Vaccine-preventable diseases among high-risk patients are a public health priority in high-income countries. Most national immunization programs have included vaccination recommendations for these population groups but they remain hard-to-reach and coverage data are poorly available. In a pilot study, we developed and tested an automated approach for identifying individuals with underlying medical conditions to feed an immunization information system (IIS).
We reviewed published recommendations on medical conditions that indicate vaccination against influenza, pneumococcal disease, meningococcal disease, hepatitis A, and hepatitis B. For each medical condition, we identified the International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis and procedure codes, the user fee exempt codes and the Anatomical Therapeutic Chemical Classification System codes and we reported these data in correspondence tables. Using these tables, we extracted three lists of patients recorded in three current data sources between 2001 and 2010 in the Apulia region of Italy: the hospital discharge registry, the user fee exempt registry, and the drug prescription registry. Using a unique personal identification number, we linked these three lists of patients with the regional IIS (2012 database), obtaining a list of patients with chronic diseases eligible for vaccination. We tested completeness, sensitivity, and positive predictive value (PPV) of this approach by asking a sample of 28 general practitioners (GPs) to evaluate the matching between a sublist of patients with clinical recommendations for influenza vaccination and the GPs individual subjects medical records.
We included a total of 1,204,496 subjects with underlying medical conditions eligible to receive any of the aforementioned vaccinations. Of these, 9% were identified in all three data sources, 18% in two sources, and 73% in one source. The completeness of this automated process in identifying GPs high-risk patients eligible for influenza vaccination was 88.9% [95% confidence intervals (95% CI): 88.1–89.8%], with a sensitivity of 69.2% (95% CI: 67.7–70.6%) and a PPV of 85.7% (95% CI: 84.4–86.8%).
The high completeness of the methodology used for identifying high-risk patients in current data sources encouraged us to apply this approach for feeding the regional IIS.