Split Adverse Events at Severity Changes

1library(yamaa)23adae <- yamaa_domain("spec.yaml")$output4adae

1import yamaa23adae = yamaa.yamaa_domain("spec.yaml").output4adae

1# R reference for the yamaa benchmark adam-adae-worsening-emergence.2# Uses the collected event records and dated findings to construct periods.3library(readr)45read_input <- function(name) {6  read_csv(paste0('/app/input/', name),7           col_types = cols(.default = col_character()),8           show_col_types = FALSE)9}10normalize <- function(x) {11  short <- !is.na(x) & nchar(x) == 16L12  x[short] <- paste0(x[short], ':00')13  x14}15ae <- read_input('ae.csv')16fa <- read_input('fa.csv')17adsl <- read_input('adsl.csv')18fa$FADTC <- normalize(fa$FADTC)19ae <- ae[order(ae$STUDYID, ae$USUBJID, as.integer(ae$AESEQ)), ]20ranks <- c(MILD = 1L, MODERATE = 2L, SEVERE = 3L)21output <- list()22previous_study <- previous_subject <- NA_character_23aseq <- 0L24for (i in seq_len(nrow(ae))) {25  event <- ae[i, ]26  subject <- event$USUBJID27  study <- event$STUDYID28  if (!identical(study, previous_study) ||29      !identical(subject, previous_subject)) aseq <- 0L30  previous_study <- study31  previous_subject <- subject32  trt <- normalize(adsl$TRTSDTM[33    adsl$STUDYID == study & adsl$USUBJID == subject34  ])[1]35  onset <- normalize(event$AESTDTC)36  ending <- normalize(event$AEENDTC)37  findings <- fa[fa$STUDYID == study & fa$USUBJID == subject &38                   fa$FALNKID == event$AELNKID, ]39  times <- sort(unique(findings$FADTC))40  sevs <- rep(NA_character_, length(times))41  grades <- rep(NA_integer_, length(times))42  sev <- NA_character_43  grade <- NA_integer_44  for (j in seq_along(times)) {45    updates <- findings[findings$FADTC == times[j], ]46    if (anyDuplicated(updates$FATESTCD)) stop('Duplicate finding at event/time')47    new_sev <- updates$FASTRESC[updates$FATESTCD == 'SEV']48    new_grade <- updates$FASTRESC[updates$FATESTCD == 'TOXGR']49    if (length(new_sev) && !is.na(new_sev)) sev <- new_sev50    if (length(new_grade) && !is.na(new_grade)) grade <- as.integer(new_grade)51    sevs[j] <- sev52    grades[j] <- grade53  }54  before <- if (is.na(trt)) integer() else which(times < trt)55  baseline_sev <- baseline_grade <- NA_integer_56  if (length(before)) {57    last <- tail(before, 1)58    baseline_sev <- unname(ranks[sevs[last]])59    baseline_grade <- grades[last]60  }61  emergent <- FALSE62  for (j in seq_along(times)) {63    if (!is.na(onset) && !is.na(trt)) {64      if (onset >= trt) emergent <- TRUE65      else if (times[j] >= trt) {66        rank <- unname(ranks[sevs[j]])67        sev_worse <- !is.na(rank) && !is.na(baseline_sev) && rank > baseline_sev68        grade_worse <- !is.na(grades[j]) && !is.na(baseline_grade) &&69          grades[j] > baseline_grade70        emergent <- emergent || sev_worse || grade_worse71      }72    }73    aseq <- aseq + 1L74    output[[length(output) + 1L]] <- data.frame(75      STUDYID = study, USUBJID = subject, ASEQ = aseq,76      AESEQ = as.integer(event$AESEQ), AEDECOD = event$AEDECOD,77      AESTDTC = onset, AEENDTC = ending, ASTDTM = times[j],78      AENDTM = if (j < length(times)) times[j + 1L] else ending,79      TRTSDTM = trt, AESEV = event$AESEV, AETOXGR = event$AETOXGR,80      ASEV = sevs[j], ATOXGRN = grades[j],81      TRTEMFL = if (emergent) 'Y' else NA_character_82    )83  }84}85dir.create('/app/output', showWarnings = FALSE)86write_csv(do.call(rbind, output), '/app/output/adae.csv', na = '')

1# Python reference for the yamaa benchmark adam-adae-worsening-emergence.2# Uses the collected event records and dated findings to construct periods.3from pathlib import Path45import polars as pl678def normalize(value):9    return value + ":00" if value and len(value) == 16 else value101112ae = pl.read_csv("/app/input/ae.csv", infer_schema=False).to_dicts()13fa = pl.read_csv("/app/input/fa.csv", infer_schema=False).to_dicts()14exposure = {15    (row["STUDYID"], row["USUBJID"]): normalize(row["TRTSDTM"])16    for row in pl.read_csv("/app/input/adsl.csv", infer_schema=False).to_dicts()17}18severity_rank = {"MILD": 1, "MODERATE": 2, "SEVERE": 3}19output = []20sequences = {}21for event in sorted(ae, key=lambda r: (r["STUDYID"], r["USUBJID"], int(r["AESEQ"]))):22    subject = event["STUDYID"], event["USUBJID"]23    trt = exposure.get(subject)24    onset, end = normalize(event["AESTDTC"]), normalize(event["AEENDTC"])25    changes = {}26    for finding in fa:27        if (finding["STUDYID"], finding["USUBJID"]) != subject:28            continue29        if finding["FALNKID"] != event["AELNKID"]:30            continue31        time = normalize(finding["FADTC"])32        values = changes.setdefault(time, {})33        test = finding["FATESTCD"]34        if test in values:35            raise ValueError("Duplicate finding at the same event and time")36        values[test] = finding["FASTRESC"]37    states = []38    sev, grade = None, None39    for time, values in sorted(changes.items()):40        if values.get("SEV") is not None:41            sev = values["SEV"]42        if values.get("TOXGR") is not None:43            grade = int(values["TOXGR"])44        states.append((time, sev, grade))45    before = [state for state in states if trt and state[0] < trt]46    baseline_sev = severity_rank.get(before[-1][1]) if before else None47    baseline_grade = before[-1][2] if before else None48    emergent = False49    for i, (time, sev, grade) in enumerate(states):50        if onset and trt:51            if onset >= trt:52                emergent = True53            elif time >= trt:54                rank = severity_rank.get(sev)55                sev_worse = (56                    rank is not None57                    and baseline_sev is not None58                    and rank > baseline_sev59                )60                grade_worse = (61                    grade is not None62                    and baseline_grade is not None63                    and grade > baseline_grade64                )65                emergent = emergent or sev_worse or grade_worse66        sequences[subject] = sequences.get(subject, 0) + 167        output.append(68            {69                "STUDYID": event["STUDYID"],70                "USUBJID": event["USUBJID"],71                "ASEQ": sequences[subject],72                "AESEQ": int(event["AESEQ"]),73                "AEDECOD": event["AEDECOD"],74                "AESTDTC": onset,75                "AEENDTC": end,76                "ASTDTM": time,77                "AENDTM": states[i + 1][0] if i + 1 < len(states) else end,78                "TRTSDTM": trt,79                "AESEV": event["AESEV"],80                "AETOXGR": event["AETOXGR"],81                "ASEV": sev,82                "ATOXGRN": grade,83                "TRTEMFL": "Y" if emergent else None,84            }85        )8687Path("/app/output").mkdir(exist_ok=True)88pl.DataFrame(output).write_csv("/app/output/adae.csv")

1/* opensas reference for yamaa's adam-adae-worsening-emergence benchmark.2   Construct periods from one AE and its collected, dated FA findings. */3data ae;4  length STUDYID USUBJID AELNKID AEDECOD AESTDTC AEENDTC AESEV AETOXGR $1024;5  infile '/app/input/ae.csv' dsd dlm=',' firstobs=2 truncover lrecl=32767;6  input STUDYID : $char1024. USUBJID : $char1024. AESEQ7    AELNKID : $char1024. AEDECOD : $char1024. AESTDTC : $char1024.8    AEENDTC : $char1024. AESEV : $char1024. AETOXGR : $char1024.;9  if lengthn(AESTDTC)=16 then AESTDTC=cats(AESTDTC,':00');10  if lengthn(AEENDTC)=16 then AEENDTC=cats(AEENDTC,':00');11run;12data fa;13  length STUDYID USUBJID FALNKID FATESTCD FAOBJ FASTRESC FADTC $1024;14  infile '/app/input/fa.csv' dsd dlm=',' firstobs=2 truncover lrecl=32767;15  input STUDYID : $char1024. USUBJID : $char1024. FASEQ16    FALNKID : $char1024. FATESTCD : $char1024. FAOBJ : $char1024.17    FASTRESC : $char1024. FADTC : $char1024.;18  if lengthn(FADTC)=16 then FADTC=cats(FADTC,':00');19  SEVRANK=.;20  if FATESTCD='SEV' then do;21    select(FASTRESC);22      when('MILD') SEVRANK=1; when('MODERATE') SEVRANK=2;23      when('SEVERE') SEVRANK=3; otherwise SEVRANK=.;24    end;25  end;26run;27data adsl;28  length STUDYID USUBJID TRTSDTM $1024;29  infile '/app/input/adsl.csv' dsd dlm=',' firstobs=2 truncover lrecl=32767;30  input STUDYID : $char1024. USUBJID : $char1024. TRTSDTM : $char1024.;31  if lengthn(TRTSDTM)=16 then TRTSDTM=cats(TRTSDTM,':00');32run;33proc sql;34  create table changes as35    select STUDYID,USUBJID,FALNKID,FADTC as ASTDTM,36      max(SEVRANK) as SEVCOLN,37      max(case when FATESTCD='TOXGR' then input(FASTRESC,best32.)38        else . end) as GRDCOL39    from fa group by STUDYID,USUBJID,FALNKID,FADTC;40  create table joined as41    select a.*,c.ASTDTM,c.SEVCOLN,c.GRDCOL,b.TRTSDTM42    from changes as c inner join ae as a43      on c.STUDYID=a.STUDYID and c.USUBJID=a.USUBJID44        and c.FALNKID=a.AELNKID45    left join adsl as b46      on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID;47quit;48proc sort data=joined;49  by STUDYID USUBJID AESEQ ASTDTM;50run;51data states;52  length ASEV $1024;53  set joined;54  by STUDYID USUBJID AESEQ;55  retain ASEVN ATOXGRN;56  if first.AESEQ then do; ASEVN=.; ATOXGRN=.; end;57  if not missing(SEVCOLN) then ASEVN=SEVCOLN;58  if not missing(GRDCOL) then ATOXGRN=GRDCOL;59  select(ASEVN);60    when(1) ASEV='MILD'; when(2) ASEV='MODERATE';61    when(3) ASEV='SEVERE'; otherwise ASEV='';62  end;63run;64/* Read the next start by scanning periods in reverse chronological order. */65proc sort data=states;66  by STUDYID USUBJID AESEQ descending ASTDTM;67run;68data periods;69  length AENDTM NEXTSTART $1024;70  set states;71  by STUDYID USUBJID AESEQ;72  retain NEXTSTART;73  if first.AESEQ then NEXTSTART='';74  if not missing(NEXTSTART) then AENDTM=NEXTSTART;75  else AENDTM=AEENDTC;76  NEXTSTART=ASTDTM;77run;78proc sql;79  create table pretreatment as80    select STUDYID,USUBJID,AESEQ,max(ASTDTM) as BASEDTM81    from periods where not missing(TRTSDTM) and ASTDTM<TRTSDTM82    group by STUDYID,USUBJID,AESEQ;83  create table baseline as84    select a.STUDYID,a.USUBJID,a.AESEQ,85      a.ASEVN as PRESEV,a.ATOXGRN as PREGRD86    from periods as a inner join pretreatment as b87      on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID88        and a.AESEQ=b.AESEQ and a.ASTDTM=b.BASEDTM;89  create table compared as90    select a.*,b.PRESEV,b.PREGRD91    from periods as a left join baseline as b92      on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.AESEQ=b.AESEQ;93  create table worsening as94    select STUDYID,USUBJID,AESEQ,min(ASTDTM) as FIRSTWRS95    from compared where not missing(TRTSDTM) and ASTDTM>=TRTSDTM and96      ((not missing(ASEVN) and not missing(PRESEV) and ASEVN>PRESEV) or97       (not missing(ATOXGRN) and not missing(PREGRD) and ATOXGRN>PREGRD))98    group by STUDYID,USUBJID,AESEQ;99  create table flagged as100    select a.*,b.FIRSTWRS from periods as a left join worsening as b101      on a.STUDYID=b.STUDYID and a.USUBJID=b.USUBJID and a.AESEQ=b.AESEQ;102quit;103proc sort data=flagged;104  by STUDYID USUBJID AESEQ ASTDTM;105run;106data result;107  length TRTEMFL $1;108  set flagged;109  by STUDYID USUBJID;110  retain ASEQ;111  if first.USUBJID then ASEQ=0;112  ASEQ=ASEQ+1;113  TRTEMFL='';114  if not missing(AESTDTC) and not missing(TRTSDTM) then do;115    if AESTDTC>=TRTSDTM or116      (not missing(FIRSTWRS) and ASTDTM>=FIRSTWRS) then TRTEMFL='Y';117  end;118run;119proc sql;120  create table final as121    select STUDYID,USUBJID,ASEQ,AESEQ,AEDECOD,AESTDTC,AEENDTC,122      ASTDTM,AENDTM,TRTSDTM,AESEV,AETOXGR,ASEV,ATOXGRN,TRTEMFL from result;123quit;124proc export data=final outfile='/app/output/adae.csv' dbms=csv replace; run;