A Machine Learning-Based Study of the Effects of Air Pollution and Weather in Respiratory Disease Patients Visiting Emergency Departments

Eu Sun Lee, Jung-Youn Kim, Young‐Hoon Yoon, Seoung Bum Kim, Hyungu Kahng, Jinhyeok Park, Jaehoon Kim, Minjae Lee, Haeun Hwang, Sung Jun ParkEmergency 2022 · DOI

TL;DR

This study uses machine learning to analyze the complex interplay between air pollution, weather, and emergency department (ED) visits for respiratory illnesses. It reveals that PM10, temperature, and steam pressure are key factors influencing these visits, with specific pollutants and weather conditions having varying impacts depending on the respiratory condition.

What Problem Does This Paper Address?

The paper addresses the challenge of predicting and understanding the surge in emergency department (ED) visits for respiratory diseases, particularly in the context of environmental factors like air pollution and weather. While previous research has explored individual links between specific pollutants or weather patterns and respiratory issues, this study aims to provide a more comprehensive, multivariate analysis. It seeks to identify which environmental factors, and with what temporal lag, most significantly contribute to respiratory disease patients seeking emergency care, thereby informing resource allocation and public health strategies.

Methods

The researchers employed a Random Forest Regressor model, a powerful machine learning technique known for its ability to handle complex relationships and identify important features. This approach was chosen over simpler decision trees to mitigate overfitting, especially given the large number of explanatory variables. The model was used to predict the number of ED visits for specific respiratory conditions based on a range of weather and air pollution data. Feature importance was calculated using impurity-based methods to rank the influence of each variable.

Datasets

The study utilized two primary datasets:

  • National Emergency Department Information System (NEDIS) database: This provided clinical and administrative data for all patients visiting EDs nationwide, with the study focusing on data from Seoul, Korea, over a 3-year period (January 1, 2015, to December 31, 2017).
  • Meteorological and Air Pollution Data:
    • Meteorological Data: Obtained from the National Climate Data Service, including temperature, precipitation, humidity, steam pressure, wind speed, and wind direction.
  • Air Pollution Data: Included measurements for carbon monoxide (CO), nitrogen dioxide (NO2), ozone (O3), PM10, PM2.5, and sulfur trioxide (SO2).

The study innovatively considered weather and air pollution data not just for the day of the ED visit but also for the preceding three days, creating a 4-day cumulative influence window.

Key Insights and Ideas

  • Multivariate Influence is Key: Respiratory ED visits are not driven by a single factor but by a complex interplay of multiple weather and air pollution variables.
  • PM10 Dominance: PM10, a larger particulate matter, appears to have a significant and immediate impact on acute respiratory conditions, influencing visits on the day of and days preceding the ED visit.
  • Temporal Lags Matter: The influence of environmental factors isn't always immediate. Steam pressure and temperature, for instance, can influence visits up to three days prior, especially when air pollution levels are high.
  • Condition-Specific Triggers: Different respiratory conditions are influenced by different environmental factors. For example, influenza and pneumonia are more sensitive to temperature and steam pressure, while acute upper respiratory infections are more strongly linked to air pollutants like NO2 and PM10.
  • Predictive Potential: The findings suggest the feasibility of developing a predictive index for respiratory ED visits, which could aid in resource management and preventative measures.

Key Results

The study analyzed data from 525,579 patients diagnosed with respiratory diseases who visited EDs. Key findings include:

  • Most Frequent Diagnoses: Acute upper respiratory infections (J00-J06), influenza (J09-J11), and pneumonia (J12-J18) were the most common reasons for ED visits.
  • Top Influencers: PM10, temperature, and steam pressure were identified as the most influential factors across several respiratory disease categories.
  • Specific Disease Impacts:
    • Pneumonia, Acute Upper Respiratory Infections, and Chronic Lower Respiratory Diseases: These top three conditions were significantly influenced by PM10.
  • Influenza and Pneumonia: Highly influenced by temperature and steam pressure.
  • Acute Upper Respiratory Infections: Primarily influenced by air pollution variables, particularly NO2 and PM10.
  • Predictive Trend: The number of respiratory disease patients visiting EDs is predicted to increase when steam pressure and temperature are low, and air pollution levels are high, particularly around day 3 prior to the visit.

Limitations and Future Work

The authors acknowledge several limitations:

  • Data Granularity: The use of pre-codified data prevented detailed analysis of clinical characteristics, prognosis, infection sources, or underlying patient conditions.
  • Study Duration: The 3-year study period might be insufficient for fully understanding the long-term effects of air pollution, especially for chronic diseases.
  • Lag Time: The chosen lag of 3 days for air pollution and weather variables was a practical decision due to the number of variables but could be extended to capture longer incubation periods (e.g., 14 days).
  • Observatory Data: The use of data from observatories near hospitals, rather than patient-specific locations, might introduce some spatial variability.

Future work could involve developing a comprehensive weather-related health index for predicting respiratory ED visits and analyzing clinical data for individual disease groups to gain deeper insights. Further research into the long-term effects of PM2.5 is also recommended.

Summary updated

Abstract

Most of the respiratory patients visiting EDs were diagnosed with acute upper respiratory infections, influenza, and pneumonia. Following temperature, steam pressure and PM<sub>10</sub> had influential relations with these diseases. It is expected that the number of respiratory disease patients visiting the emergency departments will increase by day 3 when the steam pressure and temperature values are low, and the variables of air pollution are high. The number of respiratory disease patients visiting the emergency departments will increase by day 3 when the steam pressure and temperature values are low, and the variables of air pollution are high.
BibTeX
@inproceedings{lee2022machine,
  author = {Eu Sun Lee and Jung-Youn Kim and Young‐Hoon Yoon and Seoung Bum Kim and Hyungu Kahng and Jinhyeok Park and Jaehoon Kim and Minjae Lee and Haeun Hwang and Sung Jun Park},
  title = {{A Machine Learning-Based Study of the Effects of Air Pollution and Weather in Respiratory Disease Patients Visiting Emergency Departments}},
  booktitle = {Emergency},
  year = {2022},
  doi = {10.1155/2022/4462018},
}
Figure