Secular trends in diagnostic code density in electronic healthcare data from health care systems in the Vaccine Safety Datalink project

Large observational vaccine safety studies often use automated diagnoses extracted from medical care databases to identify pre-specified potential adverse events following immunization (AEFI). We assessed the secular trends and variability in the number of diagnoses per encounter regardless of immunization status referred as diagnostic code density, by healthcare setting, age, and pre-specified condition in eight large health care systems of the Vaccine Safety Datalink project during 2001–2009. An increasing trend in diagnostic code density was observed in all healthcare settings and age groups, with variations across the sites. Sudden increases in diagnostic code density were observed at certain sites when changes in coding policies or data inclusion criteria took place. When vaccine safety studies use an historical comparator, the increased diagnostic code density over time may generate low expected rates (based on historical data) and high observed rates (based on current data), suggesting a false positive association between a vaccine and AEFI. The ongoing monitoring of the diagnostic code density can provide guidance on study design and choice of appropriate comparison groups. It can also be used to ensure data quality and allow timely correction of errors in an active safety surveillance system.