Victory in research competition for quality assurance of patient data
Researchers from Halmstad University and Lund University have won an AI competition announced by the US Food and Drug Administration, FDA. The purpose of the competition was to develop a system for tracing mislabelled cancer patient data.
Accidental mislabeling or swapping of patient samples can lead to inaccurate healthcare decisions. The problem is so great that the US Food and Drug Administration, the FDA, last autumn announced a competition to find solutions. The competition was part of former President Barack Obama’s ”Precision Medicine Initiative” from 2015. Mattias Ohlsson is a Professor in Information Technology at Halmstad University and won the competition together with colleagues from Lund University. Their solution, which is based on machine learning, can search large amounts of data and find deviations connected to incorrectly documented patient data.
Congratulations, Mattias Ohlsson! What was the biggest challenge with the task?
”Thank you! The most difficult part was the combination of so many different data types. We used data from gene and protein measurements as well as clinical data in the challenge. The fact that data came from a specific type of cancer meant that we also needed to keep track of which genes and proteins are important for this specific disease. Furthermore, the amount of data was quite limited, which also makes machine learning more difficult.
You developed an algorithm that can search large amounts of data and find mislabeled patient data. What makes this algorithm ’intelligent’?
”We may want to nuance the vocabulary somewhat, I’m not sure we can call the algorithm intelligent. We have used machine learning to analyse different types of data to determine how well they fit together. If we can predict that two different data sets for the same patient do not seem to fit together, we can claim that there may have been a confusion, that is, data has been mislabeled. Thus, the ’intelligent’ comes from using machine learning, which is part of artificial intelligence, to predict when a mistake may have occurred.”
What do you hope your solution will contribute to in the healthcare system?
”That is a difficult question! The problem with mislabeled data affects the healthcare analysis. There are many factors that affect the quality of the data that you want to analyse, where mislabeling is such a factor. Our algorithms can be used to detect and correct data errors. Quality assurance of data is the largest gain, I think. This can in turn lead to more reliable analysis, but it is quite difficult to say what it will do for the entire healthcare system.”
Text: Louise Wandel
Photo: Dan Bergmark
The winning team consists of Mattias Ohlsson from Halmstad University, Carsten Peterson, Patrik Eden and Björn Linse from Lund University and Anders Carlsson from the company Bionamic. Their method will be presented at the conference RECOMB in Washington on May 5–8, and will be published in the journal Nature Medicine together with the organisers later this year.
The FDA competition External link.
RECOMB conference External link.
Mattias Ohlsson (film) External link.
Article in Swedish:
Vinst i forskartävling för kvalitetssäkring av patientdata External link.