Code medications against WHODrug

1library(yamaa)23cm <- yamaa_domain("spec.yaml")$output4cm

1import yamaa23cm = yamaa.yamaa_domain("spec.yaml").output4cm

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 1.67from pathlib import Path89import polars as pl1011odm = pl.read_csv("/app/input/odm.csv", infer_schema=False)12coding = pl.read_csv("/app/input/coding_output.csv", infer_schema=False).with_columns(13    pl.col("CMSEQ").cast(pl.Int64)14)15whodrug = pl.read_csv("/app/input/whodrug_extract.csv", infer_schema=False)1617# One record per reported medication; the repeat number is the sequence18# number. The preferred name comes from the drug record the coder chose,19# and the class from the ATC code assigned for this use, so the same drug20# record can give a different class on different records. No coding, or a21# code not in the extract, leaves the coded variables empty.22reported = (23    odm.filter(pl.col("ItemGroupOID") == "IG.CM")24    .pivot(25        index=["StudyOID", "SubjectKey", "ItemGroupRepeatKey"],26        on="ItemOID",27        values="Value",28        aggregate_function="first",29    )30    .with_columns(31        STUDYID=pl.col("StudyOID"),32        USUBJID=pl.col("SubjectKey"),33        CMSEQ=pl.col("ItemGroupRepeatKey").cast(pl.Int64),34        CMTRT=pl.col("IT.CM.CMTRT"),35    )36    .select("STUDYID", "USUBJID", "CMSEQ", "CMTRT")37)3839cm = (40    reported.join(coding, on=["STUDYID", "USUBJID", "CMSEQ"], how="left")41    .join(42        whodrug,43        left_on=["DRUG_RECORD_NO", "ATC_CODE"],44        right_on=["DRUG_RECORD_NO", "ATC_CODE"],45        how="left",46    )47    .with_columns(48        DOMAIN=pl.lit("CM"),49        CMDECOD=pl.when(pl.col("PREFERRED_NAME").is_null())50        .then(None)51        .otherwise(pl.col("PREFERRED_NAME")),52        CMCLAS=pl.col("ATC_CLASS_NAME"),53        CMCLASCD=pl.col("ATC_CLASS_CODE"),54    )55    .with_columns(56        CMDECOD=pl.when(57            pl.col("DRUG_RECORD_NO").fill_null("") == ""58        )59        .then(None)60        .otherwise(pl.col("CMDECOD")),61        CMCLAS=pl.when(pl.col("CMDECOD").is_null())62        .then(None)63        .otherwise(pl.col("CMCLAS")),64        CMCLASCD=pl.when(pl.col("CMDECOD").is_null())65        .then(None)66        .otherwise(pl.col("CMCLASCD")),67    )68    .sort(["USUBJID", "CMSEQ"])69    .select(70        "DOMAIN",71        "STUDYID",72        "USUBJID",73        "CMSEQ",74        "CMTRT",75        "CMDECOD",76        "CMCLAS",77        "CMCLASCD",78    )79)8081Path("/app/output").mkdir(exist_ok=True)82cm.write_csv("/app/output/cm.csv")

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;