Luke Budworth | Senior Research Data Analyst
Patient Safety Research Collaborative: Rethinking Safety Intelligence for Improvement theme
Hi, I’m Luke Budworth and I’m a data analyst for the Yorkshire & Humber Patient Safety Research Collaborative. I work in the Safety Intelligence theme and contribute data science expertise to all projects across our collaborative, including our cross-cutting theme on Safety Inequity.
Our main goal in safety intelligence is to create and test new safety intelligence solutions that improve patient safety across health and care systems. But what exactly do we mean when we refer to ‘safety intelligence’? And what might potential solutions look like?
As the first of our theme’s blogs, I aim to introduce readers to the background to our work, before outlining some of our ongoing projects – primarily in our integrated data sub-theme and related to our close collaboration with Connected Bradford (cBradford) – a large local data linkage project.
Broadly, working with Connected Bradford data, we aim to understand the potential for whole systems data for patient safety intelligence, explore inequalities in patient safety based upon analysis of linked data and patient safety outcomes, and focus on developing tools to manage the risk of deterioration for patients in primary care. However, this is just a small subset of our work and I encourage viewers to read more here.
Patient safety & safety intelligence
In 2019 NHS England published their Patient Safety strategy document, highlighting that: ‘we may fail to save around 11,000 lives a year due to safety concerns [and that] the extra treatment needed following incidents may cost at least £1b. ‘
They also estimate that work to improve patient safety in the NHS can save up to 1,000 lives and £100m in care costs annually. A key strand of their strategy relates to ‘Insight’: ‘to improve understanding of safety across the whole system by drawing intelligence from multiple sources of patient safety information.’ Identifying a clear need to improve patient safety intelligence.
Patient safety intelligence in the context of the NHS refers to the strategic use of data, information technology, and analytics to improve patient safety across healthcare settings. It encompasses the systematic collection, integration, and analysis of diverse data types, ranging from electronic health records (EHRs) to incident reports and patient feedback.
This approach enables healthcare professionals to identify and understand risks, spot trends, and pinpoint opportunities for enhancing patient care. Crucial in this process is the ability to extract actionable insights from the data, which can inform the development and implementation of more effective safety protocols and practices.
Bradford, where our research centre is based, is a post-industrial city in Northern England known for its diverse, multi-ethnic population, but also significant deprivation.
In 2007 the Born in Bradford cohort was established here, recruiting 30,000 participants for monitoring and linkage of both their education and health education records, leading to many high-profile findings and changes in policy and practice.
Extending the Born in Bradford project to represent the entire local population, the Connected Bradford Whole System Data Linkage Accelerator (cBradford) was developed. Covering ~800,000 subjects, cBradford integrates pseudonymised data from five NHS Trusts, 86 GPs, and 200 schools, spanning health, education, social care, environment, and local government (Figure 1).
Figure 1: Connected Bradford overview
Notes. This gives a sense of the range of data-sources contained within the cBradford infrastructure. Note that the reported figures here fluctuate over time as access to more data is attained.
Investigating inequity in patient safety
From the patient safety intelligence perspective, cBradford represents huge potential. In the Born in Bradford cohort for instance, linkage of educational and environmental data has revealed new insights into negative associations between educational attainment and pollution levels. One could argue this is a form of ‘epidemiological’ intelligence; by integrating data across a broader set of datasets, similar insights could be derived related to patient safety (Figure 2).
In line with the NHS patient safety strategy focus – and indeed, broader NIHR and PSRC strategies – one of main pieces of collaborative work with cBradford has been to begin to explore potential inequities in patient safety based on socio-economic indicators such as wealth, health, and ethnicity.
While still in early development, in close collaboration with our lay leader and academic and clinical collaborators, I have for instance begun trying to use cBradford secondary care data to match socio-economic indicators to standardised Patient Safety Indicator (PSI) codes; measures developed by the Agency for Healthcare Research and Quality (AHRQ) to provide information on in-hospital complications and adverse events following surgeries, procedures, and childbirth. These PSIs come as standardised lists of codes in a format that can be matched to those in cBradford, and as such, frequencies can be cross-tabulated with indicators of deprivation.
Figure 2: Example potential patient safety research questions
|Evaluate the effectiveness of new safety protocols||Time Series Analysis||Track reductions in incidents after implementing new interventions over time|
|Identify areas vulnerable to high rates of surgical complications||Geospatial Analysis using GIS||Map organisations with high post-surgical infections to pinpoint potential needs for regional training or resource allocation|
|Correlate socioeconomic status with healthcare safety incidents||Network analysis||Trace patterns between low-income areas and higher rates of medication errors|
|Understand the effect of environmental factors on patient safety outcomes||Multilevel Hierarchical Modelling||Analyse area level pollutants and individual level exposures as predictors of increased rates of post operative infections|
|Explore the link between crime, external stressors and patient safety.||Survival analysis
|Investigate if patients from high-crime areas have shorter intervals to hospital readmission post-surgery|
|Assess the impact of patient education on safety outcomes||Difference-in-Differences analysis||Measure reductions in medication mismanagement cases after initiating an educational intervention|
|Compare readmission rates related to social care provision||Causal modelling/Propensity Score Matching||Contrast readmission rates of elderly patients with post-hospitalisation home care against those without|
|Identify high-risk patient pathways through the healthcare system||Latent Class Analysis||Segment patient groups by their encounters with safety incidents to pinpoint areas in care delivery that consistently pose risk|
Notes. By having a massive, interlinked dataset we can apply a range of cutting edge analytic technique to answer a broad range of patient safety-specific research questions.
Project example 2 | A tool for predicting primary care deterioration
A key benefit of cBradford is its scope to explore patient safety beyond acute care and across contexts. We will therefore use linked primary and secondary cBradford data to investigate a previously established indicator of deterioration in primary care (Figure 3) – a key target of the NHS patient safety strategy.
By using variables across linked datasets, we believe we can build predictive models that can not only accurately predict deterioration as defined in Figure 3 above, but also a range of other patient safety indicators (including those we have yet to develop – see the next section).
Once we build and test our models to establish robustness, particularly across socio-demographic indicators, these could be a useful tool in a GPs toolkit for helping them identify and proactively treat vulnerable patients – assuming that the tool was integrated into the GPs workflow in a user-friendly and useful manner (which relates to our other sub-theme work focused on implementation and improvement).
Figure 3: Example novel patient safety indicator: ‘Missed’ deterioration
Notes. Here is an example of research where the authors have managed to interlink two data-sources – primary care and emergency care data – to capture a novel patient safety indicator: ‘missed’ acute deterioration. We are currently developing a risk prediction tool to predict this and other outcomes.
Project example 3 | A scoping review of research investigating ‘whole systems integrated data’ (WSID)
While previous published reviews of the literature have evaluated and found positive impacts on patient safety of improving interoperability of electronic health records; sharing of health information between organisations; and standardisation of health information technology, no review has aimed to scope attempts to answer myriad questions related to whole systems integrated data (as visualised in Figure 4) specifically. Nor have they attempted to outline how the interlinking of data across types can be used to generate predictive indicators of patient safety risk before they occur.
So aside from developing, assessing, and user-testing tools as that above, another arm of our work is conducting a scoping review (among several others for other sub-themes) aiming to assess how best to extract insights from whole systems integrated data (WSID) – and how to develop patient safety indicators from it.
Of course, it is one thing having novel indicators of patient safety events, and well-constructed mechanisms (e.g dashboards) for having these reach stakeholders (e.g. doctors). It is another to ensure that this information changes stakeholders’ decisions to improve patient outcomes. As such, as a theme, we are aiming to ensure maximal robustness of all stages of our tools from development to implementation.
Figure 4. Integrated data across a whole system? Conceptual literature model
Notes. The red line indicates where most research has focused: evaluating the effects of the introduction of some data exchange-orientated intervention on health or safety outcomes. The yellow area represents the scope of an example WSID, the grey areas methods of standardisation to improve WSID integration, and the black box to the right outlines some other potential related research focuses aside from outcomes.
Thank you for reading the first of our theme’s blogs. I hope I have given you some insight into our work and thinking at this stage of our development. The above only relates to a subset of our work, and while our sub-themes are highly inter-linked and overlapping, it was felt best to focus on one here for brevity. I encourage anyone interested in viewing more of our work to read the links above or get in touch with me on the email above.