Timepoint Response

1library(yamaa)23rs <- yamaa_domain("spec.yaml")$output4rs

1import yamaa23rs = yamaa.yamaa_domain("spec.yaml").output4rs

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 = "Timepoint 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 1.67from pathlib import Path89import polars as pl101112def _present(value) -> bool:13    return value is not None and str(value).strip() != ""141516odm = pl.read_csv("/app/input/odm.csv", infer_schema=False)1718visits: dict[tuple, dict] = {}19tr_target: dict[tuple, list] = {}20tr_any: dict[tuple, int] = {}21tu_target_ids: dict[str, set] = {}22baseline_target_diams: dict[str, list] = {}2324for row in odm.to_dicts():25    subj, event = str(row["SubjectKey"]), str(row["StudyEventOID"])26    group, oid = str(row["ItemGroupOID"]), str(row["ItemOID"])27    value = row["Value"]28    key = (subj, event)29    if group == "IG.VISIT":30        visits.setdefault(31            key, {"studyid": str(row["StudyOID"]), "usubjid": subj}32        )[oid] = str(value) if _present(value) else None33    elif "TARGET" in group and group.startswith("IG.TR"):34        # Target tumor measurements at this visit.35        rep = str(row["ItemGroupRepeatKey"])36        tr_target.setdefault((subj, event, rep), {})[oid] = (37            str(value) if _present(value) else None38        )39        tr_any[key] = tr_any.get(key, 0) + 040    elif group.startswith("IG.TR"):41        # Non-target measurements still count as done.42        tr_any[key] = tr_any.get(key, 0) + 143    elif (44        "TARGET" in group45        and group.startswith("IG.TU")46        and event == "BASELINE"47        and oid == "IT.TU.TULNKID"48        and _present(value)49    ):50        tu_target_ids.setdefault(subj, set()).add(str(value).strip())5152# Any TR row (target or not) means the assessment has measurement records.53for (subj, event, _rep) in list(tr_target.keys()):54    tr_any[(subj, event)] = tr_any.get((subj, event), 0) + 15556# Chosen target lesions from the baseline tumor identification, else the57# baseline target measurements; none means non-target disease only.58chosen: dict[str, int] = {}59for subj in {s for s, _e in visits}:60    if subj in tu_target_ids:61        chosen[subj] = len(tu_target_ids[subj])62    else:63        baseline_reps = {rep for (s, e, rep) in tr_target if s == subj and e == "BASELINE"}64        chosen[subj] = len(baseline_reps)6566baseline_sum: dict[str, float | None] = {}67for subj in {s for s, _e in visits}:68    diams = []69    missing = False70    for (s, e, _rep), vals in tr_target.items():71        if s != subj or e != "BASELINE":72            continue73        diam = vals.get("IT.TR.LDIAM")74        if not _present(diam):75            missing = True76        else:77            try:78                diams.append(float(str(diam)))79            except ValueError:80                missing = True81    if missing or not diams:82        # No percent change can be computed from a missing baseline sum.83        # An empty baseline (non-target only) is handled as NE elsewhere.84        baseline_sum[subj] = None if (missing or not diams) and chosen.get(subj, 0) else 0.085        if chosen.get(subj, 0) == 0:86            baseline_sum[subj] = None87        elif not missing and diams:88            baseline_sum[subj] = sum(diams)89    else:90        baseline_sum[subj] = sum(diams)9192rows = []93by_subject: dict[str, list] = {}94for (subj, event), visit in visits.items():95    by_subject.setdefault(subj, []).append((event, visit))96for subj, items in by_subject.items():97    items.sort(key=lambda kv: int(kv[1].get("IT.VISIT.AVISITN") or 0))98    for seq, (event, visit) in enumerate(items, start=1):99        avisit = visit.get("IT.VISIT.AVISIT")100        avisitn = visit.get("IT.VISIT.AVISITN")101        adt = visit.get("IT.VISIT.ADT")102        has_records = tr_any.get((subj, event), 0) > 0103        if not has_records:104            # A scheduled assessment with no tumor measurement records.105            rsstresc, rsstat = None, "NOT DONE"106        elif event == "BASELINE":107            # A baseline assessment is never compared against itself.108            rsstresc, rsstat = "NE", None109        elif chosen.get(subj, 0) == 0:110            rsstresc, rsstat = "NE", None111        else:112            post_vals = [113                vals.get("IT.TR.LDIAM")114                for (s, e, _r), vals in tr_target.items()115                if s == subj and e == event116            ]117            measured = [v for v in post_vals if _present(v)]118            if len(measured) < chosen[subj]:119                rsstresc, rsstat = "NE", None120            else:121                base = baseline_sum.get(subj)122                if base is None or base == 0:123                    rsstresc, rsstat = "NE", None124                else:125                    try:126                        post = sum(float(str(v)) for v in measured)127                    except ValueError:128                        rsstresc, rsstat = "NE", None129                    else:130                        if all(float(str(v)) == 0 for v in measured):131                            rsstresc = "CR"132                        elif (base - post) / base >= 0.30:133                            rsstresc = "PR"134                        elif (post - base) / base >= 0.20:135                            rsstresc = "PD"136                        else:137                            rsstresc = "SD"138                        rsstat = None139        rows.append(140            {141                "STUDYID": visit["studyid"],142                "USUBJID": visit["usubjid"],143                "AVISIT": avisit,144                "AVISITN": avisitn,145                "ADT": adt,146                "RSSEQ": str(seq),147                "RSTESTCD": "TRGRESP",148                "RSTEST": "Timepoint Response",149                "RSSTRESC": rsstresc,150                "RSSTAT": rsstat,151            }152        )153154rs = (155    pl.DataFrame(156        rows,157        schema={158            "STUDYID": pl.String,159            "USUBJID": pl.String,160            "AVISIT": pl.String,161            "AVISITN": pl.String,162            "ADT": pl.String,163            "RSSEQ": pl.String,164            "RSTESTCD": pl.String,165            "RSTEST": pl.String,166            "RSSTRESC": pl.String,167            "RSSTAT": pl.String,168        },169    )170    .with_columns(171        pl.col("RSSEQ").cast(pl.Int64).alias("_seq"),172        pl.col("AVISITN").cast(pl.Int64).alias("_visit"),173    )174    .sort(["USUBJID", "_visit"])175    .select(176        "STUDYID", "USUBJID", "AVISIT", "AVISITN", "ADT", "RSSEQ",177        "RSTESTCD", "RSTEST", "RSSTRESC", "RSSTAT",178    )179)180181Path("/app/output").mkdir(exist_ok=True)182rs.write_csv("/app/output/rs.csv")

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='Timepoint 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;