Here you can find details of a range of current and previously-completed HSMA projects, split by the key areas the project addresses.
Click on the preview for each project to read the full details. For many completed projects, you can watch a short talk from the HSMAs who undertook the project on their plans, successes and challenges.
Mental Health
Neurodiversity
Use of Discrete Event Simulation to Tackle Long Waits and a Growing Backlog for Children Requiring Neuro Development Assessment (Autism and ADHD)
The project aimed to identify bottlenecks in the Autism and ADHD assessment pathway for children and determine changes needed to manage demand and avoid backlogs. The main issue was clinical staffing capacity. Could adding one clinician could maintain a steady state and prevent backlog growth?
Discrete Event Simulation modelling of childrens ADHD diagnosis and treatment
The project uses Discrete Event Simulation to model the children's ADHD diagnostic and treatment pathway, aiming to reduce waiting times and lists. It proposes a new pathway with preliminary diagnostic testing to ensure accurate ADHD assessments. The model will evaluate the impact of these changes and may extend to include 1:1 and group session appointments.
Using Data Science techniques to address delays in the pathway for Neurodiversity Diagnoses in Children and Young People in Devon
In Devon, around 6,000 children have waited over 52 weeks for a neurodiversity diagnosis, crucial for their education and independence. This project aims to reduce the waiting list and streamline the process using DES modelling, NLP, machine learning, and geographic modelling to identify inequalities and optimise resource allocation.
Mental Health Inpatients
Predicting Violent Incidents on Mental Health Inpatient Units
This project aimed to see if violent incidents could be reduced on hospital wards by predicting and preventing them, using machine learning. A dashboard would highlight high-risk wards to senior staff for intervention.
NHS Talking Therapies (Formerly IAPT)
Discrete Event Simulation of Cognitive Behavioural Therapy Pathway in an IAPT Service
The project used Discrete Event Simulation to model the wait list for High Intensity Cognitive Behavioural Therapy. It found that changing an evening group to a face-to-face clinic could reduce wait times, especially for face-to-face or evening 1-1 appointments.
• NHS Dorset ICB
Modelling secondary care psychological therapy flow
The project aims to improve patient flow for severe mental health treatment by assessing and projecting service user pathways under various conditions. It will provide statistical reasoning for future changes to enhance patient experience. Outputs include time in system, queue lengths, resource utilisation rates, and probabilities of exceeding thresholds for queue length, queue time, and resource utilisation.
Other Mental Health Projects
Modelling resources needed to reduce out of area placements in the Mental Health Acute Care pathway in Devon, leading to the construction of a new mental health ward
Devon Partnership NHS Trust explored changes to its Urgent Care Pathway to reduce system pressure and avoid out-of-county care. Process mapping and a simulation model identified a need for additional beds. Key findings included high occupancy rates, the necessity to reduce demand and lengths of stay, and the importance of increasing bed numbers to enhance patient care and workforce efficiency.
Diseases
Diabetes
Predicting Gestational Diabetes and other maternity-related conditions using machine learning
The project aims to use machine learning to predict gestational diabetes and other maternity-related conditions early in pregnancy, enabling timely interventions and personalised care. It will develop a predictive model, an interactive web app for clinicians to input patient data and receive predictions, and reports on the model's accuracy and effectiveness.
Cancer
Use Of Discrete Event Simulation (DES) to reduce delays in Cancer Diagnosis & Treatment
The project developed a Discrete Event Simulation model for the Colorectal Cancer pathway identify key delays and solutions. Using three years of anonymised patient records, the model highlighted diagnostic delays during investigative stages.
• University Hospitals of Morecambe Bay NHS Foundation Trust
• Dartford and Gravesham NHS Trust
DESmond: Discrete Event Simulation and Artificially Intelligent Forecasting: Modelling the 62 Day Prostate Cancer Pathway
We're missing the 62-day prostate cancer treatment target due to resource allocation issues. The project aims to develop a Discrete Event Simulation model to predict demand and identify bottlenecks using live data. This model will help reallocate resources proactively, ensuring targets are met. The outcome includes an interactive web app and dashboard for key metrics.
Optimising the location of Breast Cancer diagnostic services across Devon
Over 6000 patients are referred annually for fast-track breast symptom diagnosis at RD&E and NDDH. Increasing referrals and limited infrastructure necessitate building or extending diagnostic units. The project aims to determine optimal locations for additional services to minimize patient travel time and reduce costs, using data science to map demand and calculate travel distances and benefits.
Modelling delays in breast, head and neck cancer pathways
The project uses DES to model post-diagnosis pathways for breast and head and neck cancer. It aims to identify delays and treatment variations across England, focusing on two key steps - time from neoadjuvant SACT to surgery or radiotherapy, and time from surgery to first adjuvant treatment.
Forecasting demand in RDUH breast care services and the impact of urban development
Between 2011 and 2021, the population served by Royal Devon & Exeter Hospital grew by 13%, doubling the national average, leading to long waiting lists. This project aims to develop a forecasting tool for breast services to predict referrals and service demand, inform workforce planning, and justify infrastructure expansion. It will also predict geographical demand changes for optimal service locations.
Evaluating the use of machine learning calssifiers for the identification of the determinants of stage at diagnosis of prostate cancers using registry data
This project uses various Machine Learning methods to classify prostate cancer diagnosis stages using registry fields. The focus is on methods comparison in prostate cancer diagnosis, driven by a recent increase in diagnoses.
RALPulator : Predicting Robotic-assisted laparoscopic prostatectomy (RALP) operative times from patient letters
This project is an app that reads in patient letters ahead of surgery, using Natural Language Processing techniques to extract key information from the text, and then feeds all of that into a Machine Learning model which then predicts how long the Robotic-assisted laparoscopic prostatectomy (RALP) surgery is going to take.
COVID-19
Developing a generic vaccination service model for the COVID-19 pandemic and beyond
In winter 2020, a mass COVID-19 vaccination program required GP surgeries to deliver vaccines at an unprecedented scale. A Discrete Event Simulation model was the vaccination pathway in North Devon, predicting vaccination rates and identifying potential issues. The model was refined for safe delivery and made available for future use, aiding efficient vaccination efforts globally.
South East Regional Covid 19 Vaccination Demand & Capacity Modelling
The project aimed to create an easy-to-use model for predicting Covid vaccination demand and capacity. Despite data access barriers, the team made progress by using past vaccination data. The tool will help track vaccinations, identify problem areas, and save time and money.
• Surrey and Borders Partnership NHS Foundation Trust
Geographic and Boosted Tree Modelling of Healthcare worker vaccination uptake
There is a decreasing uptake of COVID and flu vaccinations among healthcare workers. Identifying patterns by staff uptake, gender, ethnicity, deprivation, and other factors can help. Using geographic and boosted tree modelling, along with regression analysis, can capture these patterns and provide useful data for stakeholders to address the issue.
Public Health
Smoking and Weight Management
Modelling the benefit of MECC (Making Every Contact Count) Training using agent based simulation
Making Every Contact Count (MECC) is an e-learning program for health and social care staff to promote healthy lifestyles. This project uses Agent Based Simulation to model MECC's impact on behaviors like smoking, drinking, and exercise.
• Surrey and Sussex Healthcare NHS Trust
• Somerset NHS Foundation Trust
Predictive modelling for smoking cessation success
Smoking cessation remains challenging despite public health efforts. This project aims to develop a predictive model to identify individuals likely to quit smoking based on demographics and behaviours. It will uncover key predictors and effective pathways using machine learning algorithms like logistic regression, decision trees, random forests, and neural networks, evaluating each for accuracy and interpretability.
Vaccination
Developing a generic vaccination service model for the COVID-19 pandemic and beyond
In winter 2020, a mass COVID-19 vaccination program required GP surgeries to deliver vaccines at an unprecedented scale. A Discrete Event Simulation model was the vaccination pathway in North Devon, predicting vaccination rates and identifying potential issues. The model was refined for safe delivery and made available for future use, aiding efficient vaccination efforts globally.
South East Regional Covid 19 Vaccination Demand & Capacity Modelling
The project aimed to create an easy-to-use model for predicting Covid vaccination demand and capacity. Despite data access barriers, the team made progress by using past vaccination data. The tool will help track vaccinations, identify problem areas, and save time and money.
• Surrey and Borders Partnership NHS Foundation Trust
Geographic and Boosted Tree Modelling of Healthcare worker vaccination uptake
There is a decreasing uptake of COVID and flu vaccinations among healthcare workers. Identifying patterns by staff uptake, gender, ethnicity, deprivation, and other factors can help. Using geographic and boosted tree modelling, along with regression analysis, can capture these patterns and provide useful data for stakeholders to address the issue.
Other Public Health
Understanding Excess Mortality in Dorset
The project aimed to improve access and service experience through Dorset ICS's Health Inequalities programme. It focused on defining and measuring excess mortality, identifying unexpected mortality trends, and understanding driving factors.
Identifying which patients are most at risk for an outcome across integrated neighbourhood teams
Population health management uses segmentation to categorise people by health status and needs. However, generic segments may not identify high-risk groups effectively. This project aims to create a tool to identify at-risk patient groups across different geographies, focusing on outcomes like emergency admissions, vaccination rates, and screening uptake.
Medical Specialties
Cardiac
Modelling the impact of regionalising cardiac arrest centres, leading to work to standardise best practices
Survival to hospital discharge after out-of-hospital cardiac arrest (OHCA) is around 7-8%. Analysis of the SWASFT Cardiac Arrest Registry showed survival rate variations due to different hospital services. A regional care system could improve survival rates. A collaborative project between ambulance services and hospitals aims to standardise care, addressing gaps and enhancing outcomes through shared data and best practices.
Reducing Travel Times to Treatment for Cardiac Patients in the South East of England
The project analysed travel times to understand the impact of flow of activity into London on patient travel. It identified a gap in cardiac surgery access for Kent & Medway patients, suggesting new sites could help. A Streamlit app visualised these impacts.
• NHS England
• The Strategy Unit
Colorectal
Use Of Discrete Event Simulation (DES) to reduce delays in Cancer Diagnosis & Treatment
The project developed a Discrete Event Simulation model for the Colorectal Cancer pathway identify key delays and solutions. Using three years of anonymised patient records, the model highlighted diagnostic delays during investigative stages.
• University Hospitals of Morecambe Bay NHS Foundation Trust
• Dartford and Gravesham NHS Trust
Maternity
Predicting Gestational Diabetes and other maternity-related conditions using machine learning
The project aims to use machine learning to predict gestational diabetes and other maternity-related conditions early in pregnancy, enabling timely interventions and personalised care. It will develop a predictive model, an interactive web app for clinicians to input patient data and receive predictions, and reports on the model's accuracy and effectiveness.
Neonatal
Modelling the location of neonatal critical care units in North West England
The project explored the optimal location for 22 neonatal care sites in North West England, using an algorithm that balanced travel time, distance, NICU care episodes, and admission numbers. A discrete event simulation model and dashboard were created to explore various scenarios, aiding stakeholder decision-making.
Neurology
Referral to treatment waiting times for Neurosurgical patients
The project models the neurosurgical patient pathway to predict waiting list changes and treatment wait times. It aims to add user interaction to explore how capacity adjustments affect waiting times and track patients waiting over 52 weeks each month.
Orthopaedics
Modelling strategies to reduce the elective backlog in hip surgery
This project used Discrete Event Simulation to model the hip surgery patient pathway in Exeter, assessing strategies to reduce the surgery backlog. It was a collaborative effort between the CCG and the acute provider, contributing to long-term planning and decision-making.
• NHS Devon ICB
Physiotherapy
Modelling Musculoskeletal Physiotherapy services across Exeter, Mid and East Devon
The project explores demand and access to Musculoskeletal Physiotherapy services in Exeter, Mid, and East Devon. It uses Geographic Modelling to show variations in demand and access times, using DES to optimize resource placement. A Streamlit App will be developed for ongoing service development and review.
Renal
Predicting the future demand for Renal replacement therapy
Kidney disease is projected to be the fifth leading cause of premature deaths globally by 2040. Rising demand for dialysis and transplants exceeds capacity in England. The project aims to develop a model using forecasting techniques to predict future demand and address it by increasing home dialysis or introducing a new dialysis centre.
Respiratory
Using Machine Learning to Predict Hospital Admissions and Length of Stay for Respiratory Conditions
The project aimed to predict respiratory condition admissions and length of stay using patient history and demographics. Early-stage modelling showed promising preliminary findings. Further development will include more patient history to improve predictions and extend the model's capabilities.
Rheumatology
The role of Patient Initiated Follow-up (PIFU) and ‘Digital Outpatients’ in Supporting the Elective Recovery - Can We Better Size Potential for Clearing the Backlog?
The project aimed to explore the role of Patient Initiated Follow-up (PIFU) in addressing backlogs by mapping rheumatology outpatient pathways using discrete event simulation. The team modelled resource use and redeployment, focusing on PIFU's impact on referral-to-treatment waiting lists.
A Discrete Event Simulation Model to reduce Rheumatology waiting times in Dorset
The project aimed to reduce the Rheumatology waiting list and times in Dorset using Discrete Event Simulation (DES). The team built a DES model and a Streamlit app to simulate capacity changes.
• BCP Council
Urology
Evaluating the use of machine learning calssifiers for the identification of the determinants of stage at diagnosis of prostate cancers using registry data
This project uses various Machine Learning methods to classify prostate cancer diagnosis stages using registry fields. The focus is on methods comparison in prostate cancer diagnosis, driven by a recent increase in diagnoses.
RALPulator : Predicting Robotic-assisted laparoscopic prostatectomy (RALP) operative times from patient letters
This project is an app that reads in patient letters ahead of surgery, using Natural Language Processing techniques to extract key information from the text, and then feeds all of that into a Machine Learning model which then predicts how long the Robotic-assisted laparoscopic prostatectomy (RALP) surgery is going to take.
Patient Groups
Children (Paediatrics)
Simulation modelling to test proposed models of pediatric critical care in South West England
This project used Discrete Event Simulation and Geographic Modelling to identify optimal locations for Level 2 Paediatric Critical Care Units in South West England, highlighting the need for units outside Bristol. Proposed locations were identified to better serve the population.
Use of Discrete Event Simulation to Tackle Long Waits and a Growing Backlog for Children Requiring Neuro Development Assessment (Autism and ADHD)
The project aimed to identify bottlenecks in the Autism and ADHD assessment pathway for children and determine changes needed to manage demand and avoid backlogs. The main issue was clinical staffing capacity. Could adding one clinician could maintain a steady state and prevent backlog growth?
Forecasting Demand and Length of Stay in the Emergency Department
The project aimed to use forecasting and machine learning to predict Emergency Department arrivals and length of stay. A web-based app was designed to explore trends and suggest improvements. The app forecasts attendances and stay durations, helping plan workforce needs.
Discrete Event Simulation modelling of childrens ADHD diagnosis and treatment
The project uses Discrete Event Simulation to model the children's ADHD diagnostic and treatment pathway, aiming to reduce waiting times and lists. It proposes a new pathway with preliminary diagnostic testing to ensure accurate ADHD assessments. The model will evaluate the impact of these changes and may extend to include 1:1 and group session appointments.
Using Data Science techniques to address delays in the pathway for Neurodiversity Diagnoses in Children and Young People in Devon
In Devon, around 6,000 children have waited over 52 weeks for a neurodiversity diagnosis, crucial for their education and independence. This project aims to reduce the waiting list and streamline the process using DES modelling, NLP, machine learning, and geographic modelling to identify inequalities and optimise resource allocation.
Older Adults
Using machine learning models to predict future frailty
The project uses machine learning to estimate future frailty in the Wakefield District and identify key predictive features. It aims to plan resource allocation based on evidence, using two years of linked data. The project will also explore predicting other long-term conditions and produce reports and a user interface for stakeholders.
Predicting the risk of injurious falls in older people with atrial fibrillation
Atrial fibrillation (AF) increases stroke risk, and anticoagulation reduces this risk but can cause bleeding. Despite guidelines, many clinicians avoid prescribing anticoagulants to those at risk of falls. This project explores using machine learning to predict injurious falls in older AF patients and aims to develop a tool to personalize anticoagulant treatment based on falls risk.
Women’s health
Optimising the location of Breast Cancer diagnostic services across Devon
Over 6000 patients are referred annually for fast-track breast symptom diagnosis at RD&E and NDDH. Increasing referrals and limited infrastructure necessitate building or extending diagnostic units. The project aims to determine optimal locations for additional services to minimize patient travel time and reduce costs, using data science to map demand and calculate travel distances and benefits.
Predicting Gestational Diabetes and other maternity-related conditions using machine learning
The project aims to use machine learning to predict gestational diabetes and other maternity-related conditions early in pregnancy, enabling timely interventions and personalised care. It will develop a predictive model, an interactive web app for clinicians to input patient data and receive predictions, and reports on the model's accuracy and effectiveness.
Modelling delays in breast, head and neck cancer pathways
The project uses DES to model post-diagnosis pathways for breast and head and neck cancer. It aims to identify delays and treatment variations across England, focusing on two key steps - time from neoadjuvant SACT to surgery or radiotherapy, and time from surgery to first adjuvant treatment.
Workforce
Investigating factors impacting NHS workforce retention
This project aimed to work out which factors are the biggest drivers of staff turnover using regression modelling on staff workforce figures as well as other local factors such as employment. This was turned into a dashboard for internal use.
• Yorkshire Ambulance Service NHS Trust
Geographic and Boosted Tree Modelling of Healthcare worker vaccination uptake
There is a decreasing uptake of COVID and flu vaccinations among healthcare workers. Identifying patterns by staff uptake, gender, ethnicity, deprivation, and other factors can help. Using geographic and boosted tree modelling, along with regression analysis, can capture these patterns and provide useful data for stakeholders to address the issue.
Forecasting the supply of medical doctors
The project aims to forecast the regional demand for medical doctors, considering factors like trainees and population. It will plot a graph showing the current supply versus demand. Additionally, it will simulate outcomes under different scenarios, such as changes in trainee numbers or population health, to better understand and address the gap.
Inequalities
Developing a Service Planning Decision Support Tool to Tackle Inequalities and Minimise Carbon Output
The project explored the feasibility of considering inequalities and carbon emissions in new clinic locations. Achievements include automating a Health Equity Assessment, running logistic regression models to predict appointment no-shows, and estimating patient travel carbon emissions.
• Torbay Council
• West Sussex County Council
Using Machine Learning to estimate inequities in access to hospital procedures
The project aimed to determine disparities in accessing planned hospital appointments by calculating true admission rates for high deprivation areas. They found that minimal public data and simple models were insufficient. Future work includes creating synthetic data and exploring additional features like emergency admissions.
Creating a tool to automatically generate health equity audits for Community Diagnostic Centres
The project aimed to create a tool to perform a health equity audit for Community Diagnostic Centres (CDCs) in England. The tool will identify healthcare inequalities, suggests data improvements, and supports local CDCs in understanding their impact. Developed using Python and Streamlit, it allows for easier comparison and sharing of learning across regions, with potential applications beyond CDCs.
• Reading Borough Council
Developing a tool to assess inequalities and demographic coverage of service locations
The project aimed to empower NHS service providers with a tool to understand their patient population and direct service expansion to underserved groups. It combines geospatial analysis, demographic data, and travel calculations.
• Health Innovation Wessex
• Nottingham University Hospitals NHS Trust
Mapping health inequalities, depreviation, ethnicities and crime across the UK
The aim of the project is to use ONS data around ethnicities of population, health index England, Deprivation data from QOF, and overlay the same with crime statistics at a…
Identifying potential concurrent treatment areas and services that would better support patients with multiple, complex referral to treatment (RTT) pathways.
The aim of this project is to use machine learning approaches to support analysis of patients with multiple concurrent RTT pathways, focusing particularly on healthcare…
Automating and scaling waiting list equity analysis at a trust level
This project uses a streamlit app to automates/scales waiting list equity analysis at a trust level and (optionally) service level.
eFIT: Extra funding allocation - inequality tool
The project addresses the lack of national guidance for allocating extra primary care funding by ICBs. It proposes using an equation based on deprivation scores and local needs. A Streamlit web-app tool will help ICBs allocate funds more equitably, considering various indicators and demographics, ensuring a fair distribution and reducing inequalities.
Non-NHS Project Areas
Police
What are they saying about us? An AI tool to determine the sentiment of tweets to police forces across the country, and what people are talking about
This project used AI-based Natural Language Processing to develop a dashboard predicting tweet sentiment and identifying discussion topics for police forces. It transformed Avon and Somerset Constabulary's social media can respond to public concerns.
Generating a richer understanding of relationships in crime data in order to identify opportunities to safeguard individuals and families
This project used Network Analysis to identify social relationships and enable proactive police interventions for those at risk of offending due to their links with offenders. A proof of concept showed it could identify "at risk" individuals in hours instead of months. Applied in Devon, it highlighted significant benefits for future policing, including child protection and resource savings.
Spatial Modelling of Violent Crime to Support Strategic Analysis
The project aimed to use crime data and spatial analysis for intelligence reporting. Focusing on violent crime, the team used QGIS and Python libraries to analyse geographic offence data. The model identified hotspots, coldspots, and outliers, providing statistical confidence for intelligence reports.
• Devon & Cornwall Police
• Devon County Council
Using Natural Language Processing to detect drug related content within free text
The project aimed to detect drug-related content in datasets using Natural Language Processing to automate text classification. Initial findings were promising, though tested only on dummy data. If successful, the model could be adapted for other crime types.
Using Agent Based Simulation to understand criminal adoption of new technologies
This project tests agent-based modelling, using consumer behaviour theory, criminology, and domain knowledge. The project aims to create a general methodology for digital behaviour models, applicable to specific cases, and develop methods to present findings static and interactively.
Mapping health inequalities, depreviation, ethnicities and crime across the UK
The aim of the project is to use ONS data around ethnicities of population, health index England, Deprivation data from QOF, and overlay the same with crime statistics at a…
Council
Identifying unpaid carers in Norfolk
Unpaid carers face mental health risks, loneliness, and isolation. The project will identify carer locations using geographic modelling and predict high-need areas with regression models, focusing on older populations and unpaid carers.
Predictive modelling for smoking cessation success
Smoking cessation remains challenging despite public health efforts. This project aims to develop a predictive model to identify individuals likely to quit smoking based on demographics and behaviours. It will uncover key predictors and effective pathways using machine learning algorithms like logistic regression, decision trees, random forests, and neural networks, evaluating each for accuracy and interpretability.
Modelling strategies to reduce the elective backlog in hip surgery
This project used Discrete Event Simulation to model the hip surgery patient pathway in Exeter, assessing strategies to reduce the surgery backlog. It was a collaborative effort between the CCG and the acute provider, contributing to long-term planning and decision-making.
• NHS Devon ICB
Discrete Event Simulation to model elective surgery pathways
The project created a Discrete Event Simulation to model elective surgery pathways. This tool optimises surgical pathways and provides a ready-made format for creating interactive webapps, benefiting future HSMA participants and other interested users.
Exploring the use of Machine Learning and Natural Language Processing to teach a machine to predict whether a patient is likely to be imminently admitted to hospital based on GP data and clues in GP notes
This project used Natural Language Processing to automate key information extraction from patient notes, aiming to predict imminent hospital admissions. Findings included indicators like increased note frequency, verbosity, and specific words.
• Peninsula Dental Social Enterprise
• Royal Devon University Healthcare NHS Foundation Trust
Meeting the demand of 111 for primary care services
The project modelled NHS 111 calls triaged to primary care. Simulations showed timely primary care contact could reduce 999 calls and ED attendances but would require nearly doubling primary care services.
• NHS Devon ICB
• University of Plymouth
Modelling the impact of 111 and GP access on Emergency Departments
This project will use Discrete Event Simulation to model patients’ navigation of GP vs 111 services and assess the knock-on effect for Emergency Departments.
Modelling 111 option 2 call centre
The project aims to improve two struggling 111 call centres for mental health patients in Norfolk and Suffolk. It will develop a DES model to compare staffing approaches and determine the resources needed for safe service. Additionally, a tool will be created to ensure safe rosters by inputting current rosters and forecasting call levels.
111 Downstream forecasting
This project aims to develop a model predicting the place, time, and volume of downstream activity from 111 calls, deployed as a secure API. It integrates with an existing UEC webapp, using weekly patient-level 111 call data and near-live feeds.
Population segmentation of GP-registered population in Dorset
This project aims to develop a machine learning-based population segmentation model for the GP-registered Dorset population, using multiple characteristics including healthcare utilisation. The goal is to identify segments based on care needs to inform service design, enhancing patient understanding and improving resource allocation.