The Future of Healthcare and AI

We are sitting at the cusp of major digital transformation in healthcare. Over the past few decades, we have built up the infrastructure and cross-organizational knowledge-sharing to enable AI in healthcare. Yet, the COVID-19 pandemic demonstrates that there is still further unmet potential for the healthcare sector to encompass machine learning solutions. 

But what is the latest of AI in health? How do we properly engage stakeholders along the complex value chain? What risks and dangers can we anticipate as healthcare adopts digital solutions?

To explore these questions in-depth, we gathered a group of senior healthcare executives, physicians, startup C-suite executives, faculty, and PhD/MD/MBA students for a small-group conversation about the Evolution of AI in Healthcare. 

Here are some of the major takeaways:

1. Machine learning (ML) interpretability and physician / computer scientist divide remain the biggest challenges and opportunities in integrating AI in healthcare.

Significant gaps exist between physicians and computer scientists when it comes to implementing AI solutions in healthcare. Clinicians are often focused on what is in the best interest of patients, while data scientists frame their thoughts around standard statistical models. Generally speaking, it takes a lot more than an AUC or accuracy to know whether a particular model or technology is appropriate for a given patient. In particular, many algorithmic decision support tools have proven to be clinically lousy.

To a certain extent, there is also a cultural barrier that prevents healthcare organizations from embracing AI solutions. The lack of interpretability amongst physicians is a significant barrier towards widespread clinical adoption of machine learning (ML) solutions. For the most part, ML has been presented as a black box or science fiction that maps the inputs with outputs without any causal explanations, leading to a great degree of skepticism within physicians.

"By peeling back the layers to make AI less confusing, physicians and patients alike will be able to more easily digest these solutions which could transform healthcare as we know it."

Our discussion group discussed several action steps in this area, including that when presenting AI in medicine, data scientists should focus on the clinical outcomes and impact of AI on care (eg. higher revenue collection in hospital systems, better diagnostics in oncology) and clearly demonstrate the “intermediary steps” between the input and output. By peeling back the layers to make AI less confusing, physicians and patients alike will be able to more easily digest these solutions which could transform healthcare as we know it. 

2. Healthcare investing continues to be an ever-evolving landscape, and has significant barriers to entry.

Healthcare and healthcare and AI investing is undoubtedly exploding in many different areas, including drug discovery, telehealth / mental health, diagnostics/therapeutics, etc. The question remains whether the explosion of healthcare investing reflects a genuine market need or is creating another bubble like the dot-com bubble back in 2011.

Since the start of the pandemic, a lot of money has been poured into telehealth and mental health. In some way, this is quite a frothy market as many follow-ons do not seem to add significant values. The jury is still out on whether there is enough demand in the market, and we are still exploring best practices to guarantee a consistent level of care. Broadband rollout greatly impacts the equity of access to telehealth care.

Regulatory compliance also represents a significant barrier, and factors heavily into investment decisions.

Regulatory compliance also represents a significant barrier, and factors heavily into investment decisions. For example, high regulatory demands in the medical devices market may prevent many from entering and surviving in the market, and those that do enter the market are faced with investors who are constantly trying to price in regulatory risks. The venture studio is an interesting model to help startups overcome the high barrier of regulations as founders bootstrap themselves.


3. AI-driven solutions such as natural language processing are being adopted at a very slow rate in hospitals, and lack of interoperability in electronic health records is a significant barrier.

AI-driven innovations are penetrating hospital systems at an incredibly slow rate. This is due to a combination of cultural, operational, and infrastructural challenges.

Many machine learning tools such as natural language processing (NLP), the ability for computers to understand the latest human speech terms and text, are critically needed to help optimize hospital operations. Robust NLP tools could manage the vast amounts of unstructured medical data in electronic health records (EHR), extracting insights and automating operational procedures. Earning physician trust of these solutions is a priority. ML solutions can and should be integrated into the background of everyday healthcare. For example, prior authorizations and medical imaging scans are low-hanging fruits for AI plug-ins in hospitals.

The lack of interoperability among different hospital entities and even within organizations presents itself as a significant hurdle in embracing AI innovation. Getting access to health records is very complex, as there is no standardized EHR system in the US. The rollout of programming standards in health records are great steps forward, and some providers like Redox have recently provided products in this space. However, there is still work to be done. By lowering the barrier to accessing EHR records, we can both help patients understand their own care journeys and optimize on top of them using machine learning.

4. The future of care lies in telehealth and remote monitoring.

Digital technology is revolutionizing the care delivery model, and the COVID-19 pandemic has further accelerated the rate at which organizations adopt telehealth. At the beginning of the pandemic, hospitals were given a few months to incorporate remote health solutions into their infrastructures. In India, for example, platforms not specifically tailored to healthcare such as Twilio, WhatsApp, and Venmo play a large role in virtual consults, appointment requests, electronic payments, and prescription issuing.

There is also significant adoption of AI and digital tools in patient-facing capacities, but not in the operating rooms or perioperative spaces. There is a tremendous amount of opportunity for digital resources to change the care delivery in OR and Pre-Op. Practically, there are only so many beds in a hospital and telehealth and clinical decision support tools are becoming essential to stopping the healthcare system from being overwhelmed.

When implementing these tools, it is very important to have physician buy-in. A top-down dictation by hospital administration will not be perceived very well. Patients voicing preference over these tools would help pressure organizations to implement telemedicine. Organizations like the VA, for example, went from 3000 to 40,000 per day telehealth visits due to consumer preference. A real movement in telemedicine is forming now among consumers to not waste time going back and forth to doctors’ offices when care can be delivered remotely.

Overall, healthcare is a very complex topic and several questions remain unanswered for now. We hope that by convening so many stakeholders to have conversations like these, we can push the field along.