Why the stakes are too high to not share genomic data
4 August 2016
The conversation on health data sharing – the nature and volume of data, with whom, for what intent and purposes – is firmly back on the agenda with the release of the National Data Guardian’s Review of Data Security, Consent and Opt-Outs.
The review sets out proposals to allow patients to opt-out of sharing their personal identifiable data for purposes other than their direct care. Exceptions to the opt-out are also recommended, for example when there is an overriding public interest – say for the control of infectious diseases. Although reference to genomic data is made, the review concedes that a detailed consideration was not possible within its publication timescales.
The use of genomic data in healthcare is a topic with which the PHG Foundation has been prominently engaged across a number of our projects and most recently - in collaboration with the ACGS - Data sharing to support UK clinical genetics and genomics services. Reflecting on these projects, there are at least two pertinent considerations in the context of the NDG review:
- How to distinguish ‘direct’ clinical care from supposed ‘secondary’ applications? Drawing a boundary between the two is inherently problematic when it comes to genomic data a challenge we discuss in our next blog in this series.
- The purpose of genetic / genomic data sharing – a necessity or optional extra? As we see it, data sharing is an essential aspect of routine clinical genetics / genomics service delivery – a view widely - held by genetics professionals.
Both points are closely tied to the way in which genomic data must be analyzed in order to ensure diagnostic quality.
Understanding our genome is far from straightforward
Converting genomic sequence data into a clinically actionable diagnostic report involves a number of analytical steps which rely on a combination of automated computational processes, manual input, and clinical and scientific judgement. Arguably the most challenging step in this process, and the one furthest from automation is ‘variant interpretation’; that is, evaluating the clinical significance of those variants identified in the genomic test as being potentially relevant to the patient’s condition. During this step experienced clinicians and scientists use their judgement, knowledge, and importantly pre-existing evidence in order to decide if a variant is pathogenic i.e. disease -causing or not. Genetics / genomics laboratories have a system to classify variants related to monogenic diseases; ranging from variants that are definitively ‘pathogenic’ (Class-5), to those that are ‘benign’ (Class-1). In between the two ends of this spectrum are variants whose association with disease is unclear. For this reason, Class-3 variants are referred to as ‘variants of uncertain significance’ or VUSs. Often the only way to clarify the role of a VUS in a rare genetic disorder is to pool data from unrelated patients with a similar condition.
Tackling uncertainty and delivering consistent classifications – why does it matter?
When it comes to interpreting variants, the significant challenge is in ensuring that different clinical genetics / genomics services arrive at the same - and accurate - interpretation of any given variant. A number of factors can influence the probability of consistent interpretations, including the use of a common set of guidelines to classify variants. Yet even if working to the same set of rules, differences in access to external evidence, will, and indeed do, result in discrepancies between services in classifying variants.
To be clear - failure to share data and consolidate information on variants, has implications for patient safety, equality of access to testing, and the quality of genetic services . This is because only seeing part of the available picture when trying to interpret a genomic variant results in:
- Incorrect variant interpretations - compromising patient safety and sometimes, that of their family members
- Discrepancies in variant classification between services - leading to different courses of care being recommended for patients who are carrying the same variant
- Inequalities in services - since a patient’s chances of receiving an accurate diagnosis may well depend on the access to pre-existing information their testing services has
- Delays in resolving VUS status - prolonging a patient’s diagnostic odyssey
- Delays in recognising incorrect variant classification - compromising the ongoing safety and care of patients
As genomic testing becomes more integrated into routine care and is used more widely across clinical specialities, the significance of these failures is likely to increase in terms of both scale and volume. The impact on patients is entirely foreseeable and could be profound. For this reason, action is needed now to optimise data sharing.
As we urge in our data sharing report, what is needed is a responsible and proportionate approach to optimise data sharing and one which demonstrates trustworthiness including transparency about the purpose of sharing and the risks, benefits and safeguards involved.
Our view is that the scope of potential ‘opt-outs’ should not extend to genomic data when it is shared within an NHS provider system. In such a context - and provided that the infrastructure and processes are robust enough to guard against breaches of confidentiality - opt-outs are inimical to providing safe and effective care.
This is particularly the case given that unlike other types of data which can be relatively de-identified, genomic data may be difficult to anonymise without compromising its utility. The stakes are simply far too high to not share genomic data.