Agreement between patients' self-report and medical records for vaccination: the PGRx database

Patients' self-reported vaccine exposure (PS) may be subject to memory errors and other biases. Physicians' prescription records and other medical records (MR) do not capture noncompliance with vaccination. This study compared PS with MR for influenza, 23-valent pneumococcal, and human papillomavirus (HPV) vaccines.

The Pharmacoepidemiologic General Research Extension (PGRx) database uses a network of over 300 general practitioners across France, who systematically recruit an age- and sex-stratified sample of patients (≥ 14 years old), without reference to their diagnoses or prescriptions. Patients received a structured telephone interview, combined with an interview guide listing vaccines commonly given. Patients' self-reported vaccination in the 3 years before their recruitment was compared with medical records kept by the physician or the patient.

Concordance between PS and MR was assessed for 7613 patients for whom both sources of information were available. Agreement within 3 years before the recruitment date was substantial for influenza vaccines (prevalence and bias-adjusted kappa [PABAK] = 0.74, sensitivity PS relative to MR 81.5%) and high for 23-valent pneumococcal vaccines (PABAK = 0.98, sensitivity PS 49.6) and HPV vaccines (PABAK = 0.92, sensitivity PS 91.6). In adjusted analyses, agreement varied with sociodemographic and health-related factors, particularly for influenza and 23-valent pneumococcal vaccines.

The PGRx method for drug exposure assessment is a new tool in pharmacoepidemiology that shows substantial to high agreement between PS and MR for exposure to various vaccines. Our finding of high agreement between PS and MR for HPV vaccination status in young women is a significant addition to the literature.