Solving Healthcare’s Tech Agility Problem without a Risky Overhaul  

As long as humans have been practicing medicine, the healthcare industry has been cautious in adopting new practices or tools. And rightfully so — lives are on the line.  

While AI is commonplace in many industries, healthcare faces many hurdles to implementing this technology. First and foremost: the potential risks to human health. An AI mistake could be deadly, so any uses must be vetted. Secondly, privacy and security practices limit how data can be collected and used. Organizations must establish secure digital architecture to avoid exposing personal information.  

On top of data considerations, the existing healthcare technology infrastructure is massive. The sheer scale and complexity of entrenched legacy systems filled with patient records make it extremely challenging and risky to completely replace them with new AI-powered systems. 

Healthcare experienced the effects of a total overhaul with the introduction of electronic health records (EHRs). Implementing new EHR platforms touched all systems, creating interoperability issues, disrupting workflows and introducing new errors and operational friction.  

This time around, the industry must be more targeted in its adoption. Moving forward will require modular updates to avoid total disruption.   

Healthcare’s technology agility problem

Healthcare organizations are not averse to technology; they are averse to risk. These two items may be in conflict in an environment that is experiencing rapid technology change. The healthcare’s technical infrastructure is a delicate configuration of many systems, including EHRs, lab information systems, pharmacy systems, billing platforms, etc.  Notwithstanding the fact that many organizations also have some legacy/homegrown systems that also interface with a myriad of third parties. These functions are deeply embedded in well-established processes that can’t just pivot on a dime. 

A major AI system replacement has a large blast radius, and the number of potential failure points is astronomical. A failed integration project could go so far as to bankrupt some healthcare organizations. 

The risk is the potential to “crash” the entire system, adversely impacting the organization’s ability to operate for an extended period of time. What if the system doesn’t restart as expected? What if we lose millions of patient records? What if our scheduling system is off for hours or days?  The concerns are countless, and even if everything goes smoothly, there may be significant downtime during the installation/transition process. 

All of these scenarios have the potential to hamper or even harm patient care.  

Additionally, large-scale AI implementation leaves room for oversights that lead to HIPAA violations or security breaches. Organizations must be cognizant of the regulatory and reputational repercussions related to data management, which is a significant deterrent to moving forward with AI. 

As with most other business undertakings, money is also a factor. It’s expensive to execute a full enterprise replacement; many organizations lack the capital to carry out such a major project, let alone manage any damage caused by system failures.  

While there are plenty of legitimate reasons for healthcare to be cautious about AI, the industry can’t sit on the sidelines forever. However, these organizations can crawl before they walk and run. 

The safest transformation strategy is one component at a time

If running is overhauling an entire system, crawling is farming out specific components to third-party AI providers. The legacy platforms stay in place, but information moves between them and the external tool. This approach allows organizations to add new capabilities to existing workflows without downtime and data migration.  

The first targets for AI implementation should be the most painful workflows.  

Consider the use case of prior authorizations for prescription medications, a major point of friction that can harm patient care. Providers must fill out and submit paperwork, and then a clinical team at the insurance provider reviews the request. There are plenty of potential breakdown points, such as documentation mistakes, missing information, the slow capture of clinical information. etc. Prior authorization approvals can take days to weeks, depending on the denial and appeals process, during which time patients can’t get the medicine they need. In some cases, patients never get approval. This broken process hurts long-term health outcomes.  

Technology is a natural solution to this problem, but replacing the entire system creates new risks. How can a modular approach help? Organizations can layer an AI tool on top of the existing information hub. The algorithm works with the system data without changing any infrastructure. In effect, the healthcare system has all the necessary information (patient eligibility, patient benefits, prior authorization, clinical requirements, etc.). What they typically don’t have is a system that can collect, interpret and process this information, that leverages technology to conform to their requirements.    

In the case of prior authorization management, these solutions can flag documentation problems or surface insurance-specific requirements using established data pathways, reducing paperwork and lowering the odds of denial due to administrative mistakes. This process can also guarantee 100% compliance with the prior authorization criteria using a rules-based approach to facilitate a manual clinical review.   

This targeted AI deployment does not require a wholesale system replacement and maintains existing privacy and security practices. In this model, the legacy system can remain as the system of record, while the AI module exchanges only the necessary data required to perform its task.  

These bounded use cases prove safety and value and establish a repeatable integration strategy. The next step is building a connected infrastructure that standardizes how modules connect. Eventually, these individual tools form an AI layer that works across the legacy system.  

Over time, old components get retired and are replaced by new tools. Healthcare can slowly and deliberately modernize systems without the pain of a one-time system overhaul.  

In the end, the nature of healthcare makes modular changes the most realistic strategy to minimize unintended consequences while still gaining the benefits of AI tools.