yamaa As a clinical trial statistician or programmer, challenge anything in this benchmark: describe a problem in Comments, or propose a fix with the Edit button on any source file.
- Task: Would a real study need this derivation, specified this way? Does it follow CDISC ADaM: variable names, controlled terms, record structure? Flag any rule that is ambiguous or that a sponsor would handle differently.
- Inputs: Do they look like data collected in a real trial, and do they include the situations you would expect to meet in one?
- Expected output: Would independent QC programming reproduce every value from these inputs and rules?
The other tabs beside Comments are background for assessment, not under review.
Summary
EditUnder reviewWhat this benchmark meansREADME.mdGoal: rank each subject's adverse events (AEs) from worst to
mildest: the numeric severity ASEVN orders events into a rank
SEVRANK and a dense rank SEVLVL in an Analysis Data Model (ADaM)
adverse event analysis dataset.
Input: collected Study Data Tabulation Model (SDTM) adverse
event records with the reported term and severity; the reported
severity is carried through as ASEV (empty when severity was not
reported).
Variables:
ASEVNis the numeric form ofASEV: 1 for MILD, 2 for MODERATE, and 3 for SEVERE; empty whenASEVis empty.SEVRANKis the rank of the event among the subject's events ordered byASEVN, worst first. Events with equal severity share a rank and the ranks they would otherwise fill are skipped, so a rank of 1 means the subject reported nothing worse.SEVLVLis the dense rank of the event: it numbers the distinct severities the subject reported, worst first with each severity counted once, so its largest value is how many different severities the subject has, an unreported severity counting as one more.
Note: an event with no reported severity sorts after every event with a reported severity, and events with no reported severity share one rank with each other because nothing tells them apart. A subject with a single event carries rank 1, and when none of a subject's events has a reported severity every event shares rank 1.
Inputs
Under review1 source fileinput/ae.csv
Edit12 rows| STUDYID | USUBJID | AESEQ | AETERM | AESEV |
|---|---|---|---|---|
| PILOT7 | 1 | Nausea | SEVERE | |
| PILOT7 | 2 | Fatigue | MILD | |
| PILOT7 | 3 | Pneumonitis | SEVERE | |
| PILOT7 | 4 | Rash | MODERATE | |
| PILOT7 | 1 | Headache | MODERATE | |
| PILOT7 | 2 | Insomnia | MODERATE | |
| PILOT7 | 1 | Diarrhoea | MODERATE | |
| PILOT7 | 2 | Cough | — | |
| PILOT7 | 3 | Pyrexia | — | |
| PILOT7 | 1 | Syncope | SEVERE | |
| PILOT7 | 1 | Headache | — | |
| PILOT7 | 2 | Nausea | — |
Expected output
Under review12 expected rowsexpected/adae.csv
Edit12 rows| STUDYID | USUBJID | AESEQ | ASEV | ASEVN | SEVRANK | SEVLVL |
|---|---|---|---|---|---|---|
| PILOT7 | 1 | SEVERE | 3 | 1 | 1 | |
| PILOT7 | 2 | MILD | 1 | 4 | 3 | |
| PILOT7 | 3 | SEVERE | 3 | 1 | 1 | |
| PILOT7 | 4 | MODERATE | 2 | 3 | 2 | |
| PILOT7 | 1 | MODERATE | 2 | 1 | 1 | |
| PILOT7 | 2 | MODERATE | 2 | 1 | 1 | |
| PILOT7 | 1 | MODERATE | 2 | 1 | 1 | |
| PILOT7 | 2 | — | — | 2 | 2 | |
| PILOT7 | 3 | — | — | 2 | 2 | |
| PILOT7 | 1 | SEVERE | 3 | 1 | 1 | |
| PILOT7 | 1 | — | — | 1 | 1 | |
| PILOT7 | 2 | — | — | 1 | 1 |
Comments and assessment materials
| Dataset Name | Variable Name | Variable Label | Type | Codelist/Controlled Terms | Core | Computational Method | Origin |
|---|---|---|---|---|---|---|---|
| ADAE | STUDYID | Study Identifier | character | Copy value from AE.STUDYID. | Predecessor | ||
| ADAE | USUBJID | Unique Subject Identifier | character | Copy value from AE.USUBJID. | Predecessor | ||
| ADAE | AESEQ | Sequence Number | numeric | Copy value from AE.AESEQ. | Predecessor | ||
| ADAE | ASEV | Analysis Severity/Intensity | character | MILD, MODERATE, SEVERE | Copy value from AE.AESEV. | Predecessor | |
| ADAE | ASEVN | Analysis Severity/Intensity (N) | numeric | Recode ASEV: "MILD" → "1", "MODERATE" → "2", "SEVERE" → "3"; missing values stay missing; unlisted values are errors. | Derived | ||
| ADAE | SEVRANK | Severity Rank | numeric | Rank rows from 1 using window SEVERITY_ORDER; ties share a number and the next distinct value skips the gap. | Derived | ||
| ADAE | SEVLVL | Severity Level | numeric | Rank rows from 1 using window SEVERITY_ORDER; ties share a number and the next distinct value follows without a gap. | Derived |
| Variable | Permitted values |
|---|---|
| ASEV | MILD, MODERATE, SEVERE |
| Version | Date | Author | Description | Reviewer | Sign-off |
|---|---|---|---|---|---|
| 1.3 | 2026-09-29 | yamaa docs build | Show study codelist names, external dictionaries, and ISO 8601 formats in variable rows; list the document's codelist values and metadata | ||
| 1.2 | 2026-09-29 | yamaa docs build | Classify ADaM copies from a declared input as Predecessor and recodes as Derived, and SDTM recodes of locally derived columns as Derived | ||
| 1.1 | 2026-09-29 | yamaa docs build | Resolve lookups into the donor record they select, and give every derivation shape a plain-language rule instead of its raw YAML | ||
| 1.0 | yamaa docs build | Initial generation from spec.yaml |
prompts/adam-adae-severity-rank/full.md
Edit21 linesFollowing CDISC ADaM standards, use the provided AE dataset to create an ADAE dataset with one record per adverse event.
The output dataset should contain the following columns in this order: STUDYID, USUBJID, AESEQ, ASEV, ASEVN, SEVRANK, SEVLVL
ASEV is the reported severity, and is empty when severity was not reported. ASEVN is the numeric form of ASEV: 1 for MILD, 2 for MODERATE, and 3 for SEVERE, and is empty when ASEV is empty. SEVRANK ranks the subject's events by ASEVN, worst first. Events with equal severity share a rank and the ranks they would otherwise fill are skipped, so a rank of 1 means the subject reported nothing worse. SEVLVL numbers the distinct severities the subject reported, worst first with each severity counted once. An event with no reported severity sorts after every reported one, and unreported events share one rank with each other. A subject with a single event carries rank 1, and when none of a subject's events has a reported severity every event shares rank 1.
Read the source datasets from /app/input and save the completed dataset as /app/output/adae.csv.
prompts/adam-adae-severity-rank/conventions.md
Edit16 linesFollowing CDISC ADaM standards, use the provided AE dataset to create an ADAE dataset with one record per adverse event.
The output dataset should contain the following columns in this order: STUDYID, USUBJID, AESEQ, ASEV, ASEVN, SEVRANK, SEVLVL
ASEVN is the numeric form of ASEV: 1 for MILD, 2 for MODERATE, and 3 for SEVERE. SEVRANK ranks the subject's events by ASEVN, worst first. Events with equal severity share a rank and the ranks they would otherwise fill are skipped. SEVLVL numbers the distinct severities the subject reported, worst first with each severity counted once. An event with no reported severity sorts after every reported one, and unreported events share one rank with each other.
Read the source datasets from /app/input and save the completed dataset as /app/output/adae.csv.
prompts/adam-adae-severity-rank/brief.md
Edit8 linesFollowing CDISC ADaM standards, use the provided AE dataset to create an ADAE dataset with one record per adverse event.
The output dataset should contain the following columns in this order: STUDYID, USUBJID, AESEQ, ASEV, ASEVN, SEVRANK, SEVLVL
Read the source datasets from /app/input and save the completed dataset as /app/output/adae.csv.
run.R
Edit4 lineslibrary(yamaa)adae <- yamaa_domain("spec.yaml")$outputadaerun.py
Edit4 linesimport yamaaadae = yamaa.yamaa_domain("spec.yaml").outputadaesolutions/adam-adae-severity-rank/result.R
Edit43 lines# Reference solution for the yamaa benchmark adam-adae-severity-rank (R track).## Written from the benchmark's full prompt and its inputs alone, as an agent# would write it. Harbor's oracle agent runs it to check that the prompt can# be solved and that the grader scores a correct answer 1.library(dplyr, warn.conflicts = FALSE)library(purrr)library(readr)ae <- read_csv( "/app/input/ae.csv", col_types = cols(AESEQ = col_integer(), .default = col_character()))adae <- ae |> mutate( ASEV = AESEV, ASEVN = case_when( ASEV == "MILD" ~ 1L, ASEV == "MODERATE" ~ 2L, ASEV == "SEVERE" ~ 3L ) ) |> group_by(USUBJID) |> mutate( # Competition rank, worst first; unreported sorts after every reported # one and shares one rank. Rank is one plus the count of worse events. SEVRANK = map_int(ASEVN, function(v) { if (is.na(v)) sum(!is.na(ASEVN)) + 1L else sum(ASEVN > v, na.rm = TRUE) + 1L }), # Dense rank of the distinct severities, worst first, with unreported # as one level after every reported one. SEVLVL = map_int(ASEVN, function(v) { reported <- unique(ASEVN[!is.na(ASEVN)]) if (is.na(v)) length(reported) + 1L else sum(reported > v) + 1L }) ) |> ungroup() |> select(STUDYID, USUBJID, AESEQ, ASEV, ASEVN, SEVRANK, SEVLVL)dir.create("/app/output", showWarnings = FALSE)write_csv(adae, "/app/output/adae.csv", na = "")solutions/adam-adae-severity-rank/result.py
Edit44 lines# Reference solution for the yamaa benchmark adam-adae-severity-rank (Python track).## Written from the benchmark's full prompt and its inputs alone, as an agent# would write it. Harbor's oracle agent runs it to check that the prompt can# be solved and that the grader scores a correct answer 1.from pathlib import Pathimport polars as plae = pl.read_csv("/app/input/ae.csv", infer_schema=False).with_columns( pl.col("AESEQ").cast(pl.Int64))asevn = ( pl.when(pl.col("AESEV") == "MILD") .then(1) .when(pl.col("AESEV") == "MODERATE") .then(2) .when(pl.col("AESEV") == "SEVERE") .then(3) .otherwise(None) .cast(pl.Int64))base = ae.with_columns(ASEV=pl.col("AESEV"), ASEVN=asevn)# Competition rank (worst first, gaps after ties) and dense rank of the# distinct severities. Polars ranks only non-null values, so unreported# events take one rank after every reported one.rank_min = pl.col("ASEVN").rank("min", descending=True).over("USUBJID")rank_dense = pl.col("ASEVN").rank("dense", descending=True).over("USUBJID")n_reported = pl.col("ASEVN").count().over("USUBJID")max_dense = rank_dense.max().over("USUBJID")adae = ( base.with_columns( SEVRANK=(rank_min.fill_null(n_reported + 1)).cast(pl.Int64), SEVLVL=(rank_dense.fill_null(max_dense.fill_null(0) + 1)).cast(pl.Int64), ).select("STUDYID", "USUBJID", "AESEQ", "ASEV", "ASEVN", "SEVRANK", "SEVLVL"))Path("/app/output").mkdir(exist_ok=True)adae.write_csv("/app/output/adae.csv")solutions/adam-adae-severity-rank/result.sas
Edit28 lines/* opensas reference solution for the yamaa benchmark adam-adae-severity-rank. Reads the benchmark inputs and derives the requested output. Execute with the existing opensas runtime: opensas /app/output/result.sas. */data ae; length STUDYID USUBJID AETERM AESEV $1024; infile "/app/input/ae.csv" dsd dlm="," firstobs=2 truncover lrecl=32767; input STUDYID : $char1024. USUBJID : $char1024. AESEQ AETERM : $char1024. AESEV : $char1024.;run;data graded; length ASEV $20; set ae; ASEV=AESEV; select(ASEV); when('MILD') ASEVN=1; when('MODERATE') ASEVN=2; when('SEVERE') ASEVN=3; when('LIFE-THREATENING') ASEVN=4; otherwise ASEVN=.; end;run;proc sort data=graded; by STUDYID USUBJID descending ASEVN AESEQ; run;data result; set graded; by STUDYID USUBJID descending ASEVN; retain count SEVRANK SEVLVL; if first.USUBJID then do; count=0; SEVRANK=0; SEVLVL=0; end; count+1; if first.ASEVN then do; SEVRANK=count; SEVLVL+1; end;run;proc sql; create table final as select STUDYID, USUBJID, AESEQ, ASEV, ASEVN, SEVRANK, SEVLVL from result;quit;proc export data=final outfile="/app/output/adae.csv" dbms=csv replace; run;
Comments
Signed in with GitHub; saved in repository Discussions