Five ways for AI startups in healthcare to overcome barriers to innovation
Estimated reading time: 12-14 minutes
Imagine you are a healthcare entrepreneur who has developed an algorithm that is able to detect a certain disease with 98% accuracy based on medical images. You are eager to get your solution into the hands of healthcare providers as soon as possible. Its life-saving potential drives you to put in endless hours of hard work. Yet three years later, you are still struggling to get your algorithm embedded into clinical practice, with your worries piling up as you are running out of cash.
Who would have thought it would be this hard to make your innovation work?
This is the conundrum that faces many healthcare entrepreneurs − especially in artificial intelligence (AI), where we are all exploring unchartered territory.
Building an AI solution is one thing. Getting it implemented in clinical practice is another.
Over the last three months I have had the fortune to interact with 19 startups from around the world that participated in the global Philips HealthWorks AI Startup Program, together with leading hospitals that are eager to collaborate on new solutions that help treat patients better at lower cost.
It was incredibly inspiring to see so much entrepreneurial energy. But we didn’t shun the hard questions that can make or break success further down the road. From this it became clear that the challenges are the same for all AI startups, irrespective of geography or clinical domain.
Based on what we have learned during this and previous startup programs, here are five ways to help you get your AI solution from vision to clinical practice.
1. Validate your AI solution early − both with users and buyers
The first key to successful innovation is to validate your AI solution as early as possible − not just with the people who will be using it, but also with the hospital decision makers that will be paying for it.
As an innovator, it’s only natural to start with a technology in mind. Developing an AI algorithm − and training it to perform better and better − can easily consume all your time and energy.
The risk is that you get so absorbed in thinking about the solution, that you don’t spend enough time understanding the problem. Or you may have too narrow a view of what the problem really is.
The risk is that you get so absorbed in thinking about the solution, that you don’t spend enough time understanding the problem.
How will your algorithm fit into the workflow of a radiologist? Will it act as a pre-screening tool that helps radiologists focus on the most critical cases first? Will it support a quality check afterwards? Or, if you take a broader view, could it support collaboration between clinical disciplines by tying different types of patient data together?
Think about the movations and considerations of your buyers too.
What will convince a hospital CIO or department head to invest in your solution? Will it improve patient outcomes? Will it increase patient throughput or reduce cost? What about implications for reimbursement?
Already in the second week of our program, we brought together AI startups and healthcare providers in the same room − to address these questions. To shift the mindset from technology to value creation.
Initially, this was a sobering experience for some of the startups. They had to acknowledge that they hadn’t given sufficient thought to all of healthcare’s complexities yet. But by putting their ideas to test, they now have a much better understanding of the needs they must address and the context their solutions should fit into.
2. Embed your solution into a platform to make it consumable for hospitals
Another theme that emerged from the conversations between startups and hospitals was that hospitals are increasingly looking for platform solutions, not point solutions.
Most current-day AI applications are narrow point solutions. They focus on one specific disease, or on one specific step in a physician’s workflow.
Imagine having to implement and maintain hundreds of different point applications in a hospital, for different diseases and workflow elements. This quickly becomes unmanageable. As powerful as your solution may be, a hospital may not be able to consume it.
What hospitals are actually looking for is integrated solutions − platforms that connect the dots across systems and departments to orchestrate care around the patient.
Most current-day AI applications are narrow point solutions. But what hospitals are increasingly looking for is integrated solutions − platforms that connect the dots across systems and departments to orchestrate care around the patient.
As one CIO of a large hospital network put it: “I run about 10,000 applications at 150,000 different terminals. The one thing I do not want is more applications I need to manage.”
This can be confronting when you are pouring your heart and soul into building the 10,001st application.
But the best entrepreneurs listen carefully and adapt fast to succeed.
What this CIO’s feedback really points to is an opportunity for startups and large enterprises to team up.
Startups have the agility to rapidly develop new AI solutions. Enterprises such as Philips have the platform infrastructure to make these solutions consumable for hospitals. This makes the combination of the two particularly compelling for hospitals looking to innovate.
Working with a large enterprise also allows startups to offer hospitals the required customer service and training facilities − which are essential to a successful go-to-market strategy, but hard to set up from scratch when you’re building your business.
3. Think strategically about data − across systems and workflows
A third barrier to AI innovation in healthcare is that the right data may be hard to come by.
This compels you to think carefully about models of collaboration around data. For example, a hospital may be willing to make data available, with patient consent, in exchange for access to the AI solution you are developing. But there’s not one established best practice yet. It’s worth exploring different models of collaboration, within existing legal frameworks, based on your specific data needs.
Also bear in mind that a lot of data in healthcare is unstructured, captured in notes of various kinds. Make sure to reserve enough time and get the right people on board for annotating data and preparing it for analysis. This can easily take up 80% of your time.
One additional challenge that you may run into is that data in other hospitals may be structured differently, because they have different workflows and technical infrastructure. This means your AI solution may work in one hospital, but not in another. Thinking about this early will help you adapt more easily.
Increasingly, your biggest asset will be your ability to access and connect data across hospital systems and workflows.
Learn more about Philips adaptive intelligence - applying AI in a meaningful way to improve people's lives.
Increasingly, your biggest asset will be your ability to access and connect data across hospital systems and workflows.
For example, if you are able to combine insights from medical images with existing EMR data, you can paint a more complete picture of a patient and the disease you are targeting, with a more complete understanding of the workflow you are trying to optimize. This is particularly relevant for diagnosing and treating diseases such as cancer, which relies strongly on multidisciplinary collaboration.
Start small, but think broad.
4. Develop a regulatory strategy to reconcile speed with compliance
Another source of headache for healthcare entrepreneurs is that regulatory frameworks in healthcare are strict. And they have yet to be fully clarified for AI innovation.
How to navigate these sometimes-murky regulatory waters without losing steam?
Of course, regulation is there for good reason: patient safety is paramount. The last thing you want to do is cut corners.
But the reality, if your AI solution requires FDA clearance as a medical device before you can bring it to market, it may take you years to get there. (The situation may be different depending on your target market.)
Such a long timeframe could stretch the patience of your investors to a breaking point. It also carries a huge risk of waste if you eventually discover that there is no market for your solution.
So how do you reconcile compliance with speed?
One route worth exploring is to develop your minimum viable product (MVP) in the research environment of a hospital, rather than aiming for application in clinical practice directly. This creates opportunities for testing and learning, which is so essential to lean innovation, while staying within regulatory boundaries.
One route worth exploring is to develop your minimum viable product in the research environment of a hospital, rather than aiming for application in clinical practice directly.
By collaborating with hospital researchers you are also creating ambassadors who can help you sell your AI solution to clinical practitioners once it’s more mature.
5. Think ahead to scale successfully
Once you have developed your AI solution, what will it take to scale it across settings? And how does that affect your decisions today?
As we have seen, it pays dividends to think about long-term strategy while you are developing your MVP.
Two last considerations are worth mentioning here.
First, if your ultimate vision is to scale your AI solution across markets, it is particularly important to think ahead.
For example, if you develop an algorithm based on a Chinese data set, the results do not necessarily translate to the US. There may be relevant differences between people in both countries, or you may be required to provide clinical proof based on local datasets. The two health systems have different needs, and each has their own regulatory standards. If you optimize your solution for a certain market, or only use data from that market, you may compromise your ability to enter another market.
So if you have the opportunity, familiarize yourself early with different markets, and explore the possibility of running pilots or clinical trials in different countries as a future step.
Familiarize yourself early with different markets, and explore the possibility of running pilots or clinical trials in different countries as a future step.
Finally, I was struck by what the COO of a hospital said in one of our sessions with startups. He invited them to think across clinical areas. That is, if you have developed an AI solution in one area, say cardiology, would you be able to apply elements of that solution to another area, such as oncology?
A lively discussion ensued, and that’s precisely the kind of discussion which makes it so exciting to be in the same room with startups, care providers, and colleagues from Philips. None of us has all the answers when it comes to the future of AI in healthcare. But it sure is easier to find them together.
During the Philips HealthWorks Global Breakthrough Day AI on December 12 in Eindhoven, the participating startups present their learnings to leaders in the health tech industry. Read more
About Innovation Matters
Innovation Matters delivers news, opinions and features about healthcare, and is focused on the professionals who work within the industry, as well as Philips as a cutting-edge health technology organization. From interviews with industry giants to how-to guides and features powered by Philips data, our goal is to deliver interesting, educational and entertaining content to empower and inspire all those who work in healthcare or related industries.
Alberto is the Head of Philips HealthWorks at Royal Philips. He is a seasoned executive with extensive international experience, leading multinational teams and conducting business across multiple geographies (US/Europe/Asia). HealthWorks’ mission is to boost breakthrough innovation whilst fueling a culture of entrepreneurship within Philips.