Exposure Adjusted Adverse Event Incidence

1library(yamaa)23yamaa_spec("spec_risk.yaml")$run()$save()4yamaa_spec("spec_rate.yaml")$run()$output

1import yamaa23yamaa.spec("spec_risk.yaml").run().save()4yamaa.spec("spec_rate.yaml").run().output

1subjects <- read.csv("/app/input/adsl.csv", stringsAsFactors = FALSE)2events <- read.csv("/app/input/adae.csv", stringsAsFactors = FALSE)3events$ASTDT <- as.Date(events$ASTDT)4risk <- do.call(rbind, lapply(seq_len(nrow(subjects)), function(i) {5  subject <- subjects[i, ]6  start <- as.Date(subject$TRTSDT)7  end <- as.Date(subject$TRTEDT) + 308  do.call(rbind, lapply(c("DIARRHOEA", "HEADACHE"), function(term) {9    selected <- events$STUDYID == subject$STUDYID &10      events$USUBJID == subject$USUBJID & events$AEDECOD == term &11      !is.na(events$ASTDT) & events$ASTDT >= start & events$ASTDT <= end12    onsets <- events$ASTDT[selected]13    first <- if (length(onsets)) min(onsets) else as.Date(NA)14    endpoint <- if (length(onsets)) first else end15    data.frame(STUDYID = subject$STUDYID, USUBJID = subject$USUBJID,16      TRT01A = subject$TRT01A, AEDECOD = term, FIRSTDT = first,17      EVENTN = as.integer(length(onsets) > 0),18      RISKDAYS = as.integer(endpoint - start) + 1L)19  }))20}))21groups <- split(risk, interaction(risk$STUDYID, risk$TRT01A, risk$AEDECOD,22                                  drop = TRUE))23rates <- do.call(rbind, lapply(groups, function(group) {24  count <- sum(group$EVENTN)25  days <- sum(group$RISKDAYS)26  years <- days / 365.2527  data.frame(STUDYID = group$STUDYID[1], TRT01A = group$TRT01A[1],28    AEDECOD = group$AEDECOD[1], NSUBJ = count, PERSONDAYS = days,29    PERSONYRS = years, EAIR100 = 100 * count / years)30}))31risk <- risk[order(risk$STUDYID, risk$USUBJID, risk$AEDECOD), ]32rates <- rates[order(rates$STUDYID, rates$TRT01A, rates$AEDECOD), ]33write.csv(risk, "/app/output/adrisk.csv", row.names = FALSE, na = "")34write.csv(rates, "/app/output/adinc.csv", row.names = FALSE, na = "")

1# Reference solution for the yamaa benchmark adam-adae-exposure-adjusted-incidence (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 1.67import csv8from datetime import date, timedelta9from pathlib import Path101112INPUT_DIR = Path("/app/input")13OUTPUT_DIR = Path("/app/output")14TERMS = ["DIARRHOEA", "HEADACHE"]151617def parse_date(value: str) -> date:18    return date.fromisoformat(value)192021def main() -> None:22    with open(INPUT_DIR / "adsl.csv", newline="", encoding="utf-8") as handle:23        adsl_rows = list(csv.DictReader(handle))24    with open(INPUT_DIR / "adae.csv", newline="", encoding="utf-8") as handle:25        adae_rows = list(csv.DictReader(handle))2627    # One subject record, retaining the order in which subjects appear.28    subjects = {}29    subject_order = []30    for row in adsl_rows:31        key = (row["STUDYID"], row["USUBJID"])32        first_dose = parse_date(row["TRTSDT"])33        last_dose = parse_date(row["TRTEDT"])34        if key not in subjects:35            subjects[key] = {36                "STUDYID": row["STUDYID"],37                "USUBJID": row["USUBJID"],38                "TRT01A": row["TRT01A"],39                "first_dose": first_dose,40                "last_dose": last_dose,41            }42            subject_order.append(key)43        else:44            subjects[key]["first_dose"] = min(45                subjects[key]["first_dose"], first_dose46            )47            subjects[key]["last_dose"] = max(subjects[key]["last_dose"], last_dose)4849    # Index event onset dates by subject and prespecified term.50    onsets = {}51    for row in adae_rows:52        if row["AEDECOD"] not in TERMS or not row["ASTDT"]:53            continue54        key = (row["STUDYID"], row["USUBJID"], row["AEDECOD"])55        onsets.setdefault(key, []).append(parse_date(row["ASTDT"]))5657    risk_rows = []58    for key in subject_order:59        subject = subjects[key]60        first_dose = subject["first_dose"]61        window_end = subject["last_dose"] + timedelta(days=30)6263        for term in TERMS:64            in_window = [65                onset66                for onset in onsets.get(67                    (subject["STUDYID"], subject["USUBJID"], term), []68                )69                if first_dose <= onset <= window_end70            ]71            first_date = min(in_window) if in_window else None72            event_n = 1 if in_window else 073            risk_end = first_date if event_n else window_end7475            risk_rows.append(76                {77                    "STUDYID": subject["STUDYID"],78                    "USUBJID": subject["USUBJID"],79                    "TRT01A": subject["TRT01A"],80                    "AEDECOD": term,81                    "FIRSTDT": first_date.isoformat() if first_date else "",82                    "EVENTN": event_n,83                    "RISKDAYS": (risk_end - first_dose).days + 1,84                }85            )8687    # Aggregate in the order treatments and terms first appear in ADRISK.88    incidence = {}89    incidence_order = []90    for row in risk_rows:91        key = (row["STUDYID"], row["TRT01A"], row["AEDECOD"])92        if key not in incidence:93            incidence[key] = {94                "STUDYID": row["STUDYID"],95                "TRT01A": row["TRT01A"],96                "AEDECOD": row["AEDECOD"],97                "NSUBJ": 0,98                "PERSONDAYS": 0,99            }100            incidence_order.append(key)101        incidence[key]["NSUBJ"] += row["EVENTN"]102        incidence[key]["PERSONDAYS"] += row["RISKDAYS"]103104    incidence_rows = []105    for key in incidence_order:106        row = incidence[key]107        person_years = row["PERSONDAYS"] / 365.25108        incidence_rows.append(109            {110                "STUDYID": row["STUDYID"],111                "TRT01A": row["TRT01A"],112                "AEDECOD": row["AEDECOD"],113                "NSUBJ": row["NSUBJ"],114                "PERSONDAYS": row["PERSONDAYS"],115                "PERSONYRS": person_years,116                "EAIR100": 100 * row["NSUBJ"] / person_years,117            }118        )119120    OUTPUT_DIR.mkdir(parents=True, exist_ok=True)121122    risk_fields = [123        "STUDYID",124        "USUBJID",125        "TRT01A",126        "AEDECOD",127        "FIRSTDT",128        "EVENTN",129        "RISKDAYS",130    ]131    with open(132        OUTPUT_DIR / "adrisk.csv", "w", newline="", encoding="utf-8"133    ) as handle:134        writer = csv.DictWriter(handle, fieldnames=risk_fields)135        writer.writeheader()136        writer.writerows(risk_rows)137138    incidence_fields = [139        "STUDYID",140        "TRT01A",141        "AEDECOD",142        "NSUBJ",143        "PERSONDAYS",144        "PERSONYRS",145        "EAIR100",146    ]147    with open(148        OUTPUT_DIR / "adinc.csv", "w", newline="", encoding="utf-8"149    ) as handle:150        writer = csv.DictWriter(handle, fieldnames=incidence_fields)151        writer.writeheader()152        writer.writerows(incidence_rows)153154155if __name__ == "__main__":156    main()

1/* opensas reference: first qualifying onset defines each term's risk. */2data subjects;3  length STUDYID USUBJID $80 TRT01A $20 start_text end_text $10;4  infile "/app/input/adsl.csv" dsd dlm="," firstobs=2 truncover;5  input STUDYID : $80. USUBJID : $80. TRT01A : $20.6        start_text : $10. end_text : $10.;7  TRTSDT=input(start_text,yymmdd10.); TRTEDT=input(end_text,yymmdd10.);8run;9data events;10  length STUDYID USUBJID $80 AEDECOD $20 date_text $10;11  infile "/app/input/adae.csv" dsd dlm="," firstobs=2 truncover;12  input STUDYID : $80. USUBJID : $80. ASEQ AEDECOD : $20.13        date_text : $10.;14  ASTDT=input(date_text,yymmdd10.);15run;16data population;17  length AEDECOD $20;18  set subjects;19  AEDECOD='DIARRHOEA'; output;20  AEDECOD='HEADACHE'; output;21run;22proc sql;23  create table selected as select p.STUDYID, p.USUBJID, p.TRT01A,24    p.AEDECOD, p.TRTSDT, p.TRTEDT, min(e.ASTDT) as FIRSTDT25  from population p left join events e26    on p.STUDYID=e.STUDYID and p.USUBJID=e.USUBJID27    and p.AEDECOD=e.AEDECOD and e.ASTDT >= p.TRTSDT28    and e.ASTDT <= p.TRTEDT+3029  group by p.STUDYID, p.USUBJID, p.TRT01A, p.AEDECOD, p.TRTSDT, p.TRTEDT;30quit;31data risk;32  set selected;33  EVENTN=not missing(FIRSTDT);34  if EVENTN then RISKDAYS=FIRSTDT-TRTSDT+1;35  else RISKDAYS=TRTEDT+30-TRTSDT+1;36  format FIRSTDT yymmdd10.;37run;38proc sql;39  create table risk_output as select STUDYID, USUBJID, TRT01A, AEDECOD,40    put(FIRSTDT,yymmdd10.) as FIRSTDT, EVENTN, RISKDAYS from risk41  order by STUDYID, USUBJID, AEDECOD;42  create table totals as select STUDYID, TRT01A, AEDECOD,43    sum(EVENTN) as NSUBJ, sum(RISKDAYS) as PERSONDAYS44  from risk group by STUDYID, TRT01A, AEDECOD;45quit;46data rates;47  set totals;48  PERSONYRS=PERSONDAYS/365.25; EAIR100=100*NSUBJ/PERSONYRS;49run;50proc sql;51  create table rate_output as select STUDYID, TRT01A, AEDECOD, NSUBJ,52    PERSONDAYS, PERSONYRS, EAIR100 from rates53  order by STUDYID, TRT01A, AEDECOD;54quit;55proc export data=risk_output outfile="/app/output/adrisk.csv"56  dbms=csv replace; run;57proc export data=rate_output outfile="/app/output/adinc.csv"58  dbms=csv replace; run;