Data Analysis Divergent Thinking

[Adamx090]

Hey everyone,

Let’s start throwing around ideas on how we can approach our research question:

• How can we explore the relationship between PM₂.₅ levels and cardiovascular and respiratory disease burden?

• Does this relationship look different across countries with different SDI levels? How can we test that?

• What’s the best way to look at how long-term PM₂.₅ exposure might relate to COVID-19 mortality?

Let’s try to keep our ideas connected to the data we actually have, and be mindful of any assumptions or biases. It’s easy to fall into statistical traps, so let’s question our thinking along the way.

Drop your thoughts below, curious to hear everyone's take!

[salihakalender]

Here are a few ideas for potential analyses.

1. PM₂.₅ and Disease Burden

What is it?

Understanding how PM₂.₅ levels relate to respiratory and cardiovascular disease burden (e.g., DALYs).

How to do it:

• Trend Analysis: Plot yearly PM₂.₅ and health outcomes (2010–2019) to observe changes over time.
• Correlation: Test direct associations between PM₂.₅ and DALY rates.
• Regression: Model DALY ~ PM₂.₅ + SDI + Year to control for development and time effects.

Why it matters:

• Highlights long-term exposure effects.
• Connects public health trends to environmental quality.
• Supports evidence-based policy evaluation.

2. PM₂.₅ Effects by SDI Level

What is it?

Testing whether PM₂.₅ has stronger health effects in low-SDI countries.

How to do it:

• Use regression with an interaction term:
  DALY ~ PM₂.₅ + SDI + (PM₂.₅ × SDI)
• Or compare regression results across SDI tertiles.

Why it matters:

• Exposure alone isn’t the whole story—social conditions shape vulnerability.
• Informs more targeted and equitable policy responses.

3. Lag Effect Analysis

What is it?

Evaluates whether past PM₂.₅ exposure is linked to current health burdens.

How to do it:

• Shift PM₂.₅ data backward in time (e.g., use 2010 PM₂.₅ to predict 2015 DALYs).
• Run lagged regression models for 3–5 year intervals.

Why it matters:

• Many chronic diseases have delayed onset after exposure.
• Captures real-world exposure-to-impact timelines.
• Critical for making the case for preventive public health action.

[linahKhayri]

Multiple and Panel Regression: PM2.5 Exposure and Disease Burden

Multiple Linear Regression (Cross-Country Comparison)

We will run multiple linear regressions to test whether countries with higher average PM2.5 exposure from 2010 to 2019 tend to have higher disease burdens (DALYs) for each of the following diseases:

Chronic respiratory disease

Cardiovascular disease

Stroke

Each disease will have its own regression model. After running the models, we can compare the results across diseases to identify consistent patterns or notable differences.

What we need:

One row per country

Mean PM2.5 exposure (2010–2019)

Mean SDI (2010–2019)

DALYs per disease (as rates, for consistency)

What this might tell us:

Whether countries with higher long-term exposure to PM2.5 tend to have higher disease burdens

Whether development level (SDI) plays a role in reducing or explaining that burden

Panel Regression (Over-Time Analysis Using Total DALYs)

To go beyond cross-sectional averages, we will also use panel regression with total DALYs from all key causes (chronic respiratory, cardiovascular, and stroke) to examine how disease burden changes over time in response to changes in PM2.5 within each country.

I just came across this type, and Panel regression is a technique used when analyzing data that tracks multiple entities (in this case, countries) across multiple time periods (years). It helps us explore how within-country changes such as increases or decreases in pollution relate to outcomes like disease burden.

Instead of just comparing countries at one point in time, panel regression allows us to monitor what happens within each country across the period from 2010 to 2019.

What we need:

One row per country–year (e.g., Egypt–2010, Egypt–2011, etc.)

Annual PM2.5 values

Annual SDI values

Annual total DALYs

We will create a DALY_total column by summing the DALYs for chronic respiratory disease, cardiovascular disease, and stroke.

What this tells us:

Whether increases in pollution within a country over time are associated with increases in disease burden

Whether development (SDI) and fixed country-specific traits influence that relationship

A time-aware understanding of the link between pollution and health outcomes

[FalaqMajeed]

COVID-19 Mortality and Long-Term PM2.5 Exposure Analysis

We can run a simple multiple linear regression to test whether countries with higher average PM2.5 exposure from 2010 to 2019 experienced higher COVID-19 mortality. This method helps us examine if long-term air pollution is associated with worse COVID-19 outcomes at the country level.

What we need:

One row per country

Mean PM2.5 exposure (2010–2019)

Mean SDI (2010–2019)

COVID-19 mortality (either total deaths or mortality rate)

What this tells us:

Whether countries with higher long-term PM2.5 levels also had higher COVID-19 deaths

Whether socio-demographic development (SDI) explains or modifies this relationship

Provides a broad overview of how air pollution and development level might have influenced COVID-19 impacts across different countries

It’s a straightforward first step to explore the connection between long-term pollution exposure and COVID-19 mortality across diverse countries.

Before running the regression, we can calculate Pearson correlation coefficients to see how strongly and in what direction:

PM2.5 exposure relates to COVID-19 mortality

SDI relates to COVID-19 mortality

PM2.5 exposure relates to SDI

This helps us understand if there is any linear association between variables and gives us an initial sense of the relationships we’ll test more formally with regression.

[Adamx090]

I am agreeing with most of what I am seeing here so far. So I will just make some additions.

• Include all countries for which data is available. This will improve trend visibility and increase the size of the dataset, which is especially valuable if we found a good predictive model later. A larger dataset will allow for a more effective split between training and validation sets.
• When analyzing PM₂.₅ exposure, we should focus on the estimates derived from cities only. Using the “total” or “rural” calculations may dilute the average exposure and lead to misleading conclusions, especially since health burdens tend to appear only after certain concentration thresholds are crossed, which are more likely to occur in cities.
• For the health metrics, we should use rates rather than absolute numbers or percent (percent measures the cause contribution to all causes), as they reflect the burden relative to population size, enabling fairer comparisons across countries. For the measure itself, DALYs are preferred since they capture both mortality and disability. An exception might be COVID-19, where using deaths may be more appropriate due to the acute and high-mortality nature of the disease.
• We should be cautious about timing and causality. When assessing changes over time, we should avoid making direct conclusions based on PM₂.₅ levels and disease burden recorded in the same year. Health impacts from air pollution often appear after a time lag, so interpreting year-to-year relationships should be done carefully. That said, examining long-term trends may still uncover meaningful patterns or insights.