(.default = col_character())27)2829ref_map <- setNames(suppressWarnings(ymd(dm$RFSTDTC)), dm$USUBJID)3031wide <- odm |>32  mutate(visit_key = paste(SubjectKey, StudyEventOID, StudyEventRepeatKey, sep = "|")) |>33  select(visit_key, SubjectKey, StudyOID, StudyEventOID, ItemOID, Value) |>34  pivot_wider(names_from = ItemOID, values_from = Value, values_fn = first)3536sv <- wide |>37  rowwise() |>38  mutate(39    dates = list(na.omit(c(IT.VS.VSDTC, IT.LB.LBDTC, IT.EX.EXDTC))),40    SVSTDTC = if (length(dates) == 0) NA_character_ else min(dates),41    SVENDTC = if (length(dates) == 0) NA_character_ else max(dates)42  ) |>43  ungroup() |>44  filter(!is.na(SVSTDTC)) |>45  transmute(46    DOMAIN = "SV",47    STUDYID = StudyOID,48    USUBJID = SubjectKey,49    VISIT = IT.TV.VISIT,50    VISITNUM = suppressWarnings(as.numeric(IT.TV.VISITNUM)),51    VISITDY = suppressWarnings(as.integer(IT.TV.VISITDY)),52    SVSTDTC,53    SVENDTC,54    SVSTDY = vapply(55      seq_len(n()),56      function(i) study_day(ref_map[[USUBJID[i]]], suppressWarnings(ymd(SVSTDTC[i]))),57      integer(1)58    ),59    SVENDY = vapply(60      seq_len(n()),61      function(i) study_day(ref_map[[USUBJID[i]]], suppressWarnings(ymd(SVENDTC[i]))),62      integer(1)63    ),64    TAETORD = suppressWarnings(as.integer(IT.TV.TAETORD)),65    EPOCH = IT.TV.EPOCH,66    SVUPDES = IT.TV.UPDES67  ) |>68  arrange(USUBJID, SVSTDTC) |>69  group_by(USUBJID) |>70  mutate(SVSEQ = row_number()) |>71  ungroup() |>72  arrange(USUBJID, SVSEQ) |>73  select(74    DOMAIN, STUDYID, USUBJID, SVSEQ, VISIT, VISITNUM, VISITDY, SVSTDTC,75    SVENDTC, SVSTDY, SVENDY, TAETORD, EPOCH, SVUPDES76  )7778dir.create("/app/output", showWarnings = FALSE)79write_csv(sv, "/app/output/sv.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-sv-subject-visits (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"""Build the SDTM SV dataset from the DM and ODM source datasets.89One record per subject per study event occurrence.  Visit start and end10dates are the earliest and latest of the form dates carried by the11visit's VS, LB and EX forms; study days are counted from the subject's12RFSTDTC in DM with no day zero.13"""1415import csv16from datetime import date1718INPUT_DIR = "/app/input"19OUTPUT_DIR = "/app/output"2021COLUMNS = [22    "DOMAIN", "STUDYID", "USUBJID", "SVSEQ", "VISIT", "VISITNUM",23    "VISITDY", "SVSTDTC", "SVENDTC", "SVSTDY", "SVENDY", "TAETORD",24    "EPOCH", "SVUPDES",25]2627# Item groups whose dates bound the visit.28FORM_GROUPS = {"IG.VS", "IG.LB", "IG.EX"}293031def parse_date(text):32    return date.fromisoformat(text)333435def study_day(day, reference):36    delta = (day - reference).days37    return delta + 1 if delta >= 0 else delta383940def main():41    # Reference start dates from DM, keyed by (STUDYID, USUBJID).42    rfstdtc = {}43    with open(f"{INPUT_DIR}/dm.csv", newline="") as handle:44        for row in csv.DictReader(handle):45            rfstdtc[(row["STUDYID"], row["USUBJID"])] = row["RFSTDTC"]4647    # Group the ODM items by subject and study event occurrence.48    events = {}49    order = []50    with open(f"{INPUT_DIR}/odm.csv", newline="") as handle:51        for row in csv.DictReader(handle):52            key = (53                row["StudyOID"],54                row["SubjectKey"],55                row["StudyEventOID"],56                row["StudyEventRepeatKey"],57            )58            if key not in events:59                events[key] = {"tv": {}, "dates": []}60                order.append(key)61            event = events[key]62            if row["ItemGroupOID"] == "IG.TV":63                event["tv"][row["ItemOID"]] = row["Value"]64            elif row["ItemGroupOID"] in FORM_GROUPS and row["Value"]:65                event["dates"].append(row["Value"])6667    records = []68    for key in order:69        study_oid, subject_key, _, _ = key70        event = events[key]71        tv = event["tv"]72        dates = sorted(event["dates"])73        svstdtc = dates[0] if dates else ""74        svendtc = dates[-1] if dates else ""75        reference = rfstdtc.get((study_oid, subject_key), "")76        svstdy = svendy = ""77        if reference:78            ref_date = parse_date(reference)79            if svstdtc:80                svstdy = study_day(parse_date(svstdtc), ref_date)81            if svendtc:82                svendy = study_day(parse_date(svendtc), ref_date)83        records.append({84            "DOMAIN": "SV",85            "STUDYID": study_oid,86            "USUBJID": subject_key,87            "VISIT": tv.get("IT.TV.VISIT", ""),88            "VISITNUM": tv.get("IT.TV.VISITNUM", ""),89            "VISITDY": tv.get("IT.TV.VISITDY", ""),90            "SVSTDTC": svstdtc,91            "SVENDTC": svendtc,92            "SVSTDY": svstdy,93            "SVENDY": svendy,94            "TAETORD": tv.get("IT.TV.TAETORD", ""),95            "EPOCH": tv.get("IT.TV.EPOCH", ""),96            "SVUPDES": tv.get("IT.TV.UPDES", ""),97            "_sort": svstdtc,98        })99100    # SVSEQ numbers each subject's visits from 1 in SVSTDTC order.101    by_subject = {}102    for record in records:103        by_subject.setdefault(record["USUBJID"], []).append(record)104    sequenced = []105    for subject in by_subject:106        visits = sorted(by_subject[subject], key=lambda r: r["_sort"])107        for index, record in enumerate(visits, start=1):108            record["SVSEQ"] = index109            sequenced.append(record)110111    with open(f"{OUTPUT_DIR}/sv.csv", "w", newline="") as handle:112        writer = csv.DictWriter(handle, fieldnames=COLUMNS)113        writer.writeheader()114        for record in sequenced:115            writer.writerow({name: record[name] for name in COLUMNS})116117118if __name__ == "__main__":119    main()

1/* opensas reference solution for the yamaa benchmark sdtm-sv-subject-visits.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 odm;12  length StudyOID MetaDataVersionOID SubjectKey StudyEventOID StudyEventRepeatKey ItemGroupOID ItemGroupRepeatKey ItemOID Value $1024;13  infile "/app/input/odm.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;14  input StudyOID : $char1024. MetaDataVersionOID : $char1024. SubjectKey : $char1024. StudyEventOID : $char1024. StudyEventRepeatKey : $char1024. ItemGroupOID : $char1024. ItemGroupRepeatKey : $char1024. ItemOID : $char1024. Value : $char1024.;15run;1617proc sort data=odm out=ordered; by StudyOID SubjectKey StudyEventOID StudyEventRepeatKey; run;18data visits;19  length VISIT VNUM VDAY ORD EPOCH SVUPDES $1024;20  retain VISIT VNUM VDAY ORD EPOCH SVUPDES;21  set ordered;2223  by StudyOID SubjectKey StudyEventOID StudyEventRepeatKey;24  if first.StudyEventRepeatKey then do;25    VISIT=''; VNUM=''; VDAY=''; ORD=''; EPOCH=''; SVUPDES='';26  end;27    if ItemOID='IT.TV.VISIT' then VISIT=Value;28  if ItemOID='IT.TV.VISITNUM' then VNUM=Value;29  if ItemOID='IT.TV.VISITDY' then VDAY=Value;30  if ItemOID='IT.TV.TAETORD' then ORD=Value;31  if ItemOID='IT.TV.EPOCH' then EPOCH=Value;32  if ItemOID='IT.TV.UPDES' then SVUPDES=Value;33  if last.StudyEventRepeatKey then output;34run;35proc sql;36create table dates as select StudyOID,SubjectKey,StudyEventOID,StudyEventRepeatKey,min(Value) as SVSTDTC,max(Value) as SVENDTC from odm where ItemOID in ('IT.VS.VSDTC','IT.LB.LBDTC','IT.EX.EXDTC') and not missing(Value) group by StudyOID,SubjectKey,StudyEventOID,StudyEventRepeatKey;37create table joined as select a.*,b.SVSTDTC,b.SVENDTC,c.RFSTDTC from visits 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 left join dm c on a.StudyOID=c.STUDYID and a.SubjectKey=c.USUBJID; quit;38data prepared; set joined; length DOMAIN $2 STUDYID USUBJID $1024; DOMAIN='SV'; STUDYID=StudyOID; USUBJID=SubjectKey; VISITNUM=input(VNUM,best32.); VISITDY=input(VDAY,best32.); TAETORD=input(ORD,best32.); SVSTDY=.; SVENDY=.; if not missing(RFSTDTC) then do; SVSTDY=input(SVSTDTC,yymmdd10.)-input(RFSTDTC,yymmdd10.); if SVSTDY>=0 then SVSTDY+1; SVENDY=input(SVENDTC,yymmdd10.)-input(RFSTDTC,yymmdd10.); if SVENDY>=0 then SVENDY+1; end; run;39proc sort data=prepared; by STUDYID USUBJID SVSTDTC StudyEventOID StudyEventRepeatKey; run;40data result; set prepared; by STUDYID USUBJID; retain SVSEQ UNSCH;41if first.USUBJID then do; SVSEQ=0; UNSCH=0; end; SVSEQ+1;42if StudyEventOID='UNSCH' then do; UNSCH+1; VISITNUM=99+UNSCH/10; VISITDY=.; TAETORD=.; end; else SVUPDES=''; run;43proc sql;44  create table final as select DOMAIN, STUDYID, USUBJID, SVSEQ, VISIT, VISITNUM, VISITDY, SVSTDTC, SVENDTC, SVSTDY, SVENDY, TAETORD, EPOCH, SVUPDES from result;45quit;46proc export data=final outfile="/app/output/sv.csv" dbms=csv replace; run;