BACKGROUND: The Sentinel Initiative was established in 2008 to monitor the safety of FDA-regulated medical products. We evaluated the positive predictive value (PPV) of ICD-9 codes for post-vaccination febrile seizures to identify optimal algorithms for use in post-market safety surveillance.
METHODS: We identified ICD-9 diagnosis codes for fever and seizures in the emergency department or inpatient setting after vaccinations of interest from July 1, 2010 to June 30, 2011. Medical record review was conducted to verify febrile seizure events.
RESULTS: Of 216 potential febrile seizures identified with one or more seizure codes (the broadest algorithm), 152 were chart-confirmed (i.e., documentation of fever within 24 h of seizure or clinician diagnosis of febrile seizure; PPV 70%, 95% CI 64, 76%). Two codes specific for febrile seizures produced the highest PPV (PPV 91%, 95% CI 85, 95%) and accounted for 140 confirmed febrile seizures. In the absence of febrile seizure codes, other seizure codes yielded much lower PPVs, regardless of the presence of fever codes.
CONCLUSIONS: Our results indicate that ICD-9 diagnosis codes in the inpatient and emergency department settings have high predictive value for identifying febrile seizures within the Sentinel Distributed Database. While the PPV of the algorithm based on any diagnosis code for seizure is moderate, the algorithm limited to febrile seizure codes has a high PPV (>90%) and captures the vast majority of confirmed cases identified by the broadest algorithm, suggesting that the narrower algorithm limited to febrile seizure codes may be preferred.