Which causes of death are predictable? A machine learning analysis
Principal Investigator:
Professor Alexandre Chiavegatto Filho
Approved Research ID:
30140
Approval date:
February 26th 2020
Lay summary
Being able to predict which individuals are more likely to die is one of medicine?s most recurring challenges. The recent expansion of new machine learning algorithms is a promising way to predict mortality, especially in studies with a wide variety of individual characteristics such as is the case of the UK Biobank. Our analyzes will be able to identify: - Which of the main causes of death can be predicted with higher accuracy, using all baseline characteristics. - Which machine learning algorithms have a higher predictive accuracy for each of the main causes of death. The results of the study may be applied directly by healthcare services to help improve decisions about patient hospitalization and the intensity of care provided for high-risk patients. Our study will test the performance of a few dozen of the most popular machine learning algorithms to predict all-cause mortality, and mortality for each of the most frequent causes of death, using all the baseline characteristic of the individuals. We plan to include the full cohort in our analysis.