Learning to care
Using machine learning to improve the efficiency of telemonitoring of long-term illness.
The prevalence of long term illness worldwide is set to double by 2030. Telehealth was strongly championed as a solution to this problem. However, major trials suggest that it does not save time and resources.
The effectiveness of telehealth relies on early identification of problems to enable intervention thereby preventing deterioration or hospitalisation.
However, current algorithms (based on symptoms counting and simple thresholds) are poor predictors of deterioration and generate many false alerts causing increased workload. There is a desperate need to make these predictive systems more efficient.
Using daily recorded symptom and physiological data (~0.5 million daily recordings) linked to admission data from our COPD trial, we used novel machine learning techniques to create personalised, time series based, predictive algorithms that will greatly increase the efficiency and cost effectiveness of telemonitoring.
This project used machine learning to enable sophisticated predictive algorithms to provide clinicians with estimated risks of deterioration based on individual characteristics and previous presentations.
Results will be available in summer of 2018.
Learning to care: Using machine learning to improve prediction of COPD admissions (European Respiratory Journal, 2015)
|Funder||University of Edinburgh MRC Confidence in Concept Scheme|
|Chief Investigator||Prof Brian McKinstry|
|Study Researchers||Mary Paterson|