The data ecosystem: fine-tuning surgical performance
A data-lead at Proximie discusses the power of the platform’s software-first approach and the company’s ambitions to “establish a global ecosystem that is compliant and able to collect data from a wide variety of sources, from many different regions, patient types and surgeon backgrounds.”
It takes two weeks, from introduction to integration, for Proximie to get into any hospital, anywhere in the world. That’s one of the many things that sets Proximie apart; it’s very nimble in terms of being able to access the market and surgeons in order to quickly establish a data stream coming out of the operating room. Working with huge companies, it can take many months — if not in some cases a year — to put in place the GDPR and privacy agreements needed, but when you look at the Proximie solution it’s really very simple.
Essentially, Proximie is providing surgeons with a compliance ecosystem that works well with hospital staff because it doesn’t change the operating room; we’re not offering a piece of equipment that they have to formulate their whole day around — it fits easily into the ecosystem of any operating room.
It works across medical device companies because they want to accelerate revenue orders across hospitals, and they are interested in collecting and analysing data about their workflow and how surgeons operate their devices. So we’re aligned across both groups, and everybody is quite happy to pull Proximie into the operating room — it’s much more of a pull than a push — and that’s what’s given us a lot of traction.
Surgical data is currently very limited; what goes into the operating room is typically the Electronic Medical Record with the patient’s name, comorbidities and maybe some preoperative images — it’s a really slim data set without much richness to it. What comes out of the operating room is three or four sentences dictated by the surgeon saying “we did this, this and this,” and then the patient may or may not be followed up at a later date. The surgical event is based entirely on the surgeon’s skill set, and their experience of having performed a certain procedure.
For example, there’s a surgeon in Celebration, Florida — just outside of Disney World — who removes 1000 prostates every year. If I’m a patient with prostate cancer living in Kansas or some other part of the world where the state of healthcare isn’t as robust, I’ll probably have access to a surgeon who removes 10 to 15 prostates per year. Which is fine, unless you happen to be a patient that has a certain type of BMI, condition or lymph node state; you might run into trouble because your nearest surgeon doesn’t have the same collective knowledge base as the surgeon in Florida.
“Proximie delivers equal access across the board to that same quality of care, whether you’re in middle America or North Africa; the way we’re doing that is through video, and video is a very rich data set.”
Typically, people don’t consider video to be data; usually they’re thinking about structured data like demographic information and date of birth. But video data has a huge amount of content that has all manner of applications, and baseball is a great example of this. There’s a company that can set up multiple cameras in a baseball stadium, so that for any given baseball pitch they can record the angle of departure from the hand, the angle of the arrival at the glove, the speed of the ball and determine whether it’s a slider, curveball or fastball. But they go beyond categorising the pitch; they can tell the exact angle the ball was travelling at any point on the trajectory of its flight. That’s real data that’s never existed before, about the split second between the ball leaving the pitcher’s hand and arriving in the receiver’s glove; you’ve created a data centre on that event that didn’t exist before. In surgery, the operation is the flight of the ball, and we’re not just recording cases but we’re able to analyse things from those cases — looking at each step of the procedure, or how far along in the procedure certain measures are taken and so on.
Alongside data collection, it’s vital to ensure that the data is diverse and, when it comes to medical data, ensure that it is global. The ‘racist soap dispenser’ is a simple illustration of the potential unfortunate consequences of a narrow data set; you have two guys trying to use an automated soap dispenser in a bathroom, and it dispenses soap for the person with light skin but not for the person with darker skin — and that’s an uncomplicated example of not having tested the dispenser using a diverse range of skin tones. Now look at that dynamic in a medical context; if a small company wants to develop an algorithm that will identify polyps during a colonoscopy, they would look at the publicly available data sets and develop their algorithm based around that. The problem is those data sets are limited, and typically it’s going to be a white, middle-aged male data set, and that’s what’s going to determine diagnostic capability — underlying data that is in some way biased. Instead, we need to establish a global ecosystem that is compliant and able to collect data from a wide variety of sources, from many different regions, patient types and surgeon backgrounds.
Once you have those two components in place — rich data, from diverse sources — there are some exciting applications from a medical perspective.
“Right now, in the analogue environment of the operating room, it’s very difficult to say exactly what happened in terms of surgical technique.”
If we are able to measure the procedure in any number of ways, however — not just the overall procedure, which is a very general metric, but each step in the procedure and the efficiency of the instruments — then you can break it down to the level of efficiency of the surgeon’s every movement. And it’s not just the surgeon you’re focussed on, but everything that happens around the patient, so that includes being able to analyse how well the operative team is working and collaborating together in the operating room — the setup and the way they provide the instruments actually improves the patient’s treatment and how long they’re in the procedure, because they’re so well-coordinated as a team. For example, there are papers published that show a staff that has been operating together for an extended period of time actually achieves better outcomes for the patient.
“Once you are measuring the data around these procedures down to the millisecond, that’s what gives you the ability to improve efficiency and overall quality of care even further — and that provides a competitive advantage.”
This sort of data is hugely useful to medical device companies as well; how can they make their devices better if they’re only collecting data sets from their own devices? They need a broad and diverse data set that provides insights from many different procedures and many different devices so they can integrate those learnings into their own research and development. If you take the Hugo surgical robot from Medtronic as an example, they just announced they’ll be integrating machine learning into that robot — but by what technique, and what are the data sources that were fed into that? Are they global data sources, or is it just based on the robotic data? Was it narrowed to that medical device company, or was it across every type of procedure everywhere in the world?
So the next step to improve is to use a Proximie-like data set, which is agnostic to the device and agnostic to the demographic in the region, and feed that into how insights are developed — making the device smarter by integrating insights and intraoperative guidance.
At the same time, there is some data that medical device companies simply don’t have access to about their own devices. If a medical device company deploys a device and it’s implanted in 30 seconds, that is not currently something that is known based on the existing data. The average time for implanting the device might be 30 seconds, but then in some cases it might take 45 seconds, and understanding why it was 15 seconds slower in certain instances would be a very important deviation for medical device companies to understand, because time is a crucial dimension in the operating room; the longer things take, the higher the possibility that something might go wrong. With this sort of rich data at their disposal, medical device companies would be able to say what sort of technique should be used to most efficiently implant a device, or precisely where an incision should be made to achieve a better outcome for patients. Hospitals are interested in this data too, because their margins are razor thin, so they’re interested in saving time while keeping quality of care high.
“Identifying which areas of the operating theatre are performing less efficiently is integral to that, but there’s currently no way to extract the necessary data.”
The environmentalist John Muir said, “When we try to pick out anything by itself, we find it hitched to everything else in the universe.” This is something that’s important to bear in mind when you’re developing a data ecosystem — you can’t just pick out different, isolated bits of information and assume you have the whole picture.
We need to actually do a good job of establishing a broad ecosystem informed by the widest variety of information possible, because that is what provides the basis for efficiencies, algorithms and techniques that can provide the highest quality of care for the highest number of people, and the knowledge to do so as effectively as possible.