Speedy data improves traumatic brain injury monitoring A pilot programme involving NHS Greater Glasgow and Clyde and the University of Glasgow’s IDEAS research group has incorporated Aridhia’s data modelling and analytics expertise to investigate real-time data in patients with traumatic brain injuries. Working with anonymised data from patients in the Neuro-Intensive Care Unit (Neuro-ICU) at the Institute of Neurological Sciences in New South Glasgow University Hospital, the collaboration aimed to improve the down-sampling of high-frequency patient data to allow additional clinical algorithms to be applied in the future, helping to inform clinical decision-making and ultimately improve clinical practice. Traumatic Brain Injury (TBI) is a growing public health problem
worldwide. The World Health Organisation predicts that TBI will become a major cause of death and disability by 2020. Many survivors have serious and long-term morbidity, placing a significant burden on family and caregivers, so it is imperative that innovative approaches to the collection and analysis of patient data are used to deliver improvements in the diagnosis, planning and assessment of clinical treatment. The IDEAS group wanted to investigate ways to swiftly translate physiological and clinical data modelling into clinical practice. They had developed a down-sampling algorithm using the R statistical package that performs peakto-peak waveform signal processing on Neuro-ICU data. However, running the algorithm on a sample of data for a single patient collected over a 12day period took 16 hours to process, an unrealistic timescale for researchers who need to tune an algorithm or develop and compare different models. There was therefore a need to reduce the time taken to down-sample patient data, so that data collected could be used to inform the medical response to TBI at the bedside. A sample of ‘vital signs’ data, collected from the NeuroICU, along with an existing algorithm, was deployed into Aridhia’s AnalytiXagility data science platform and
24
UK Lifescience Industry Magazine
safe haven. These time series data were collected across seven channels, spanning over 470 million rows. Approaching the problem in three steps, the collaborative team firstly reviewed and optimised the algorithm for a single processor, secondly the data storage model, and thirdly the parallel computation of the algorithm. By employing advanced data processing techniques designed to determine the optimal section size from the data sample, reduce the number of iterations of the algorithm, and exploit the multiple processor cores available in the AnalytiXagility platform by running the algorithm in parallel, the team was successful in reducing the time taken to process the original data sample from 16 hours to 48 minutes. The success of the pilot study has led to further funding being awarded by Innovate UK in order to expand the progress and translate the work done with Neuro-ICU waveform data into a real-time clinical setting. This more detailed project, CHART-ADAPT, will see further development of a data and analytics framework that will enable clinical teams in Neuro-ICU to run physiological algorithms on highfrequency patient data, a significant innovation to address the traditional barriers of translating research into services. By exploiting clinical data in a real healthcare setting, it is hoped that clinicians will be able to make earlier decisions about treatment, and that through reduced length of hospital stay and in-hospital mortality, it will lead to more cost-effective healthcare delivery.
www.aridhia.com