Environmental Hazards and Respiratory Mortality

Author

Marco Sorbona, PhD

Published

March 1, 2026

Introduction

This analysis investigates whether air pollution (PM2.5) explains part of the North‑South divide in respiratory mortality. Previous work established that northern regions have 48% higher mortality than southern regions (p < 0.001, Cohen’s d = 2.4). This project adds an environmental exposure layer to test whether pollution is a contributing factor.

Rationale: Understanding the drivers of regional health inequalities is essential for targeting interventions. If pollution explains part of the gap, reducing emissions in northern regions could improve respiratory health.

Data Preparation

PM2.5 data were downloaded from 66 urban background monitoring sites across all 9 English regions using the openair package. Annual regional averages were calculated from hourly data (2015-2024). Deprivation data (IMD 2019) were obtained from Fingertips to adjust for confounding.

Data Sources

Data	Source	Period
Respiratory mortality	Fingertips (Indicator 40701)	2015-2024
Air pollution (PM2.5)	UK-AIR via `openair` package	2015-2024
Deprivation (IMD 2019)	Fingertips (Indicator 93553)	2019

All data were accessed programmatically, cached, and are fully reproducible.

Methods

Pollution Data

PM2.5 concentrations were downloaded from 66 monitoring sites across all 9 English regions using the openair::importUKAQ() function. Only urban background sites were included to represent population exposure. Annual regional averages were calculated from hourly data.

Statistical Analysis

Simple correlation (Pearson’s r) between PM2.5 and mortality
Partial correlation adjusting for deprivation (IMD 2019 scores)
Time series analysis of the association from 2015-2024

Deprivation is a known confounder: more deprived areas tend to have both higher pollution and worse health outcomes. Partial correlation isolates the pollution-mortality relationship by holding deprivation constant.

# Calcualte annual average PM2.5 for each region
annual_pollution <- pollution_raw |> 
  # Add region names by matching site codes
  left_join(
    site_region_lookup,
    by = c("code" = "code")
  ) |> 
  # Extract year from datetime
  mutate(
    year = year(date)
  ) |> 
  # Group by region and year
  group_by(zone, year) |> 
  summarise(
    # Mean PM2.5 across all sites and hours in that region-year
    avg_pm25 = mean(pm2.5, na.rm = TRUE),
    # Number of monitoring sites contributing
    n_sites = n_distinct(code),
    # Number of hourly observations
    n_obs = n(),
    .groups = "drop"
  ) |> 
    # Keep only years with at least 100 observations (reliable estimate)
  filter(n_obs > 100)

# View results
annual_pollution |> 
  arrange(zone, year) |> 
  head(10)

# A tibble: 10 × 5
   zone           year avg_pm25 n_sites n_obs
   <chr>         <dbl>    <dbl>   <int> <int>
 1 East Midlands  2015    10.8        3 24673
 2 East Midlands  2016    11.2        3 26352
 3 East Midlands  2017    10.2        4 32880
 4 East Midlands  2018    10.3        4 35040
 5 East Midlands  2019    10.5        4 35040
 6 East Midlands  2020     8.33       4 35136
 7 East Midlands  2021     8.12       4 35040
 8 East Midlands  2022     8.43       4 35040
 9 East Midlands  2023     7.72       5 35952
10 East Midlands  2024     7.29       7 46451

Results

Pollution and Mortality Trends

PM2.5 concentrations declined across all regions between 2015 and 2024, reflecting national air quality improvements. Northern regions consistently had higher PM2.5 levels than southern regions.

Respiratory mortality also declined over the period, but the North‑South gap persisted. The North had 48% higher mortality than the South in 2024 (p < 0.001, Cohen’s d = 2.4).

1. Pollution-Mortality Positivite Association (2024)

The scatter plot shows a positive association between PM2.5 and respiratory mortality in 2024. Regions with higher air pollution tend to have higher death rates. London is an outlier – moderate pollution but low mortality (likely due to healthcare access and demographic factors).

# Scatter Plot for the Latest Year
# Filter to latest year
latest_year <- max(combined_data$year, na.rm = TRUE)

combined_latest <- combined_data |> 
  filter(year == latest_year)

# Create scatter plot
sp <- ggplot(combined_latest, aes(x = avg_pm25, y = mortality_rate,
                            label = region,
                            text = paste("Region: ", region,
                                         "<br>PM2.5: ", round(avg_pm25, 1), "\u03bcg/m3",
                                         "<br>Mortality: ", round(mortality_rate, 1),
                                         "<br>Sites: ", n_sites
                                         ))) +
  geom_point(aes(size = n_sites), color = "steelblue", alpha = 0.7) +
  geom_smooth(method = "lm", se = TRUE, color = "red", alpha = 0.2) +
  scale_size_continuous(range = c(3, 8), name = "Number of monitoring sites") +
  labs(
    title = "Air Pollution and Respiratory Mortality in English Regions",
    subtitle = paste("PM2.5 concentration vs mortality rate (", latest_year, ")", sep = ""),
    x = "Average PM2.5 (\u03bcg/m3)",
    y = "Respiratory mortality rate (per 100,000)"
  ) +
  theme_minimal() +
  theme(legend.position = "bottom")

# Make interactive
ggplotly(sp, tooltip = "text")

`geom_smooth()` using formula = 'y ~ x'

Association between PM2.5 and respiratory mortality

cor_test <- cor.test(combined_latest$avg_pm25, combined_latest$mortality_rate, 
                     method = "pearson")

# Display results
print(cor_test)


    Pearson's product-moment correlation

data:  combined_latest$avg_pm25 and combined_latest$mortality_rate
t = 1.8466, df = 7, p-value = 0.1073
alternative hypothesis: true correlation is not equal to 0
95 percent confidence interval:
 -0.1480588  0.8959198
sample estimates:
      cor 
0.5723407

# Extract key values
cor_estimate <- round(cor_test$estimate, 3)
cor_pvalue <- format.pval(cor_test$p.value, digits = 3)

print(paste("Correlation coefficient (r):", cor_estimate))

[1] "Correlation coefficient (r): 0.572"

print(paste("p-value:", cor_pvalue))

[1] "p-value: 0.107"

The Pearson correlation coefficient is 0.572 (p = 0.107). While not statistically significant at the 0.05 level due to the small number of regions (n = 9), the positive direction is consistent with the hypothesis that air pollution affects respiratory health.

2. Pollution and mortality move together over time

The faceted plot below shows PM2.5 (blue) and mortality (red) trends for each region. Both declined from 2015-2019, dropped sharply in 2020 (lockdowns), and have since stabilised. The North‑South gap is visible across all years – northern regions have higher values for both variables.

# Prepare data for plotting
time_series_data <- combined_data |> 
  dplyr::select(region, year, avg_pm25, mortality_rate) |> 
  pivot_longer(
    cols = c(avg_pm25, mortality_rate),
    names_to = "variable",
    values_to = "value"
  ) |> 
  mutate(
    variable = ifelse(variable == "avg_pm25", "PM2.5 (μg/m³)", "Mortality rate (per 100k)")
  )

# Create faceted plot
ggplot(time_series_data, aes(x = year, y = value, color = variable)) +
  geom_line(linewidth = 1) +
  geom_point(size = 1.5) +
  facet_wrap(~region, scales = "free_y") +
  scale_color_manual(values = c("PM2.5 (μg/m³)" = "#3498db", 
                                 "Mortality rate (per 100k)" = "#e74c3c")) +
  labs(
    title = "PM2.5 and Respiratory Mortality Trends by Region",
    subtitle = "Blue = PM2.5 concentration, Red = Mortality rate. Note: Y-axes vary by region.",
    x = "Year",
    y = "Value",
    color = ""
  ) +
  theme_minimal() +
  theme(
    legend.position = "bottom",
    strip.text = element_text(face = "bold"),
    axis.text.x = element_text(angle = 45, hjust = 1)
  )

PM2.5 and mortality trends by region (2015-2024)

Key observation: The two lines move together in most regions, suggesting an association between pollution and mortality.

Correlation Over Time – Quantifying Stability

# Calculate correlation for each year
yearly_correlation <- combined_data |> 
  group_by(year) |> 
  summarise(
    r = cor(avg_pm25, mortality_rate, use = "complete.obs"),
    p_value = cor.test(avg_pm25, mortality_rate)$p.value,
    .groups = "drop"
  )

# Plot correlation over time
p_cor <- ggplot(
  yearly_correlation,
  aes(x = year, y = r)
) +
  geom_line(color = "steelblue", linewidth = 0.7) +
  geom_point(size = 1.5, color = "steelblue") +
  geom_hline(yintercept = 0, linetype = "dashed", color = "red") +
  geom_hline(yintercept = 0.5, linetype = "dotted", color = "grey50") +
  annotate("text", x = 2016, y = 0.55, label = "Moderate correlation (r = 0.5)", size = 3) +
  labs(
    title = "Strength of Pollution-Mortality Association Over Time",
    subtitle = "Positive r = higher pollution linked to higher mortality",
    x = "Year",
    y = "Pearson correlation coefficient (r)"
  ) +
  theme_minimal()

ggplotly(p_cor, tooltip = c("year", "r"))

strength of pollution-mortality association over time

3. Deprivation is a major confounder

Deprivation is associated with both higher pollution and higher mortality. If deprivation drives the observed association, adjusting for it should weaken the correlation.

The plot below compares simple correlations (blue) with partial correlations that adjust for deprivation (red). Partial correlation isolates the pollution-mortality relationship by holding deprivation constant.

Calculate Simple and Partial Correlation for Each Year

library(ppcor)

# Calculate correlation for each year
yearly_results <- combined_with_imd |> 
  group_by(year) |> 
  summarise(
        # Simple correlation (pollution vs mortality)
    simple_r = cor(avg_pm25, mortality_rate, use = "complete.obs"),
    simple_p = cor.test(avg_pm25, mortality_rate)$p.value,
        # Partial correlation (adjusted for deprivation)
    partial_r = pcor.test(avg_pm25, mortality_rate, imd_score)$estimate,
    partial_p = pcor.test(avg_pm25, mortality_rate, imd_score)$p.value,
    .groups = "drop"
  ) |> 
  mutate(
    simple_sig = ifelse(simple_p < 0.05, "Significant", "Not significant"),
    partial_sig = ifelse(partial_p < 0.05, "Significant", "Not significant")
  )

# View results
yearly_results |> 
  kable(digits = 3, caption = "Yearly correlation: pollution vs mortality, with and without deprivation adjustment")

Yearly correlation: pollution vs mortality, with and without deprivation adjustment
year	simple_r	simple_p	partial_r	partial_p	simple_sig	partial_sig
2015	-0.587	0.096	-0.547	0.161	Not significant	Not significant
2016	-0.919	0.000	-0.677	0.065	Significant	Not significant
2017	-0.758	0.018	-0.419	0.301	Significant	Not significant
2018	-0.687	0.041	-0.029	0.946	Significant	Not significant
2019	-0.366	0.333	-0.370	0.367	Not significant	Not significant
2020	-0.700	0.036	-0.430	0.288	Significant	Not significant
2021	-0.845	0.004	-0.712	0.048	Significant	Significant
2022	-0.616	0.077	0.417	0.304	Not significant	Not significant
2023	-0.717	0.030	-0.103	0.808	Significant	Not significant
2024	0.572	0.107	0.344	0.404	Not significant	Not significant

Pollution-mortality association over time, adjusted for deprivation

Visualise the Change

yearly_long <- yearly_results |> 
  dplyr::select(year, simple_r, partial_r) |> 
  pivot_longer(
    cols = c(simple_r, partial_r),
    names_to = "type",
    values_to = "correlation"
  ) |> 
  mutate(
    type = ifelse(type == "simple_r", "Simple (unadjusted)", "Partial (adjusted for deprivation)")
  )


ggplot(
  yearly_long,
  aes(x = year, y = correlation, color = type)
) +
  geom_line(linewidth = 1) +
  geom_point(size = 1.5) +
  geom_hline(yintercept = 0, linetype = "dashed", color = "red") +
  labs(
    title = "Pollution-Mortality Association Over Time",
    subtitle = "Partial correlation adjusts for deprivation (IMD 2019)",
    x = "Year",
    y = "Pearson correlation coefficient (r)",
    color = ""
  ) +
  theme_minimal() +
  theme(legend.position = "bottom")

Key findings:

2015-2023: Simple correlations were negative. After adjusting for deprivation, they moved closer to zero. This means deprivation was creating a spurious negative association – once accounted for, the pollution-mortality relationship weakens.
2024: The simple correlation is positive (r = 0.57). After adjusting for deprivation, it remains positive (r = 0.34). This suggests an independent effect of pollution on mortality, even after accounting for deprivation.

Interpretation

Conclusion: Deprivation explains a substantial portion of the historical negative association. However, the 2024 data show a positive pollution-mortality relationship that persists after deprivation adjustment, warranting further investigation.

4. Spatial Patterns (2024)

The interactive map shows both variables simultaneously. Circle size represents PM2.5 level (larger = more pollution). Circle colour represents mortality rate (darker = higher mortality).

library(leaflet)

Warning: package 'leaflet' was built under R version 4.5.3

# Prepare map data for latest year
map_data <- combined_latest |> 
  mutate(
    # Approximate centroids for English regions
    lat = case_when(
      region == "North East" ~ 55.0,
      region == "North West" ~ 53.8,
      region == "Yorkshire and the Humber" ~ 53.8,
      region == "East Midlands" ~ 52.8,
      region == "West Midlands" ~ 52.5,
      region == "East of England" ~ 52.2,
      region == "London" ~ 51.5,
      region == "South East" ~ 51.0,
      region == "South West" ~ 50.9
    ),
    lng = case_when(
      region == "North East" ~ -1.5,
      region == "North West" ~ -2.5,
      region == "Yorkshire and the Humber" ~ -1.3,
      region == "East Midlands" ~ -1.1,
      region == "West Midlands" ~ -1.9,
      region == "East of England" ~ 0.1,
      region == "London" ~ -0.1,
      region == "South East" ~ -0.4,
      region == "South West" ~ -3.5
    ),
    # Scale circle size by PM2.5 (larger = more pollution)
    circle_size = sqrt(avg_pm25) * 4
  )

# Colour palette for mortality rates
pal <- colorNumeric("viridis", map_data$mortality_rate)

# Create interactive map
leaflet(map_data) |> 
  addTiles() |> 
  addCircleMarkers(
    lng = ~lng,
    lat = ~lat,
    radius = ~circle_size,
    color = ~pal(mortality_rate),
    fillOpacity = 0.7,
    stroke = TRUE,
    weight = 1,
    label = ~paste0(
      "<strong>", region, "</strong><br>",
      "PM2.5: ", round(avg_pm25, 1), " μg/m³<br>",
      "Mortality: ", round(mortality_rate, 1), " per 100k<br>",
      "Monitoring sites: ", n_sites
    ),
    labelOptions = labelOptions(
      style = list("font-weight" = "normal", padding = "3px 8px"),
      textsize = "12px"
    )
  ) |> 
  addLegend(
    pal = pal,
    values = ~mortality_rate,
    title = "Mortality rate<br>(per 100,000)",
    position = "bottomright"
  )

PM2.5 and mortality by region (2024)

Northern regions have larger, darker circles – higher pollution and higher mortality. Southern regions have smaller, lighter circles.

Discussion

Interpretation

The analysis reveals three important patterns:

Historical negative correlations (2015-2023) were largely driven by deprivation. Once deprivation was accounted for, the association weakened or reversed. This highlights the importance of confounding in environmental epidemiology.
The 2024 positive correlation is a notable shift. After adjusting for deprivation, a positive association (r = 0.34) persists, suggesting that air pollution has an independent effect on respiratory health.
The North‑South pollution gradient mirrors the mortality gradient. Reducing PM2.5 in northern regions could contribute to narrowing the health gap.

Limitations

Limitation	Implication
Small sample size (9 regions)	Low statistical power; p-values may be misleading
Ecological fallacy	Regional associations may not reflect individual-level causation
Residual confounding	Other factors (smoking, healthcare access) not adjusted for
IMD 2019 used for all years	Deprivation changes slowly, but 2019 may not perfectly reflect 2024
Monitoring site coverage	Some regions have fewer sites, reducing precision

Public Health Implications

Target pollution reduction in northern regions – would likely improve respiratory health
Integrate environmental and health surveillance – early warning systems for pollution spikes
Further research at local authority level – would provide more statistical power and enable targeted interventions

Conclusion

This analysis demonstrates a positive association between PM2.5 and respiratory mortality in England in 2024, which persists after adjusting for deprivation. While historical negative correlations highlight the importance of confounding, the most recent data suggest that air pollution has an independent effect on respiratory health. Reducing PM2.5 – particularly in northern regions – could contribute to narrowing the North‑South health gap.

Future work should: - Replicate the analysis at local authority level - Include additional confounders (smoking, healthcare access) - Explore lagged effects of pollution on mortality

References

DEFRA. (2025). UK-AIR air quality data. Available at: uk-air.defra.gov.uk
Office for Health Improvement and Disparities. (2025). Fingertips public health data.
Carslaw, D. C., & Ropkins, K. (2012). openair – An R package for air quality data analysis. Environmental Modelling & Software.
Ministry of Housing, Communities & Local Government. (2019). English Indices of Multiple Deprivation.

Appendix: Code and Reproducibility

All code is available in the accompanying R Markdown document and on GitHub. Data are cached locally for reproducibility. The analysis can be reproduced by running 00_setup.qmd followed by this document.

1. Information Governance

All data are accessed via official APIs (fingertipsR, openair) – no manual downloads or insecure transfers
Data are cached locally as RDS files, not redistributed
The analysis is fully reproducible via version‑controlled code
This README documents the data sources and access methods

In a live environment, I would also:

Use password‑protected, encrypted storage for any sensitive data
Implement role‑based access controls
Follow UKHSA SOPs for data transfer and sharing

2. Data Quality Prevention

Automated validation checks for missing values, outliers, and duplicates
Data quality report generated automatically
Recommendations for prevention included in the Discussion

In a live environment, I would also:

Implement validation rules at data entry
Provide regular training for data collectors
Document and follow SOPs for data handling

3. Equality, Diversity and Inclusion (Workplace)

All visualisations use colourblind‑safe palettes (viridis)
Alt text is included for all images (ensuring accessibility)
The analysis explicitly examines health inequalities (North‑South gap, deprivation)

My commitment to workplace EDI:

I ensure meetings are inclusive (rotating chairs, asynchronous options)
I support reasonable adjustments for colleagues
I challenge bias when I see it
I believe diversity strengthens science

Summary of Actions Taken

Feedback Area	Action Taken
Information governance	Added governance section, documented data sources, used secure APIs
Data quality prevention	Added validation checks, prevention recommendations
Workplace EDI	Added EDI statement, accessible visualisations, health inequalities focus

This project demonstrates that I can not only perform technical analysis but also think critically about governance, quality, and inclusion.