Target Lesion Response at Each Tumor Assessment

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

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

1# Reference solution for the yamaa benchmark sdtm-rs-timepoint-response (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(tidyr)1011odm <- read_csv(12  "/app/input/odm.csv",13  col_types = cols(.default = col_character())14)1516visits <- odm |>17  filter(ItemGroupOID == "IG.VISIT") |>18  pivot_wider(19    id_cols = c(StudyOID, SubjectKey, StudyEventOID),20    names_from = ItemOID,21    values_from = Value,22    values_fn = function(x) x[[1]]23  )2425tr_target <- odm |>26  filter(grepl("TARGET", ItemGroupOID, fixed = TRUE),27    startsWith(ItemGroupOID, "IG.TR")) |>28  pivot_wider(29    id_cols = c(StudyOID, SubjectKey, StudyEventOID, ItemGroupRepeatKey),30    names_from = ItemOID,31    values_from = Value,32    values_fn = function(x) x[[1]]33  )3435tr_any <- odm |>36  filter(startsWith(ItemGroupOID, "IG.TR")) |>37  count(SubjectKey, StudyEventOID, name = "n_tr")3839tu_target <- odm |>40  filter(grepl("TARGET", ItemGroupOID, fixed = TRUE),41    startsWith(ItemGroupOID, "IG.TU"),42    StudyEventOID == "BASELINE",43    ItemOID == "IT.TU.TULNKID",44    !is.na(Value) & Value != "") |>45  distinct(SubjectKey, Value)4647response_one <- function(subj, event, avisitn, base_sum, n_chosen, post_diams, has_records) {48  if (!has_records) {49    return(list(rsstresc = NA_character_, rsstat = "NOT DONE"))50  }51  if (event == "BASELINE") {52    return(list(rsstresc = "NE", rsstat = NA_character_))53  }54  if (n_chosen == 0) {55    return(list(rsstresc = "NE", rsstat = NA_character_))56  }57  measured <- post_diams[!is.na(post_diams) & post_diams != ""]58  if (length(measured) < n_chosen) {59    return(list(rsstresc = "NE", rsstat = NA_character_))60  }61  if (is.na(base_sum) || base_sum == 0) {62    return(list(rsstresc = "NE", rsstat = NA_character_))63  }64  post <- sum(suppressWarnings(as.numeric(measured)))65  if (all(suppressWarnings(as.numeric(measured)) == 0)) {66    return(list(rsstresc = "CR", rsstat = NA_character_))67  }68  if ((base_sum - post) / base_sum >= 0.30) {69    return(list(rsstresc = "PR", rsstat = NA_character_))70  }71  if ((post - base_sum) / base_sum >= 0.20) {72    return(list(rsstresc = "PD", rsstat = NA_character_))73  }74  list(rsstresc = "SD", rsstat = NA_character_)75}7677subjects <- unique(visits$SubjectKey)78out <- list()79for (subj in subjects) {80  chosen <- nrow(tu_target |> filter(SubjectKey == subj))81  if (chosen == 0) {82    chosen <- nrow(tr_target |>83      filter(SubjectKey == subj, StudyEventOID == "BASELINE"))84  }85  base_rows <- tr_target |>86    filter(SubjectKey == subj, StudyEventOID == "BASELINE")87  base_diams <- base_rows$IT.TR.LDIAM88  base_sum <- if (chosen == 0) {89    NA_real_90  } else if (any(is.na(base_diams) | base_diams == "")) {91    NA_real_92  } else {93    sum(suppressWarnings(as.numeric(base_diams)))94  }95  subj_visits <- visits |>96    filter(SubjectKey == subj) |>97    arrange(suppressWarnings(as.integer(IT.VISIT.AVISITN)))98  seq <- 0L99  for (i in seq_len(nrow(subj_visits))) {100    seq <- seq + 1L101    event <- subj_visits$StudyEventOID[[i]]102    n_rec <- tr_any |>103      filter(SubjectKey == subj, StudyEventOID == event) |>104      pull(n_tr)105    has_records <- length(n_rec) > 0 && n_rec[[1]] > 0106    post_diams <- tr_target |>107      filter(SubjectKey == subj, StudyEventOID == event) |>108      pull(IT.TR.LDIAM)109    resp <- response_one(110      subj, event, subj_visits$IT.VISIT.AVISITN[[i]], base_sum, chosen,111      post_diams, has_records112    )113    out[[length(out) + 1]] <- tibble(114      STUDYID = subj_visits$StudyOID[[i]],115      USUBJID = subj,116      AVISIT = subj_visits$IT.VISIT.AVISIT[[i]],117      AVISITN = subj_visits$IT.VISIT.AVISITN[[i]],118      ADT = subj_visits$IT.VISIT.ADT[[i]],119      RSSEQ = seq,120      RSTESTCD = "TRGRESP",121      RSTEST = "Target Response",122      RSSTRESC = resp$rsstresc,123      RSSTAT = resp$rsstat124    )125  }126}127128rs <- bind_rows(out) |>129  arrange(USUBJID, suppressWarnings(as.integer(AVISITN))) |>130  select(131    STUDYID, USUBJID, AVISIT, AVISITN, ADT, RSSEQ, RSTESTCD, RSTEST,132    RSSTRESC, RSSTAT133  )134135dir.create("/app/output", showWarnings = FALSE)136write_csv(rs, "/app/output/rs.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-rs-timepoint-response (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 1.67import csv8import os9from collections import defaultdict1011INPUT = "/app/input"12OUTPUT = "/app/output"1314COLUMNS = [15    "STUDYID", "USUBJID", "AVISIT", "AVISITN", "ADT", "RSSEQ",16    "RSTESTCD", "RSTEST", "RSSTRESC", "RSSTAT",17]181920def load_items(path):21    """Pivot the long ODM extract into per-group item dictionaries."""22    groups = {}23    order = []24    with open(path, newline="") as handle:25        for row in csv.DictReader(handle):26            key = (27                row["StudyOID"], row["SubjectKey"], row["StudyEventOID"],28                row["StudyEventRepeatKey"], row["ItemGroupOID"],29                row["ItemGroupRepeatKey"],30            )31            if key not in groups:32                groups[key] = {33                    "study": row["StudyOID"],34                    "subject": row["SubjectKey"],35                    "event": row["StudyEventOID"],36                    "group": row["ItemGroupOID"],37                    "items": {},38                }39                order.append(key)40            groups[key]["items"][row["ItemOID"]] = (row["Value"] or "").strip()41    return [groups[key] for key in order]424344def to_number(text):45    if text in (None, ""):46        return None47    try:48        return float(text)49    except ValueError:50        return None515253def main():54    records = load_items(os.path.join(INPUT, "odm.csv"))5556    visits = defaultdict(list)          # subject -> [visit dicts]57    chosen_targets = defaultdict(set)   # subject -> {lesion ids chosen}58    target_meas = defaultdict(dict)     # (subject, event) -> {lesion: ldiam}59    any_meas = defaultdict(bool)       # (subject, event) -> any tumor record6061    for rec in records:62        subject = rec["subject"]63        items = rec["items"]64        if rec["group"] == "IG.VISIT":65            visits[subject].append({66                "event": rec["event"],67                "avisit": items.get("IT.VISIT.AVISIT", ""),68                "avisitn": items.get("IT.VISIT.AVISITN", ""),69                "adt": items.get("IT.VISIT.ADT", ""),70            })71        elif rec["group"] == "IG.TUTARGET":72            link = items.get("IT.TU.TULNKID", "")73            if link:74                chosen_targets[subject].add(link)75        elif rec["group"] == "IG.TRTARGET":76            link = items.get("IT.TR.TRLNKID", "")77            if link:78                key = (subject, rec["event"])79                any_meas[key] = True80                target_meas[key][link] = to_number(items.get("IT.TR.LDIAM", ""))81        elif rec["group"] in ("IG.TRNT", "IG.TUNT"):82            any_meas[(subject, rec["event"])] = True8384    rows = []85    study_of = {}86    for rec in records:87        study_of.setdefault(rec["subject"], rec["study"])88    for subject in sorted(visits):89        subject_visits = sorted(90            visits[subject],91            key=lambda v: (to_number(v["avisitn"]) is None,92                           to_number(v["avisitn"]) or 0),93        )94        if not subject_visits:95            continue96        baseline_event = subject_visits[0]["event"]97        chosen = chosen_targets.get(subject, set())98        baseline_sum = None99        if chosen:100            base = target_meas.get((subject, baseline_event), {})101            measured = [base[lesion] for lesion in chosen102                        if base.get(lesion) is not None]103            if len(measured) == len(chosen):104                baseline_sum = sum(measured)105106        for seq, visit in enumerate(subject_visits, start=1):107            event = visit["event"]108            done = any_meas.get((subject, event), False)109            rsstresc = ""110            rsstat = ""111            if not done:112                rsstat = "NOT DONE"113            elif event == baseline_event:114                rsstresc = "NE"115            elif not chosen:116                # Non-target disease only.117                rsstresc = "NE"118            else:119                meas = target_meas.get((subject, event), {})120                measured = {lesion: meas[lesion] for lesion in chosen121                            if meas.get(lesion) is not None}122                if len(measured) < len(chosen):123                    rsstresc = "NE"124                elif baseline_sum is None or baseline_sum == 0:125                    rsstresc = "NE"126                else:127                    total = sum(measured.values())128                    if all(value == 0 for value in measured.values()):129                        rsstresc = "CR"130                    else:131                        change = 100.0 * (total - baseline_sum) / baseline_sum132                        if change <= -30.0:133                            rsstresc = "PR"134                        elif change >= 20.0:135                            rsstresc = "PD"136                        else:137                            rsstresc = "SD"138            rows.append({139                "STUDYID": study_of.get(subject, ""),140                "USUBJID": subject,141                "AVISIT": visit["avisit"],142                "AVISITN": visit["avisitn"],143                "ADT": visit["adt"],144                "RSSEQ": seq,145                "RSTESTCD": "TRGRESP",146                "RSTEST": "Target Response",147                "RSSTRESC": rsstresc,148                "RSSTAT": rsstat,149            })150151    os.makedirs(OUTPUT, exist_ok=True)152    with open(os.path.join(OUTPUT, "rs.csv"), "w", newline="") as handle:153        writer = csv.DictWriter(handle, fieldnames=COLUMNS)154        writer.writeheader()155        writer.writerows(rows)156157158if __name__ == "__main__":159    main()

1/* opensas reference solution for the yamaa benchmark sdtm-rs-timepoint-response.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data odm;6  length StudyOID MetaDataVersionOID SubjectKey StudyEventOID StudyEventRepeatKey ItemGroupOID ItemGroupRepeatKey ItemOID Value $1024;7  infile "/app/input/odm.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input StudyOID : $char1024. MetaDataVersionOID : $char1024. SubjectKey : $char1024. StudyEventOID : $char1024. StudyEventRepeatKey : $char1024. ItemGroupOID : $char1024. ItemGroupRepeatKey : $char1024. ItemOID : $char1024. Value : $char1024.;9run;1011proc sort data=odm out=ordered; by StudyOID SubjectKey StudyEventOID; run;12data visits;13  length AVISIT VISITORDER ADT $1024;14  retain AVISIT VISITORDER ADT;15  set ordered;16  where ItemGroupOID='IG.VISIT';17  by StudyOID SubjectKey StudyEventOID;18  if first.StudyEventOID then do;19    AVISIT=''; VISITORDER=''; ADT='';20  end;21    if ItemOID='IT.VISIT.AVISIT' then AVISIT=Value;22  if ItemOID='IT.VISIT.AVISITN' then VISITORDER=Value;23  if ItemOID='IT.VISIT.ADT' then ADT=Value;24  if last.StudyEventOID then output;25run;26proc sort data=odm out=ordered; by StudyOID SubjectKey StudyEventOID ItemGroupRepeatKey; run;27data targetforms;28  length DIAM $1024;29  retain DIAM;30  set ordered;31  where ItemGroupOID='IG.TRTARGET';32  by StudyOID SubjectKey StudyEventOID ItemGroupRepeatKey;33  if first.ItemGroupRepeatKey then do;34    DIAM='';35  end;36    if ItemOID='IT.TR.LDIAM' then DIAM=Value;37  if last.ItemGroupRepeatKey then output;38run;39data targets; set targetforms; DIAMETER=input(DIAM,?? best32.); run;40proc sql;41create table chosen as select StudyOID,SubjectKey,count(distinct Value) as NCHOSEN from odm where ItemGroupOID='IG.TUTARGET' and StudyEventOID='BASELINE' and ItemOID='IT.TU.TULNKID' and not missing(Value) group by StudyOID,SubjectKey;42create table baseline as select StudyOID,SubjectKey,count(*) as NBASE,count(DIAMETER) as NBASEDONE,sum(DIAMETER) as BASE from targets where StudyEventOID='BASELINE' group by StudyOID,SubjectKey;43create table measures as select StudyOID,SubjectKey,StudyEventOID,count(DIAMETER) as NMEAS,sum(DIAMETER) as TOTAL,max(DIAMETER) as MAXDIAM from targets group by StudyOID,SubjectKey,StudyEventOID;44create table done as select distinct StudyOID,SubjectKey,StudyEventOID from odm where ItemGroupOID in ('IG.TRTARGET','IG.TRNT');45create table joined as select a.*,b.NCHOSEN,c.NBASE,c.NBASEDONE,c.BASE,d.NMEAS,d.TOTAL,d.MAXDIAM,e.StudyEventOID as DONEEVENT from visits a left join chosen b on a.StudyOID=b.StudyOID and a.SubjectKey=b.SubjectKey left join baseline c on a.StudyOID=c.StudyOID and a.SubjectKey=c.SubjectKey left join measures d on a.StudyOID=d.StudyOID and a.SubjectKey=d.SubjectKey and a.StudyEventOID=d.StudyEventOID left join done e on a.StudyOID=e.StudyOID and a.SubjectKey=e.SubjectKey and a.StudyEventOID=e.StudyEventOID; quit;46data prepared; set joined; length STUDYID USUBJID RSTESTCD RSTEST RSSTRESC RSSTAT $1024; STUDYID=StudyOID; USUBJID=SubjectKey; AVISITN=input(VISITORDER,best32.); RSTESTCD='TRGRESP'; RSTEST='Target Response'; RSSTRESC='NE'; RSSTAT=''; NSELECTED=coalesce(NCHOSEN,NBASE,0);47if missing(DONEEVENT) then do; RSSTRESC=''; RSSTAT='NOT DONE'; end;48else if StudyEventOID ne 'BASELINE' and NSELECTED>0 and NMEAS>=NSELECTED and NBASEDONE=NBASE and not missing(BASE) and BASE ne 0 then do;49if MAXDIAM=0 then RSSTRESC='CR'; else if (BASE-TOTAL)/BASE>=0.30 then RSSTRESC='PR'; else if (TOTAL-BASE)/BASE>=0.20 then RSSTRESC='PD'; else RSSTRESC='SD'; end; run;50proc sort data=prepared; by STUDYID USUBJID AVISITN; run;51data result; set prepared; by STUDYID USUBJID; retain RSSEQ;52  if first.USUBJID then RSSEQ=0;53  RSSEQ+1;54run;5556proc sql;57  create table final as select STUDYID, USUBJID, AVISIT, AVISITN, ADT, RSSEQ, RSTESTCD, RSTEST, RSSTRESC, RSSTAT from result;58quit;59proc export data=final outfile="/app/output/rs.csv" dbms=csv replace; run;