The Accuracy of Hospital Discharge Diagnosis Codes for Major Birth Defects: Evaluation of a Statewide Registry With Passive Case Ascertainment.

Context: Birth defects prevention, research, education, and support activities can be improved through surveillance systems that collect high-quality data.

Objective: To estimate the overall and defect-specific accuracy of Florida Birth Defects Registry (FBDR) data, describe reasons for false-positive diagnoses, and evaluate the impact of statewide case confirmation on frequencies and prevalence estimates.

Design: Retrospective cohort evaluation study.

Participants: A total of 8479 infants born to Florida resident mothers between January 1, 2007, and December 31, 2011, and diagnosed with 1 of 13 major birth defects in the first year of life.

Main Outcome Measures: Positive predictive value: calculated overall (proportion of FBDR-identified cases confirmed by medical record review, regardless of which of the 13 defects were confirmed) and defect-specific (proportion of FBDR-identified cases confirmed by medical record review with the same defect) indices.

Results: The FBDR's overall positive predictive value was 93.3% (95% confidence interval, 92.7-93.8); however, there was variation in accuracy across defects, with positive predictive values ranging from 96.0% for gastroschisis to 54.4% for reduction deformities of the lower limb. Analyses suggested that International Classification of Diseases, Ninth Edition, Clinical Modification, codes, upon which FBDR diagnoses are based, capture the general occurrence of a defect well but often fail to identify the specific defect with high accuracy. Most infants with false-positive diagnoses had some type of birth defect that was incorrectly documented or coded. If prevalence rates reported by the FBDR for these 13 defects were adjusted to incorporate statewide case confirmation, there would be an overall 6.2% rate reduction from 82.6 to 77.5 per 10 000 live births.

Conclusions: A statewide birth defects surveillance system, relying on linkage of administrative databases, is capable of achieving high accuracy (>93%) for identifying infants with any one of the 13 major defects included in this study. However, the level of accuracy and the ability to minimize false-positive diagnoses vary depending on the defect.