Quantifying clinical data quality using relative gold standards

As the use of detailed clinical data expands for strategic planning, clinical quality measures, and research, the quality of the data contained in source systems, such as electronic medical records, becomes more critical. Methods to quantify and monitor clinical data quality in large operational databases involve a set of predefined data quality queries that attempt to detect data anomalies such as missing or unrealistic values based on meta-knowledge about a data domain. However, descriptive data elements, such as patient race, cannot be assessed using these methods. We present a novel approach leveraging existing intra-institutional databases with differing data quality for the same data element to quantify data quality for descriptive data. Using the concept of a relative gold standard, we show how this method can be used to assess data quality in enterprise clinical databases.