Versions Compared

Old Version 33

changes.mady.by.user Nancy Brucken

Saved on

compared with

New Version Current

changes.mady.by.user Karin LaPann

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.
Comment: Updated header to ADQRS BEST PRACTICE

Decisions

TopicADQRS BEST PRACTICE
Choice of Standards for Development
  • Standards can be developed for QS, RS or FT domain instruments, as priorities dictate.
Dataset Structure and Naming Conventions
  • Do not include QS, FT or RS in the ADaM dataset names.  That releases 2 more characters for use in the dataset names.  The GDS-

...

  • SF data set will be ADGDSSF.
  • Whether to include responses to all items in the ADQRS datasets, or only summary records being analyzed will be decided on a per-instrument basis, so this information will be included in the ADQRS supplements on a per-instrument basis.  However, the FDA has expressed a preference for including both the individual items and summary records in the same dataset.  They recalculate all of the scores to compare against values provided by the sponsor, and it's easier for them if both individual items and summary records are all in the same dataset.
  • Data sent from a vendor with all scores computed should be handled as SDTM data, and additional ADaM datasets created, even if score calculations are not replicated. There are differences in variable names between SDTM and ADaM, and the vendor datasets will probably not contain ADSL variables required for analysis.
  • Each instrument should be stored in a separate dataset as a best practice, for ease of analysis and review.  However, this is not an ADaM requirement, so the final decision on whether to submit individual instrument datasets or a single "ADQS" dataset is left with the sponsor.
  • A list of expected variables for ADQRS datasets will not be created, since these supplements are meant to serve as examples.
PARCATy/PARAM/ PARAMCD
  • Include PARCAT1 in ADQRS datasets in case multiple instruments are subsequently combined into a single dataset.  PARCAT1 contains the name of the instrument, and may come from

...

  • --CAT.
  • PARCAT2 is allowed if needed, and should contain the subscale name (may come from

...

  • --SCAT, but is driven by analysis/reporting needs).
  • Bring

...

  • --TEST and --TESTCD directly into PARAM and PARAMCD, respectively, for maximum traceability back to QS/FT/RS.
  • Create new PARAM/PARAMCD values for subscale and total scores, and where items from different SDTM versions are combined for analysis purposes, such as CHART-SF, where the paper and interview versions are summarized together.  Submit requests for new PARAM/PARAMCD values to the QRS CT group.
  • When creating PARAMCD values for subscale scores, keep the SDTM

...

  • --TESTCD value, and abbreviate any published subscale codes.  Select values for PARCATy and PARAMCD such that the items are grouped in a logical order.
  • Use a completely different PARAMCD value for the computed total score than the SDTM

...

  • --TESTCD values.  That makes it easier to spot in the data set, and also distinguishes it from a collected total score that may be stored in SDTM.  For the GDS-SF, this will be called GDS02TS.  A new Controlled Terminology request should be submitted for computed score records.
AVAL/AVALC
  • Keep

...

  • --ORRES, and use AVALC only if it differs from

...

  • --ORRES.

...

  • Conditionally branched (formerly "logically skipped") items are now being captured in the QRS supplements (2021). We recommend that they are carried over from SDTM for traceability. In this case, the AVAL and AVALC is null, unless someone needs to summarize a count. In the latter case, we suggest to add AVALC of '

...

  • Conditionally Branched Item' so that it may be counted. These items do not need to be carried over to additional summary datasets.
  • Sometimes the same question can be used on different instruments that are related;  if the data is being pooled across studies for integration, and some of the items are the same, it may be possible to combine data from those items via the author codes (question codes pre-printed on the instruments).  Author codes should be included in ADaM if present in SDTM.
Analysis and Baseline Flags
  • Set ANLxxFL and ABLFL only on total score and sub-score records, unless individual items are also summarized.
  • Baseline definitions are analysis-specific, and thus are outside the scope of ADQRS supplements.  Baseline values may still be identified in the examples, but those are not intended to endorse a specific baseline definition.
Treatment Variables
  • Include at least one treatment variable (TRTP/TRT01P/TRTA/TRT01A) in examples for now, to conform to the latest ADaM IG.
Total and Subscale Scores
  • Document instrument scoring instructions in define.xml.
  • Even if SDTM has computed a total score, keep a separate ADaM total score record in the analysis dataset. Also keep the total score from SDTM which has a different parameter.
  • There is no clear standard terminology for describing subscores/subtotals.  The ADQRS subteam has adopted as subscore as the default term to use in the event that there is no clear term used for this concept in the published source document. Otherwise, the ADQRS supplement should use whatever terminology is present in the source documentation.
  • Use of "concept domain" and "dimension" in a supplement should be consistent with any published scoring instructions or other references. However, if "dimension" appears in the published references, add "(concept domain)" in parentheses to the supplement the first time "dimension" is used so an individual reviewing and familiar with that terminology will understand that the two terms can be used interchangeably (or vice versa, if "concept domain" or "domain" appears in the published references).
Imputations
  • Only include missing value imputations for individual items in ADQRS examples if imputation methodology has been described in the published instrument documentation.  If the published documentation for the scoring algorithm does not include missing value imputation, that should be noted in the supplement.


Tabled Items

  • How much manipulation can be done on AVAL for records coming from SDTM before a new parameter should be used instead of copying QSTEST/QSTESTCD --TEST/--TESTCD into PARAM/PARAMCD?  For example, what if SDTM only collects character values (<=15 minutes, 15-30 minutes, etc.), but numeric codes are required for analysis- can those go into AVAL?  What would be imputed- the character or the numeric value?  Does this change if we are transforming or reversing collected scores for analysis?
  • PARCAT1 holds the instrument name, and PARCATy holds the subscale name(s) associated with the item.  How should items contributing to multiple subscale calculations be modeled?
  • Recommendation has been made to publish these best practices in a more formal format than this Wiki page.  Will revisit in 2023, once more supplements have been published and this is felt to be more stable.
  • What should be included in an ADQRS supplement when an instrument such as SF-36 has been scored by an outside vendor in SDTM?  Do we need to develop a full ADQRS supplement, or only a Readme file?  Is it acceptable/safe to assume the vendor is handling missing values appropriately?