2  Analysis Results Standard Document

https://wiki.cdisc.org/display/ARSP/Analysis+Results+Standard+User+Guide+v1.0

3 Introduction

4 Background and Purpose

Large trials and studies generate many analysis results in the form of tables, figures, and written reports. Historically, a typical workflow for producing analysis results involves the end user generating the display in a static format such as RTF or PDF from the Analysis Data Model (ADaM) dataset (see Figure 1). The Analysis Results Metadata (ARM) for Define-XML (available at https://www.cdisc.org/standards/foundational/define-xml/) is then created retrospectively to provide high-level documentation about metadata relating to the analysis displays and results. However, there is no formal model or structure to describe analysis results and associated metadata, leaving a gap in standardization. The current process is expensive and time-consuming, lacks automation and traceability, and leads to unnecessary variation in analysis results reporting.

Example of Current Work Flow

The goal for the future state of analysis results is that they are machine-readable, easily navigable, and highly reusable. The aim in creating the ARS was to provide a logical model that fully described analysis results and associated metadata to support

  • automated generation of machine-readable results data;

  • improved navigation and reusability of analysis and results data;

  • storage, access, processing, and reproducibility of results data; and

  • traceability to the study protocol, statistical analysis plan (SAP), and to the input ADaM dataset.

The ARS Model has several possible implementations, including leveraging analysis results metadata to aid in automation as well as representing analysis results as data in a dataset structure. The creation of an ARS technical specification could be used to support automation, traceability, and the creation of data displays. An analysis results dataset could support reuse and reproducibility of results data. Figure 2 is an exampleof how the ARS Model could be used in a modernized workflow that shifts the focus from retrospective reporting to prospective planning.

Analysis results play a crucial role in the drug development process, providing essential information for regulatory submission and decision making. However, the current state of analysis results reporting is suboptimal with limited standardization and poor traceability. Currently, analysis results (e.g., tables, figures, listings) are often presented in static, PDF-based reports that are difficult to navigate and vary among sponsors. Moreover, these reports are expensive to generate and offer limited reusability. The CDISC Analysis Results Standard (ARS) Model and accompanying user guide have been developed to support automation, consistency, traceability, and reuse of results data. 

The ARS Model is currently not considered to be a replacement for the ARM for Define-XML standard (available at https://www.cdisc.org/standards/foundational/define-xml/). The ARM for Define-XML meets a regulatory need and has not been modified. However, components have been added to the ARS Model to facilitate the creation of ARM for Define-XML, including the reason and purpose of each analysis and documentation references for both analyses and outputs. The ARM for Define-XML was developed for the purpose of submitting to regulatory agencies to provide traceability for a given analysis result to the specific ADaM data used as input to generating the analysis result. The ARM for Define-XML is often created retrospectively and only for key analyses. In contrast, the ARS Model is intended to leverage analysis results prospectively to enable automation, reusability, and traceability. 

The creation and use of the ARS Model is based on the assumption that input analysis datasets will be “analysis-ready,” as defined in ADaM v2.1 Section 3.1, Fundamental Principles (https://www.cdisc.org/standards/foundational/adam/). This means that ARS metadata components are designed only to define and describe the minimal additional processing needed to produce results from analysis-ready analysis datasets; they are not intended to describe more complex data manipulations (e.g., transformations, transpositions).