Embracing AI to Enhance Healthcare

In simple terms, AI refers to studying and mimicking the ways and methods to make machines think and act like human beings. A few years ago Elon Musk, who has been at the forefront of innovation with his involvement in companies like Tesla, SpaceX, and Neuralink, dubbed Artificial Intelligence as the “biggest existential threat” and its usage akin to ”summoning the demon”. Bill Gates, not so long ago, termed AI as being “both promising and dangerous”. While the jury is still out on what is the likely impact AI will have on the world in the coming decades and the nature of the impact, there can be little doubting of the promise it holds of having an impact in one one of the most significant aspects of human existence and well-being: healthcare. In this blog, we aim to explore the paths and the possibilities for AI-based solutions for healthcare.

How can healthcare develop better AI-based solutions?

They say the devil is in the details. While AI has been thrown around as a buzzword for a while now, it is imperative to consider the various nuances that can help in building world-class AI-based products. Below are some of the pointers that can help in shaping such products that have a lasting impact:

• Problem versus Product: Unlike a traditional product, a product manager has to be focused on the problem being intended to solve as opposed to the focus on the product itself, since there would not be an identified product at the beginning. Even within healthcare, there are a large number of areas where AI-based solutions can fit in. And one of the greatest challenges would be to find the right area with the best opportunities.

  A lot of research will need to be done by the product managers and data scientists to find the right target area. While it will be key to define a long-term vision at the beginning, the finer details around strategies will have to be fluid. Similarly, while it is important to define how the impact will be measured - these can be the number of lives that may be saved, the number of diseases that can be detected early, user satisfaction, patient satisfaction, decreasing costs, decreasing time, etc. , it will be crucial to be flexible on the specific use case - for example - whether the focus will be on the detection of breast cancer or glaucoma, or AIDS.

  Milestones can be developed as the research moves forward, but there should be readiness to pivot to cause the greatest impact. As it applies to pretty much any product development in the real world, constraints related to resources and skillsets would mean that opportunity cost will be a crucial factor in making such decisions. So, if the center of attention is solving problems related to cancer, it can often inherently mean that problems with lung or eye diseases will get placed on the back burner.

• Laser-sharp focus on users and patients: This holds for AI-based products as much as it does for traditional products, if not more. Understanding the challenges of users and patients is of prime significance to zero in on the right focus areas. It is paramount that not only the problems of users and patients are well understood, but also their preferences in terms of potential solutions are kept under focus. For example, if the focus is on helping glaucoma patients get diagnosed early, it should be the main focus to understand what are the possible results or solutions that can be useful to such patients. As life expectancy continues to increase with the advent of modern medicines, it is key to keep in consideration that old-age-related health issues will become a major healthcare challenge that can be a focus of AI-based healthcare solutions.

• Building the right algorithms: The data set on which an algorithm is tested and trained should be a representative data set with a minimum amount of bias or skew, and cover a large set of cases. The algorithm should not only tell the chances of a disease but also recommend the right action in case the disease is detected. Often a health complication can be a result of multiple factors. For example, the reduction in vision in old age patients can be because of diabetes or glaucoma, or AMD (Age-related Macular Degeneration). The algorithm intended for solving vision problems caused by glaucoma should incorporate ways to discount the effect of other causes, to as much extent as possible.

  The algorithm should be able to prove that it could solve the intended problem and be evaluated in the right way using a retrospective data set. Once that is done, further progress can be made by moving to a prospective data set and analyzing an algorithm for real-time care. For instance, the algorithm should not only be able to predict that a person has glaucoma, but it should also be able to predict the different stages with varying severities. Similarly, a study claimed that 80% of eye diseases related to AMD (Age-Related Macular Degeneration) can be cured or prevented if the diagnosis can be done early. A good AI algorithm for AMD should be able to achieve that level of accuracy and consistency.

• Generalization and Limitations: Once an algorithm has been trained on an existing data set and has been shown to perform reliably in both retrospective and prospective scenarios for data sets related to the same source, it is necessary to understand the challenges around its generalization for all kinds of healthcare scenarios, and whenever found out otherwise, the limitations should also be pointed out. It is possible, for example, that demographic and genetic factors may lead to a discrepancy in applying an algorithm from one geography to another, or from one age group of people to another. The endeavor should be to refine the algorithm to such a state that generalization across a wide set of data sets and scenarios is feasible.

• Ethical and Regulatory aspects: In traditional product management, one gets relatively rapid feedback (either directly through customers or through tracking metrics in real-time or almost real-time) and can make a change (do an AB test or a usability test on a website, for example) and within a few days one can know which of those considered options is the best. But in research for AI-based solutions, one is thinking about where an algorithm would be used in a product and what impact could it have in the long term. Since many of these AI-based changes may need to be validated and certified from a regulatory and ethical standpoint and can have legal repercussions, the need to navigate through the maze of complexities for ratifying such a solution is prominent.

Areas of impact through AI-based solutions

The range of areas in healthcare where AI can make a great impact is immense. Let us look at some of the areas below:

1. Better diagnosis and better utilization of medical information: There is a great potential for not only cutting down on costs but also saving time in concluding a diagnosis. Providers, payers, and patients - all are slated to benefit in this kind of scenario. Detection of life-threatening diseases like cancer in its early stages can greatly enhance the chances of recovery for the patients. Many vicious blood diseases can be countered effectively by predictions and recommendations made by the algorithms trained on large sets of data. Scanning of various images can give early insights into potential areas and can help doctors recommend the right approach: whether a patient needs to be urgently referred or kept under observation or should go for routine tests at regular intervals.

   As per some estimates, the healthcare industry loses 100 billion USD of highly valuable information every year among the maze of trillion data points. AI can power even more effective genetic sequencing and image processing to help predict a potential list of people who may get a disease. Verification and classification of a set of symptoms can also be powered through these AI algorithms. In a world, where there is an ever-increasing openness to discussing and expressing mental health issues, AI may have a pertinent role in the early recognition and treatment of such issues.

2. Simplification of patient journey: A lackluster customer service mechanism and complicated paperwork affected over 35 million patients in 2016. There can be numerous operational challenges that can be overcome through AI applications that can ensure Emergency Room availability and keep a tab on waiting periods at a healthcare organization. The crucial time that often gets wasted in hopping from one hospital to another for availability reasons can often prove to be the difference between life and death. The processing of surveys filled out by patients to comprehend the health challenges or concerns of a patient and direct them to the right healthcare professional can also be a great area of impact.

3. Creation and evolution of effective drugs: AI algorithms designed and enhanced on a huge set of medication-related data sets can be used not only to predict the development of new drugs but also to find the greater utility of existing drugs. The enriched capabilities can help us come up with the formula for a life-saving drug in weeks or months, as opposed to years and decades. This can have an unprecedented impact in areas like cancer treatment or brain-related complications. The ability to discover links between various chemicals with a shorter turnaround would mean that various permutations can be tested on an expeditious basis to arrive at the result. In a world razed by the pandemic caused by a coronavirus, wouldn’t we have loved it if the vaccine formulas had been figured out within weeks? Or wouldn’t we like to feel assured that if/when the next pandemic comes around (one surely hopes it doesn’t), we would have to worry about combating it as much as we worry about completing a straightforward house chore? The cost of developing and taking a drug to the clinical trial stage is 2.6 billion USD currently, one can very well imagine the potential for cost reduction that quality AI-based solutions can provide in the long run.

4. Robotics and Surgery: A dog is often said to be a man’s best friend. But, arguably, robots who have traditionally been used to accomplish mundane and repetitive tasks, can stake a claim to that title. Better still, if those robots are powered by AI that can help doctors conduct super complex surgeries at a faster pace with more precision and control than what is possible for a human being, these robots will be a life-saving friend.

Peeping into the Future

AI-based solutions can be the next big leap that empowers Innovaccer to fulfill its mission of connecting and curating the world’s healthcare information to make it accessible and useful. AI may not be the magic wand that it is often claimed to be, but still has immense potential and promise for the road ahead. We will do well to remember that there are many inventions and creations around us that would have been considered impossible at the time their idea was conceived by someone.

We can strive towards reaching a stage when healthcare organizations confirm to Innovaccer that their data points show that more cancer patients recovered because they were diagnosed early based on AI-based solutions of Innovaccer. And look forward to a surgeon narrating a story of how a robot powered by Innovaccer AI helped in performing a complicated transplant, which was helpful not only in increasing precision but also in freeing up his time that could be used to tend to more patients. And that fewer patients are having to wait to get their appointments or to get admitted because the system could predict much earlier in advance they would be needing a particular kind of treatment shortly. How about a world in which a cure for a deadly virus can be found within weeks, instead of months or years, or, in some cases, decades or never? The possibilities seem endless and exciting. As the cliche goes, the sky's the limit for Innovaccer. This limit too may be breached when we have humans settled on Mars and require healthcare services!

We often talk about Mission Apollo at Innovaccer as a part of the huge effort and focus to make all Innovaccer applications even more scalable and robust. A couple of years(or maybe even months) down the line, we may be talking about a certain Mission Curiosity with the intent to see how much we can push the limits of AI boundaries to solve healthcare problems for mankind. To get there and beyond, we seek to keep the fire burning!