Originally posted on The Horizons Tracker.
Machine learning technologies have become increasingly adept at making medical predictions, whether it’s diagnosing illnesses or predicting the success of treatments. The latest evidence of its potency comes via a study1 conducted by the University of Oxford that saw machine learning used to predict whether cystic fibrosis patients should be referred for a lung transplant.
The study found that the new AI-based method led to a 35% improvement in the accuracy of these predictions compared to existing statistical methods.
The condition, which is believed to affect over 10,000 people in the UK alone, causes a number of challenging symptoms affecting the whole body. Lung transplants are the last resort treatment option, but it comes with serious risks of complications, and the number of donors is very low.
The researchers tested whether machines can do a better job by extracting anonymized data from the UC Cystic Fibrosis Registry. This data was used to develop a prognostic model to assess the risk involved in undergoing a lung transplant.
The model, known as AutoPrognosis, was able to achieve a positive predictive value of 65%, which compares to the best practice currently in operation of just 48%. The team believe that their work can be invaluable in improving the ability of clinicians to make accurate decisions, whilst they also believe it could help with other conditions as well.
“While machine learning has proven successful in making predictions in a clinical setting, its deployment in practice has been limited. The outcomes of our research with the Cystic Fibrosis Trust demonstrate that with the right in-depth expertise, anonymised data from a large population, and input from clinicians, we can create algorithmic methods to support clinicians in their day-to-day decision-making,” the researchers say.
Article source: Using Machine Learning To Help Treat Cystic Fibrosis.
- Alaa, A. M., & van der Schaar, M. (2018). Prognostication and Risk Factors for Cystic Fibrosis via Automated Machine Learning. Scientific reports, 8(1), 11242. ↩