1     General Considerations

This section describes the implementation of the analysis dataset for QRS instruments as based on the Basic Data Structure (BDS) described in the Analysis Data Model Implementation Guide (ADaMIG). The ADaMIG can be found on the CDISC website at https://www.cdisc.org/standards/foundational/adam. The BDS was chosen as the model for analysis datasets of QRS instruments because it supports a wide variety of analyses, including summaries of total score by timepoint as well as change from baseline by timepoint. This data structure allows the ADaM variables PARAM and PARAMCD to come directly from Study Data Tabulation Model (SDTM) variables RS.RSTEST and RS.RSTESTCD, respectively, for individually captured questions, thus ensuring traceability between SDTM and ADaM data. If other analyses are required that are not supported by the BDS, then consideration would be needed as to the appropriate structure. Analysis requirements always drive the design of the analysis datasets.

This ADaM supplement is based on the Acute Physiology and Chronic Health Evaluation II (APACHE II) Clinical Classification Supplement Version 1.1 to the SDTM Implementation Guide for Human Clinical Trials (SDTMIG). 

Note that source SDTM score values may not be suitable for analysis because the values may need to be imputed or transformed. In addition, recalculation of collected values may be needed per analysis rules. In these cases, the derivation of AVAL may not be a simple copying of RS.RSSTRESN. 

These specific analysis implementation details for the APACHE II are meant to be used in conjunction with ADaMIG. All completed questionnaires, ratings, and scales documentation can be found on the CDISC website at: https://www.cdisc.org/foundational/qrs.

1.1   Representations and Warranties, Limitations of Liability, and Disclaimers

This document is a supplement to the ADaMIG and is covered under Appendix C of that document, which describes representations, warranties, limitations of liability, and disclaimers. Please see Appendix C of the ADaMIG for a complete version of this material.

CDISC does not endorse any QRS instrument and does not specify how to conduct clinical assessments, protocols, or analyses. CDISC disclaims any liability for your use of this material.

Include this line only for supplements that the US FDA will be reviewing: Although the United States Food and Drug Administration (US FDA) has provided input with regard to this supplement, this input does not constitute US FDA endorsement of any particular instrument.  

1.2   Copyright Status

CDISC believes this instrument to be in the public domain, but you should perform your own assessment. CDISC has included the APACHE II in the CDISC library of QRS data standards supplements. There may be many versions of this instrument in the public domain or copyrighted. CDISC has chosen to use the version attached to the QRS supplement to the SDTMIG, and the QRS supplement to the ADaMIG is based on the same version.

Note: CDISC Controlled Terminology is maintained by National Cancer Institute (NCI) Enterprise Vocabulary Services (EVS). The most recent versions of the SDTM and ADaM controlled terminologies should be accessed through the CDISC website (https://www.cdisc.org/standards/terminology).

1.3   Known Issues

The Acute Physiology and Chronic Health Evaluation (APACHE II) is a severity of disease measurement and mortality estimation tool for adult patients applied within 24 hours of admission to an intensive care unit (ICU). The instrument consists of individual items that are summed up to create a total score. This instrument is generally collected once at Screening/Baseline and used as a predictor of mortality rather than an outcome measure. Individual items on this instrument are summed to create a total score but this total score is generally collected as part of the instrument with no instructions provided for handling missing data. In the majority of cases, where APACHE II is only assessed once and the collected total score used as a baseline covariate, the SDTM RS domain may be sufficient to support the analysis requirements.

2     Rules and References

The primary source used in the development of this supplement is Knaus WA, Draper EA, Wagner DP, Zimmerman JE. APACHE II: a severity of disease classification system. Critical Care Medicine 1985; 13(10):818–29. The development of the GDS SHORT FORM is discussed in "Geriatric Depression Scale (GDS): Recent Evidence and Development of a Shorter Version"[2] by Drs. Sheikh and Yesavage. Validation and possible interpretations of the GDS SHORT FORM are discussed in "The Short Form of the Geriatric Depression Scale: A Comparison With the 30-Item Form."[3]

All links are valid and referenced information is current as of the date of this ADaMIG QRS supplement. This supplement does not make recommendations for the handling of missing data or adoption of imputation methods.

3     Scoring Software

There is no known publicly available scoring software as of the date of this ADaMIG QRS supplement.

4     Introduction to the Instrument   

(Example text. This section should be very specific to your QRS instrument)

The APACHE II, is a severity-of-disease classification system that uses a point score based on the initial values of 12 routine physiologic measurements, age, and previous health status. Scores range from 0 to 71 and higher scores correspond to more severe disease and a higher risk of death. The composite score is based on findings represented in other SDTM domains such as Laboratory Test Results (LB) and Vital Signs (VS).

5     Key Data Checks

The computed total score consists of integer values ranging from 0 to 71. The classification system consists of 11 physiology measures based on a numeric rating scale (0-4) where 0 reflects values within the normal range and an increase in the numeric rating scale represents an increase in the abnormality of the physiology measure in either the high or low direction.

6     Example

6.1   Background

Analyses based on ADaM datasets may vary depending on the sponsor, therapeutic area, and the needs of regulatory authorities in their decision-making processes. This fictional example is provided for illustrative purposes only. The ADaM dataset describing the analysis results from an instrument is based on the protocol and statistical analysis plan (SAP) for the individual study.

The example dataset described here is intended to support a variety of analyses, ranging from simple univariate summary statistics on both the actual total scores and change in total score from baseline by visit to other types of by-visit and repeated-measure analyses. For that reason, standard ADaM BDS variables such as AVAL (analysis value) and CHG (change from baseline) have been included in the example. In addition, because scores can be classified into categorizations of depression severity, frequency of subjects reporting each category may also be generated.

In the example presented here, it is assumed that the final GDS SHORT FORM analysis dataset (ADGDSSF) contains all of the original records, plus a record for the computed total score for each patient and assessment timepoint. The date of the assessment—and time of the assessment, if collected—should also be included in the dataset. For traceability, raw item-level responses from the SDTM QS.QSORRES variable, stored separately from the analysis variable AVAL, are included. Derivation type (DTYPE) should be used to clearly indicate those records where missing values or other information has been imputed. If useful for analysis, a summary analysis dataset which includes only the computed total score records may also be created.

This example is based on Yesavage's short form English scoring,[1] which states that if 5 or fewer instrument items are missing, the values of the missing items are imputed as the average of the non-missing responses. If the imputation procedure results in a total score that is not a whole number (e.g., 4.5, 10.2), the total score should be rounded up to the next whole number (i.e., round up to a more severe level of depression). For example, if 3 of the 15 items are missing and the total score based on the 12 non-missing items is 4, then impute the value of each of the 3 missing items as 4/12 = 1/3, for a total score of 4 + (3 * 1/3) = 4 + 1 = 5 (see the Parameter Value Level List - ADGDSSF [AVAL] table). If more than 5 items on the GDS SHORT FORM are missing, no item-level imputation is performed and a total score is not calculated. 

6.2   Example Analysis Dataset, Variable and Value-level Metadata

The ADGDSSF Dataset Metadata table shows an example of how the GDS SHORT FORM may be implemented in ADaM. The dataset metadata is a recommendation for the dataset name, label, structure, and keys. These can be modified for sponsor-specific standards and analyses. 

(Use the table macros to insert these tables. they reside at the end of the document like an appendix)

The ADGDSSF Variable Metadata table contains an example of analysis variable metadata for a subset of possible variables available for the GDS SHORT FORM. Example value-level metadata are found in the Parameter Value Level List - ADGDSSF [AVAL] metadata table. Derivations for AVALCAT1 and AVAL included in the Derivation column were obtained from the scoring instructions  for the GDS SHORT FORM[1] and are described in the Introduction to the Instrument section of this supplement. The ADGDSSF Variable Metadata table does not include all of the possible variables that may be present in the Subject-Level Analysis Dataset (ADSL) and copied to the BDS structure analysis dataset, with the exception of the required STUDYID and USUBJID. For a complete list of available ADSL variables, consult the ADaMIG. Some ADSL variables (e.g., INVNAM) were included in this example for stratification, and are not required. Additional covariates or other variables may be added as needed for analysis. There is no specific ordering of variables recommended in this supplement. The SDTM variables were placed next to the corresponding derived ADaM variables for illustration purposes only. 

Update the below table according to the current QRS Short_Name instrument.  

The following SDTM variables can be copied into the corresponding equivalent ADaM variables. Note that the SDTM variable values can be converted to mixed case as needed for analysis purposes.  The Action column shown in the table below is optional, and can be removed if all values for these variables are simply copied over from SDTM.  For supplements where the SDTM values are modified, please provide a description of what was done.  For example, in the CHART-SF supplement, PARAM/PARAMCD are modified by replacing the version number with 'A', so that action is described here as "Harmonizing the two methods of administration (Paper and Interview) for individual items.  Fragment representing version number is changed to an 'A'."

All controlled terminology for instruments in both SDTM and ADaM is associated with a specific version of the instrument. When integrating responses from an instrument across studies, it is important to know which version of the instrument was used for each study. When multiple versions of the same instrument exist in an integration, a pooled analysis of responses across different versions can only be performed if determined to be clinically and statistically acceptable. In that case, sponsor-defined terminology, similar to CDISC Controlled Terminology but without the reference to a specific version, should be used for PARAMCD, PARAM and PARCATy.

6.3   Example Analysis Dataset 

The GDS SHORT FORM Summary Analysis Dataset table shows the terminology and variables for implementing the example analysis dataset for this instrument in the BDS. This specific example is taken from a multicenter, randomized, double-blind, active-controlled parallel design study conducted in elderly patients with major depressive disorder. The purpose of the analysis is to compare the effect of Drug A to Drug B on depression in elderly patients using the GDS SHORT FORM to construct an endpoint. All analyses are conducted on an intent-to-treat basis, and the planned duration of treatment is 5 months. The baseline measurement is defined as the last non-missing value collected at or prior to the start of treatment; change from the last non-missing post-baseline value to baseline is used to construct an endpoint. The SAP specifies that planned collected visits, including baseline measurements, are to be used in analyses and displays of results, whereas unscheduled visits appear only in listings. The total score records for 1 subject were collected at 3 visits, plus a derived last observed value (LOV) endpoint record for the GDS SHORT FORM. Visit 1 is remapped to AVISIT = "BASELINE". There are 2 visits at AVISIT = "VISIT 2". However, because 1 visit was unscheduled, it is not used in the analysis and the ANL01FL value is null. Visits 5 and 6 are missing because the subject discontinued early due to an adverse event unrelated to treatment.

Update the below dataset and title according to the current QRS Short_Name instrument.  

GDS SHORT FORM Complete Analysis Dataset

The GDS SHORT FORM Complete Analysis Dataset table shows all of the individual question responses for this example, as well as the computed total score. This example is considered to be the final analysis dataset for submission. Computed total scores can appear after the individual item responses, with appropriate labels for identification.

Update the below dataset and title according to the current QRS Short_Name instrument.  

GDS SHORT FORM Summary Analysis Dataset

Due to the number of items on the instrument and the fact that only the total score records are required for analysis, the separate summary dataset shown in the following example, which contains only the total score records, may also be created.  QSSEQ was not included in this example because it would be null for all computed total score records. QSORRES is not shown because only derived total score records are displayed.

7   References

1. Yesavage J. Geriatric depression scale. Stanford/VA/NIA Aging Clinical Research Center (ACRC) Website. Accessed May 6, 2020. http://web.stanford.edu/~yesavage/GDS.html

2. Sheikh JI, Yesavage JA. Geriatric depression scale (GDS): recent evidence and development of a shorter version. Clin Gerontol. 1986;5(1-2):165-173. doi:10.1300/J018v05n01_09

3. Burke WJ, Roccaforte WH, Wengel SP. The short form of the geriatric depression scale: a comparison with the 30-item form. J Geriatr Psychiatry Neurol. 1991;4(3):173-178. doi:10.1177/089198879100400310

End of Document