Standardize Non-Numeric Lab Results

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

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

1# Reference solution for the yamaa benchmark sdtm-lb-character-results (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(purrr)10library(stringr)1112odm <- read_csv(13  "/app/input/odm.csv",14  col_types = cols(.default = col_character())15) |>16  mutate(Value = na_if(Value, ""))1718form_order <- c(19  "FO.LB_PROT" = 0L, "FO.LB_CK" = 1L, "FO.LB_GLUC" = 2L,20  "FO.LB_KETON" = 3L, "FO.LB_CREAT" = 4L21)22form_test <- c(23  "FO.LB_PROT" = "PROT", "FO.LB_CK" = "CK", "FO.LB_GLUC" = "GLUC",24  "FO.LB_KETON" = "KETON", "FO.LB_CREAT" = "CREAT"25)26test_name <- c(27  PROT = "Protein", CK = "Creatine Kinase", GLUC = "Glucose",28  KETON = "Ketones", CREAT = "Creatinine"29)3031standardize <- function(raw) {32  if (is.na(raw)) return(NA_character_)33  low <- str_to_lower(str_replace_all(str_trim(raw), " ", ""))34  if (low %in% c("negative", "neg")) return("NEGATIVE")35  if (low %in% c("trace", "tr")) return("TRACE")36  if (low %in% c("1+", "+1", "1plus")) return("1+")37  if (low %in% c("2+", "+2", "2plus")) return("2+")38  raw39}4041forms <- odm |>42  distinct(StudyOID, SubjectKey, FormOID, FormRepeatKey) |>43  filter(FormOID %in% names(form_order))4445one_form <- function(study, subj, formoid, formrep) {46  sub <- odm |>47    filter(48      SubjectKey == subj, FormOID == formoid, FormRepeatKey == formrep49    )50  get <- function(oid) {51    hit <- sub |> filter(ItemOID == oid)52    if (nrow(hit) == 0) return(NA_character_)53    hit$Value[[1]]54  }55  raw <- get("IT.LB.RESULT")56  if (is.na(raw)) return(tibble())57  code <- form_test[[formoid]]58  stresc <- standardize(raw)59  stresn <- suppressWarnings(as.numeric(raw))60  if (!is.na(stresc) && stresc %in% c("NEGATIVE", "TRACE", "1+", "2+")) {61    stresn <- NA_real_62  }63  if (str_starts(str_trim(raw), "<") || str_starts(str_trim(raw), ">")) {64    stresn <- NA_real_65  }66  nrind <- if (str_starts(str_trim(raw), "<")) {67    "LOW"68  } else if (str_starts(str_trim(raw), ">")) {69    "HIGH"70  } else {71    NA_character_72  }73  unit <- get("IT.LB.LBORRESU")74  tibble(75    DOMAIN = "LB",76    STUDYID = study,77    USUBJID = subj,78    ORDER = form_order[[formoid]],79    REP = as.integer(formrep),80    LBTESTCD = code,81    LBTEST = test_name[[code]],82    LBORRES = raw,83    LBORRESU = unit,84    LBSTRESC = stresc,85    LBSTRESN = stresn,86    LBSTRESU = unit,87    LBNRIND = nrind,88    LBDTC = get("IT.LB.LBDTC")89  )90}9192lb <- pmap_dfr(93  list(forms$StudyOID, forms$SubjectKey, forms$FormOID, forms$FormRepeatKey),94  one_form95) |>96  arrange(STUDYID, USUBJID, ORDER, REP) |>97  group_by(STUDYID, USUBJID) |>98  mutate(LBSEQ = row_number()) |>99  ungroup() |>100  select(101    DOMAIN, STUDYID, USUBJID, LBSEQ, LBTESTCD, LBTEST, LBORRES,102    LBORRESU, LBSTRESC, LBSTRESN, LBSTRESU, LBNRIND, LBDTC103  )104105dir.create("/app/output", showWarnings = FALSE)106write_csv(lb, "/app/output/lb.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-lb-character-results (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 LB dataset from the ODM extract in /app/input.89One record per laboratory result (per form occurrence). Character10(dipstick) results are folded to their standardized spellings, numeric11results carry LBSTRESN, and censored results (< / >) drive LBNRIND.12"""1314import re1516import polars as pl1718INPUT = "/app/input"19OUTPUT = "/app/output"2021# Collection order of the tests, with their short and long names.22TEST_ORDER = {23    "FO.LB_PROT": (1, "PROT", "Protein"),24    "FO.LB_CK": (2, "CK", "Creatine Kinase"),25    "FO.LB_GLUC": (3, "GLUC", "Glucose"),26    "FO.LB_KETON": (4, "KETON", "Ketones"),27    "FO.LB_CREAT": (5, "CREAT", "Creatinine"),28}2930# Known dipstick spellings (matched case-insensitively) and the31# standardized result each folds to.32DIPSTICK_FOLDS = {33    "NEGATIVE": "NEGATIVE",34    "NEG": "NEGATIVE",35    "TRACE": "TRACE",36    "TR": "TRACE",37    "1+": "1+",38    "+1": "1+",39    "1 PLUS": "1+",40    "2+": "2+",41    "+2": "2+",42}4344PLAIN_NUMBER = re.compile(r"^[+-]?\d+(\.\d+)?$")4546COLUMNS = [47    "DOMAIN",48    "STUDYID",49    "USUBJID",50    "LBSEQ",51    "LBTESTCD",52    "LBTEST",53    "LBORRES",54    "LBORRESU",55    "LBSTRESC",56    "LBSTRESN",57    "LBSTRESU",58    "LBNRIND",59    "LBDTC",60]616263def standardize(collected):64    """Fold known dipstick spellings; keep anything else as collected."""65    return DIPSTICK_FOLDS.get(collected.strip().upper(), collected)666768def main():69    odm = pl.read_csv(f"{INPUT}/odm.csv", infer_schema_length=0).fill_null("")7071    # Gather the items of each form occurrence.72    occurrences = {}73    subject_order = []74    for row in odm.iter_rows(named=True):75        key = (row["SubjectKey"], row["FormOID"], row["FormRepeatKey"])76        if row["SubjectKey"] not in subject_order:77            subject_order.append(row["SubjectKey"])78        occ = occurrences.setdefault(79            key,80            {81                "STUDYID": row["StudyOID"],82                "USUBJID": row["SubjectKey"],83                "FormOID": row["FormOID"],84                "FormRepeatKey": row["FormRepeatKey"],85                "LBORRES": "",86                "LBORRESU": "",87                "LBDTC": "",88            },89        )90        if row["ItemOID"] == "IT.LB.RESULT":91            occ["LBORRES"] = row["Value"]92        elif row["ItemOID"] == "IT.LB.LBDTC":93            occ["LBDTC"] = row["Value"]94        elif row["ItemOID"] == "IT.LB.LBORRESU":95            occ["LBORRESU"] = row["Value"]9697    # Order records: subject (first appearance), then collection order98    # of the test, then form repeat.99    records = sorted(100        occurrences.values(),101        key=lambda o: (102            subject_order.index(o["USUBJID"]),103            TEST_ORDER[o["FormOID"]][0],104            int(o["FormRepeatKey"]),105        ),106    )107108    out = {name: [] for name in COLUMNS}109    seq_by_subject = {}110    for occ in records:111        _, testcd, test = TEST_ORDER[occ["FormOID"]]112        orres = occ["LBORRES"]113        unit = occ["LBORRESU"]114        stresc = standardize(orres)115        stresn = stresc if PLAIN_NUMBER.match(stresc.strip()) else ""116        if stresc.startswith("<"):117            nrind = "LOW"118        elif stresc.startswith(">"):119            nrind = "HIGH"120        else:121            nrind = ""122        seq_by_subject[occ["USUBJID"]] = seq_by_subject.get(occ["USUBJID"], 0) + 1123        values = {124            "DOMAIN": "LB",125            "STUDYID": occ["STUDYID"],126            "USUBJID": occ["USUBJID"],127            "LBSEQ": seq_by_subject[occ["USUBJID"]],128            "LBTESTCD": testcd,129            "LBTEST": test,130            "LBORRES": orres,131            "LBORRESU": unit,132            "LBSTRESC": stresc,133            "LBSTRESN": stresn,134            "LBSTRESU": unit,135            "LBNRIND": nrind,136            "LBDTC": occ["LBDTC"],137        }138        for name in COLUMNS:139            out[name].append(values[name])140141    # Blank values go in as nulls so they are written as empty fields.142    for name in COLUMNS:143        if name != "LBSEQ":144            out[name] = [v if v != "" else None for v in out[name]]145    lb = pl.DataFrame(out).select(COLUMNS)146    lb.write_csv(f"{OUTPUT}/lb.csv")147148149if __name__ == "__main__":150    main()

1/* opensas reference solution for the yamaa benchmark sdtm-lb-character-results.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 FormOID FormRepeatKey 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. FormOID : $char1024. FormRepeatKey : $char1024. ItemGroupOID : $char1024. ItemGroupRepeatKey : $char1024. ItemOID : $char1024. Value : $char1024.;9run;1011proc sort data=odm out=ordered; by StudyOID SubjectKey FormOID FormRepeatKey; run;12data wide;13  length LBORRES LBORRESU LBDTC $1024;14  retain LBORRES LBORRESU LBDTC;15  set ordered;1617  by StudyOID SubjectKey FormOID FormRepeatKey;18  if first.FormRepeatKey then do;19    LBORRES=''; LBORRESU=''; LBDTC='';20  end;21    if ItemOID='IT.LB.RESULT' then LBORRES=Value;22  if ItemOID='IT.LB.LBORRESU' then LBORRESU=Value;23  if ItemOID='IT.LB.LBDTC' then LBDTC=Value;24  if last.FormRepeatKey then output;25run;26data prepared; set wide; if not missing(LBORRES); length DOMAIN $2 STUDYID USUBJID LBTESTCD LBTEST LBSTRESC LBSTRESU LBNRIND TOKEN $1024;27DOMAIN='LB'; STUDYID=StudyOID; USUBJID=SubjectKey; LBSTRESU=LBORRESU; LBSTRESC=LBORRES; LBNRIND=''; LBSTRESN=.; TOKEN=lowcase(compress(LBORRES,' '));28if TOKEN in ('negative','neg') then LBSTRESC='NEGATIVE'; if TOKEN in ('trace','tr') then LBSTRESC='TRACE'; if TOKEN in ('1+','+1','1plus') then LBSTRESC='1+'; if TOKEN in ('2+','+2','2plus') then LBSTRESC='2+';29if prxmatch('/^[+-]?([0-9]+(\.[0-9]*)?|\.[0-9]+)([eE][+-]?[0-9]+)?$/',strip(LBORRES)) and not (LBSTRESC in ('NEGATIVE','TRACE','1+','2+')) then LBSTRESN=input(LBORRES,best32.);30if substr(strip(LBORRES),1,1)='<' then LBNRIND='LOW'; if substr(strip(LBORRES),1,1)='>' then LBNRIND='HIGH';31if FormOID='FO.LB_PROT' then do; LBTESTCD='PROT'; LBTEST='Protein'; TESTORD=1; end;32else if FormOID='FO.LB_CK' then do; LBTESTCD='CK'; LBTEST='Creatine Kinase'; TESTORD=2; end;33else if FormOID='FO.LB_GLUC' then do; LBTESTCD='GLUC'; LBTEST='Glucose'; TESTORD=3; end;34else if FormOID='FO.LB_KETON' then do; LBTESTCD='KETON'; LBTEST='Ketones'; TESTORD=4; end;35else if FormOID='FO.LB_CREAT' then do; LBTESTCD='CREAT'; LBTEST='Creatinine'; TESTORD=5; end; else delete;36REP=input(FormRepeatKey,best32.); run;37proc sort data=prepared; by STUDYID USUBJID TESTORD REP; run;38data result; set prepared; by STUDYID USUBJID; retain LBSEQ;39  if first.USUBJID then LBSEQ=0;40  LBSEQ+1;41run;4243proc sql;44  create table final as select DOMAIN, STUDYID, USUBJID, LBSEQ, LBTESTCD, LBTEST, LBORRES, LBORRESU, LBSTRESC, LBSTRESN, LBSTRESU, LBNRIND, LBDTC from result;45quit;46proc export data=final outfile="/app/output/lb.csv" dbms=csv replace; run;