Clean Data Combined with Federated Learning Keys to Maximizing AI Efforts

The promise of AI in healthcare hinges on a fundamental requirement: high-quality data. Without it, even the most advanced algorithms will fail to deliver meaningful results, according to Sonya Makhni, MD, Medical Director of Applied Informatics at Mayo Clinic Platform.

“Data quality isn’t always the most exciting topic at conferences, but it’s the foundation of everything we do in AI,” Makhni said, who also serves as a hospitalist at Mayo Clinic Rochester. “Accuracy, integrity, and validity are non-negotiable if we want to build reliable models.”

Scroll down to watch or listen to the full interview; or subscribe to healthsystemCIO on your favorite podcasting channel

Makhni noted that many health systems struggle with inconsistent data due to M&A, disparate EHRs, and varying documentation practices. “You can’t apply a great algorithm to bad data and expect good results,” she said. “Standardization and governance are essential.”

Health systems generate enormous amounts of data daily, from patient encounters to imaging and lab results. However, much of this data exists in different formats and structures, making it difficult to use effectively in AI-driven applications.

One of the biggest barriers to AI adoption is that many health systems have incomplete, inconsistent, or siloed data. “Even within a single health system, data from different hospitals might not be standardized,” Makhni noted. “This inconsistency creates challenges in developing AI models that can provide accurate, reliable insights.”

Beyond interoperability, there is also the issue of data governance. Health systems must ensure that data is not only high-quality but also secure and used in compliance with regulatory standards such as HIPAA. “Privacy and security must always be top of mind,” Makhni said. “We need to enable innovation while maintaining the trust of our patients.”

Data Needs Direction

Given the complexity of managing health data, Makhni noted the importance of strong data governance and suggested that a Chief Data Officer (CDO) could be a beneficial role.

“Without high-quality data, AI efforts will struggle. Leadership must ensure that data is well-structured, secure, and usable,” she said.

The role of a CDO is to oversee data strategy, ensure interoperability, and establish governance frameworks that facilitate AI adoption. “Health systems need someone who is responsible for the integrity of their data. Without that leadership, it’s difficult to build sustainable AI initiatives,” Makhni said.

Beyond leadership roles, she emphasized that collaboration is key. “Different stakeholders bring different perspectives, and that diversity of thought leads to better decision-making,” she said. “It’s about striking the right balance—having the right experts at the table without creating unnecessary bureaucracy.”

The Benefits of Federation

Makhni also discussed federated learning, a decentralized approach to AI training that allows health systems to collaborate without transferring sensitive patient data.

“The process of data centralization could lead to risks,” she explained. “Federated learning enables us to train models across multiple institutions while maintaining strict security controls.”

This approach is particularly important for ensuring AI models are generalizable across diverse patient populations. “If you’re only training an AI model on data from one hospital or one region, it might not work well elsewhere,” she said. “Federated learning helps us build more robust, unbiased models.”

However, implementing federated learning requires investment in infrastructure, including standardized data models and robust security protocols.