Big data, big promise and big issues

In their article, Craig et al.¹ eloquently describe how current Australasian ED administrative data sets do not address quality of care and benchmarking and research opportunities. This is unsurprising because these data sets were not designed for these purposes. I agree that the clinical data that could be available would provide valuable insights into the quality of care, areas for improvement and opportunities for research. I admire the ambition of this initiative, but big data comes with big issues.

Any analysis of big data is only as good as the data entered. For maximum effectiveness and validity, data need to be clean, complete, accurate and formatted consistently. The differences in the design and implementation of health information systems (including electronic medical records [EMRs]) challenge data quality and consistency. For example, not all EMR systems require procedures be specifically captured and there may be differences in coding sets and how they are used.

Current administrative data sets are jurisdiction-based and government-owned. Development of a national/binational data set will need the participation of governments. With the potential political impacts of comparisons between jurisdictions, obtaining government support will not be easy.

Also, a data set of the size generated by this initiative will demand sophisticated stewardship, curation and data governance. Who will ‘own’ the data? Who will decide how it is used and by whom? Governments will want a stake in ownership which will open the possibility of influence in project selection and reporting.

The privacy and consent issues are complex and full discussion is beyond this editorial's remit. The authors' assertion that the use of routinely collected healthcare data for quality improvement and research is generally acceptable to people may be true when people are asked about data in general. It may be less so when asked about the use of their data. Australian evidence suggests that while most patients attending ED expect that data are used in this way, about 20% report that this use without consent will not be acceptable to them; a majority will prefer a consent requirement.² How this can be made workable is challenging, especially if re-identifiability for data linkage is included.

I do not agree that data collection without consent will be acceptable under legislation. Previous approaches, such as use of privacy notices stating that information may be used for quality improvement and research, are unlikely to be acceptable under privacy legislation.³ In my experience, jurisdictions vary in their interpretation of what can be defined as a ‘directly related secondary purpose’ for use of health information, a potentially valid exemption from requiring consent. Also, bundled consent – ‘bundling’ together multiple requests for an individual's consent to a range of collections and uses of personal information, without giving them the opportunity to choose which collections and uses they agree to – is no longer acceptable.³

Big data can also create a divide in the health research community between those who have the access, tools and resources to engage in this work and those who do not.⁴ How do we ensure that the interests of those who can use the data are aligned with the interests of ED patients and are for the public good?

Big data will not replace conventional research, although it is tempting to think so to avoid the time and cost of traditional medical research.⁵ Hypotheses generated from big data will need confirming in the real world of clinical medicine with its complexity and the impact of patient preferences.

In summary, for Emergency Medicine, big data has big promise but also big challenges. How those challenges are approached and resolved will determine the impacts – both scientific and sociopolitical – of its use.