Study on natural language processing and machine learning to develop oncology decision aid for multiple myeloma in Alberta
In Canada, approximately 2,900 people were diagnosed with multiple myeloma in 2017. Unique, targeted therapies are currently in development for multiple myeloma and will likely alter the disease course and survival rates.
A critical step in the delivery of targeted therapies is the accurate identification of eligible patients. Testing for multiple myeloma starts with histopathology and the results of that inform further testing. Because of this, histopathology results must be accurate and easily combined with other patient characteristics. In Alberta, histopathology results are presented to physicians as written pathology reports, where extracting the relevant information may be time-consuming. Moreover, studies show that the traditional narrative style of reporting may lead to omission of clinical information and may be subject to misinterpretation.
Medlior Health Outcomes Research has embarked on a proof-of-concept study that will employ natural language processing to extract the relevant data elements from free-text histopathology reports and create a searchable, structured dataset which can be combined with other electronic health data. What is most exciting about this study is that this data will then be used to develop machine learning algorithms with the potential to perform tailored survival and risk assessments, predict patient response and relapse, and identify a clinical course with the highest efficacy and lowest cost.
Medlior has partnered with the University of Calgary and Mount Royal University to collaborate with two students to conduct this research study and is currently underway for the fall semester 2019. This will be an ongoing study, Medlior will be sending updates once the study concludes.