Code medications against WHODrug

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

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

1# Reference solution for the yamaa benchmark sdtm-cm-whodrug-coding (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)15coding <- read_csv(16  "/app/input/coding_output.csv",17  col_types = cols(CMSEQ = col_integer(), .default = col_character())18)19whodrug <- read_csv(20  "/app/input/whodrug_extract.csv",21  col_types = cols(.default = col_character())22)2324# One record per reported medication; the repeat number is the sequence25# number. The preferred name comes from the drug record the coder chose,26# and the class from the ATC code assigned for this use, so the same drug27# record can give a different class on different records. No coding, or a28# code not in the extract, leaves the coded variables empty.29reported <- odm |>30  filter(ItemGroupOID == "IG.CM") |>31  pivot_wider(32    id_cols = c(StudyOID, SubjectKey, ItemGroupRepeatKey),33    names_from = ItemOID,34    values_from = Value,35    values_fn = first36  ) |>37  mutate(38    STUDYID = StudyOID,39    USUBJID = SubjectKey,40    CMSEQ = as.integer(ItemGroupRepeatKey),41    CMTRT = `IT.CM.CMTRT`42  ) |>43  select(STUDYID, USUBJID, CMSEQ, CMTRT)4445cm <- reported |>46  left_join(coding, by = c("STUDYID", "USUBJID", "CMSEQ")) |>47  left_join(whodrug, by = c("DRUG_RECORD_NO", "ATC_CODE")) |>48  mutate(49    DOMAIN = "CM",50    CMDECOD = if_else(51      is.na(DRUG_RECORD_NO) | DRUG_RECORD_NO == "" | is.na(PREFERRED_NAME),52      NA_character_, PREFERRED_NAME53    ),54    CMCLAS = if_else(is.na(CMDECOD), NA_character_, ATC_CLASS_NAME),55    CMCLASCD = if_else(is.na(CMDECOD), NA_character_, ATC_CLASS_CODE)56  ) |>57  arrange(USUBJID, CMSEQ) |>58  select(DOMAIN, STUDYID, USUBJID, CMSEQ, CMTRT, CMDECOD, CMCLAS, CMCLASCD)5960dir.create("/app/output", showWarnings = FALSE)61write_csv(cm, "/app/output/cm.csv", na = "")

1# Reference solution for the yamaa benchmark sdtm-cm-whodrug-coding (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 an SDTM CM dataset from ODM reported medications, the coder's8WHODrug coding output, and the WHODrug extract.910One record per reported medication (the IT.CM.CMTRT item of each IG.CM11item group in ODM). The coded variables come from the WHODrug extract row12for the drug record and ATC code the coder assigned to that record; they13stay blank when the reported name is not yet coded.14"""1516import polars as pl1718INPUT_DIR = "/app/input"19OUTPUT_DIR = "/app/output"2021COLUMNS = [22    "DOMAIN",23    "STUDYID",24    "USUBJID",25    "CMSEQ",26    "CMTRT",27    "CMDECOD",28    "CMCLAS",29    "CMCLASCD",30]313233def read_csv(path):34    # Read every column as text so values are kept exactly as written.35    return pl.read_csv(path, infer_schema_length=0)363738def main():39    odm = read_csv(f"{INPUT_DIR}/odm.csv")40    coding = read_csv(f"{INPUT_DIR}/coding_output.csv")41    whodrug = read_csv(f"{INPUT_DIR}/whodrug_extract.csv")4243    # One record per reported medication: the CMTRT item of each CM item44    # group. CMSEQ is the item group's repeat key within the subject.45    meds = odm.filter(pl.col("ItemOID") == "IT.CM.CMTRT")46    cm = meds.select(47        pl.lit("CM").alias("DOMAIN"),48        pl.col("StudyOID").alias("STUDYID"),49        pl.col("SubjectKey").alias("USUBJID"),50        pl.col("ItemGroupRepeatKey").cast(pl.Int64).alias("CMSEQ"),51        pl.col("Value").alias("CMTRT"),52    ).sort(["USUBJID", "CMSEQ"], maintain_order=True)5354    # The drug record and ATC code the coder assigned to each record.55    cm = cm.join(56        coding.select("STUDYID", "USUBJID", "CMSEQ", "DRUG_RECORD_NO", "ATC_CODE")57        .with_columns(pl.col("CMSEQ").cast(pl.Int64)),58        on=["STUDYID", "USUBJID", "CMSEQ"],59        how="left",60    )6162    # Preferred name and ATC class for that drug record together with the63    # assigned ATC code; the same drug record can carry a different class64    # for a different assigned ATC code.65    cm = cm.join(66        whodrug.select(67            "DRUG_RECORD_NO",68            "ATC_CODE",69            "PREFERRED_NAME",70            "ATC_CLASS_NAME",71            "ATC_CLASS_CODE",72        ),73        on=["DRUG_RECORD_NO", "ATC_CODE"],74        how="left",75    )7677    cm = cm.rename(78        {79            "PREFERRED_NAME": "CMDECOD",80            "ATC_CLASS_NAME": "CMCLAS",81            "ATC_CLASS_CODE": "CMCLASCD",82        }83    )8485    # Uncoded values stay null and are written as empty fields.86    cm.select(COLUMNS).write_csv(f"{OUTPUT_DIR}/cm.csv", null_value="")878889if __name__ == "__main__":90    main()

1/* opensas reference solution for the yamaa benchmark sdtm-cm-whodrug-coding.2   Reads the benchmark inputs and derives the requested output.3   Execute with the existing opensas runtime: opensas /app/output/result.sas. */45data coding_output;6  length STUDYID USUBJID DRUG_RECORD_NO DRUG_CODE ATC_CODE $1024;7  infile "/app/input/coding_output.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;8  input STUDYID : $char1024. USUBJID : $char1024. CMSEQ DRUG_RECORD_NO : $char1024. DRUG_CODE : $char1024. ATC_CODE : $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;1617data whodrug_extract;18  length DRUG_RECORD_NO PREFERRED_NAME DRUG_CODE ATC_CODE ATC_CLASS_CODE ATC_CLASS_NAME $1024;19  infile "/app/input/whodrug_extract.csv" dsd dlm="," firstobs=2 truncover lrecl=32767;20  input DRUG_RECORD_NO : $char1024. PREFERRED_NAME : $char1024. DRUG_CODE : $char1024. ATC_CODE : $char1024. ATC_CLASS_CODE : $char1024. ATC_CLASS_NAME : $char1024.;21run;2223proc sort data=odm out=ordered; by StudyOID SubjectKey ItemGroupRepeatKey; run;24data wide;25  length CMTRT $1024;26  retain CMTRT;27  set ordered;28  where ItemGroupOID='IG.CM';29  by StudyOID SubjectKey ItemGroupRepeatKey;30  if first.ItemGroupRepeatKey then do;31    CMTRT='';32  end;33    if ItemOID='IT.CM.CMTRT' then CMTRT=Value;34  if last.ItemGroupRepeatKey then output;35run;36data medications; set wide; length DOMAIN $2 STUDYID USUBJID $1024; DOMAIN='CM'; STUDYID=StudyOID; USUBJID=SubjectKey; CMSEQ=input(ItemGroupRepeatKey,best32.); run;37proc sql; create table result as select a.*,c.PREFERRED_NAME as CMDECOD,c.ATC_CLASS_NAME as CMCLAS,c.ATC_CLASS_CODE as CMCLASCD from medications a38left join coding_output b on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.CMSEQ=b.CMSEQ39left join whodrug_extract c on b.DRUG_RECORD_NO=c.DRUG_RECORD_NO and b.DRUG_CODE=c.DRUG_CODE and b.ATC_CODE=c.ATC_CODE; quit;40proc sql;41  create table final as select DOMAIN, STUDYID, USUBJID, CMSEQ, CMTRT, CMDECOD, CMCLAS, CMCLASCD from result;42quit;43proc export data=final outfile="/app/output/cm.csv" dbms=csv replace; run;