CLINICAL RESEARCH

Beyond CDISC: Taking the next step in clinical data standardisation

Kit Howard, Kestrel Consultants Inc.

The battle to prove the value of standards in clinical trials has been won. Between the FDA’s encouragement of CDISC’s SDTM1 submission standards and companies’ recognition of the improved metrics that standards provide, regulated industry has finally accepted the need for change. They join many others, including CDISC’s CDASH2 project, the National Cancer Institute’s caBIG, HL-7’s messaging standards and more. These projects all primarily focus on harmonising the technical aspects of the data, such as variable names and dataset structures. Although there is overlap between them that should be addressed, they are laying critical groundwork for data sharing in both regulated and academic research settings.

There is another less obvious standardisation requirement, and that is the need to establish consistency in the processes and rules used to handle the data, ie the intent of the data. High quality data are “fit for use”, meaning that everyone from the clinicians to the data managers to the statisticians have a similar understanding of how the data should be used. One can never know prospectively all the uses to which data may be put, but if the assumptions made during definition, collection, storage, and cleaning of the data are consistent and documented, the data are more likely to be used appropriately. Few organisations have developed methodologies for defining and documenting these assumptions. Most have documents that describe how data will be handled within functional areas, but there is little crossfunctional harmonisation or understanding of the reasons for the rules.

An example drawn from the author’s experience involves the 17-item Hamilton Depression Scale. When individual items on the scale are missing, some statisticians use the average of the remaining responses to impute the missing value and calculate a total score, while others believe the scale is invalid if any responses are missing and a total score cannot be calculated. There is nothing in the structure of the collected data that indicates what rules were followed. Additionally, Data Management knows that the HAMD-17 is completed during the subject visit, and therefore missing items cannot be retrieved and no queries will be generated. Unless they understand its impact, they have no reason to look for patterns of missing data that could prompt site retraining. In this case, the project clinical scientists and medical writers were not aware that data were being imputed, and thus assumed that the values were complete and valid, which may or may not have been the case.

The author has seen numerous cases where different therapy areas in the same company, and even teams working on different indications within one compound, have used different approaches. In most organisations, functional areas operate in fairly isolated silos that are only partially breached by cross-functional study teams. If the data handling rules are not uniformly applied throughout the clinical trials by all functional areas, the data may not accurately reflect what really happened and the trials may not ultimately be comparable.

One solution is to develop an additional layer of standards for each data domain (eg adverse events, demography, HAMD-17) that document generic data rules. These standards should be developed using a cross-functional team drawn from across the organisation, and the results should be stored in a centrally-accessible location. They should cover all safety and efficacy data used in every trial, and should include definitions of the following information:

• Purpose: the purpose of collecting the data domain, to provide an idea of how the data are used.
• Protocol: the information that the protocol should contain pertaining to the data domain. It can include specific wording for a protocol or a description of the points that should be in the protocol.
• Clinical study report: the treatment of the data in the clinical study report and where they are expected to appear.
• Statistical considerations: the considerations when including the data domain in a statistical analysis plan, including any assumptions, eg for handling missing data.
• Data displays: the presentation of the data in the clinical study report, including data listings, summary tabulations and graphical displays. Sample layouts of the most common data presentations should be provided.
• Data capture and reporting datasets: the structure and variables in the data capture and reporting datasets and the relationship between them. Common derivations should be included.
• Data capture: sample case report form/electronic CRF layouts, CRF completion guidelines and annotated CRFs.
• Data quality: guidelines for defining and assuring quality, including data listings for manual review, monitoring guidelines, computerised edit checks and medical monitoring guidelines.

These should be accompanied by the references used to establish the rules, especially the applicable regulatory guidelines. In addition, key decisions made regarding the structure or handling of the data should be documented, along with the rationale for the decision. There must also be a commitment to managing the standards over time and to providing the appropriate governance to hold all functional areas accountable for the implementation and adherence to them.

This approach has many advantages. First and foremost, it ensures that each functional area treats the data consistently, and that analysis and conclusions are more likely to be valid. Each functional area achieves a better understanding of the requirements of the others, and can streamline their activities accordingly. If implemented correctly, these standards do not limit scientific exploration, and are flexible enough to accommodate necessary variations while being structured enough to prevent accidental or capricious changes.

These points speak primarily to data quality, but obviously one of the greatest benefits is increased reusability and the resulting savings in time and resources. Indeed, a recent report suggested that implementing technical standards throughout the clinical trials process can produce savings of up to 60%3. Streamlining and harmonising the associated processes can only improve this result.

The process of developing and implementing standards is neither simple nor quick. Many organisations are just beginning CDISC SDTM implementation, and the thought of embarking on more standardisation seems overwhelming. Without process standards, however, we run the risk of creating data repositories that are consistently structured but analytically flawed, and if that occurs we will have lost the war.

KIT HOWARD, MS, CCDM

Kit Howard is the owner and Principal of Kestrel Consultants and has over 20 years of experience in the industry.

She specialises in cross-functional clinical data and process standards for industry and academia, and also has experience in clinical data management, project management, programming and outsourcing.

Kestrel Consultants
410 Rose Drive
Ann Arbor MI 48103 USA
Tel: 001 734 576 3031
Fax: 001 734 761 6586
Kit@KestrelConsultants.com
www.KestrelConsultants.com
www.kestrelconsultants.blogspot.com

Notes
1. CDISC is a consortium that has developed the Study Data Tabulation Model (SDTM) standards that define the structure of data in electronic regulatory submissions.
2. CDISC is also sponsoring a project to define data collection parameters, called Clinical Data Acquisition Standards Harmonisation (CDASH)
3 Gartner / CDISC standards evaluation project “CDISC Standards Enable Reuse without Rework” available from gartner.com, or from Carol Rozwell, VP, carol.rozwell@gartner.com, or from CDISC at www.cdisc.org

back