Yuriy Klochan/123RF
 

What Are The Challenges To AI Adoption In Healthcare? 26 Experts Share Their Insights

  • 23 September 2019
  • Sam Mire

What would you do to make your doctors visits and overall healthcare experience less of a headache? Would artificial intelligence and other technologies look more appealing as a healthcare tool if they could save you tangible money and time? Before this is possible, the challenges to AI adoption in healthcare have to be faced head-on.

These industry professionals shared with us the greatest hurdles to AI adoption in healthcare. Here's what they said:

1. Randy Hamlin, Vice President and Segment Leader for Point-of-Care Ultrasound at Philips

The number one challenge to adoption in healthcare is development. If you think about AI in other industries, it’s relatively easy to access the information needed to train the algorithm. But in healthcare, this information is highly regulated. Take ultrasound. There’s a lot of regulation around the ownership of images. That makes it trickier to get the sheer volume of images necessary to train, design, develop and validate a safe, effective, and meaningful AI-enabled device for various clinical applications.”


2. Laura Marble, VP, IT at Blue Cross Blue Shield of Michigan

“AI, Machine Learning and Deep Learning all are dependent on the data integrity itself. Health care organizations have always managed large amounts of data, but not necessarily in a digitized format. They face added complexity when modernizing processes due to strict patient data privacy regulations. It can prove difficult to cleanse data similar to how a non-health care organization might do to optimize the integrity of the insights. Fortunately, most health care organizations are prepared for such compliance challenges when transitioning to disruptive technologies such as AI. Health care organizations should study AI integrity in other industries to better understand the blueprints for success.”


3. Niven Narain, Co-Founder, President & CEO of BERG

“The important thing is that the algorithm and/or the AI has to be embedded in a dashboard format for anyone to use effectively in their day to day job. That transition is essential to making AI an everyday component in healthcare. Large provider networks and hospital systems have been leading the way in adopting and developing cutting-edge computational tools, but usage of these tools in all provider settings remains to be accomplished.”


4. Antony Edwards, COO of Eggplant

“The challenge is all about risk and patient/doctor acceptance. Handing over critical illness diagnosis to a machine is something not many are prepared to do. Despite being positioned as augmenting doctors, i.e., proposing some options that the human can dismiss or use as a starting point. Many see this as a slippery slope that they ‘don't want to start on. It's all about confidence, and that's a challenge AI has in every mission-critical application.”


5. Maxim Ivanov, CEO and co-founder of Aimprosoft

Industry conservatism that is compiled of many factors. Let’s start with complexity to convince industry stakeholders about the positive return on their investments. This is hard to do without multiple real cases. You need some pioneers to go this path. The next stumbling block is complex stakeholder relationships because you have to convince not only the executive board but the whole clinic’s management chain, including doctors and patients. Unrealistic expectations and a lack of AI and ML skills also add to the challenges of AI adoption. 

Another hinder is data security and compliance with regulations. Data sharing among multiple databases requires special efforts to comply with the HIPAA and FDA. 

So-called “black box difficulty” is a controversial point that stands for impossibility to understand AI’s technical workings: how a system has come to such conclusions that will be a basis for patient treatment.”


6. Jennifer Hill, Chief Operating Officer at Remedy Analytics

“Understanding that it’s not a magic wand yet. Like many new technologies, AI has tremendous potential when used correctly based upon where the company, the problem and the technology is within its lifecycle. When all things are aligned properly, there is a greater chance of success, which ideally leads us to a solution for the next hardest problem.”


7. Rana Gujral, CEO at Behavioral Signals

“AI has a very broad application spectrum when it comes to healthcare, ranging from medical hardware in ERs( surgery robots) to assistive patient care(monitoring and assisting in recovery), and several in between. Every single application comes with a lot of challenges. The main challenge would be official regulations and approval. Just because someone builds an AI application does not mean it can be used in healthcare.

People's lives depend on proper procedures. Tests and trials need to be run; Doctors need to adopt it and recommend it; Statewide medical associations need to get on board and the State has to issue regulations based on their proposals. Although technology is running at very high speeds, healthcare innovation is not following at the same pace and rightfully so. We can not play with human lives and testing takes time and a lot of data accumulation. Some AI solutions are less invasive and can be tested safely in less time, like social robots with speech and emotion recognition; others, like medical imaging diagnosis, are critical and need time and a lot of data.”


8. Sanket Shah, Clinical Assistant Professor at the University of Illinois at Chicago's Masters of Science in Health Informatics and Health Information Management

“If I can select a 1a and 1b, the answer to this is organizational readiness and costs associated with AI adoption. Organizations must have the expertise, infrastructure, and buy-in from senior leadership to successfully implement an AI program. Well-defined goals and objectives for the use of AI are going to be critical and corresponding results must be delivered quickly. There are many organizations that are already there, but there is quite a bit that is not. It’s not “mainstream” because there is still a learning curve when it comes to AI.

Organizations are evolving and investing heavily in data science and next-generation analytics inclusive of AI and Machine Learning(ML). The other aspect of this is the cost. It’s still a heavy investment for most to undertake right now. As start-up costs continue to fall, wide-spread adoption will certainly increase.”


9. Susan Wood, CEO of VIDA

“Obtaining high quality datasets to train algorithms continues to be a major challenge for many AI vendors.  For small companies in particular, this can be a huge barrier; at VIDA we have an exceptional advantage as we’ve been building our 100,000+ imaging database for over 10 years which includes expert analyst annotated data from over 60 academic and clinical trials and clinical use.”


10. David Maman, CEO, CTO, & Co-founder of Binah.ai

“While healthcare organizations are awash with traditional technologies like monitoring machines and newer tech like computer-guided operating room tools, they still face aging infrastructures that are not technologically capable of integrating a wide range of solutions.

In addition, regulatory requirements such as HIPAA and budget issues keep them from adopting AI at a faster pace. But once organizations and governments will start to adopt and dedicate funds for these projects, they will be quick to understand the immense advantages and savings these technologies can bring, and everybody will follow.”


11. Emi Gal, co-founder and CEO of Ezra

“I think the #1 challenge to AI adoption in healthcare is actually twofold: individuals feeling threatened by the prospect of AIs “taking” their jobs, and ensuring that AIs are well-tested enough to bar them from making errors that could gravely harm patients.”


12. Dr. Anuj Shah MD, founder of Apex Heart and Vascular Care

“The #1 challenge to artificial intelligence adoption in healthcare is regulation. Compared to the use of artificial intelligence in nonmedical settings, the stakes in healthcare are high. Diagnosis and treatment decisions can potentially have deadly consequences if mistakes are made. As a result, rigorous testing and benchmarking of artificial intelligence algorithms will be necessary. This will likely be a costly and time-consuming process to ensure optimal patient safety. This would also require studies looking at proof of concept in larger populations and looking at accuracy of AI in true community settings outside of the research population.”


13. Sean Lane, CEO of Olive

“The #1 challenge to AI adoption in healthcare is that hospital leaders hear the buzz around AI and feel as if they should be adopting it in their organization, but oftentimes, they don’t know where to start. Hospitals need a vendor partner with healthcare expertise who will help them quickly identify candidate AI processes, and stand them up using industry best practice. Our customers find most success when they start with processes that are high-volume, repetitive and error-prone. Applying AI to these workflows first lead to quick wins with big impacts.”


14. Jane Kaye, Healthcare Finance Consultant at HealthCare Finance Advisors

Cost. As a costly investment for a healthcare organization, implementing AI automation technology brings with it pressure to demonstrate an immediate return on investment. And implementing AI technology is an iterative process, creating a variety of organizational impacts that aren’t immediately quantifiable. Organizations need to commit financially to invest in AI technology, and must accept that other organizational changes will result from these technological changes.”


15. Dekel Gelbman, CEO of FDNA

“Without a doubt, getting the doctors’ buy-in through proper integration with workflow. It is critical to design and position AI systems with the value of augmenting clinicians’ capabilities by aiding in the aggregation of and sifting through mass amounts of data. When it comes to changing workflow to integrate AI, we should start by respecting existing workflows, rather than challenging them. Trust is built over time, not achieved in a single battle. Physicians require validation of technology before adoption. Humans need to stay in charge of making decisions in healthcare.”


16. Neal Liu, Co-Founder and CTO of uCare.ai

“A pain point in encouraging healthcare professionals to adopt AI is the second-mover strategy. No one wants to take the risk of implementing the technology first and rather see a few success stories before jumping onboard themselves. There is nothing wrong with this strategy other than the risk of being left behind.”

 


17. AJ Abdallat, CEO of Beyond Limits

“Healthcare must fully embrace technology AI. Routine healthcare administration is hampered by outdated, manual, inefficient processes that lead to poor financial outcomes for providers. Cognitive AI can help organizations connect the dots while shifting from fee-for-service reimbursement to value-based care.

Healthcare AI technology can reduce the friction, errors and cost in registration, scheduling, charge capture, health information management, and billing and collections. The goal is to reduce the number of denied insurance claims, speed explanation of benefits (EOB) reconciliation, improve the quality of information, streamline denial management, and automate processes.”


18. Nagi Prabhu, Chief Product Officer at Solutionreach

As AI relates to patient communication and engagement, it’s fear of change — on the part of both staff and patients. The dominant concern there often revolves around the balance between technology and human interaction. It’s easy to get excited about AI’s potential in PRM (and rightfully so!), but it’s also important for providers not to push the automation too far and create a less-than-ideal patient experience. The question providers are asking is “How does AI best complement my engagement strategy while maintaining human, authentic patient relationships?”


19. Russell Glass, CEO, Ginger

“The biggest challenge to AI adoption in healthcare is the quality and relevancy of the data that is used to train AI systems. The increasing volume of healthcare data is staggering, as it will experience an annual compound growth of 36% through 2025, brought on by emerging technology, chatbots, medical imaging, and other healthcare advancements. However, without the presence of high-quality data that has been meticulously curated, building AI and machine learning into any clinical workflows will be difficult. Healthcare data can be subjective, inaccurate, and often siloed, which can make it difficult to establish a patient’s full healthcare profile and make informed decisions on those recommendations. Currently, clean data sets that fit the bill are also difficult to obtain and access, due to very real privacy concerns and HIPAA regulations.”


20. Kevin Harris, CEO of CureMetrix

One of the major issues facing AI is trust. Doctors and patients must be able to trust in the efficacy of the AI software. They need to know they can count on the AI application to deliver accurate results substantially equivalent to or better than the current standard of care.  Performance transparency is key to building this trust. No different than a drug or medical device, healthcare users of AI need to know where the AI works, where it doesn’t, and how and when to safely use AI solutions.

Another key issue facing AI is fear among potential users, like radiologists. Some have expressed concern that AI’s computer algorithms will take the place of their clinical expertise. Many in the industry, however, do not hold this view. Doctors with an eye to the future see AI as an asset – one that will aid radiologists, not replace them. 

A third challenge to AI adoption is concern over company quality. Healthcare customers need to know they are working with a quality AI company – one that is stable, secure, HIPPA compliant and with products that are FDA cleared, such as our cmTriage workflow solution. They’ve got to know that the company is going to be here five years from now. AI companies have to demonstrate that level of quality to potential customers in order for the industry to advance.”


21. Shantanu Nigam, CEO of Jvion

Adopting AI requires a change in mindset. AI should not be viewed in traditional technology terms. Rather, it is critical that consumers of AI within healthcare understand it as a way to augment and focus their efforts. AI outputs, when created the right way, should be simple, elegant, and easy to consume. They should direct attention to the areas of greatest impact and should reduce distraction. The technology adoption environment within healthcare is hindered by change fatigue resulting from years of Meaningful Use mandates and—in some cases—promises that haven’t come true. Driving the mindset needed to adopt AI within this kind of setting is challenging at best.

Noise from unproven solutions in the market adversely affect any change in mindset and can cause more patient harm than help. New venture investments advance technology but also put a lot of pressure on new entrants to quickly meet revenue and hit sales targets. New solutions should spend a few years within a health system validating their AI before rushing to the market. Further, it is critically important that any AI solution include a methodology for evaluating performance across resources and patient impact.”


22. Chris Bouton, founder & CEO of Vyasa Analytics

“The #1 challenge to the adoption of deep learning algorithms in healthcare is the same challenge as we’ve had previously in the healthcare space, data access. For a number of valid and important reasons data, is kept very secure in the healthcare arena. Deep learning algorithms, though, require large amounts of clean normalized data to be effective. Hopefully, we will see continued momentum toward standardization and anonymized open access to data sets for training and research purposes.”


23. Keith Figlioli, General Partner at LRVHealth

“Given all this interesting news, there is still the reality of bias in the models and false positives. I think this will be overhang on AI/ML for quite some time in healthcare. If you look at the IBM Watson set of issues, this will be an ongoing debate.”

 


24. Patrick Gauthier, Director of Healthcare Solutions at Advocates for Human Potential

“Privacy. Recent massive breaches of PHI (tens of millions of people affected) are adding up and causing people to doubt whether or not their information is protected. Clearly, it is not adequately protected. AI works because it is fed information. No info, no AI.”


25. Haza Newman, co-founder and CEO of Geras Solutions

“Once again, I do not believe there is a #1 challenge. I think there are two significant challenges within AI adoption in healthcare.

The first is maintaining the status quo in care and continuing analogue or dated care models established many decades ago. Thus it is incredibly important to find and work with carers or specialists that believe in using ICT to transform the way patients receive care. Using such tools should improve efficiency and guarantee better results while giving care units more time to focus on the patient and understanding the root causes of whatever they require assistance with. These tools should be so easy to use that it feels like something they would normally do, without having an extra load of work to maintain with the new solution. Thus, the professionals are the ones that should be involved in the processes of creating such tools to ensure their use and acceptance. I believe using these tools will lower misclassifications and provide the right individuals, who require help now, receive care while maintaining a watch over those that do not need that care immediately.

The second major challenge is regulation. I believe that regulation is incredibly important. It is in place to maintain and ensure the safety of individuals that require care, but manoeuvring regulation bodies can be complicated when each market is unique, and each has a different body one must work with to ensure legal and regulatory parameters are maintained. Is it a municipal issue, is it a state issue or is there perhaps another acting government body? This can often be costly and difficult to manoeuvre without assistance or a network that understands the given market.

One other perspective I reflect on often is that it is often difficult to lobby for change when procedures and regulation have already been established for such a long time in a particular country. Thus, it is often exciting to focus on developing countries that are still establishing those regulations and are perhaps more open to new ideas or technology. You must appeal to the decision-makers and work with them to ensure acceptance and adoption of AI into healthcare.”


26. Joe Polaris, Senior Vice President of Product and Technology at R1 RCM 

Technology teams within most healthcare organizations are overly siloed and stretched too thin, especially as they still utilize outdated technology – like fax machines – in their daily workflow. They simply don’t have the in-house capacity to participate directly in the design and development of AI-enabled technologies to help solve their challenges. To be truly successful, healthcare providers need a specialized partner who understands both the technologies and the unique nuances of intricate healthcare workflows, such as the revenue cycle.”


Have expert insights to add to this article?

Share your feedback and we'll consider adding it to the piece!

ADD YOUR INSIGHTS

About Sam Mire

Sam is a Market Research Analyst at Disruptor Daily. He's a trained journalist with experience in the field of disruptive technology. He’s versed in the impact that blockchain technology is having on industries of today, from healthcare to cannabis. He’s written extensively on the individuals and companies shaping the future of tech, working directly with many of them to advance their vision. Sam is known for writing work that brings value to industry professionals and the generally curious – as well as an occasional smile to the face.

Comments

COMMUNITY