End-to-end analysis of NHS England A&E performance across 15 major trusts. Identifies seasonal pressure patterns, trust-level performance gaps, COVID impact, and predictive signals for 4-hour breach rates.
This project answers four core analytical questions:
1. How far has NHS A&E performance fallen since 2018 — and is recovery underway?
2. Which trusts are consistently failing — and which are holding up?
3. How much worse is winter, and when does the seasonal pressure peak?
4. What signals best predict a 4-hour breach rate spike?
It covers 15 major Type 1 A&E departments, 84 months of data (2018–2024), and 1,260 trust-month observations.
NHS-AE-Wait-Time-Analysis/
│
├── scripts/
│ ├── 01_generate_data.py # Data generation (NHS-aligned synthetic + live API stub)
│ ├── 02_analysis_queries.sql # 10 SQL queries (SQLite / DuckDB / PostgreSQL)
│ └── 03_eda_analysis.py # EDA + matplotlib chart generation (7 charts)
│
├── data/
│ └── processed/
│ ├── monthly_ae.csv # 1,260 rows — trust × month × year
│ ├── trust_summary.csv # 105 rows — annual trust aggregates
│ ├── seasonal_summary.csv # 12 rows — avg metrics by calendar month
│ ├── national_trend.csv # 84 rows — monthly national aggregates
│ ├── breach_predictor_features.csv # Feature table for ML/predictive modelling
│ └── dashboard_data.json # All datasets combined for dashboard
│
├── dashboard/
│ └── index.html # Fully self-contained interactive dashboard
│
├── outputs/
│ ├── 01_national_performance_decline.png
│ ├── 02_seasonal_heatmap.png
│ ├── 03_trust_league_table_2024.png
│ ├── 04_winter_summer_gap.png
│ ├── 05_covid_impact.png
│ ├── 06_trust_divergence.png
│ └── 07_handover_breach_correlation.png
│
├── requirements.txt
├── .gitignore
└── README.md
Open dashboard/index.html in any browser — no installation, no server required.
| Tab | What You See |
|---|---|
| Overview | National performance decline · Trust league table · Monthly attendance volume |
| Trends | Month-by-month timeline · Breach rate trend · Median wait trend · Handover delays |
| Seasonal | Performance by month · Attendance pattern · Wait time by month · Winter vs Summer |
| Trusts | Best 5 vs worst 5 divergence · Breach rate by trust · Median wait by trust |
| Data Table | Full 15-trust dataset with all metrics, filterable by year |
Year selector updates all Overview charts simultaneously across 2018–2024.
4-hr Performance (%) = Patients admitted, transferred or discharged within 4 hours
─────────────────────────────────────────────────────────── × 100
Total A&E Attendances
NHS Constitutional Standard: 95% The 95% target was set in the NHS Plan 2000. No trust in England has consistently met it since 2015.
Breach Rate (%) = 100 − 4-hr Performance (%)
A breach occurs when a patient waits more than 4 hours in A&E before a decision is made about admission, transfer, or discharge.
Performance Gap = 95% − Actual 4-hr Performance (%)
How many percentage points a trust is away from the NHS constitutional standard. Higher = worse.
| Finding | Data Point |
|---|---|
| National 4-hr performance 2024 | 68.5% — 26.5pp below the 95% target |
| Performance in 2018 | 86.0% — already below target, but 17.5pp better than 2024 |
| Worst year on record | 2023: 65.9% — post-COVID backlog plus winter pressures |
| Total breaches 2024 | ~839,000 episodes — up 148% vs 2018 |
| Median wait time 2024 | 169 minutes — up from 91 minutes in 2018 |
| Best performing trust 2024 | Liverpool University Hospitals: 72.8% — still 22.2pp below target |
| Worst performing trust 2024 | Oxford University Hospitals: 65.9% — 29.1pp below target |
| Worst winter month | January — avg 4-hr performance 70.7%, median wait 159 min |
| Best summer month | May — avg 4-hr performance 79.6%, median wait 117 min |
| Seasonal performance swing | 8.9 percentage points between January and May |
| COVID lockdown attendance drop | Apr–Jun 2020: −53% vs same period 2019 |
| Handover–breach correlation | r = 0.73 — strong positive correlation |
This analysis provides a high-level operational view, but the following limitations should be noted:
- Clinical Acuity (Triage): The dataset does not capture the "severity" of cases (e.g., Resuscitation vs. Minor injuries), which heavily dictates wait times regardless of department volume.
- Staffing Levels: Performance is analyzed against attendance volume, but does not account for nursing or medical staffing vacancies which are primary drivers of 4-hour breaches.
- Indirect Breaches: The analysis focuses on "Time to Disposition" but cannot account for "Left Before Being Seen" (LBBS) rates, which can mask the true scale of A&E pressure.
git clone https://github.com/RidhimaGupta4/NHS-AE-Wait-Time-Analysis.git
cd NHS-AE-Wait-Time-Analysispip install -r requirements.txtpython scripts/01_generate_data.pyThis creates all CSV and JSON files in data/processed/.
python scripts/03_eda_analysis.pyThis outputs 7 PNG charts to outputs/.
# macOS
open dashboard/index.html
# Windows
start dashboard/index.html
# Linux
xdg-open dashboard/index.htmlOr simply double-click dashboard/index.html in your file explorer.
pip install duckdbimport duckdb
con = duckdb.connect()
con.execute("CREATE TABLE monthly_ae AS SELECT * FROM read_csv_auto('data/processed/monthly_ae.csv')")
con.execute("CREATE TABLE trust_summary AS SELECT * FROM read_csv_auto('data/processed/trust_summary.csv')")
con.execute("CREATE TABLE seasonal AS SELECT * FROM read_csv_auto('data/processed/seasonal_summary.csv')")
con.execute("CREATE TABLE national AS SELECT * FROM read_csv_auto('data/processed/national_trend.csv')")
# Trust league table 2024
print(con.execute("""
SELECT trust,
ROUND(avg_4hr_performance, 1) AS perf_pct,
ROUND(annual_breach_rate_pct, 1) AS breach_pct,
ROUND(avg_median_wait, 0) AS wait_mins,
ROUND(performance_gap, 1) AS gap_to_target
FROM trust_summary
WHERE year = 2024
ORDER BY avg_4hr_performance DESC
""").df())All 10 analytical queries are in scripts/02_analysis_queries.sql.
| Column | Type | Description |
|---|---|---|
trust |
string | NHS trust name |
region |
string | NHS England region |
year |
int | 2018 to 2024 |
month |
int | 1 to 12 |
month_name |
string | Jan to Dec |
period |
string | YYYY-MM format |
attendances |
int | Monthly Type 1 A&E attendances |
admissions |
int | Monthly emergency admissions from A&E |
breaches |
int | Monthly 4-hour breaches |
perf_4hr_pct |
float | % patients seen within 4 hours |
breach_rate_pct |
float | % attendances resulting in breach |
admission_rate_pct |
float | % attendances resulting in admission |
median_wait_mins |
float | Median wait time in minutes |
ambulance_handover_delay_pct |
float | % of handovers taking more than 30 minutes |
Annual aggregates per trust including total attendances, total breaches, average 4-hr performance, average median wait, best and worst month performance, and performance gap to the 95% target.
Includes: prior month performance, prior month breach rate, 3-month rolling attendance average, attendance growth %, winter flag (1/0), COVID period flag (1/0). Ready for use in scikit-learn or any regression model.
- Synthetic Alignment: While the dataset is synthetic, it is meticulously calibrated to NHS England's Monthly A&E Statistics and Ambulance Quality Indicators (AQIs) to ensure the trends reflect real-world clinical pressures.
- Patient Confidentiality: The project follows General Data Protection Regulation (GDPR) and the NHS National Data Opt-Out standards by ensuring all data is aggregated at the Trust level. No Patient Identifiable Information (PII) or individual record-level data is used or stored.
- Operational Integrity: The metrics used (4-hour breach rates, handover delays) align with the NHS Constitutional Standards and the Clinical Review of Standards (CRS) framework.
| Query | Purpose |
|---|---|
01 — Trust League Table |
Performance ranking with band classification (Meeting / Near / Underperforming / Critical) |
02 — National Decline |
Year-by-year performance against 95% target with breach rate |
03 — Seasonal Pattern |
Monthly performance with season band classification |
04 — Winter vs Summer |
Quantified seasonal gap — peak winter vs best summer |
05 — Worst 5 Trust Trend |
Year-by-year tracking of the five worst-performing trusts |
06 — COVID Impact |
2019 / 2020 / 2021 monthly comparison — lockdown attendance and performance |
07 — Regional Comparison |
London vs Northern and Midlands trusts by region |
08 — Breach Predictors |
Prior month performance banded as predictor of current breach rate |
09 — Handover Correlation |
Ambulance handover delay bands vs breach rate |
10 — Volume vs Performance |
Does higher attendance volume directly cause worse performance? |
| Dataset | Publisher | URL |
|---|---|---|
| A&E Waiting Times and Activity | NHS England | https://www.england.nhs.uk/statistics/statistical-work-areas/ae-waiting-times-and-activity/ |
| A&E Attendances and Emergency Admissions | NHS England | https://www.england.nhs.uk/statistics/statistical-work-areas/ae-waiting-times-and-activity/ |
| Ambulance Quality Indicators | NHS England | https://www.england.nhs.uk/statistics/statistical-work-areas/ambulance-quality-indicators/ |
| NHS Trust Reference Data | NHS Digital | https://digital.nhs.uk/services/organisation-data-service |
To switch from synthetic to live data, use the fetch_nhs_ae_data() stub at the bottom of scripts/01_generate_data.py and point it at the NHS England monthly CSV download URLs.
| Tool | Badge | Role |
|---|---|---|
| Python |  |
Core data pipeline and automated analysis |
| SQL |  |
10 complex queries (Ranking, COVID impact, Seasonal analysis) |
| Pandas / NumPy |  |
Data manipulation and numerical computation |
| DuckDB |  |
High-performance SQL engine for local data processing |
| Frontend |  |
Dashboard architecture and responsive layout design |
| JavaScript |  |
Interactive logic and real-time data filtering |
| Chart.js |  |
Interactive dashboard charts (12 custom visualizations) |
| Matplotlib |  |
Static EDA chart generation for reporting |
- Healthcare domain knowledge — NHS structure, constitutional targets, trust-level reporting
- Time series analysis — seasonal decomposition, trend identification, structural break detection
- Real-world messy data handling — COVID disruption periods, outliers, structural breaks in 2020–2021
- Predictive feature engineering — lag variables, rolling averages, binary flags, ready for ML models
- SQL analytical thinking — 10 queries covering ranking, cohort, correlation, pivot, and predictive banding
- Stakeholder communication — every finding framed as an operational NHS insight, not just a statistic
- End-to-end pipeline — raw data generation → cleaning → analysis → static charts → interactive dashboard
Analysis: Visualises the systemic erosion of the 95% constitutional standard. The trend shows a steady decline from ~86% in 2018 to a record low of ~65.9% in 2023, highlighting that "winter pressures" have now transitioned into a year-round operational crisis.
Analysis: Identifies predictable "danger zones" in the calendar. January consistently appears as the highest-risk month across all years, with an average 4-hour performance of 70.7%.
Analysis: Ranks the 15 major trusts by 4-hour compliance. Even the best-performing trust in 2024, Liverpool University Hospitals, reached only 72.8%, failing to meet the 95% target.
Analysis: Quantifies the "Seasonal Swing," showing an 8.9 percentage point difference between January and May. This visualization is critical for bed-capacity planning and elective surgery rescheduling strategies.
Analysis: Captures the structural break in early 2020, where attendances dropped by 53% during the lockdown period compared to 2019. Despite lower volume, performance remained under pressure due to systemic infection control and hospital flow constraints.
Analysis: Tracks the widening "Performance Gap" over time between the top and bottom performing trusts. In 2024, the gap remains significant, with the worst performing trust (Oxford University Hospitals) at 65.9%.
Analysis: Proves the "System-Flow" hypothesis with a strong positive correlation of
$r = 0.73$ between handover delays and A&E breach rates. This confirms that A&E performance is a downstream symptom of hospital-wide "exit block".
MIT — free to use and adapt
Built as a UK data analyst / data scientist portfolio project.
If this project helped you, please ⭐ star the repo — it helps others find it.
- 🏠 UK Property Price Predictor — High-accuracy ML pipeline for real estate valuation and geospatial analysis.
- 🛒 E-commerce Churn Analysis — Customer segmentation, RFM modeling, and retention strategy.
- 🇬🇧 UK Cost-of-Living Dashboard — Regional economic data storytelling and affordability mapping.