A regular refrain from the current debate on integrated systems is the need for shared information, better analytics and intelligence to support delivering integrated care.
In this guide, we draw out resources and examples that are relevant to analysts and evaluators who need to understand the impact of integrated services. It draws on methods we have developed over a number of years, including pioneering analytical processes using linked data, new measurement techniques and evaluating innovations that were intended to improve integration.
We have pulled out the most relevant resources from our research reports into one place – making it easier to know more about each and to find out what has been tried and known to be useful, to avoid reinventing the wheel.
The resources have been grouped under four headings, but these overlap, and can be linked together for greater impact.
The terms integrated care and integration have taken on different meanings over time, as policy has shifted towards “integrated care systems”. The NHS England Planning Guidance for 2018/19 refers to integrated systems as “health and care organisations that voluntarily come together to provide integrated services for a defined population”.
What indicators can I use to measure integrated care?
A first step in understanding integrated care is to identify the aspects of the service that are important, and translate these into indicators that can be used to assess change over time or variation between populations or systems. For example, indicators can be used to track change over time, to compare different parts of a health system, or to benchmark with similar systems elsewhere in England.
With collaborators on the evaluation of integrated care pioneers, we have identified potential indicators for integrated care. After testing, a subset of indicators has proved feasible to use for monitoring at pioneer level on an ongoing basis (see the next table).
While this is a useful starting point, there are notable gaps in what can be measured using routine data. They include: staff and patient experience of integration, community health service measures, information on self-funded social care, and services delivered outside of the public sector.
The next table runs through different indicators of integrated care – click on each source for more information.
These integrated care indicators shown were designed to bring together information across a whole system.
A range of other measures will also be relevant for particular care settings (see the next table). For example, we have tested the feasibility of a range of measures to assess aspects of integrated care at GP practice level, and identified potential measures and data sources.
Measures of disparity between mental and physical health could also be used to understand integration between mental and physical health services.
This next table shows our relevant Nuffield resources on measures of integrated care – click on each to find out more.
How can I make the most of routine data sources? And what can we learn from linking data?
At the Nuffield Trust we have worked extensively on methods to exploit routine data sources, either individually, or linking data from different services (see the next table).
Using linked data has become more complex over time, due to changes in information governance requirements, but learning from earlier analysis of linked data can inform future work that may be possible as new sources of integrated data become available.
Work in progress on exploiting routine data includes identifying conditions that could indicate how harm to patients has occurred, tools for undertaking a comprehensive geriatric assessment in hospital, and using routine data to research prisoners’ health.
This next table shows our relevant Nuffield resources on using routine data sources – click on each to find out more.
What are the strengths and limitations of predictive models? How can they be used for evaluation?
Predictive models have been identified as a key tool for developing more integrated models of care. Predictive risk models can be used to identify people at high risk of undesirable events, by analysing historic data to identify correlations between the event of interest and explanatory variables. The most comprehensive models tend to use multiple data sources, including primary care data.
Risk models enable analysts to take account of differences in populations, and at the Nuffield Trust we have used predictive models extensively to evaluate service innovations and new care models (see the next table).
While risk models can successfully identify people at high risk of particular events, such as emergency admissions, or future high service use and costs, there is currently limited evidence that identifying high-risk patients has ultimately led to service improvements and improved outcomes. It has proved challenging to identify and implement effective interventions to improve care for those groups of high-risk individuals.
This next table shows our relevant Nuffield resources using predictive models – click on each to find out more.
Intelligent monitoring: how can we identify when a change has occurred?
Monitoring change over time or against an expected trajectory then raises the question of how to identify when a change has occurred?
Statistical process control methods support proactive monitoring, and provide alerts to identify when a change in trend may have happened. They also mean we can see whether changes in trends have occurred retrospectively.
Tracking differences between intervention and control groups over time is a key part of evaluating new services.
This next table shows our relevant Nuffield resources on intelligent monitoring – click on each resource to find out more. See the previous table too.
The challenges and opportunities
Using data collected for administrative or clinical purposes for measurement brings a number of well documented challenges.
We would highlight three areas as particularly likely to impact your work on integrated care.
As well as the usual challenges of unpicking changes in recording and coding practices from real differences, there are particular problems with how new patient pathways and care models are captured using routine data.
They include innovations such as specialist ‘advice and guidance’ services as a substitute for outpatient referrals, one-stop clinics combining diagnostics and assessment, and new urgent care pathways. The standard units of acute activity do not capture these services in a consistent way across the country.
Translating rich clinical data into measures
As more parts of the patient record become electronic, there are opportunities to use this digital information for analysis. But experience from analysing primary care data, for which the patient record has been electronic for a number of years, highlights a number of issues.
First it is difficult to identify events and standard units of activity, such as to distinguish between information captured during a face-to-face appointment, a telephone call with the patient, a clinical-clinical call about the patient, or information added by administrative staff.
Second, although clinical terms are used to capture information, there is often inconsistency in how this is done – between and within practices.
Third, a large part of the valuable information is held within free-text parts of the notes, which presents both analytical and governance issues.
Finally, no discussion of challenges would be complete without referring to information governance. Using patient data is vital to improve health and care for everyone, and good information governance helps to enable it, but there are often challenges faced with data access. Measuring integrated care often brings different patient data together to get a proper view of their care – requiring data to be linked at the individual level.
This raises numerous data privacy issues. While there are technical solutions to address linkage and enable analysis using de-identified data, there is also a greater need for engagement with patients and the public to address growing concerns about how ‘their’ data is handled and used, and to demonstrate the benefit that the use of patient data in research has on improving health services.
Encouragingly, progress is being made on sources of information on integrated care. Patient level data sets are now flowing for secondary mental health services, and (most recently) community services, which will address a significant gap.
It is also now recognised nationally that the benefits from integrating clinical records go well beyond improvements in direct patient care, and provide the basis for driving transformation of local health systems, if the integrated information can be made available for analysis.
The move to digital health records in acute care is generating a huge amount of data, with potential for analysis based on much greater detail on both patients’ clinical condition and on the resources deployed during treatment. Added to this, patient-generated information is also being captured as part of some shared health records – providing an opportunity to understand the health and care delivered by patients and carers themselves.
There is also cause for optimism about analytical methods. In the medium term, machine learning and artificial intelligence methods are likely to have significant impact.
More immediately, there are significant opportunities – with the benefit of data linking patients care across multiple settings – to shift existing analysis methods to use a patient-and-population, rather than a service, perspective.
Examples include analysing variations in care along pathways, or taking account of case-mix information, and developing indicators of potential harm, such as a routine hospital admission followed by a sharp increase in a patient’s use of health care services. There is also huge learning from applying evaluation methods developed in research contexts to enable more rapid evaluation of innovations.
Two additional areas need to be mentioned. The first is the need for detailed work on how existing and emerging data sources could be used to understand resource use, to enable more robust analysis of the financial implications of service changes, and improve understanding of costs of delivering care for individual patients and cohorts within populations. This will be essential to develop and validate alternative funding models for integrated care.
The second is the need for more measures and information on patients’ social situation (such as levels of support and isolation) and general wellbeing, which impacts on their ability to manage their own health and care, and the type of care they need from health and social care services.
Integrating health information with other public sector data could support this, such as by using housing information to understand the links between service use, household composition and social support.
The combination of policy drivers to understand population health, along with digital transformation, provide exciting opportunities to develop analytics to support integration of health and care. We at the Nuffield Trust will be continuing to develop methodologies to support our own work, and sharing our learning for use by the wider health and care system.