Using data and analytics to underpin better health and care

David Maguire (The King’s Fund): I'd recommend checking out the slides presented by Ivana Bartoletti (Head of Privacy and Data, Gemserv, and Co-Founder, Women Leading in AI Network) at our 2019 emerging technologies conference for a fuller exploration of the ethical issues around AI. According to Ivana, the three main things to look out for are narrow or poorly trained datasets, a lack of diversity in the developers of AI systems and discrimination in how data is collected or who it's collected from. To counter this, clear and powerful governance frameworks need to be established alongside acknowledgement of potential issues while developing datasets (understanding who is at risk locally of exclusion, for example).

David Maguire (The King’s Fund): There is a longstanding issue in the NHS around duplication of effort in data collection and as Ming Tang touched on in the session, there are national efforts under way to try and streamline some of this. There are often duplicate systems operating on a historical basis that would require significant investment of resource to replace. Hopefully as data entry and submission become more standardised in the coming years, we'll see less of this duplication of effort.

David Maguire (The King’s Fund): This question gets to a key point in communicating the story of analysis well. Often it's the narrative around statistics that can make data feel powerful, rather than the magnitude of the numbers presented. At the Fund, we often try and include the personal impact of an issue alongside the quantitative analysis we produce to try and add more power to the arguments we present. I'd recommend our recent work on inequalities and inclusion among staff in the NHS as an example of how hard data can be blended together with lived experiences to form a narrative and call to action.

Rob O’Neill (University Hospitals of Morecambe Bay NHS Foundation Trust): I suggest approaching this from a system level – have a clear vision and strategy for the system (eg, integrated care partnership, integrated care system or other group of organisations that share a combined mission) and a clear analytical/digital delivery plan where data sharing is a prerequisite to success. With this in place I'd undertake robust stakeholder engagement to ensure buy-in to the vision/deliverables (sell the benefits to patient care, outcomes, service improvement and so on) and finally use a digital framework to make the administration process of setting up, and signing off, data sharing agreements as easy and light-touch as possible – while being secure, transparent and fully auditable.

Nick Ward (Qlik): There are a range of considerations to be covered in order to achieve this. Clearly one fundamental need is for data-sharing agreements between organisations, which can be tough to put in place due to the fear of sharing. What is needed is a champion, both in terms of lead organisation and a specific individual, who can help the various parties understand the benefits and value of sharing data – make it a positive conversation about potential, rather than a negative one focused on risk. When that is established a platform needs to be put in place that can host the data coming from the various parties, whether this is a shared environment or led by one particular organisation. The next question is of access. With many different organisations needing access, the platform is unlikely to be held in a private domain so Cloud ought to be considered. Authentication will also need to be evaluated and single sign-on may not be possible for all. Once those fundamentals are covered then the correct technology solution can be put in place that allows governance to be applied while at the same time providing freedom for the audience to explore the data and find the insights that will drive improvement.

Rob O’Neill (University Hospitals of Morecambe Bay NHS Foundation Trust): They can. I think there's a two-part response to this. First, we have a responsibility as data professionals to ensure that any analysis we present is fit for purpose. Investing in staff training and education is very important and helps to create learning cultures within organisations and systems. Professional standards are another aspect of this. Second, organisational culture is key, in the context of driving a shift to organisations that are truly data driven. I believe that a key part of success in this area will come down to how confident stakeholder groups are in being able to understand data, interpret analysis, articulate requirements and so on. It feels to me a bit like a contract – as data professionals we need to commit to providing high-quality fit-for-purpose analysis and clinicians and operational staff should commit to being confident in interpreting what we provide and taking action as a result. The concept of turning data into insight, and insight into action is very important.

Nick Ward (Qlik): Graphs and charts serve one purpose, to present the data and metrics they are configured with. Analytics can only be misleading if incorrectly designed or interpreted. Therefore, it is imperative that subject matter experts (SMEs) are involved in the design process, or better still, in charge of and directly responsible for it. Complex epidemiological situations need to be presented in an appropriate way, with the necessary level of context and relevance detailed within a dashboard view. Any single chart can provide a range of different perspectives, a collection of analytics presented in combination will provide the full story and clarity on the situation being explored. It may be the case that instructions are provided as a pre-cursor to viewing the dashboard, such that the non-SME user understands how to interpret the dashboard correctly. Natural language explanations can be configured to present alongside the charts to help the user understand what they are seeing. And a good story-telling capability that explains the findings within the charts, graphs and dashboards that can be easily created by an SME will help the wider audience understand the details.

Rob O’Neill (University Hospitals of Morecambe Bay NHS Foundation Trust): We need to focus on predictive analytics, and the evidence tells us that this should be embedded at all levels of a learning health organisation. I think we're at a crossroads for health care, we're seeing a lot more organisations invest in dedicated data science resource, which is fantastic. However, there arguably isn't either the budget or the resource available to embed this across all organisations. What we can all do, however, is both seek to up-skill and transform analytical teams into teams focused on predictive analysis and to partner with the leading technology firms that are delivering services within this space. Automated machine learning is growing exponentially as a technology, I've implemented software (DataRobot) at University Hospitals of Morecambe Bay NHS Foundation Trust and we've seen an exponential increase in our team’s ability to develop and deploy predictive models – the acceleration is significant and we're importantly augmenting our staff’s ability to model, as opposed to replacing it. I predict that what happened with the business intelligence (BI) explosion 10-years or so ago is happening with artificial intelligence (AI) and machine learning at the moment, and within health care we'll see seismic shifts towards this technology over the next few years.

Nick Ward (Qlik):This is a question of trust. Many people trust their gut instincts over what the data is telling them, as they don’t trust the data. Potentially data quality issues will contribute to this but in the majority of cases it’s a lack of education and confidence around how to interpret the data that creates the mistrust. Data Literacy enablement will be a big factor in the drive towards data-driven decision-making, and this generally needs to start from the top – if the leadership and senior management aren’t citing the data and showing by example then the rest of the organisation will fail to do so. Another part of the approach is to provide a range of interfaces with which to access the data. Not everyone is comfortable looking at dashboards and reports so allowing users to enter the data through a variety of interfaces will aid adoption. These could include embedding the analytics into the operational systems such that the two exist side by side and become part of the user’s daily routine. Other approaches could be chatbots, alerting and mobile access. The more options that are provided, the more likely it is that the users will find one they feel comfortable using.

Nick Ward (Qlik): In a word, transparency. We, of course, have the ability to use the data that is provided through opt-in to act as a sample of the wider view. However, the more data we have the more accurate we can be. Thus, the question becomes how do we increase the quantity of opt-ins that are granted? This comes through transparency. It is far more likely that a person will choose to share their data if they understand how it is going to be used and see benefits in that, but infinitely more so if they see how it can provide value to themselves. Therefore, we must explore ways in which we can use data to provide value to people who share their data. What if we could expose the contrast in understanding of a patient’s condition when attending a new service, between opt-in and opt-out? Or demonstrate how a patient doesn’t need to relay their details multiple times through various interactions if they’ve opted in? I would suggest this would be a better approach than simply making the best of the data we receive permission to use.