HACA 24 Recordings

Main Presenter/s: Michael Cheetham
Additional Presenter/Co-Author/s: Adam Woodgate, Martin Patrick Griffiths, Ben Bloom

Assaults recorded in ED represent a substantially different set of events to those recorded by the police. Since 2014, data on assault location and mechanism has been collected in Emergency Departments in England under the Information Sharing to Tackle Violence standard (ISTV) and is used to identify geographical hotspots, temporal trends and changes in the nature of violence. This informs various violence reduction activity. Implementation of this nationally has been challenging because geocoding has been done manually at a local level. The objective of this study is to test the value of an automated geocoding algorithm.

Using data on all those presenting at a Barts Health UEC service in 2022 and 2023 recorded as having been assaulted (n= 6,998), this presentation describes a pilot natural language processing methodology for converting this free text to a point location. The methodology uses a combination of regular expressions (to identify postcode); phonetic matching using refined Soundex to a place name list (to identify streets and places) and supervised machine learning using subword vectors generated by fastText (to identify assaults happening at home).

Up to 93% of the test data (n= 1,730) have the location identified as well as the entry allows. Of the errors the majority are entries that should have been matched which are identified as not being matchable. This performance is good enough for most operational uses of ISTV data.
The intention is to refine this methodology to make this data available quickly across England and enable ISTV implemented.

Main Presenter/s: Ehsan Taati
Additional Presenter/Co-Author/s: Lee Robertson

Keywords: Knowledge extraction, Natural Language Processing (NLP), Suicide attempts, Mental Health Interventions, Large Language Model

Aim:
This research aims to develop an automated pipeline to identify and extract reports of suicide attempts from police records. Using advanced Natural Language Processing (NLP) techniques, we intend to streamline the text labelling process by implementing prompt engineering with a large language model.

Background:
Manually analysing and labelling large volumes of textual data in the UK Police's record systems, specifically flagging incidents as suicide attempts, is labour-intensive and time-consuming. This slows the broader analysis and interpretation of critical information.

Methodology:
In developing our automated pipeline, we first employ text normalisation to standardise language from the police reports. Utilising prompt engineering with Llama2, we improve the pipeline’s ability to identify information about suicide attempts. Human evaluators with police and mental health expertise subjectively assess the pipeline, guiding iterative refinements in text normalisation and prompt engineering. This qualitative evaluation ensures the system aligns with nuances important to police and mental health. This integrated approach aims to create a sophisticated system to effectively identify and extract reports of suicide attempts from police records.

Results
Subjective evaluation, including calculating the number of correct and incorrect predicted labels in the dataset, demonstrated the system's accuracy. Efficiency was underscored by comparing to manual labelling efforts, while expert assessments emphasized high accuracy (83%). Dataset analysis confirmed the system's robust performance in identifying diverse instances of suicide attempts.

Conclusion:
In summary, our automated system utilizes advanced NLP and prompt engineering to effectively identify suicide attempt incidents within police reports. Validation through comparisons to manual labelling and expert evaluation confirms the system's efficiency and accuracy. This research contributes a valuable tool integrating technology, law enforcement, and public health, with the potential to enable timely and precise knowledge extraction to inform mental health interventions.

Main Presenter: Eladia Liliana Valles Carrera
Additional Presenter/Co-Author/s: Alice Tapper, Daniel Goldwater, Matthew Taylor

Background:
Health and care service feedback mechanisms are crucial for continuous improvement and patient satisfaction. Every year we receive thousands of reviews for NHS services which must adhere to strict guidelines on personal information, abuse and discrimination. Moderating these reviews manually is a resource-intensive, time-consuming, and expensive task that hinders user experience.

Aim:
This project aimed to automate the content moderation process for written reviews on NHS.UK by using natural language processing (NLP) techniques to improve efficiency, scalability, and user satisfaction.

Methods:
Each rule on the content moderation policy was tackled separately, leveraging regex for the simpler rules to more sophisticated NLP tools like Named Entity Recognition and Part of Speech tagging for identifying names and descriptions. We trained three models using logistic regression, BERT, and SVM classifiers. These models were trained to identify suicidal and self-harm related content, to distinguish between personal experiences and generic comments, and a complaints model to flag reviews requiring formal escalation. The text in the reviews was transformed into a numerical representation using various embedding methods to capture the textual patterns and the semantic meaning of the reviews.

Results:
The implementation of the automoderation tool has enhanced the moderation process. It has reduced the time from several days to near real-time, decreased operational costs, and it offers the potential to scale the moderation service. The models have performed at par with, or even surpassed manual moderation.

Conclusions:
The use of advanced analytics in the form of NLP has proven to be a powerful tool in addressing the challenges of moderating written reviews for health and care services. This project exemplifies how innovative solutions can lead to substantial improvements in service delivery, efficiency, and set a precedent for future applications of NLP to enhance public health services.

Main Presenter: Sarah Louise Blundell
Additional Presenter/Co-Author/s: Andrew Lavelle

Background:
The national competency framework launched October 2023. The framework covers Data Analysts, Data Scientists and Data Engineers, looking not only at the competencies required within each discipline but also wider competencies in Leadership, Working in Projects and Behaviours. The Framework can be further nuanced by incorporating a Specialism dimension and an environment domain.

Aim:
To hear how its adoption has impacted on the development of data and analytics staff. How the framework will be expanded in 2024/25, and how the framework forms the foundation of the Data and Analytics Academy.

Methods:
The approach throughout the Framework development has been collaborative, engaging with colleagues across health and social care. Each development stage has been done in collaboration with experts with the data and analytics community to ensure a broad range of views could be collated.

Results:
By the HACA2024 Conference we will be able to showcase the impact on adoption on staff skill assessment and development planning in adopter organisations. Currently there are 36 organisations adopting the NCF. (March 24)

Conclusion:
The reality is there will be no conclusion for the Framework. It will continue to be dynamic and evolve and adapt as the requirements and demands on the data professional workforce changes. By raising awareness, increasing interest and maximising engagement with the Framework, we hope to create a truly integrated, informative and constructive tool that will support analytical development and professionalisation for individuals, teams, organisations and collaborative system level intelligence functions.

Presenter: Nuha Bazeer

Background: In recent years, record numbers of nurses trained outside of the UK and EU (overseas-trained nurses) have moved to the UK and are now working in the NHS, filling vacancies Whilst UK nurse “inflow” is well evidence, outflow to other countries is more difficult to track, with limited evidence on the indicators prior to nurses leaving the UK, or the workforce.

Aim: To identify trends in the potential outflow of UK registered nurses between 2018/19 and 2022/23 , and explore their connection to changes in the global labour market.

Methods: Nursing and Midwifery Council (NMC) registered nurses seeking to work abroad must apply for a Certificate of Current Practicing Status to prove their practicing status to foreign registration authorities.

The number of nurses applying for a certificate is therefore a reasonable proxy for potential outflow and was used to calculate the likelihood of UK registered nurses seeking employment abroad. Trends in application numbers were assessed against the length of time on the register before application, their training country, and the country they applied to, relative to the total number of nurses on the register.

Results: In 2022/23, overseas-trained nurses were 7 times as likely to apply than UK-trained nurses and accounted for 70% of the almost 12,000 overall applicants. Almost 25% of overseas-trained nurse had 3 or fewer years on the UK register before applying, compared to around 3% in 2018/19. Over 80% of applications were for potential destination countries with attractive labour market conditions, policies, and remuneration: Australia, New Zealand, and the USA.

Presenter: Alexander Lawless

Background:
General practices in England are currently facing major workforce pressures with falling numbers of full time General Practitioners (GPs) and rising demand for services. There is interest in understanding whether skill mix changes have improved access to general practice, increased the range of general practice services available to patients, or reduced use of other parts of the health care system.
This study aims to identify clusters in practice staffing levels using routine general practice consultations and to assess distribution of clusters within practice deprivation quintiles.

Methods:
This cross-sectional study utilises Clinical Practice Research Datalink (CPRD) data to derive practice-level daily full time equivalent (FTE) rate per 1000 registered patients by staff group. Consultation data grouped by staff type were aggregated to medical, nursing, registered healthcare professionals (HCP’s), healthcare assistants (HCA’s) or administrative roles. Practice FTE totals by staff group were analysed using Latent Profile Analysis (LPA) methods to identify clustering. Practice Index of Multiple Deprivation (IMD) quintile was linked to examine the relationship between deprivation and practice staffing configurations.

Results:
Analysis is currently in progress. All results will be completed and available by June 2024 in line with NHS England funding. Distribution of practices across latent profiles will be explored. Deprivation profiles of our staffing clusters will be generated and assessed, similarly, the geographical distribution within our profiles will be assessed to correlate geography and primary care staffing configurations.

Implications:
This analysis operationalises data on various staff groups and creates a composite measure that incorporates comparative levels of staffing within the practice. The utility of a composition measure that describes staffing can be compared against crude measures such as the proportion of staff at a practice that are medics. The next logical application is to examine the relationship between staffing mix and patient outcomes, for example secondary care admissions for avoidable conditions.

Presenter: Anastasiia Zharinova
Additional Presenter/Co-Author/s: Sarah Blincko, Rob Scott

With funding challenges and growing demand on health services, it is more important than ever to ensure that the NHS is run efficiently. While NHS acute productivity is well researched, there is limited knowledge about how productive the General Practice sector is and what factors contribute to its productivity the most.

This work investigates different approaches to estimating productivity of General Practices. We explored two metrics: total factor productivity and labour productivity. We linked multiple datasets, such as GP Workforce Data, ONS Demographics, NHS Payments to GP, GP Appointments Dataset and others. We applied regression modelling to analyse how General Practice characteristics, patient characteristics and other factors affect productivity in general practices. To ensure quality of the model and mitigate data limitations, we built both panel models and cross-sectional models and conducted additional robustness checks. In addition, we investigated quality metrics to test feasibility of quality adjustments and evaluate if there is a correlation between quality and productivity of General Practices. Over the course of the project, we worked with Primary Care policy teams and data teams to ensure both a correct interpretation of data insights and clear practical outcomes from analytical insights.

This analysis can inform policy decisions around additional support for General Practices. It can also support practices when making operational decisions about optimal skill mix of staff or addressing population needs. In addition, this work contributed to Primary Care Data Strategy and can inform future analysis in this area, for example, productivity of pharmacies.

Main Presenter: Colleen Dempster
Additional Presenter/Co-Author/s: Emilia Dulaj Kmiecik

The COVID-19 Health Protection Surveillance team at Public Health Agency (PHA), Northern Ireland, produce a weekly, public facing, COVID-19 Epidemiological Bulletin. Initially the production of the bulletin used various systems such as Access, Excel and Word, which where embedded together. This manual process was time consuming and error-prone due to the limitations of Microsoft. There was a risk that accurate data, critical to the surveillance of the ongoing pandemic, might not be available. This would impact decision making on testing policies, case management and vaccination campaigns.

It was crucial that we identified and implemented a more resilient and robust way of producing the COVID-19 Epidemiological Bulletin. We started by bridging the gap between Access and Microsoft by writing R scripts. This protected us from the immediate loss of data. We streamlined the process to R Markdown, moving away from Excel and Word. We worked with other teams moving to a cloud-based data analytics platform. We created various R-Studio-Connect apps to produce automated outputs that have made reporting processes more efficient and reduced the likelihood of error. This was a long process as we had to map out all data pipelines coupled with extensive data validation steps.

Revolutionising the production of the COVID-19 Epidemiological Bulletin was the result of a successful cross-departmental collaboration project lasting around 18 months. This project required staff upskilling and developmental work to take place in parallel, whilst ensuring information governance was in place and continuing to respond to the acute demands of the pandemic. Investing this time, allowed automation of reports which has reduced person hours, increased accuracy by removing any chance of human error and has enhanced the skill set of our staff. This work has been used as a template for other workstreams, thereby further improving efficiencies, allowing staff to work in a more agile and collaborative way.

Main Presenter: Bethany Thompson
Additional Presenter/Co-Author/s: Tom Frost

Keywords: dashboard, visualisation, metrics, ROI, impact, health inequalities

Introduction:
In response to Long Term Plan commitments, Alcohol Care Teams are being implemented in secondary care in England. Analysis of patient level data was required to monitor activity and outcomes, to identify health inequalities and to support the financial sustainability case for the programme.

Methodology:
Analysis of the Patient level data collection was used to create an Alcohol Dependence Dashboard which allows Trusts, ICBs and Regions to see their data through a variety of metrics. There has also been a particular focus on health inequalities within the dashboard with breakdowns of all metrics by different demographics (age, gender, ethnicity, deprivation). The Return on Investment Tool was developed to support the case for continued investment in alcohol services at sub-national level. It is based on hospital data and insight from a case study in the NICE Quality and Productivity collection.

Conclusions and recommendations for further research:
Both the dashboard and ROI tool were well received by Trusts, ICBs and Regions and allowed them to show and quantify the benefits of ACTs. The dashboard has been updated multiple times with improved functionality, since first publication (based on user feedback) and now the collection is in its 3rd year, the volume of data is providing important insights into inequalities at local and national level. Webinars were held to showcase the dashboard and the ROI tool and gave an overview of how the products worked.

Presenter: Matthew Eves

Main Presenter: Robert Watkins
Additional Presenter/Co-Author/s: Adrian Harvey

Activity at University Hospitals Leicester has increased by 13% since 2018/19, for children. To understand and address this increase in demand, a small team was commissioned to undertake analysis of rates of activity and possible links to health inequalities. The team included data analysts, Paediatric ED consultants, Public Health professionals and commissioners. The findings are to be presented to the ICB board.

Main Presenter: Jamie Gale
Additional Presenter/Co-Author/s: Elisa Santoro, Neil Jackson

Purpose:
• The Integrated Performance Report (IPR) is used by the South East regional MHLDA Clinical Delivery Team for internal meetings and briefings, requests for information and discussions with ICBs.
• A survey was recently circulated to those who receive the IPR. Feedback was it too long, too many metrics, difficult to interpret the charts.
• Overall, stakeholders found it challenging to get the intelligence they required, so it was agreed to re-design the report.

Methods:
• Meetings took place between programme leads and analysts to work collaboratively on the new report.
• The structure of the report and metrics were defined by the group based upon a patient pathway approach, considering business needs and regional priorities (i.e. aligned to the Regional and the NHS Long Term Plan).
• Any duplicate metrics were identified and consolidated.
• With the restructure of NHS England, the MH and LDA programmes were merged into one; additional meetings took place to integrate the LDA metrics within the report.
• Manual data collection processes were reviewed via a task and finish group; new metrics were added and those no longer required were stood down.
• As the report became established, it was automated using R Studio and the ‘plot the dots’ Statistical Process Control chart package.

Results:
• Clear exception-based narrative enabled evidence-based and higher quality decision making.
• One version of the truth for the metrics provided, using existing data sources when possible.
• Simplified report with fewer more concise slides and no duplicate metrics
• Reduced analyst time to produce the report and make changes.
• Appendices with additional detail was built into the report, with no additional resource required.
• Manual data collections streamlined reducing the data burden placed upon ICBs and providers.

Main Presenter: Kathryn Marie Lupton
Additional Presenter/Co-Author/s: Michael Richardson

How do we get better at integrating analytics and research into national policy development? How do we co-produce analytical and policy tools in a meaningful and iterative way with end users?

This presentation will draw on emergent learning from the piloting of the National Quality Board’s Quality Early Warning Signs Framework, being led by NHS England. The QEWS framework has been designed to support providers, systems, regional and national teams to identify, mitigate and manage early signals about the quality of care, relating to leadership, culture, governance and other considerations.

The approach is underpinned by a dashboard hosted on Model Health System, and a framework signposting to qualitative data and wider sources of information. It is currently being piloted in four Integrated Care Boards and being overseen by a Working Group with representatives across system, regional and national teams.

By drawing on the learning from these pilots, this presentation will explore:
- The case for building analytics and social research into policy and strategy development BAU
- How we can do this by the conversations we have and how we work together across system, regional and national geographies, with people using services, clinicians policy makers and analysts
- Key ingredients, enablers and challenges to achieving this, including culture, leadership and behaviours
- What difference this can make to how policies are adopted and the difference they make, with insights into how we spread the learning.

Main Presenter: Ruth Green
Additional Presenter/Co-Author/s: Kelly Bishop, Joanne Bayliss

This presentation will focus on the work NHS ML BI Consultancy Unit are leading, in collaboration with NHS ML Nursing and Urgent Care Team, to support an Integrated Care Board (ICB) to understand the potential opportunities for improvements in cancer pathways.

Background:
As referenced in the national NHS priorities and operational planning guidance for 2023/24, the significance of early cancer diagnosis is an NHS priority. One of the biggest actions the NHS can take to improve cancer survival is to diagnose cancer earlier.

Funded by the regional Cancer Alliance, the ICB identified the need for a deep dive cancer pathway analysis. The areas of focus of this in-depth review were late-stage cancer diagnosis and a review of primary care cancer referrals.

Aim:
The aim was to identify and investigate variations in cancer pathways. The insight gained informed recommendations to improve the patient experience by optimising referral to diagnosis timescales, and improving early diagnosis, where possible and appropriate.

Approach / Key message:
Our joint approach of clinical and analytical teams working closely together was key to the delivery of the project, which encompassed clinical audit as part of the analysis. Working in partnership with the ICB, primary care and acute hospital provider cancer services teams within the ICB geography, meant we could incorporate a wide source of data and information, and obtain the required input from clinical specialists as part of a case study approach.

Adopting a mixed method analysis enabled a rich study to develop, providing the evidence needed to inform recommendations for improvement and transformation of cancer pathways across the integrated care system, with the potential to be adopted across other ICBs / regions.

Main Presenter: Hadi Modarres
Additional Presenter/Co-Author/s: Thomas Henstock, Divya Balasubramanian, Dimitris Pipinis, Gursimran Thandi, Achut Manandhar, Rupert Chaplin

Introduction:
Timely cancer diagnosis is critical for improving patient outcomes, yet late diagnosis remains a challenge associated with limited treatment options and increased mortality. The NHS aims to enhance early cancer detection by 2028, a goal further complicated by COVID-19 related disruptions. Identifying high-risk sub-populations for certain cancers, characterised by vague symptoms and low incidence rates, is particularly difficult. Leveraging advanced machine learning techniques offers promise in addressing this challenge within the NHS. This study utilises NHS England's patient-level data and data science capabilities to predict cohorts at elevated risk of future cancer diagnoses with the aim of targeting and tailoring interventions to sub-populations that are at higher risk.

Method:
A dataset comprising 29 million individuals over 40 in England was compiled, integrating various datasets including the master patient index, population segmentation data, 111 calls, hospital attendances, and mortality records. Features capturing healthcare interactions (e.g. number of 111 calls, number of hospital attendances), demographic, socioeconomic, and clinical diagnosis variables were developed. Machine learning algorithms were trained to predict future cancer risk for all cancers and specific priority sites identified by the cancer programme.

Results:
The trained models achieved a performance of 0.79 AUC, comparable to existing literature. Notably, interactions with the healthcare system, such as A&E attendances, emerged as significant predictors of future cancer risk. Identified high-risk cohorts exhibited up to a tenfold increase in cancer incidence compared to baseline population rates.

Discussion:
This study highlights the potential of secondary care data coupled with machine learning in identifying high-risk cohorts for targeted interventions. The incorporation of primary care and cancer registry data could further enhance predictive accuracy, representing a key area for future research within this project.

Main Presenter: Kathie McPeake
Additional Presenter/Co-Author/s: Kelly Dixon

35,000 people are living with cancer in Lincolnshire. This is predicted to rise to over 45,000 by 2030. A cancer diagnosis can have a profound effect on a person, with not just physical health impacted. People’s mental health, finances, relationships and work can also be negatively affected. Evidence shows that people who have access to good, personalised support before and after a cancer diagnosis often have better outcomes than those who don’t.
The Lincolnshire Living with Cancer Programme is an innovative, collaborative, co-produced programme which is transforming the way in which people affected by cancer access personalised support. We are creating a better and sustainable future for supporting people living with cancer, involving and integrating all relevant parts of the health and social care system, using the assets we already have, supporting people in the place they would like and in the way they would like, and placing people at the centre of everything we do. Data drives our programme, and we measure impact by synthesising quantitative and qualitative data.
At the start of the programme, qualitative data in the form of patient stories told us what needed changing to improve patient experience. It informed our approach to transformation and what we needed to do. In our delivery phase, we need to understand the impact we’re having. We use quantitative and qualitative data to do this. Our data analysts have developed a dashboard bringing together all quantitative data relevant to the programme from different organisations in the Lincolnshire system, which shows the difference we’re making. They have also developed a methodology of measuring the impact of qualitative data (case studies) which quantifies patient experience and enables us to express patient experience as a return on investment for our ICB Investment Panel. We also use data to inform our service delivery.

Main Presenter: Brandon Chapman
Additional Presenter/Co-Author/s: Joe Turner, Jo Davis

With a small data science team now in place at RCHT, we have both developed tools and worked on getting them used in operational practice. We have done this because of the ability of data science to open up new ways of understanding the range of information available to the hospital, especially on the predictive side, enabling the operational and clinical teams to take more informed actions.
To do this, we have developed a range of new products working with the clinical and operational teams. Successes have included a machine learning based DNA predictor which can predict with good accuracy which patients are most likely not to attend. Working with our outpatient transformation team, we have been contacting these patients to bring about an actual reduction in non-attendances.
We have also developed a suite of predictive tools in urgent care, for instance:
- An estimate of individual risk of emergency department admissions
- Urgent care alert system which generates an overall ED pressure score based on the indicators most related in practice to ED crowding
- Assessing the impact of seasonal tourism on emergency and trauma theatres
- Assessing demand for blood in emergency operating.
The results to date have included getting some tools in decision-making, for instance delivering over 200 outpatient slots’ worth of benefit, but also an improvement in our understanding of how we improve take-up of what we develop.
Learning points have included the importance of project definition, addressing questions that operational teams want answered, as well as identifying operational questions which are boundaried enough to lend themselves to actions dependent on what the tools show. Also, the importance of designing tools with an eye of how they will ultimately be used (eg high specificity for the DNA tool to focus limited resource for calling patients).

Main Presenter: Dr Zella King
Additional Presenter/Co-Author/s: Alison Clements

In a health system under pressure, with hospitals operating at capacity, difficult decisions are made daily. Operational managers must decide when to initiate escalation measures like discharging patients earlier, cancelling elective admissions or diverting ambulances because patient care is no longer safe.

These decisions depend on having a solid picture of current capacity and imminent pressures. Gaining that picture often involves manual data collection, validation of data in the Electronic Health Record (EHR), frequent meetings with ward managers, and the use of simple heuristics to predict end-of-day bed state. With such activities, the promise of EHRs to unleash greater productivity and better insight remains unrealised. This is unfortunate both for hospitals themselves and for the national, regional and sector bodies who are seeking real-time oversight of urgent and emergency care through System Control Centres and the Operational Pressures Escalation Levels (OPEL) framework

At HACA 2023 we presented a predictive model of short-term demand for emergency beds, which is in daily use at UCLH. That model provides one tool to inform decision-making about escalation. However, for the full picture, the ‘supply side’ is also important. In this joint talk from an operations director and a data scientist, we introduce models that predict short-term discharge activity at different levels of aggregation and certainty. When combined with predictions of demand, and based on real-time data, such models give operational managers better insight about where capacity needs to be freed up, and when to initiate escalation measures.

We discuss the data challenges and some of the realities of hospital life that make this important work difficult. We lay out a vision of how predictive modelling of short-term capacity can supplement the view provided by an EHR, and how this can lead to better informed decision-making about the provision of urgent and emergency care.

Presenter: Jane Kirkpatrick

Background:
Our team develops and maintains an A&E Forecasting tool which supports operational decision making.
We make daily 21-day ahead forecasts of non-elective admissions via A&E for 121 trusts in England. These are presented on a user-facing dashboard, with forecast explanations and historic accuracy metrics.

Rationale of our work:
Since this tool was stood up during the pandemic, we were working to tight timelines to deliver a product that met users’ top priorities. We now have an opportunity to improve the tool and expand its use. We have heard from users the improvements they would find useful. We want to share our code, and have refactored to improve code usability.

Actions:
This presentation will focus on the continuous development of the Bayesian hierarchical model we use to forecast – new features tried and lessons learnt. It will be an honest presentation of the reality of data science tool development – sometimes you try things, and they don’t work as planned!
Recently we have worked on:
- Providing forecasts with breakdown by site. To enable this we explored normalisation approaches for input data. We found this did not lead to expected performance improvements.
- Forecasting for attendances as well as admissions. This involved adapting the model, tuning parameters of prior distributions and modularisation of code.
- Exploring if data on respiratory infections could improve predictions.
- Refactoring and maintaining codebases for modelling and deployment.

Outcome:
“I think you’ve got something which is ground-breaking and incredibly useful.”
Associate Director for Performance Information, Luton & Dunstable Hospital
Maintaining and improving an analytical tool is an important part of its lifecycle. This work has allowed us to better meets users’ needs, improved user buy-in and supported working towards publishing our code.

Main Presenter: Melissa Co
Additional Presenter/Co-Author/s: Freya Tracey, Kathryn Marszalek

In January 2023, the Prime Minister pledged that ‘NHS waiting lists will fall and people will get care more quickly’. In October 2023, we analysed what it would take to achieve this pledge before the general election. The NHS elective care waiting list has been growing since 2013 and increased sharply during the pandemic because elective care was suspended. Even after lockdowns were lifted, the waiting list continued to grow because the number of treatments completed had not caught up with the steady stream of new referrals.
We created four projections for how the waiting list would change up to January 2025 with and without the effect of industrial action (strikes by consultants and junior doctors) and varying productivity (increases/decreases in completed treatments). To help illustrate the uncertainty in the system, we also created an interactive calculator using R Shiny to allow readers to explore their own alternative scenarios.
The politicised backdrop – industrial action, the upcoming general election, the critically stretched NHS – added challenges beyond the analytical modelling itself. We had to balance our commitment to objectivity and accuracy while accepting we could not model the full, continually changing system.
We will discuss our analytical approach, including our choices when modelling the effect of industrial action and where and why we decided to allow users to make their own decisions. With an additional 9 months of data available by the time of the presentation, we will also compare our projections to reality and reflect on why they may differ.
National and local healthcare analytics will continue to face challenges working on uncertain and politically-sensitive topics – and this is increasingly likely as pressures on the NHS mount. We aim to share the lessons we have learned and our thoughts on how to manage these challenges.

Main Presenter: Kamil Barczak

The growing NHS waiting list has been in the spotlight for years. Cuts to the NHS budget, the growing elderly population and the effects of 2020 COVID-19 pandemic have all contributed to record breaking growth of those waiting to be seen and the duration of the referral to treatment times (RTT). A report by the Health Foundation UK suggests that the NHS waiting list is expected to peak at more than 8 million by summer 2024 . Coinciding with long RTT times are deaths of patients on waiting lists. There’s a real concern that longer RTT mean more deaths and increased disease impact. Apart from reducing the RTT, methods to prioritise those who are most vulnerable to the effects of increased RTT and can be effective and minimise the impact of long waiting times.

Our aim was to develop a risk stratification tool using machine learning to identify patients most likely to die within 12 months of their pathway journey.

Retrospective open pathways and death data was collected, cleaned and temporalized. A recent 12-month period was chosen to track deaths within 12 months of starting a pathway. A previous 12-month period was used for training. Any information prior to the beginning of the training data was used to calculate metrics of interest such as death rate by pathway. We used a Synthetic Minority Over-sampling Technique (SMOTE) and random under sampling to reduce the impact of data imbalance . We also used a logistic regression-based feature selection method to reduce the dimensionality of the training data and modelled risk of death per pathway using a logistic regression model, treating each record as a unique sample.

Model outputs have been integrated into a Power BI dashboard and deployed onto Cornwall’s platform. We plan to collect feedback from the client in the upcoming weeks.

Presenter: Shaun Rowark

For the past 25 year the National Institute for Health and Care Excellence’s (NICE) core purpose has been to help practitioners and commissioners get the best care to people fast, while ensuring value for the taxpayer. While this remains constant for NICE how we make decisions has changed to ensure we focus on what matters most, provide useful and useable advice, and constantly learn from data and implementation.
In this lightning talk I will describe how NICE has adapted to the proliferation of real-world data and the development of real-world evidence methods and how we use this data and methodology to make decisions for the development of our products. These products focus on national guidance as well as whether medicines are approved for patient access in England. I will also cover future considerations for NICE, such as patient reported data from wearables, synthetic data and data generated by artificial intelligence.

Main Presenter: Neil Bendel
Additional Presenter/Co-Author/s: Eliza Varga

Measures of socio-economic inequalities are important markers for health and social policy and for society in general. It is important that indicators and tools which purport to measure socio-economic inequalities are both accurate and presented in a way that minimises the chances of misinterpretation and supports better quality decision making.

Drawing on the earlier work of ScotPho and PHE, the Manchester Measuring Inequalities Toolkit is designed the fill the need for an easy to use, interactive training package which illustrates and explains some of the most commonly used methods for measuring health and other inequalities, including simple measures of inequality gaps, regression-based inequality measures (e.g. Slope Index of Inequality), and Lorenz-curve-based measures, and how these should be calculated, visualised and interpreted. Through this, the Toolkit will help to improve the development and monitoring of interventions to address socio-economic inequalities by helping information analysts and policy makers to produce and utilise more statistically rigorous and accurate outputs describing the scale and nature of the inequalities in a local area.

This presentation will describe in more detail the purpose of the Measuring Inequalities Toolkit, how it was commissioned and the initial content. It will go on to explain how selected modules were piloted by a group of analysts from a range of different organisations and how the content was then revised following the feedback provided. Finally, it will describe the next steps in the development of the toolkit and how this fits in with the wider programme of work to monitor the impact of Making Manchester Fairer – the City’s action plan for tackling health inequalities.

Presenter: Anna Pickford

The Healthier and Fairer Programme within Northeast and North Cumbria is a whole health and care system transformational programme of prevention, healthcare inequalities, and social and economic inequality projects to improve population health.

The development of the programme has included the use of analytics from inception, adopting modelling and population health management approaches, a strong culture of evaluation and joint working across system partners.

The Business Intelligence function has supported with 5 core components;

Intelligence and Insight – Providing evidence-based advice on emerging themes for the population, producing Healthcare Needs Assessments, delivering training on identifying inequalities within wider Programmes of work, advising how to measure and demonstrate impact.

Analytics and dashboard development – Providing cohort analysis for specific projects, creating a model of estimate need for increased weight prevalence, producing Power BI dashboards for workstreams to aid ongoing monitoring of key population issues specific to their agenda.

Performance – Demonstrating progress against national and local priorities such as CORE20plus5 agenda, ICB priority strategies and national policy.

Research and Evidence – Working in partnership with universities to evaluate specific projects. Writing quantitative protocols, undertaking analysis and contributing to the writing of academic and board papers.

Data quality and data flows – identifying appropriate data flows and where not available, working with providers to create new flows.

We currently have 5 self-service workstream dashboards live which are available to all system partners including Local Authorities, Acute and Mental Health providers and OHID.

The dashboards include; Alcohol Prevention, Healthy Weight and Treating Obesity, CVD prevention, Tobacco Control, Public Health and Prevention in Maternity and an Overarching Healthier and Fairer dashboard.

We have embedded routine performance reports to ICB executive, 4 evaluations are in progress and have completed three Healthcare Needs Assessments relating to the prevention agenda for NENC ICB.