AI in Healthcare: Data Governance Becomes Critical
Generative artificial intelligence (AI) is rapidly transforming healthcare, offering the potential to boost productivity and improve patient care. However, AI’s reliability is directly tied to the data it’s trained on, making healthcare data governance more vital than ever.
A recent survey from Amazon Web Services (AWS) and Harvard Business Review highlighted concerns among chief data officers across various industries regarding their data assets. The survey found that 52% of respondents considered their organization’s preparation for generative AI as “inadequate,” and 39% identified data-related issues as the primary obstacle to effectively scaling AI.
According to Thomas Godden, enterprise strategist with AWS, the healthcare industry’s regulatory framework uniquely positions it to leverage AI effectively. “Data governance is fundamentally the bedrock for ensuring patient safety,” notes Godden, who previously served as CIO for Foundation Medicine. “Healthcare organizations have already needed to clean and control their data. So, in a lot of ways, they’re better positioned for AI than other industries.”
Why AI Makes Healthcare Data Governance More Complex
Data governance encompasses the policies and standards that guarantee data is high-quality, accessible, secure, and trustworthy. The substantial data volumes needed by AI-supported technologies have increased the complexity of data governance in several key respects:
Keeping Data Sets Updated
Healthcare data is constantly changing. AI training models must reflect these changes to maintain accuracy. As Godden points out, “If you’re not updating the models daily or weekly, you’re going to miss things that are happening in the world and with your patients.”
Removing Biases
Data can contain biases tied to factors such as gender, race, and socioeconomic status. Susan Laine, chief field technologist at Quest Software, emphasizes the necessity for data teams to have systems in place to identify and remove these biases from the training data. “Data problems will only be amplified when fed into AI for things like diagnoses and treatment recommendations,” she warns.
Identifying Responsibility and Accountability
When an AI-driven decision leads to an adverse outcome, identifying who is responsible—the developer, the user, or the system itself—becomes a crucial challenge. Laine states, “If you don’t have transparency around what’s happening with your data, then you won’t know the true source of the problem or where a fix is needed.”
The Benefits of AI and Data Governance
A strong data governance framework guarantees that AI models receive high-quality information, mitigating risks. “Data governance is like having a glass box around the AI,” says Laine. “It provides transparency into what’s feeding the AI model and who has touched that data.”
At the same time, AI can aid data management by enforcing policies and analyzing security patterns. For instance, AI can monitor and check that sensitive patient data is accessed and handled correctly. Chatbots can streamline the end-user experience by helping analysts sort and interpret data from large datasets more efficiently.
In addition, machine learning tools can assist healthcare organizations in leveraging larger data influxes. AI processes and learns from collected data automatically, enabling continuous system improvement.
Setting Realistic Expectations for AI Data Governance
A common pitfall, according to Godden, is that leaders believe they must overhaul all of an organization’s datasets before realizing value from an AI tool. Instead, he advises adjusting expectations and starting with smaller objectives: “Identify a business opportunity and focus on governing and cleaning only the data you need to solve that specific problem.”
It is essential to clearly communicate the organization’s values, ensuring that employees understand them. This offers vital guidance, so that when a data anomaly occurs, employees can accurately identify and fix it, aligned with the company’s expectations. “AI models are going to have biases, and corrections will come down to individuals making value calls,” says Laine. She cautions that healthcare systems need to recognize AI’s imperfections. Human intervention is crucial, especially when determining the root cause of data anomalies. “If I were a doctor, I would feel more reassured knowing a data governance team is behind the scenes verifying that the data makes sense,” Laine says.
Who Should Lead AI Data Governance?
Data governance efforts are generally headed by the chief data officer, supported by data quality analysts and architects. Furthermore, prompt specialists are increasingly being used to improve the training of AI training models. Laine stresses that data management experts should lead the way when starting with AI and data governance. “These are the people who understand how the data moves and changes. I think relying on their expertise is key to an organization getting it right.”
Godden adds that when establishing a healthcare AI program, the policies and procedures that will govern the technology should be crafted by a diverse team. This team should encompass IT and data teams; medical professionals; and personnel from the legal, marketing, and HR departments. “You need everyone involved in building and using the AI to understand it and have their antennas up,” Godden says, emphasizing that all team members have a role in monitoring AI for inconsistencies. “This is not an IT problem. This is an everyone problem.”
