How to find health data

In the second part of our PCN data series Sophie Hodges, Lead Client Service Manager at the Health Economics Unit, Midlands and Lancashire Commissioning Support Unit gives an insight into available sources of data, where to find them and, crucially, what to watch out for

The Covid-19 pandemic drove a rapid increase in appreciation across health and care of the importance and impact of making data-led decisions. As a result, more and more people are looking for data and insight to help them make better decisions, drive more effective partnerships across different parts of the system and improve outcomes for patients. But what is out there and how can you access it?

What data is available to me?
If you’re relatively new to data analysis, it’s understandable that you may not know where to begin. There is a wealth of data available, but it must be said that there is also a lot of diversity in terms of what different sources have to offer. For example, some datasets may go down to patient-level, while others are large, aggregated datasets – both serving a different purpose. We also have access to some tools which sit a level or so above an aggregated dataset.

For this reason, it’s important to understand the question you are trying to answer and where you are going to begin your analysis. This can be a challenging and lengthy process where you investigate the data and find it’s not answering your question, or that the question itself changes. Therefore, diving headfirst into all the data on offer isn’t always helpful. There is so much information out there, it’s always wise to start by focusing on one area. That isn’t to say that, once you are further down the line, it’s never possible to be broad.

Also, from an ethical point of view, we have a responsibility to respect the privacy of patients and ensure we limit our access to only those situations where it is justified. From a practical point of view, we know that adding data which doesn’t help us answer our question tends to skew analytical models and can lead to incorrect reasoning – a phenomenon known as the “curse of dimensionality”.

Some key data sources and platforms you may wish to consider include the Office for Health Improvement and Disparities’ (OHID) fingertips tool, the Office for National Statistics (ONS) and, of course, Quality and Outcomes Framework (QOF) results. These are all big, aggregated datasets, but can be helpful for understanding the whole population and establishing overall needs. Also, you can be confident in how the data has been interpreted in big tools, which is helpful to avoid looking into datasets where you may not understand definitions or local context.

QOF in particular can help you to think more broadly about how your practice or PCN approaches the management of long-term health conditions, prevention and detection and wider public health issues. For other projects you may need to access a more granular, patient-level dataset, perhaps from your own practice or PCN, or using national sources such as hospital episode statistics (HES) or Secondary Uses Service (SUS). However these are not publicly available, and you may need to look into getting access granted.

What do you need to be aware of when looking for a data to answer your question?
Data is extremely powerful. This means it can help us do a lot of good but, used incorrectly, it can also cause more issues than it solves. For example, health data is known to contain poor coding of ethnicity, which can lead to biases and risks skewing your analysis away from those most in need. This is one of several things that must be considered when making data-led decisions. Other factors to consider include:

Understanding the original purpose of the dataset
No dataset is created without a purpose and that purpose will impact the shape of the data it contains, such as what was collected and how, the definitions used, and the level of accuracy. It can be helpful to understand if there was any incentive behind the data collection, as incentivised data, such as HES or SUS, tends to be more complete. For example, some parts of the cancer pathway are measured against targets, while others are not. Therefore, you may find that some data items related to this are more accurately and consistently captured than others.

Another common issue across health and care is that there isn’t always a standard approach to how we record information. For example, it’s more than likely that each practice in your network records certain data differently due to a lack of consistent terms in systems. It can vary too between GPs within a practice, and even individual GPs from one day to the next. In fact, I once worked with a PCN which had an incredible 78 different codes for appointment types (face to face, telephone, video call) thanks to these variations between practices and clinicians. This lack of standardisation can make it difficult to do solid data analysis without expert support and local input.

Interpretation of data definitions
The way in which measures are defined is a constant source of disagreement and confusion throughout health and care. This poses a significant risk, as it’s simply not possible to do good analysis without a clear grasp of the definitions of the data items you’re looking at and even what they were during previous collections.

For example, on the surface it should be easy to agree on a standard definition of an ICU ‘bed’ being in use. However, it’s not as simple as it seems. There are different bed types, equipment can be moved around and shared between patients and who decides when the clock starts and stops? There is a richness behind every measure. Some are easier to understand than others, but we need to get under the skin of the measures to be able to properly use the insight they provide.

Where can you get support?
In an ideal world, data analysis is best attempted by qualified analysts who truly understand health care data and its peculiarities, in collaboration with those who appreciate the origin of the data and what it means in real life. In our previous article, we talked about the varied roles that make up the healthcare analytics profession. But if you don’t have an analyst embedded within your PCN, where can you go to start such a partnership?

Looking ahead, all integrated care systems will be required to have an intelligence function but, clearly, it is early days in many areas, and most are unlikely to be well-established at this point in time. For more information on how intelligence functions should be formed, you can take a look at these design principles devised by the Midlands Decision Support Network and Strategy Unit. Regardless, your first stop is still likely to be your local ICS, or your local authority’s public health team, who will help to connect you with their intelligence team in whatever stage of maturity it has reached.

They will be able to direct you to the most suitable person or team to support your analysis. As you build a relationship, they should be able to help to harness your local expertise and combine this with their knowledge of the techniques and approaches available to make the best of the data on offer. This partnership approach to driving data-led decisions helps to join up the data journey all the way from collection through to a decision and can foster a strong and productive relationship between clinicians and analysts – leading to improved services and better outcomes for patients. Clearly it represents a significant time and effort investment and won’t always be an easy process as you build mutual understanding across the life of each new project.

Organisations like the Health Economics Unit are committed to raising the profile of health care analytics, boosting the skills of analytical professionals, and encouraging this vital partnership approach to decision making. Our team proudly offers a deep level of expertise across the entire intelligence profession, including data engineering, analytics, data science, health economics and population health management. We have worked with many organisations to apply our expertise in periods of transition and fill temporary gaps in resource or knowledge, and often operate alongside local teams to deliver specific projects, simultaneously upskilling their analysts for the future. Every day we see the overwhelmingly positive impact of strong clinician-analyst relationships in improving the way the health service views and uses data and are keen to support this in any way we can.

By Sophie Hodges, Lead Client Service Manager at the Health Economics Unit, Midlands and Lancashire Commissioning Support Unit