Is healthcare AI’s future held back by bad data?
After a peak of investment in 2021, many healthcare AI investors are slowing down and becoming more selective, prioritizing solutions that show real-world lasting impact. This shift reflects the high stakes of healthcare, in which the cost of error is steep and proving effectiveness is critical. Often, the challenge isn't the AI tool itself but instead the fragmented, inconsistent, and siloed data that power it. Unless we solve this data dilemma, AI’s full potential will remain out of reach.
Inconsistent data quality, privacy concerns, and limited interoperability all hinder AI’s potential. That’s why strong data governance is critical to getting AI right. By aligning data practices with organizational priorities and regulatory standards, governance helps maintain the integrity and responsible use of healthcare data.
AI itself can also be part of the solution, with emerging technologies helping us improve how we capture, structure, connect, and secure data. We can begin to unlock AI’s full potential across the healthcare ecosystem by using AI to strengthen the foundation it relies on through several efforts:
Starting with accurate, usable data at the point of care. Electronic health records are a valuable tool, but usability challenges can hinder their efficiency. Emerging AI tools such as ambient listening can help us move away from click lists and manual entry to natural, real-time documentation.
Enhancing how we structure and standardize data. Existing code sets and standards have helped, but there’s still work to do. Clinical coding, for example, can be overly complex and inconsistent, affecting data quality and slowing workflows. AI, especially large language models, can streamline coding by suggesting context-relevant options and help improve accuracy.
Generating synthetic data offers a powerful way to fill data gaps. Although the Health Insurance Portability and Accountability Act (known as HIPAA) and related regulations protect patients’ privacy, they limit access to the real-world data that most effectively train AI models. Patients’ limited internet access creates additional gaps, leaving digitally disconnected communities potentially misrepresented or overlooked in AI models. Synthetic data generation can help bridge these gaps by creating artificial data that mimic real patient information without compromising privacy.
Employing techniques like privacy preserving record linkage helps us create a fuller picture of health in safe ways. Privacy preserving record linkage brings together information from across different systems, such as those used in housing, education, and free-standing clinics, without sharing patients’ identifiers, and it provides a more complete view of a patient without compromising their information security.
"With strong governance and high-quality data, we can unlock broader adoption of AI at scale—from routine administrative tasks to more advanced, high-impact applications," writes Ngan MacDonald.