Data Management Beyond Clinical Trials
By Steve Cutler, COO, ICON plc [NASDAQ: ICLR]
Electronic Health Records (EHRs) exist to aid the delivery of care and administration of health systems. EHRs were not designed to collect data for use in clinical trial research. However, we are seeing an ever growing appetite for sponsors and institutions wishing to accelerate the uses of EHR data. In the future, it is envisaged that the number of classic clinical trials will contract and the use of EHR or Real World Evidence (RWE) data will expand. EHRs contain longitudinal data representing the health of patients over an extended period of their lifetime, which can provide valuable insights into how medicinal products are working in the real world. So it makes perfect sense to tap into the largest databases to gather this data.
However, how can sponsors and regulators be sure that the data collected using EHRs, meets the high standards currently used in clinical research? The FDA has responded to this question with their draft guidance, Use of Electronic Health Record in Clinical Investigations, May 2016. However, FDA’s acceptance of data from clinical investigations for decision-making purposes depends on FDA’s ability to verify the data quality and integrity during its on-site inspections and audits. Sponsors are responsible for assessing the validity, reliability, and integrity of any data used to support a marketing application for a medical product.
Even if the data is not being used to support an application for a medicinal product, when collecting data for which you wish to publish your results, it is important that the validity, reliability, and integrity of the data are not questionable.
To ensure that the validity, reliability and integrity of data generated in clinical trials is of high quality, we use Clinical Data Management (CDM) practices. The CDM process begins with the end in mind. Clinical trials are designed to answer specific questions and the CDM process is designed to deliver valid and reliable data for statistical analysis. The acronym ALCOA is used in clinical trials and other regulated industries to ensure data integrity. ALCOA relates to data, whether paper or electronic, and is defined as attributable (who generated/ changed the data), legible (readable), contemporaneous (time-stamped), original (source data) and accurate (free from errors). The FDA considers ALCOA a fundamental part of the data collection life cycle when using data from EHRs, but herein lie the challenges. Thinking about this, unless the EHR is certified by the Office of National Co-ordinator (ONC) program, it seems unlikely that EHR data will meet ALCOA standards in their present format.
The extent of the problem collecting data from disparate systems is obvious in the following example from Sweden; there are eight quality registries using thirty five individual technical platforms and there is no interoperability or standardization between systems, which prevents valuable analysis via data pooling.
Clinical trials are designed to answer specific questions and the CDM process is designed to deliver valid and reliable data for statistical analysis
By contrast, the Netherlands has a single platform strategy so there is conformity in the way data is ingested, stored and standardized. In addition there is a three year process in place to allow each provider time to ensure that their data meets the quality standards before it is considered for reporting.
Ideally, using data from ONC certified technology is preferable, however, this is a U.S. initiative and it is more likely that data will be solely mined from EHR’s which are not certified by ONC and will require understanding of the source data and defined clinical data management processes to ensure the validity, reliability and integrity of the data. The eClinical Forum’s eSource Readiness Assessment (eSRA) is a very useful tool, which provides the minimum requirements for self-assessment for systems using data that may be included in a clinical trial. Going back to the fundamentals of ALCOA, sponsors will have to ensure that system access is limited to authorized users, those authors are identifiable, there is an audit trail and records are available for inspection and the privacy and security of patient data is safeguarded.
The future of using EHR and other real world data lies in the process, semantic (ability to understand the data), and technical interoperability between systems supporting clinical trials following ALCOA principles to ensure the integrity, validity, and reliability of the data that has been mapped and extracted or exchanged. Integration activities must also comply with legal and regulatory requirements to protect patient’s privacy and security of data. This in itself is a serious undertaking for any organization.
Our experience to date collecting data from EHRs/patient reported systems indicates you will need to develop a robust version-controlled data dictionary, which clearly defines each data item you wish to collect, supporting definitions, response options, the timing of when data was collected and the reporting source. We have also noted that using coding standards such as ICD10 for missing data may improve data quality. You will need to have a clear understanding of how data is collected in the source systems from which data is being accessed or accepted from. Nuances in clinical practices introduce variability, so you must allow for this in your data transfer specification/ mapping tool. This means spending time speaking to the data managers at each provider to ensure that the data extracted meets the data dictionary definition. A gap analysis and pilot testing helps to identify the degree of missing data and indicates where there are issues with validity. Structural checks should be run over data received from each provider to ensure that the data extracted meets the data specification. This does not however tell you that they mapped the data item correctly and even with the best instructions, people will have their own interpretations. Therefore a process or data audits will need to be implemented.
Classic data cleaning will no longer be applicable as the data you receive is source data. Unless the data provider is able to correct data within the EHR system after receiving feedback about the issues, the expectation is that missing or odd values will be carried through the process. It will then be a clinical and statistical decision as to whether or not records are included in the analysis.
Notably the integrity of some data points can only be checked with visual review; for example, patient-reported outcome instruments need to have been migrated faithfully using the ePRO consortiums best practice guidelines. This implies the need to access patient-facing materials and verify that the instrument has not been changed.
Another challenge is the variability of languages. Case Report Form data is generally captured in English language. The EHR systems usually capture data in local language and therefore this needs to be taken into consideration.
This is a new era for clinical data management and new processes and methods will be necessary to manage this new data source to ensure the validity, reliability and integrity of the data, as there is little point in collecting data if it cannot be trusted to meet the quality standards reliably.