Divergent Thinking - Data Collection

[Husayn01]

This discussion is dedicated to coordinating the data collection phase for our research question:

_

How do respiratory and cardiovascular health outcomes vary across global cities categorized by their dominant air pollutant?

_

Objective:

To identify, collect, and organize relevant datasets that will enable us to:

• Assess air quality indicators across multiple global cities.

• Determine the dominant pollutant in each city (e.g., PM2.5, NO₂, O₃).

• Link pollution levels to respiratory and cardiovascular health outcomes such as hospital admissions, mortality rates, or disease prevalence.

Key Tasks:

• Compile a list of reliable data sources (WHO, OpenAQ, Global Burden of Disease, local health departments, etc.)

• Identify target cities or regions to include in the study.

• Gather air pollution data (preferably categorized by pollutant type and time).

• Collect health outcome data for respiratory and cardiovascular conditions.

[salihakalender]

Website Review: Kaggle – Global Air Pollution Data

🔗 https://www.kaggle.com/datasets/sazidthe1/global-air-pollution-data

This dataset can be a valuable asset for the data collection phase of our research question:

How do respiratory and cardiovascular health outcomes vary across global cities categorized by their dominant air pollutant?

✅ Why this dataset is useful:

1. Covers Multiple Cities Worldwide:

The dataset includes air quality data from many global cities, which aligns with our objective to assess and compare urban air pollution.

2. Categorized by Pollutant Type:

Key pollutants like PM2.5, PM10, NO₂, SO₂, CO, and O₃ are tracked—helping us determine the dominant pollutant in each city.

3. Time Series Format:

Pollution data is provided over time, allowing us to analyze trends and correlate with health outcomes (hospital admissions, mortality rates, etc.).

4. Complements Health Data Sources:

While this dataset focuses on pollution levels, it can be linked with external health data (e.g., Global Burden of Disease, WHO databases) to explore respiratory and cardiovascular impacts.

6. Structured and Ready to Use:

Data is provided in CSV format, making it easy to integrate into our processing pipeline.
Pollution data is provided over time, allowing us to analyze trends and correlate with health outcomes (hospital admissions, mortality rates, etc.).

[Husayn01]

Great Job Saliha, just went through the dataset, it is robust and vast. It is worth being considered as a potential source for tackling our research question, however i could not find the column represntative of time, for the time series analysis you proposed, could i have missed it?

[salihakalender]

Yes, it covers the year 2024 and does not include a time or date column - no monthly, daily, or historical breakdown is shared. It appears to be a single-year snapshot, not a time-series dataset.

[linahKhayri]

💡Possible Backup if We Can’t Find City-Level Health Data

Hey! Just wanted to share something I came across while searching a bit around I hadn’t heard of it before, but it might help if we don’t find direct or consistent health data at the city level.

It’s called the Global Burden of Disease (GBD) study, led by IHME It compiles disease estimates (like for asthma, COPD, stroke, heart disease, etc.) and allows you to compare across countries or regions using standardized health metrics like:

DALYs (Disability-Adjusted Life Years)

YLLs (Years of Life Lost)

Deaths per 100,000, etc.

On their interactive tool (VizHub) on the website, we can:

Choose the exact metric you want (like DALYs YALLs or whatever)

Select the specific disease category (e.g. cardiovascular or respiratory diseases)

And even directly filter by risk factor, like air pollution including ambient PM2.5, ozone, and more.

BUT The latest full data set is from 2021, which seems to be the most recent release

here's the links:

🔗 GBD VizHub (explore and download): https://vizhub.healthdata.org/gbd-results/
🔗 GHDx (raw dataset archive): https://ghdx.healthdata.org/

Might be worth keeping this in mind if we hit a dead end on city-level raw health data it could still help! I’ll keep researching thoroughly tomorrow and explore available datasets and sources.
If anyone already knows something about GBD or these estimates or has insights on how to work with it, that would be super helpful!

[Adamx090]

That's somewhat reassuring, I love backups. I heard Evan mentioning a proxy data in the workshop, I wonder if this counts as one.

[salihakalender]

Thank you so much for sharing this - really helpful! I agree, the GBD study is a solid resource and the features you pointed out (like filtering by disease category or specific risk factors such as PM2.5) can definitely add value to our analysis or help as a strong backup.

That said, for our current needs, we're specifically looking for health data from 2024, and it seems the latest available data on the GBD platform is from 2021. So while it's a great reference, I hope we can find similar data that's available for 2024.

[Adamx090]

WHO Air Quality Database

A comprehensive dataset with over 40,000 records of PM2.5, PM10, and NO₂ concentrations across global cities. The data represents annual mean concentrations collected between 2010 and 2022 from official, government-verified monitoring stations.

✅ Pros

• Published by a trusted global health authority (WHO)
• Transparent methodology and documentation
• Data sourced from validated government stations
• Ideal for city-level yearly analysis
• Includes metadata on data quality, method, and coverage
• Large dataset, suitable for robust statistical analysis
• Easy to cite in academic or public health research

⚠️ Cons

• Limited to only PM2.5, PM10, and NO₂
• Latest entries stop at 2022
• Data coverage varies by country, some regions may have less or outdated records

[FalaqMajeed]

Can you share the link for this data please

[Adamx090]

The title is the link 😅

[FalaqMajeed]

ohhhhh! sorry for that, my bad

[FalaqMajeed]

To be honest, I haven’t found anything truly convenient so far most of the datasets were either incomplete or behind a paywall. So I suggest that we either narrow our focus to specific countries (which makes it easier to find reliable health data), or go with Linah’s idea of using the Global Burden of Disease (GBD) dataset. But I’ll keep searching and will share any updates I find.

[linahKhayri]

For me on the health concerns, it seems like finding data on the cases (especially at city level) is a bit harder than expected. I couldn’t really find anything good so far!

Also, I think we should be extra clear that our goal is just to identify patterns or associations, and never try to draw causal conclusions or causations. These health outcomes are influenced by so many other factors and making strong causal claims would definitely be beyond our scope.
and also I was hoping we could finalize our datasets by Monday, so we can start cleaning and processing everything from Monday or Tuesday.

[FalaqMajeed]

yeah couldn't imagine that we will be stuck in the health data, we need an expertise asap lol

[salihakalender]

I found a great report titled "State of Global Air 2024 – A Special Report on Global Exposure to Air Pollution and Its Health Impacts, with a Focus on Children’s Health". Although I was excited to see "2024" in the title, the data it presents actually covers the years 1990 to 2021. The report is a collaborative effort between the Health Effects Institute and the Global Burden of Disease project. It provides a comprehensive global overview of air pollution exposure (PM2.5, NO₂, ozone) and its impact on respiratory and cardiovascular health, with a special focus on children under five.

🔗 https://www.stateofglobalair.org/resources/report/state-global-air-report-2024

[linahKhayri]

Thanks @salihakalender for sharing this we should definitely look into it!

[linahKhayri]

Hi team! 👋

Falaq and I wanted to share an update based on the datasets we've found so far and we’d want your feedback and thoughts.

It's not about changing the main research question or objectives.
We’re still focused on:

• Respiratory and cardiovascular health outcomes

• Their relationship with dominant air pollutants

• And geographic comparisons across cities or countries

The only shift is in the timeframe, from a strict 2024 focus to a range from 2018 to 2021, which fits better with available data (like IHME’s GBD and WHO’s pollution data that @Adamx090 shared).

Why the Change in these Years?
We couldn’t find a reliable dataset so far for 2024 health burden data (like DALYs or deaths by disease & risk factor). The most recent available data from IHME is for 2021.

We chose 2018–2021 on purpose not just as a workaround, but to:

Capture real health and pollution trends over time

Include the impact of COVID-19, which is a major shift worth analyzing

Broaden our scope and add value, instead of focusing on a single (unavailable) year

We also wanted to acknowledge that while these years might sound like ancient history, they actually let us study meaningful shifts before, during, and after the pandemic.

suggested Question:

How have respiratory and cardiovascular health outcomes changed between 2018 and 2021 across countries grouped by their dominant air pollutant, and what insights can be drawn about pollutant-specific health risks before, during, and after the COVID-19 pandemic?

⭕Possible Extension: Predicting 2024

Since we can get 2024 pollution concentrations, how about thinking of building a basic prediction model to forecast 2024 health outcomes based on trends from 2018–2021?

This would be clearly marked as exploratory but it adds a forward-looking piece to our project that could be really interesting.

Data We Have So Far (I'll share the files on Slack) :

WHO ambient air pollution data

IHME GBD estimates (2018–2021) (for DALYs/deaths by disease and risk factor)

[Husayn01]

Hi @linahKhayri, well done you guys, great idea you have here. I just went through the dataset proposed by @Adamx090 myself. Your suggestions are valid, and i strongly agree with you. This may be our only viable course of action at this point, considering the challenges we've been facing lately regarding the data sourcing.

I have an addition though, how about we work with a wider range of the dataset, say from 2010 - 2021 from @Adamx090 proposed dataset and adjust the filter of yours to cover the same year range, then build a model that predicts a varable, llike you have suggested, since models tend to perform better when they are built with large datasets.

I am yet to fully grasp the nature of your own dataset (IHME GBD estimates), so ill reserve further comments for now. I also feel our Research question may become overly complicated if we factor in the COVID 19 pandemic. Simplicity sometimes is key. Hence i suggest we stick to our current question. Just a suggestion tho, let hear what others have say.

Well done again

[salihakalender]

From what I understand, if we use @Adamx090’s dataset, we can obtain pollutant data for the extended period of 2010–2021. Additionally, we can access health outcomes up to 2021 from WHO. Based on this, we can confidently proceed with the analysis using this existing data range, and therefore, I believe there is no strong need to build a predictive model specifically for 2024.

[linahKhayri]

Hey @Husayn01 , thanks a lot for the thoughtful input really appreciate it! 🙏

You're absolutely right that using a wider range like 2010–2021 gives us more room for proper training, testing, and validation if we move forward with a prediction model. I think that's a strong advantage.

Just to clarify one thing on the dataset (IHME GBD): I purposely chose to focus on “air pollution” as a general risk factor rather than specifying PM2.5 or NO2, etc... just to keep the scope broad at first and hopefully uncover any strong patterns between dominant pollutants and the respiratory/cardio health burdens by ourselves so it wouldn't be just analysis

And also I agree with your point on covid, I can see how it could make the research question feel too overloaded. We didn’t mean to center it, but more like use it as context for interpreting trend shifts, especially since there might be a drop down in pollution levels during lockdown periods, which might show up in the health data. Could even be a sub-finding or a note, not necessarily a core objective. But we should hear other's inputs and findings and then we can stick with our original question or reframe accordingly.

[linahKhayri]

@salihakalender The prediction model could just be an additional layer we explore later on and not something we need to focus on right away, but it might be an interesting if things went well

[FalaqMajeed]

Husayn01 Absolutely, just as Linah mentioned COVID-19 isn’t the main focus of our project. It’s more of a contextual note to highlight how the trends may have shifted during the pandemic period. Including this perspective can actually support us if we choose to add an exploratory layer later, like forecasting 2024 outcomes. So it’s not central to our question, but it can enrich the analysis without complicating the core objective.

[salihakalender]

@linahKhayri and @FalaqMajeed Thank you for considering this thoughtful addition to our research.

Adding a 2024 prediction component can be a valuable and forward-looking extension to our project — as long as we clearly mark it as exploratory. This transparency ensures that our findings are interpreted as preliminary insights rather than definitive conclusions. The benefits are clear: we can leverage the availability of recent PM2.5 data, apply simple regression or trend analysis methods, and generate a basic forecast of health outcomes (e.g., DALYs or mortality) for 2024. This not only enriches our project with a future-oriented perspective but also enhances its relevance for policymakers by offering insights into potential upcoming health risks linked to air pollution trends.

[linahKhayri]

Thanks so much @salihakalender 🙌 and I completely agree, the 2024 prediction part will stay exploratory, since we don’t have the actual 2024 health data yet to validate.

Also, in the GBD estimates we focused on general air pollution rather than just PM2.5 to keep things broad and allow us to spot any patterns between dominant pollutants and health outcomes which is in line with what we originally intended

[salihakalender]

Another suggestion from my side: instead of using 2024 pollutant data, we could use EPA’s AQS (Air Quality System) data for the years 2018–2021 to align directly with the available health outcome data.

✅ Potential Advantages:

Temporal consistency: Health and pollution data would cover the exact same years, improving the reliability of our comparisons.

Data quality: EPA data is regularly validated and publicly accessible, with high-resolution daily and annual summaries.

Geographic coverage: Enables city- or county-level analysis within the U.S., which could support detailed insights.

This trade-off might make our models stronger for U.S.-based analysis and help maintain internal consistency across the dataset. Since the EPA data covers only the United States, this approach would be most suitable if the study focuses solely on U.S. cities.

You can explore and download the data here:

https://aqs.epa.gov/aqsweb/airdata/download_files.html#Daily
https://aqs.epa.gov/aqsweb/airdata/download_files.html#Annual

Alternatively, we could also consider using the datasets shared by @Adamx090, which may offer broader geographic coverage or complementary data.

https://www.who.int/publications/m/item/who-ambient-air-quality-database-%28update-jan-2024%29

[linahKhayri]

Thank you @salihakalender thats a great suggestion and I really appreciate it! I'm not familiar tho with the AQS but I’ll check it out and explore the datasets you shared.
But since our original goal was more global, I still feel like sticking with datasets that allow for cross-country comparisons could help us capture broader patterns

But thats a strong suggestion too and if the EPA data offers valuable insights and fits well with our direction, I’m open to where the group wants to take this! Let’s see what others think

[salihakalender]

Review of the Study: Impact of Air Pollution on Human Health

📎 Kaggle Notebook Link
📂 Dataset Link

This study uses publicly available data to examine the health effects—particularly mortality—of air pollution across countries, with a focus on PM2.5 and indoor/outdoor pollution sources.

• The data is presented on an annual basis.
  ➡️ 1990 – 2019

• Each row represents a unique combination of country and year.

• In some datasets that include age group information, an additional column specifying the age group is also included.

📊 Available Datasets

1. absolute-number-of-deaths-from-ambient-particulate-air-pollution.csv
   Total deaths by country and year from PM2.5 and outdoor air pollution.

2. death-rate-from-air-pollution-per-100000.csv
   Age-standardized death rates per 100,000 people from all sources of air pollution.

3. death-rates-from-air-pollution.csv
   General death rates from air pollution by cause and country over time.

4. deaths-from-indoor-air-pollution-by-age.csv
   Indoor air pollution deaths by age group, year, and country.

5. outdoor-pollution-death-rate.csv
   Death rates due to outdoor pollution only, by country and year.

6. outdoor-pollution-rates-by-age.csv
   Outdoor pollution-related deaths, segmented by age group.

7. share-deaths-indoor-pollution.csv
   Percentage of total deaths attributable to indoor air pollution by country/year.

✅ Pros

• Rich Mortality Data

Covers both indoor and outdoor air pollution-related deaths, focusing on respiratory and cardiovascular causes.

• Temporal and Demographic Dimensions

Includes breakdowns by year and age group, useful for limited trend and vulnerability analysis.

• Structured & Ready-to-Use

Provided in CSV format; clean structure allows merging with external datasets like WHO or GBD.

• Useful for Regional Comparisons

Enables analysis of pollution effects by region or country and classification of dominant pollutant types.

❌ Cons

• Lacks City-Level Data

All data is aggregated at the country or regional level; city-specific analysis is not possible directly.

• Limited Health Metrics

Focuses only on mortality. Hospital admissions, disease prevalence, or chronic condition data are not included.

• Pollutant Scope Focused on PM2.5

Other urban pollutants like NO₂ and O₃ are underrepresented or missing, limiting full pollutant classification.

🌍 How This Content Supports Our Project

Research Question:
How do respiratory and cardiovascular health outcomes vary across global cities categorized by their dominant air pollutant?

• Country-Level Data Availability:

The dataset fits a proxy strategy—grouping cities by national pollution profiles—to compensate for missing city-level data.

• Health Outcomes Focus:

Strong coverage of mortality outcomes related to air pollution, especially respiratory and cardiovascular-related deaths.

• Temporal and Age Group Dimensions:

Enables high-level trend and demographic analysis using year and age group filters.

• Pollution Type Coverage:

While focused mostly on PM2.5, the dataset distinguishes between indoor and outdoor pollution effects.

• Linkage Potential:

GBD (Global Burden of Disease) data is publicly accessible and highly compatible with this dataset.

Provided by the Institute for Health Metrics and Evaluation (IHME), GBD data can be accessed through the GBD Results.

Users can filter by country, year, disease type (e.g., COPD, ischemic heart disease), age group, and sex, and download the results in CSV format.

📂 Available metrics include:

Mortality rates from air pollution-attributable diseases

Years of Life Lost (YLL) due to respiratory and cardiovascular conditions

Prevalence and incidence of specific diseases related to air quality

📌 This makes it feasible to link pollution exposure data with relevant health outcomes for statistical analysis such as correlation or regression modeling at the country level.