First Principles Thinking for Healthcare IT: AI Integration and Life Sciences

Organizations across the entire healthcare ecosystem have been betting big on AI. The excitement is justified. Implementing these technologies can save a lot of time and money to do a lot of wonderful things. 

Sadly, however, implementing AI can just as easily waste a lot of time and money to do a lot of stupid things.

One of the worst things any organization can do to its data architecture is automate processes to improve the wrong problem. Not only does it waste time and resources, it increases bloat and entrenches unnecessary distractions and hurdles to function and progress. It is safe to say these compounding ill-effects are already pretty familiar to anyone who has ever worked with healthcare IT, — payers, providers, pharma, biotech . . . No one is immune. 

Foolish AI use could easily make this state of affairs even worse, misdirecting effort toward glitzy functions no one needs and costly features no one uses. So while it may seem counterintuitive, when it comes to effective use of AI, you actually shouldn’t start with AI. 

You have to start with identifying the problem you are trying to solve.

Shift in perspective

Back in college, I studied civil engineering, where Aristotle’s “first principles thinking” is canon for generating efficient processes and optimal outcomes. The approach involves breaking down complicated problems into basic foundational elements and only then reassembling them to achieve your objective. And having an objective is key.

In civil engineering terms, why would you erect an expensive steel suspension bridge atop a perfectly functional flat stretch of highway? Even if it’s the strongest and most impressive bridge ever constructed, no one benefits from using it, so it serves no purpose. 

In the real world, every organization has computing and data management systems. AI is a powerful and impressive new capability organizations naturally want to incorporate into those systems. But regardless of capability, it has to deliver real-world benefits to be of any value whatsoever. 

So you must start with an appropriate definition of the problem aligned to the desired outcome. Then you can systematically address the relevant components and actual process involved. And you cannot weigh it down with all the old processes that you put in place because of past technology requirements and limits. Question everything. Computer science legend Grace Hopper once said the most dangerous phrase is “We’ve always done it that way” — and it’s worth noting she was talking about data processing when she said it. 

Challenge every assumption and preconception, eliminate anything unnecessary, strip everything down to its barest form and function for your purpose. This ensures you understand actual necessities for addressing a real problem. That should dictate data strategy going forward, and that focuses AI integration on delivering value.

First principles in life sciences AI use

Language and text-related generative-AI is currently one of the more mature forms of the technology (and I am not talking about chatbots). To illustrate smart use, let’s zero in on healthcare’s life sciences sector for first-principles thinking examples in problem-solution focused AI integration. 

Consider a pharmaceutical or med-tech device company, and how they build out a manufacturing process for a new drug or medical device. That process needs a design for physical manufacturing and materials management, as well as for meeting regulatory requirements for every aspect of production. This guides establishing the actual manufacturing site all the way from testing individual pieces of equipment, to testing sections of equipment, to facilitating ongoing testing of the entire facility. That process is called “commissioning, qualification, and verification,” and it can involve hundreds of thousands of pages of documentation. In layman’s terms, the level of documentation represents tons of work.

The role of documentation is incredibly important because it validates all that testing and supplies a science-based understanding that the process is working correctly, materials are being produced appropriately, and that it will all pass inspection and secure FDA approval for market distribution.

FDA approval is the prize, the laborious process of proper documentation is required to achieve it.

So a clear engineering goal for deriving value from AI integration in this context would be automating the production of vetted and correctly formatted commissioning, qualification, and verification documentation that meets FDA standards. Data translating every aspect of build-out and testing procedures, along with data detailing the minutia of all the various FDA requirements for each aspect of that process, can be leveraged to feed a large-language model (LLM) and generative AI-engine that ensures appropriate documentation is automatically and continuously correctly collected and produced. That would save countless human work hours! 

On top of that, an organization’s depth of expertise and institutional knowledge on the business processes involved in pharmaceutical or medical device manufacture can also be ingested into this model to further refine the sophistication of documentation management and development, thus increasing competitive advantage from a financial perspective. Obviously, humans will still have to review the documentation, but the difference is in who (or rather, what) is preparing the documentation consistently and accurately and how much time and effort is saved. The point is that the AI integration is focused on addressing the “right” problem — documentation burden — where it delivers practical and significantly valuable improvement. 

If that sounds a bit esoteric, how about AI tools for supplying pre-screening processes for clinical trials in a way that fits right into existing patient case-review operations at physician practices. This type of capability is incredibly helpful for, say, rural clinicians who may support several 1000 patients over several 1000 miles and simply do not have the human resources available to provide informed and time-sensitive research. 

The right AI model applied to that problem exponentially enhances their ability to match patients to potentially life saving treatments faster and more effectively. That can actually save lives, and it represents exactly how we want these new technologies to save time and money and do wonderful things.

The only “trick” required for truly successful AI integration — in life sciences or any other facet of the healthcare industry —  is clarity of purpose. First-principles thinking is an excellent way to ensure your effort and investment actually align with and produce desired outcomes — and real value.

Photo: Overearth, Getty Images

Chris Puuri, VP, Global Head of Healthcare and Life Sciences at Hakkōda, uses his intimate understanding of healthcare IT and regulatory challenges to solve problems in data and analytics unique to healthcare. With over 18 years of experience as a data architect for organizations spanning medical systems, pharma, payers, and biotech companies, Chris has built, integrated, and launched data solutions for some of the country’s largest healthcare organizations.

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.