“McLaren has been around for 52 years,” explained chief medical officer Adam Hill, pictured. “For 50 of those years, we’ve been racing fast cars every other weekend. For 27 years, we’ve been building super cars and hyper cars for the road through a number of different automotive companies.
“But, equally, for 27 years we’ve had a technology company, now called McLaren Applied Technologies (MAT), that was established to develop sensors. Sensors at that time were still large and rather expensive but had got to a size where they were capable of capturing data in F1 cars.”
Today, MAT not only sells sensors, but also telemetry and engine control units, data visualisation and simulation software for use in a range of motorsport series, including F1, Nascar and Indycar.
“Building upon that motorsport electronics systems foundation, MAT started to look at other sectors and that’s when, 11 years ago, we had our first foray into health,” Hill added.
In healthcare, MAT concentrates on providing processing platforms. The company has recently been deploying physiological sensors on its drivers and pit crew during training to optimise performance.
“We send our teams to race in foreign climates, across multiple time zones, in hot weather, cold weather, humidity and different times of the day and night. That can create a whole host of physiological stresses for the team, which we can then analyse in order to mitigate their impact,” Hill said.
For this, MAT uses a processing platform developed initially for the pharmaceutical company Pfizer in 2006 as a proof of concept wireless monitor for mothers in their third trimester of pregnancy. Now being deployed for a range of clinical health trials, it is currently in its fourth generation.
According to Hill, the platform is derived from the time series data visualisation tool that MAT uses in motor sport. Called ATLAS, it builds on the same set of competencies required to capture, analyse and derive insight from time series data.
“It is highly configurable,” Hill claimed. “It would take an engineer to deploy it and set it up and capture data appropriately. But, equally, that engineer can deploy this platform to support a whole host of applications.”
The prototyping platform allows multiple sensors to be plugged in and captures, in a non-invasive fashion, physiological parameters, which are sent to a gateway.
“The gateway has a number of different channels which can capture data,” explained Hill. “It has a PCB at the centre and a radio device which can communicate and push data into the cloud.” There, algorithms can analyse the data.
“MAT has a data science team that develops algorithms manually across data sets that we have captured,” said Hill.
“Typically, algorithm development is in the cloud and, once those algorithms have been validated, the team wants to reduce or improve their computation of efficiency, so they can reside on the gateway, or even on the sensor or the data-capturing device.”
According to Hill, MAT was the first to run time series analyses in real time, with large amounts of data, in order to derive insight.
The difference between ATLAS and the healthcare processing platforms is said to be that ATLAS can consume many more channels of data; is significantly more complex; and requires more computation. The healthcare platform, on the other hand, is a lighter version, with limited channels to capture data.
One application of the platform is at the Birmingham Children’s Hospital (BCH), where it monitors children with heart problems. The biotelemetry platform – called LifeInsight – captures time series data from sensors, such as a heart rate monitor, on children in a cardiac step down unit, which is a hospital unit providing care between that of an intensive care unit and a normal ward.
According to Hill, the platform can fuse up to eight channels of time series data, although it is not currently using that amount in the BCH project.
The data is fed to a system that runs a series of ‘what if’ analyses across the data set in order to determine whether the vital signs are likely to end up out of the thresholds set for a particular child.
“Of the last 650 patients with whom we’ve deployed this system, we have captured a number of adverse clinical events before the clinical staff would have ordinarily identified them with the more conventional vital signs monitoring system,” affirms Hill.
There were several challenges involved with this project.
The first was making the legacy IT systems interoperable with the platform. The second was managing information governance around the data set.
“Clearly, the nurses and doctors need to understand from whom the data is collected, whereas those that are architecting the solution or performing a maintenance upgrade on the system need access to the data, but shouldn’t have access to identifiable data in any shape or form,” said Hill.
A similar partnership between Google DeepMind and the NHS is developing an app that will produce medical alerts for clinicians. Patient data will be processed according to an algorithm and relevant information delivered as alerts to the doctors and nurses.
The hospital has agreed to share five years of historical data on patients, as well as real-time information on their status. Previous efforts by the UK Government to create a medical database floundered over concerns about medical confidentiality.
According to Hill, the final challenge was the necessary shift in clinical thinking. Medical practitioners generally rely upon raw data sets in order to derive insight and rarely rely upon decision support tools that use statistical methods on which to base clinical judgments.
“There’s a shift in the way in which clinicians think, from being very deterministic or standard operation procedure driven through to being probabilistic in the way in which they consider clinical intervention,” he added.
This shift in thinking is demonstrated by the clinicians’ acceptance of LifeInsight.
A degree of tailoring is required for different therapeutic conditions. As Hill professed: “You can’t just throw technology at a human and hope to capture the relevant information.”
He discussed one case where the platform had to be adapted to measure characteristics of motor neurological disease.
“To show that a drug has an effect on a disease, you need to be able to measure characteristics of that disease. So we focus on the discovery of novel biomarkers which would allow us to better study the drug’s capabilities.
“A biomarker we have discovered of disease progression in motor neurological disease is voice deterioration. It’s not necessarily something that can be picked by a clinician from day-to-day or week-to-week. But by following this biomarker’s progress in a clinical trial with a sensor, or a mechanism for data capture, scientists know whether their drug is working or not.”
