At the TEDxTelAviv 2018, you said that "algorithms will see us soon." Six years later, some countries are still struggling with implementing EHR. Do we underestimate the efforts needed for fundamental changes in healthcare?
We knew it would be exceedingly challenging because innovation at its core is the science of change management, and changing existing practices is always hard.
Your observation that we're still struggling with digital transformation more in some countries than others is quite correct. However, I'm not sure that every country understands the urgent need to transform their healthcare systems. One reason is that we intuitively perceive the risks associated with innovative approaches but tend to discount the enormous risks of the status quo.
Healthcare, as compared to other industries, is a conservative, fragmented, and highly regulated industry. In some cases, this is for good reasons: It's traditional because it has to be careful – medicine has always been a cautious profession with complex business models. Nevertheless, we've seen many shifts recently, from telemedicine that boomed during the pandemic to AI slowly taking its way to doctors' offices.
Kotter described the eight steps of change management, and the first one is always a sense of urgency. In healthcare, we don't have this sense or don't always realize the platform is already burning. The trick is that without a sense of urgency, change will not happen, regardless of the other seven steps.
Healthcare today is, in most countries, a non-sustainable industry. The need and costs are increasing, the quality varies from clinic to clinic, and error rates are unacceptable. We face overdiagnosis, overtreatment, and then underprovision of the proper care at the right time. We miss a lot of opportunities to improve care when we stick to the status quo.
If health systems don't transform, they will implode since they are in a dangerous vicious circle: increasing population needs and budget cuts lead to staff overburden and burnout, leading to workforce shortages and reduced quality, among others.
What aspects of the current policy are hindering digitalization? Is it a lack of political will or external resistance in the healthcare community?
First, we must dispel some old notions that have become truisms but may no longer be valid. A great example can be drawn from the results of our recent GIHF-AI Study 2023, "Trust in the use of health data – a comparison between Germany and Israel." Many of my colleagues in Germany and Israel were always under the impression that the German public is very conservative and would prefer their health data not to be collected and processed. At the same time, Israelis are considered very liberal regarding data sharing. This was the justification I had heard many times from those who felt Germany was falling behind in digital.
To check if it was true, under ELNET's German-Israeli Health Forum for Artificial Intelligence (GIHF-AI), we did a study on a controlled sample of over 2000 citizens from both countries.
It turned out our assumptions were wrong: There is a general high level of willingness (82.4% in Germany, 81.4% in Israel) to establish anonymized datasets of patient information for research purposes in both countries, and even higher willingness for using data to inform the doctor's work. In both countries, only less than 10% are categorically against.
I encourage revising old beliefs and verifying biases by reaching for data.
Secondly, policymakers increasingly come to realize that a solid digital infrastructure is the basis for any successful, sustainable, effective, high-quality healthcare system of the present, not just the future. If you don't invest now in high-speed internet, Electronic Health Records, and data exchange platforms, you will be unable to provide adequate care in the immediate term. The proper infrastructure lets the data flow. And without data, policymakers, physicians, and patients are flying blind.
Thirdly, you should have a robust system for monitoring, measuring, and reporting healthcare outcomes, which is a solid incentive to enhance healthcare data infrastructure and, at the end of the day, to improve healthcare quality. For that to work, you also need infrastructure and access to data.
Presently, healthcare prioritizes quantity over quality in reimbursement, while prevention is not reimbursed. Put plainly, there is no business model for quality and prevention. Must healthcare undergo substantial reforms to adopt digital innovations?
These reforms, when appropriate, and the digital transformation must happen in parallel.
The infrastructure must be established to support preventive medicine, high-quality reactive care, error reduction, increased system efficiency, and better resource utilization. This digitization brings numerous benefits, including improved patient service and immediate care. Simultaneously, addressing health policies and incentives is vital. Without proper incentives, stakeholders may not engage in proactive and preventive care, hindering its adoption.
Healthcare professionals want to do their job well, but conflicting business models are not helpful. No doubt, there is a need for systemic changes alongside infrastructure and workforce training.
I'm optimistic we are on the right path. Firstly, growing evidence demonstrates the economic benefits of digitization and machine learning-driven care. Secondly, decision-makers are increasingly aware of these benefits, leading to increased resource allocation for digitization and innovation. Finally, the public's demand for technology-driven healthcare improvements further fuels my optimism.
But there are reasons to worry. No one knows how many errors are performed daily in a healthcare system, how many missed opportunities for better lines of treatment exist but are ignored, or how many errors of omission in diagnostic interpretation occur.
And this ignorance is not bliss; it leads to complacency and resistance to change. There's a famous saying about innovation: The key to innovation is creative dissatisfaction." I have a burning sense of urgency to improve care quality, effectiveness, and efficiency. And that helps.
So, what have we done wrong in modernizing the health sector over the past few years?
Well, one of the things is that we didn't pay enough attention to usability and the user experience of healthcare IT systems. Tools and platforms were aimed at serving administrators. Poorly designed in many cases, they made the daily practice of physicians and nurses worse instead of improved.
This can and needs to be corrected. Digital platforms should be developed with the involvement of providers having providers' experience in mind, front and center.
Additionally, we need to establish feedback loops and outcome monitoring. Physicians should receive something in return if they invest time in curating their electronic health records. It's not enough to send data into a black hole for administrative use – there should be a clinical feedback loop providing meaningful information back to them. This is the cornerstone of every platform we create at Clalit.
Lastly, in every healthcare system, we see too many pilots and too few follow-up efforts at systematic implementation at scale. Spending resources on small-scale pilots without considering how or if they can be implemented at scale is a waste of time and energy. Before starting a small-scale project, a large-scale implementation plan must be delineated.
Let's talk about data. Can you give some examples of how access to data and data science saved lives during the COVID-19 pandemic?
We've done two things that have shown the importance of data. One was how we used AI predictive tools to provide real-time warnings to patients about the risk of severe COVID-19. We called 200,000 patients as early as March 2020 to tell them there was a new disease surging and they might be at risk according to our predictive model that was based on our existing predictive models for influenza complications, adapted to COVID-19 using mortality data from Italy and China that were first to experience mortality surges.
Those 200,000 people we told: "There's a new disease coming, be careful, stay at home. And if you want medical care, don't come to the overburdened clinic. Call us, and we'll either give you online care or come to your home."
That saved so many lives. We later used this predictive model to provide Paxlovid proactively to patients at risk. We found patients infected a day before and identified those at high risk for complications. We knew that they only had five days to get the drug, so we told them, "We're going to bring the drugs to your doorstep so you can take them in time to prevent the disease complications."
Another example is that we have been the source of real-world evidence to the world on whether the vaccines work and to what extent. The first publication in the New England Journal of Medicine about vaccine effectiveness of 94% came from us. It was the first real-world study. Some decisions made in the United States and around the world about providing a third dose were not based on a clinical study but on clinical data that showed waning immunity and the effectiveness of the third vaccine in reducing morbidity and mortality burdens.
You are a member of the United Nations AI Advisory Board. What are the most discussed AI-related issues on a high political level?
While many focus on the risks of AI and "Terminator"-like scenarios, I am less concerned with a futuristic dystopic vision where machines take over the world and more worried about the real challenges our planet is facing and the role AI will take in improving or further deteriorating them.
We should push forward on the potential uses of AI to improve life and well-being towards the Sustainable Development Goals (SDGs), especially in healthcare and education, without creating blockades that inhibit their use for good while nevertheless trying to tackle potential risks in the near and more distant future. When we discuss risks, I am more worried about, for instance, the impact of AI on misinformation, fake news, and erosion of trust in democratic societies than of future AGI (artificial general intelligence) taking over. We should have a balance, which the advisory board has portrayed in its interim report.
Yes, there are risks, and we must consider guardrails and governance. But we must also look at the potential proper uses and their benefits. We should push forward on broader, more equitable availability of talent, computation, and data without crippling scientific and practical advancement.
As we put forward new AI-driven models, we should always pay attention to fairness and equity to ensure that they can be used by various subgroups of the local and global population without introducing unfair algorithmic biases.
As the Deputy Director General and Head of the Innovation Division of Clalit Health Services – Israel's largest healthcare organization – tell us what innovation Clalit recently introduced that impressed you most.
One significant innovation is our AI platform for general practice in primary care. Primary care is the most crucial healthcare subsector, yet it is often underutilized, underfinanced, and lacks proper attention and infrastructure in many healthcare systems. To support the work of primary care physicians, we have developed an AI-driven platform called C-Pi (Clalit Proactive-Preventive Interventions).
This system provides every physician with an easily accessible, one-screen interface that offers fully explainable AI insights. Nothing is a black box. Every recommendation is broken down to the medical logic behind it, allowing physicians to trust it gradually. The platform provides two types of insights. It identifies which patients should be brought in for proactive appointments to prevent potential future issues. This prioritization is based on many data points, risk factors, and potential negative outcomes.
Then, the platform evaluates the care currently provided to each patient against personalized ideal care defined within the system utilizing carefully crafted clinical algorithms. It highlights treatment gaps, missed diagnostic procedures, and necessary treatment adjustments. For example, for diabetic patients, the platform generates around 9 million recommendations nightly for half a million diabetic patients, providing primary care with specialist-level recommendations to improve care quality and prevent errors.
Another notable initiative is the introduction of AI-driven decision support systems in radiology and pathology. These systems help identify imaging studies that require early attention and provide additional insights to clinicians, ensuring thorough examination and reducing the risk of oversight.
These examples show how AI can significantly enhance healthcare delivery, particularly in primary care and diagnostic fields, improving patient outcomes and reducing errors.
That's impressive. Could you give one piece of advice for healthcare organizations willing to be as innovative as Clalit is?
Start by identifying the most significant pain points within your organization that everyone acknowledges, is aligned with, and is amenable to change through advanced technology and data-driven tools.
Then, assess the existing solutions to determine whether and which technology will likely address these issues effectively.
Additionally, make sure you invest in the data infrastructure that would enable a myriad of tools to be integrated into your system, as many solutions require a foundational level of data and technology infrastructure. Without it, progress will be limited.
Once you've established the necessary infrastructure, prioritize change management and technology introduction based on what is truly important to your organization, administrators, and providers. Focus on areas where there is consensus that change is urgently needed to ensure successful implementation. Finally, invest in making the solution user-friendly to its intended users and plan for continuous cycles of tweaking to achieve user satisfaction. Otherwise, you will have yet another pilot in your organization's "pilot graveyard."
