Electronic Health Reporter

By Joachim Roski, Principal, Booz Allen Hamilton and Kevin Vigilante, Executive Vice President and Chief Medical Officer, Booz Allen Hamilton.

As artificial intelligence (AI) continues to transform our world, the healthcare sector stands to substantially gain from AI. The ability to compute a massive amount of information quickly has promising implications for delivering better health outcomes, improved healthcare operations, or expanded research capabilities. However, AI is not immune to the risks that accompany the adoption of any new technological advancement – risks that may cause healthcare professionals to doubt its reliability or trustworthiness.

Fortunately, there are steps available that can help build a strong AI foundation, improving the quality of life for patients and healthcare professionals alike. If you’re looking to implement AI in your healthcare organization, here are a few things to keep in mind:

Understand your Data

If a data is not of high quality or is not representative of the intended group of study, the conclusions based on that data will likely be flawed. To accurately anticipate how a flu outbreak will impact hospital visits among a community population, it’s essential to account for variables that can affect all or part of the relevant population. For example, if a significant subset of the population is more susceptible to serious flu symptoms (e.g., based on pre-existing conditions), this needs to be taken into account by the hospital(s) in question. Given that algorithmic models are often designed to grow and expand with compounding data sets, any previous mistakes can snowball and lead to significant long-term bias or inaccuracy in the results.

As such, it’s critically important to build an AI foundation on a robust and comprehensive data acquisition and management operating system, complete with careful and consistent oversight of AI algorithms. Such a system should be supported by compliance and monitoring protocols to ensure that data is securely flowing. Having a clear understanding of where relevant data comes from and how it was collected is critical to ensure AI algorithms create meaningful output.

Accounting for the Unaccountable

No matter how strong your model may be, it is virtually impossible to consider every eventual variable within an AI system. An obvious example is the COVID-19 pandemic, which introduced such massive levels of institutional disruption that no AI algorithm could have accurately predicted the true extent of the resources needed or the potential human cost. COVID massively disrupted usual care patterns – such as elective surgeries and regularly scheduled check-ups – to a degree that would not have been predicted by an AI model trained on pre-COVID data.

Such limitation can at times be overcome with flexible modelling parameters. This elastic approach allows systems managers to quickly integrate corrections and additional factors based on unanticipated changes (e.g., very different demand for different healthcare services) as they develop in real-time. Ideally, these modelling parameters are both intuitive and transparent, so stakeholders and professionals can reduce any possible lead time in updating the processes.

Securing Your Systems

There is perhaps no faster way to lose the trust of patients and employees than by experiencing a widespread data breach. If not safeguarded properly, AI can be as vulnerable to malicious actors as human operators.

A crucial component of launching or improving an AI strategy is a comprehensive, foundational control system that accounts for the most common areas of IT risk, such as human error. It’s critical to partner across an organization to implement governance and data management processes that allow for multi-tiered permissions that ensure proper safeguarding of sensitive data, including personally identifiable information (PII) or protected health information (PHI).

Growing Alongside the Solution

AI’s contribution to different use cases continues to evolve. With the right foundations in place, AI developers and users can take advantage of AI’s innovation potential, maintain user trust in AI, and drive improved outcomes through better solutions.