Things to Know Before Implementing AI at Scale in Healthcare Organizations

A survey from Optum shows that 83% of healthcare organizations have implemented an AI strategy, and 56% of healthcare executives accelerated AI implementation due to the pandemic.

But in the same survey, many executives stated they were facing several barriers to scaling AI and automation.

Download our free ebook, Maximizing the Opportunities in Digital Health Tech

Discover how you can leverage AI in health tech to build a digital healthcare institution.

Executives consider slow time to implement, lower ROI than expected and staff constraints as the top challenges to deploying AI at scale.

 

The Optum research states that

“Given that automation is increasingly necessary but difficult to implement and scale, not all organizations will succeed. As several executives noted when interviewed, poorly executed automation can even set the organization back and make driving organizational buy-in for future automation projects more difficult. That makes having the right partner all the more important.”

Types of AI Use Cases in Healthcare and Difficulty Level of Implementation

 

AI use cases for healthcare organizations exist in all areas, including:

  • Administrative and business processes
  • Patient engagement and experience processes
  • Clinical support processes
  • Population and public health processes

There are also different types of AI applications, such as:

  • Task automation: automates rules-based high-volume processes in static environments
  • Pattern recognition: uncovering data patterns and helping staff in static environments
  • Contextual reasoning: understanding the context for data and acting in dynamic environments

The difficulty level of implementation, the time to deploy and the ROI depend on both the use case and the AI application.

Deploying Task Automation AI Applications in Healthcare

Task automation applications like composing referral letters or patients self-booking appointments based on preferences or severity triaging are task-oriented and straightforward. Implementing these types of AI applications is the easiest, usually completed in a few weeks and quickly generates savings for the healthcare institute.

Deploying Pattern Recognition and Contextual Reasoning AI Applications

Unlike rules-based solutions, like task automation, pattern recognition AI solutions operate based on probabilities. For AI solutions to “recognize” patterns reliably and validly, they have to be “trained” by programmers and subject matter experts on large data sets of relevant data.

Training an AI application is technically complex and requires appropriate input from staff, who also need to be trained on its use.

Further, probability-based algorithms come with risks and healthcare institutions implementing them need organizational and clinical tolerance for risk. They have to be prepared for the consequences associated with the AI solution getting the answer “wrong.” AI algorithms also need to produce ethically sound and explainable results, which will also affect implementation time and costs.

Hence, implementing pattern recognition or contextual reasoning-based AI applications are much more complex and resource-intensive.

Implementing AI at Scale Requires Leadership and Investment

Healthcare enterprises that have reached maturity in AI deployment have found that scaling AI solution development and deployment across their organization generates even more value than simply using AI in a few instances.

While there is enormous value in deploying AI organization-wide, there are also challenges. Healthcare institutions should take a long-term view of AI deployment to make the project sustainable. For instance, they know that they need:

  • Long-term plan: takes a systematic, enterprise-wide approach for identifying, vetting and resourcing new ideas for AI solutions instead of ad hoc individual projects selected for prioritization.

  • Modernizing systems: scaling enterprise-wide AI solutions will be impossible with legacy systems. Different AI solutions rely on dispersed, widely diverse applications and data sets usually managed in separate data systems. This means healthcare institutions need modern data systems and work with data scientists to efficiently access and analyze this data to build the required algorithms.
  • Integrating expertise: as the requirements change with AI applications. Healthcare organizations have to put together a cross-functional team consisting of staff, clinicians, physicians, data managers, other technical and subject matter experts — instead of AI generalists who can help — to lead their AI project seamlessly.

As organizations move from implementing isolated AI solutions to deployment at scale across their enterprise, organizational culture often gets in the way. Healthcare executives have to carefully assess and help lead their organizations to AI maturity by becoming a data-driven organization and cultivating a culture of learning.

Sign up for the MIT SAP Leading Health Tech Innovation course to learn how to bring your health tech idea to life from the founding faculty chair of MIT's Mind+Hand+Heart Initiative, Prof. Rosalind Picard, who co-founded AI health techs Affectiva and Empatica.

MIT SAP Leading Health Tech Innovation is delivered as part of a collaboration with MIT School of Architecture + Planning and MIT Sloan and Esme Learning. All personal data collected on this page is primarily subject to the Esme Learning Privacy Policy.

 

© 2021 Esme Learning Solutions. All Right Reserved.