Lung and bronchus cancer (LBC) is one of the most common causes of cancer death worldwide, accounting for 11.6% of all cancer deaths in 2018. In both France and the United States, it is the leading cause of cancer death, and while smoking plays a prominent role in the disease, pollution and socioeconomic conditions are also risk factors. A team from the University of Buffalo sought to understand why these factors had different consequences depending on where the patients lived. They published the results of their research, titled ” Explainable artificial intelligence to explore spatial variability in lung and bronchial cancer mortality rates in the contiguous United States,” in the journal Scientific Reports in December 2021.

This study identifying key risk factors for AML mortality using Explainable Artificial Intelligence (XAI) brought together an interdisciplinary team:

• Zia U. Ahmed, PhD, Database/Visualization Specialist at the UB RENEW Institute;
• Kang Sun, PhD, senior faculty member at the UB RENEW Institute and assistant professor of civil, structural, and environmental engineering in the UB School of Engineering and Applied Sciences;
• Michael Shelly, PhD, environmental/ecological economist at the UB RENEW Institute;
• Lina Mu, PhD, MD, associate professor of epidemiology and environmental health in the UB School of Public Health and Health Professions.

Zia U. Ahmed states of the research:

“The results matter because the United States is a spatially heterogeneous environment. There is a wide variety of socioeconomic factors and education levels – essentially, one size does not fit all. Here, the local interpretation of machine learning models is more important than the global interpretation. “

Lina Mu adds:

“The study can be a model for integrating artificial intelligence into an epidemiological study. It can also serve as an example of using predictive models when studying cancer. It can greatly help in identifying high-risk areas where the cancer registry is not available. “

The study

The UB team applied explainable artificial intelligence (XAI) on a stack set machine learning model framework to explore and visualize the spatial distribution of known risk factor contributions to lung and bronchus cancer mortality rates in the contiguous United States. To develop stack ensemble models, she used generalized linear model (GLM), random forest (RF), gradient boosting machine (GBM), extreme gradient boosting machine (XGBoost).

Zia U. Ahmed explains:

“XAI in local interpretation is still lacking, especially with respect to environment and science. “

Risk factors explored by the study represented lifestyle variables including smoking, socioeconomic status (poverty rate), demographics, air pollution, physical environment, biophysical factors as well as health insurance.

The study notes that smoking rates were related to poverty levels and race/ethnicity, with Hispanics, for example, smoking less than whites. It also shows a strong relationship between poverty, and therefore lack of access to care, and AML mortality rates in the United States.

With respect to air pollution, the researchers examined the pollutants nitrogen dioxide (NO2), sulfur dioxide (SO2), ozone, and particulate matter and their spatial variability in relation to lung and bronchus cancer mortality rates.

Smoking and poverty were found to be the two main risk factors for lung and bronchial cancer. While these results were expected, this study demonstrates a strong potential for the implementation of explainable artificial intelligence to complement or replace traditional spatial regression models.

Article source: Ahmed, ZU, Sun, K., Shelly, M. et al. Explainable artificial intelligence (XAI) to explore spatial variability in lung and bronchus cancer (LBC) mortality rates in the contiguous United States. Sci Rep 11, 24090 (2021). https://doi.org/10.1038/s41598-021-03198-8

Translated from Des chercheurs de l’Université de Buffalo identifient les principaux facteurs du cancer du poumon grâce à l’IA