mres projects pages (#307)

* mres projects pages

* remove reference

* remove significant

Author: amaiaita

docs/images/mh-in-leeds.png

@@ -0,0 +1,28 @@
+---
+title: 'Understanding delays in elective pathways'
+summary: 'A case study of lithotripsy using NHS England data and process mining and machine learning techniques'
+tags: ['SECONDARY CARE', 'FORECASTING', 'PROCESS MINING', 'MODELLING', 'STRUCTURED DATA', 'PYTHON', 'PYSPARK', 'COMPLETE']
+---
+*This project was completed by Evelyn Koon, Analyst, in the Elective Analysis Team, as part of the Data Science MRes at the University of Leeds.*
+
+## Background
+The UK’s National Health Service (NHS) provides publicly funded healthcare free at the point of use. In 2025, NHS England introduced the 'Reforming Elective Care for Patients' (NHS England and Department of Health and Social Care, 2025) initiative, aimed at enhancing the delivery of elective services and reducing waiting times. Urology was one of the first five specialties chosen for reform due to its large and growing waiting list (NHS England and Department of Health and Social Care, 2025). Using national data, this project examines patient pathways for Extracorporeal Shock Wave Lithotripsy (ESWL) within Urology, aiming to understand why some patients remained on the waiting list for over 18 weeks.
+
+## Methods
+
+NHS England has introduced a new dataset, the Elective Patient Pathway (EPP), designed to capture activities within the elective care journey. This project applied process mining techniques to the national dataset, extending the scope beyond a single site. Directly Follows Graphs (DFGs) were used with input from NHS experts, new variables were derived and further examined using machine learning techniques to evaluate the association of these new variables on elective waiting times. Additionally, conformance analysis was used to compare individual hospitals with the longest and shortest durations on the elective waiting list, highlighting variations in care processes.
+
+## Results 
+
+Incorporating four of the additional activity types, which were non-lithotripsy appointments and assigning sequential identifiers to lithotripsy treatments (e.g., ESWL_1  or lit_1 for the first session), enabling the discovery of more meaningful care pathways. The machine learning models identified two of these as the most predictive features when they were tested. These were the other last appointment, which occurred last out of all non-lithotripsy appointments, and the first other outpatient routine appointment.
+Patients who waited less than 18 weeks generally followed similar sequences to those who waited longer, though extended delays were concentrated in the early stages of the pathway, particularly after being added to the waiting list. Conformance analysis across hospitals revealed that more complex pathways, with more activity, took longer to complete. Much of this extended activity occurred after patients left the waiting list, indicating the potential impacts of more complicated pathways on service provision.
+
+## Conclusions
+
+By combining national data with clinical expertise, this study uncovered variation in patient pathways, particularly for those with complex needs navigating the lithotripsy pathway. Process mining and machine learning revealed disparities in waiting list durations and activities that influence time a patient waited for their treatment, while conformance analysis highlighted variations in patient pathways. Grounding the dataset in real-world clinical practice ensured operational relevance to improve NHS performance. These insights can support frontline decision-making and offer promising applications in the NHS, especially with the development of new pathway datasets.
+
+Output|Link
+---|---
+Code and Documentation - private while under development|[Link](https://github.com/nhsengland/PM4_elective_care.git)
+
+#

docs/mres/elective-pathways-delays.md

@@ -0,0 +1,24 @@
+---
+title: 'Investigating unknown ethnicity records in NHS emergency care data: patterns, predictors and missingness '
+summary: 'Supporting the improvement of ethnicity recording in emergency care data by: (a) investigating variation in unknown ethnicity records, (b) exploring patterns of missingness between ethnicity and other variables, and (c) identifying the variables most important in the recording of an unknown ethnicity. '
+tags: ['PUBLIC/POPULATION HEALTH', 'EMERGENCY CARE', 'CLASSIFICATION', 'MODELLING', 'STRUCTURED DATA', 'PYTHON', 'R', 'COMPLETE']
+---
+*This project was completed by Grace Dean, Senior Analyst, in the South East Data & Analytics Team Team, as part of the Data Science MRes at the University of Leeds.*
+
+High quality ethnicity data is necessary for tackling health inequalities. This project looked at unknown ethnicity recording (where the ethnicity is either Null, Not Stated or Not Known) and was focused on three objectives: 
+
+1.	Investigating variation in unknown ethnicity recording in ECDS across different organisation types, patient demographics and attendance characteristics.
+2.	Looking at the patterns of missingness between ethnicity and other variables. 
+3.	Identifying the factors which most contribute to ethnicity being recorded as unknown. 
+
+## Main Findings
+
+1.	Variation in unknown ethnicity recording was observed across demographics and attendance characteristics, with younger patients, those requiring low-level emergency care and those with fewer attendances more likely to have an unknown ethnicity record. Differences were also observed across organisation type, with higher rates of unknown ethnicity recording in independent sector organisations and community trusts compared to acute trusts. 
+2.	Missing data in other ECDS variables was more common and more extensive when ethnicity was unknown. Increases in missingness were observed in variables relating to the investigations and treatments received by the patient, the severity of the patient’s condition, the chief complaint on arrival and place of discharge.
+3.	Provider site code was identified as the most important variable in recording ethnicity as unknown, and a subset of acute trusts and independent sector organisations were identified as disproportionately contributing to unknown ethnicity records.
+
+Output|Link
+---|---
+Code and Documentation - private while under development|[Link](https://github.com/nhsengland/ethnicity-coding-ECDS)
+
+#

docs/mres/ethnicity.md

@@ -8,13 +8,19 @@ A showcase of some of these projects can be found in these pages.
 
 - [Prediction of CVD Onset](cvd-onset.md)
 - [Supporting Dementia Diagnosis](dementia.md)
+- [Understanding Delays in Elective Pathways](elective-pathways-delays.md)
+- [Investigating unknown ethnicity records in NHS emergency care data](ethnicity.md)
 - [Predicting GP Staff Turnover](gp-turnover.md)
 - [Menopause-Related Diagnosis Coding](menopause.md)
+- [Environmental Impacts on Mental Health in Leeds](mh_leeds.md)
 - [Impact of midwife-led continuity of carer on birth outcomes](midwives.md)
 - [Predicting perinatal depression](perinatal.md)
 - [Poor pregnancy outcomes in women with type 2 diabetes](pregnancy-outcomes.md)
 - [Process Mining on Patient Pathways in Healthcare](process-mining.md)
+- [Relationship between Psychotropic Medication Usage and Talking Therapies Treatment Outcomes](psychotropic-meds-talking-therapies.md)
+- [Factors related to Stillbirth Outcomes](stillbirths.md)
 - [UTI Surgery Risk Predictions](uti.md)
+- [Predicting Winter Pressures on Emergency Admissions](winter-pressures.md)
 
 
 !!! note

docs/mres/index.md

@@ -0,0 +1,17 @@
+---
+title: 'Mental health and its relationship with noise and air pollution and the deprivation index in Leeds'
+summary: 'Mental health and its relationship with noise and air pollution and the deprivation index in Leeds at two geographical scales.'
+tags: ['MENTAL HEALTH', 'MACHINE LEARNING', 'LIKAGE', 'REGRESSIONS', 'GEOGRAPHICAL ANALYSIS', 'LINKAGE', 'RESEARCH', 'STRUCTURED DATA', 'PYTHON', 'SQL', 'GIS', 'COMPLETE']
+---
+*This project was completed by Chloe Peaker, Net Zero Estates Senior Insight Manager, in the Estates - Soft FM Team, as part of the Data Science MRes at the University of Leeds.*
+
+![test](../images/mh-in-leeds.png)
+
+There is a growing mental health crisis in the UK that is exacerbating an already strained healthcare system and there is also pressure for continuous growth leading to more and more urbanisation. This study therefore explores the relationship between environmental factors and mental health referrals for Leeds for the financial year 2023/24 using two geographical scales: 1 km square grid and Lower Super Output Areas (LSOAs). 
+
+This study shows that noise pollution alone showed a weak positive correlation when aggregated to 1 km but this was diminished when the same data was aggregated to LSOA level. A gender disparity was found warranting further investigation and potentially more proactive outreach to males. The analysis also highlighted key cleaning steps for the Patient Level Answering New Questions (PLANQ) dataset and that more expansive noise, 
+air and Index of Multiple Deprivation (IMD), variables should be included for future analysis. As expected rates of mental health referrals were found to correlate with the IMD, though limitation in IMD granularity and outdated data raises concern for its use in transport and health policy. 
+
+Air pollutants including PM2.5, PM10, ozone, carbon monoxide, benzene, sulphur dioxide and nitrogen oxides, and noise pollution, were found to account for 16% of variance in mental health referrals, suggesting there is a correlation but no causal link was investigated, a geographic map of residual errors was also presented. These results offer valuable insights for future research particularly in a new geographic domain explored using geographical statistics. The results also hold relevance for policy development across both the healthcare and transport domain.
+
+#

docs/mres/mh_leeds.md

@@ -0,0 +1,36 @@
+---
+title: 'Exploring the relationship between psychotropic medication usage and Talking Therapies treatment outcomes: a retrospective cohort study'
+summary: 'Exploring medication use as a predictor of reliable recovery following completion of a course of Talking Therapies, using logistic regression and causal analysis'
+tags: ['PRIMARY CARE', 'MENTAL HEALTH', 'MACHINE LEARNING', 'MODELLING', 'RESEARCH', 'STRUCTURED DATA', 'R', 'PYSPARK', 'SQL', 'COMPLETE']
+---
+*This project was completed by CHRISTINE WELLS, Analyst, in the Modelling & System Analytics Team, as part of the Data Science MRes at the University of Leeds.*
+
+NHS Talking Therapies (NHSTT) provide psychological support to people with anxiety and depressive disorders. The ideal outcome, following a completed course of treatment (minimum 2 sessions) is to achieve ‘reliable recovery’. Psychological therapy may sometimes be supplemented by prescribing psychotropic medication such as antidepressants. At each appointment patients are asked whether they are prescribed and taking any psychotropic medication, although no specific details of any medication is collected and responding is not mandatory. Whilst this information has sometimes been included as a predictor in machine learning-based analysis of therapeutic outcomes, there has been less focus on understanding how outcomes differ between medicated and non-medicated groups. 
+
+## Objectives: 
+
+* To use national NHSTT data to train and test a logistic regression model that included medication status (taking/not) as a predictor of the odds of achieving reliable recovery. 
+* Use propensity score matching to undertake causal analysis of the impact of medication on reliable recovery. 
+
+## Design: 
+A retrospective cohort study of 2½ years’ worth of patients who had completed treatment. 
+
+## Setting: 
+Data was extracted from the NHS England NHSTT database for patients who had completed treatment between 1st January 2022 and 31st May 2025. 
+
+## Participants: 
+The cohort was N = 1,120,839 for the logistic regression (669,142 = taking medication, 451,697 = not taking medication) and N = 54,802 for the matching analysis (N = 27,401 per group). 
+
+## Interventions: 
+The exposure was whether patients were taking psychotropic medication during treatment or not, and whether they received high or low intensity therapy. 
+
+## Primary outcome measure: 
+Whether patients achieved reliable recovery by completion of treatment. Results: Logistic regression on the cohort showed that medication was associated with reduced odds of recovery. Logistic regression on the matched groups suggested:
+
+1. a small but beneficial effect of medication, although this result was non-significant, and 
+2. taking medication in combination with high intensity therapy was more beneficial than high intensity therapy alone. 
+
+## Conclusions: 
+The results support previous research regarding other predictors associated with increased odds of recovery, including number of treatment sessions and lower levels of deprivation. Whilst the medication information contained in the NHSTT dataset is limited, there is scope to expand this analysis through linkage to prescribing data. Enhancing our understanding of how medication status relates to therapeutic outcomes has important implications for interpretation of NHSTT recovery rates.
+
+#

docs/mres/psychotropic-meds-talking-therapies.md

@@ -0,0 +1,19 @@
+---
+title: 'Study Relationship Between Continuity of Midwife Care, Deprivation, Ethnicity, and Stillbirth Outcomes'
+summary: 'To investigate the association between a data-driven measure of care continuity, socioeconomic deprivation, ethnicity, and other key factors with the risk of stillbirth, using the Maternity Services Dataset maintained by NHS England.'
+tags: ['SECONDARY CARE', 'MACHINE LEARNING', 'CLASSIFICATION', 'EXPLAINABILITY', 'RESEARCH', 'STRUCTURED DATA', 'PYTHON', 'COMPLETE']
+---
+*This project was completed by Ketankumar Italiya, Senior Data Architect, in the Metadata, Data Quality & Standards Team, as part of the Data Science MRes at the University of Leeds.*
+
+In my capacity as a Data Engineer, my previous experience working with the NHS Maternity Services Data Set (MSDS) and the research conducted by Chris Roebuck inspired me to pursue further investigation into this domain. I was further encouraged by the collaboration and sponsorship of two relevant senior staff. Their support provided the necessary confidence to move forward with this strategic initiative.
+
+The analysis was conducted on a filtered cohort of 773,786 pregnancies (including 2,233 stillbirths) where the number of unique care activities was greater than two, ensuring a sufficient basis for continuity analysis.
+
+We found a mix-trend relationship between the care continuity cohorts and the stillbirth rate. The stillbirth rate was lowest (0.186%) with 'Fragmented Care' (care from >6 midwives), rising to 0.339% with 'Good Continuity' (≤3 midwives) and peaking at 0.416% with 'Low Continuity' (≤6 midwives). The analysis by subgroups revealed that the impact of ethnicity, deprivation, and maternal age on stillbirth risk varied significantly across these continuity cohorts. For example, after multivariable adjustment, the odds ratio for stillbirth among Black women (compared to White women) was highest in the 'Low Continuity' cohort (OR 1.806). Predictive models achieved high accuracy (0.9972) but Area Under the Curve (AUC) is 0.500, indicating no predictive power is not helpful.
+
+Output|Link
+---|---
+Published paper on the Impact of midwife continuity of carer on stillbirth rate and first feed in England
+|[Link](https://www.nature.com/articles/s43856-025-01025-z)
+
+#

docs/mres/stillbirths.md

@@ -0,0 +1,18 @@
+---
+title: 'Predicting Winter Pressures on Emergency Admissions in England Using Machine Learning'
+summary: 'A 4-year national study across all NHS Trusts in England using Daily Emergency Care Data Set (ECDS) linked with UK Met Office weather data to predict winter pressures on Emergency admissions using Machine Learning to improve planning and response in the Emergency Department.'
+tags: ['SECONDARY CARE', 'EMERGENCY CARE', 'PUBLIC/POPULATION HEALTH', 'MACHINE LEARNING', 'RESEARCH', 'LINKAGE', 'MODELLING', 'STRUCTURED DATA', 'TIME SERIES', 'PYTHON', 'SQL', 'COMPLETE']
+---
+*This project was completed by Joby Jose, Senior Analytics Developer, in the Enterprise Data Visualisation Team, as part of the Data Science MRes at the University of Leeds.*
+
+Winter pressures present a continued challenge to the National Health Service (NHS) in England, as cold weather causes an increase in Emergency Department (ED) admissions. This study investigates the relationship between winter weather conditions and ED admissions across all NHS trusts in England. It aims to quantify the influence of temperature, rainfall, and snowfall on ED admissions and assess the predictive value of weather features using machine learning.
+
+Daily Emergency Care Data Set (ECDS) records were linked with UK Met Office weather data, assigning mean daily temperature, rainfall, and snowfall to each trust. Feature engineering included lagged weather variables, calendar effects, and public holidays. An Extreme Gradient Boosting model was trained and optimised using GridSearchCV with time-series cross-validation and interpreted using SHapley Additive exPlanations (SHAP).
+
+The model achieved an RMSE of 18.78 and MAE of 10.6 admissions per trust per day. SHAP analysis showed that temperature, and its lagged values were the strongest predictors, with colder weather linked to higher admissions. Subgroup analysis showed diagnosis-specific temperature effects. Circulatory admissions increased 4-6 days after temperatures of 2-12°C; infectious diseases and injuries responded to shorter lags; and respiratory admissions peaked 7 days after temperatures below -2°C. Rainfall had a moderate influence and snowfall had a marginal influence. These findings support the integration of weather-based forecasting into NHS planning tools.
+
+Output|Link
+---|---
+Code available on the GitHub: | MRes_Project/Predicting_Winter_Pressures.ipynb at main jobyjose6/MRes_Project 
+
+#

docs/mres/winter-pressures.md

@@ -187,13 +187,19 @@ nav:
     - MRes Projects:
       - Prediction of CVD Onset: mres/cvd-onset.md
       - Supporting Dementia Diagnosis: mres/dementia.md
+      - Understanding delays in elective pathways: mres/elective-pathways-delays.md
+      - Investigating unknown ethnicity records in NHS emergency care data: mres/ethnicity.md
       - Predicting GP Staff Turnover: mres/gp-turnover.md
       - Menopause-Related Diagnosis Coding: mres/menopause.md
+      - Mental Health Environmental Impacts Leeds: mres/mh_leeds.md
       - Impact of midwife-led continuity of carer on birth outcomes: mres/midwives.md
       - Predicting perinatal depression: mres/perinatal.md
       - Poor pregnancy outcomes in women with type 2 diabetes: mres/pregnancy-outcomes.md
       - Process Mining on Patient Pathways in Healthcare: mres/process-mining.md
+      - Relationship between psychotropic medication usage and Talking Therapies treatment outcomes: mres/psychotropic-meds-talking-therapies.md
+      - Relationship Between Factors and Stillbirth outcomes: mres/stillbirths.md
       - UTI Surgery Risk Predictions: mres/uti.md
+      - Predicting Winter Pressures on Emergency Admissions: mres/winter-pressures.md
   - Articles:
     - articles/index.md
   - Playbooks: playbooks.md