Registry Data Quality in 2025: Still More Art Than Science, but AI Is Knocking

Seyyedeh Fatemeh Mousavi Baigi, Seyyed Mohammad Tabatabaei, Abbas Sheikhtaheri, et al.

Summary: A registered systematic review — a study that methodically searches and synthesizes existing research — analyzed 119 studies examining how data quality in health registries is defined, measured, and improved, with particular attention to artificial intelligence and machine learning approaches. The review found that most quality assessments focus on whether records are complete and accurate, while largely ignoring whether the people in a registry are representative of the broader patient population. Traditional rule-based checks and audits remain the most common quality tools. AI and machine learning applications are growing — particularly for spotting unusual entries and linking records across datasets — but they are still selective and face real barriers related to how systems connect with each other and how data are structured.

Why it matters: Registries are increasingly used to support regulatory decisions and to generate the kind of real-world evidence that influences treatment guidelines and market approvals. If the data in those registries are incomplete or unrepresentative, the conclusions drawn from them can be quietly distorted. The finding that representativeness — whether a registry genuinely reflects the full range of patients with a condition — is rarely assessed is important: a registry that captures mainly patients treated at specialist centers, for example, may paint a misleading picture of how a treatment works in ordinary clinical practice.

What to keep in mind: Systematic reviews inherit the limitations of the studies they include. The 119 studies analyzed here use varying definitions of data quality and inconsistent reporting standards, making it difficult to draw firm general conclusions. The review's finding that AI tools are underused is credible, but it also reflects a practical reality: automating fuzzy quality judgments — such as whether a clinical note makes sense in context — is much harder than automating clearly defined tasks. The authors' call for quality processes that are built into registry operations from the start, rather than applied after the fact, is directionally sound but not yet backed by evidence showing that this approach delivers better outcomes in practice.

Relevance: Patient Registries | Real-World Data | Data Quality | Clinical Informatics

Source: https://pubmed.ncbi.nlm.nih.gov/42813616/

---