Practices and patients selection
Data for this study were derived from the routine electronic medical records of GP practices that participate in the Netherlands Information Network of General Practice (LINH), a national network of around 90 general practices, who are representative of all Dutch general practices with respect to geographical distribution and degree of urbanization [7]. All Dutch citizens are enlisted as patients in a family practice, so the population listed in a general practice can be used as the denominator in epidemiological studies. The GP is the first professional to contact for health problems and a referral to the secondary health care system (hospitals and medical specialists) can only be made by the GP. The LINH practice population consists of more than 350,000 registered patients, representative of the Dutch population with respect to age and gender. The database holds longitudinal data on morbidity, prescriptions and referrals. Clinical diagnoses are coded using the ICPC (International Classification of Primary Care) coding system [8]. Drugs are coded according to the Anatomical Therapeutic Chemical (ATC) classification [9]. GP practices were included when they provided data on the registration of claimed services, ICPC-coded diagnoses, and ATC-coded prescriptions for the full calendar year of 2009. Only patients enlisted with the general practice during the whole year 2009 were included. Data collection within the LINH network is carried our according to Dutch legislation on privacy. Each patient is coded with an anonymous administrative number. The key to this coding number is only with the general practitioner. The privacy regulation of the LINH network was approved by the Dutch Data Protection Authority. According to the Dutch Central Committee on Research Involving Human Subjects, obtaining informed consent is not obligatory for observational studies and approval by the Medical Ethical Committee was not necessary.
Data collection
All contacts (claims) of patients with the GPs were considered. First, patients having a prescription of oseltamivir (ATC-code J05AH02) between August 8 and December 31, 2009 were considered. No prescriptions of zanamivir were registered. When there was more than one oseltamivir prescription, only the first one was taken into account. We looked for all diagnoses within 7 days before or after the date of prescription, in order to take into account registration delays. However, in 88% of oseltamivir prescriptions, the diagnosis was registered at the same day. We specifically considered diagnoses of influenza (ICPC-code R80, which will be used by GPs for patients with influenza-like illness in the context of an influenza epidemic, or upon virological confirmation of an influenza virus), another acute respiratory infection (ARI, defined by ICPC-codes R74 - acute upper respiratory infection, R78 - acute bronchitis/bronchiolitis or R81 - pneumonia), and non-specified virus infection (A77, which can be used by GPs for patients with influenza-like illness outside the influenza season).
Patients who did not receive oseltamivir were evaluated for a diagnosis of influenza. When more than one episode was present, the date of the first consultation was taken. Recommended treatment of high risk patients with oseltamivir was based on age (<2 years or ≥60 years), co-morbidity (cardiac disease, respiratory disease, diabetes mellitus, chronic renal insufficiency, reduced resistance against infections and children up to 18 years using salicylates), a complicated course of illness (the occurrence of pneumonia within 7 days before or after the diagnosis of influenza or ARI, or the need to prescribe antibiotics – not for urinary infections – within these days), or a third trimester pregnancy (an oseltamivir prescription or an influenza diagnosis between 20 to 34 weeks after a first contact related to pregnancy, based on the assumption that pregnant women usually have their first pregnancy-related contact with their GP between weeks 6 and 20; a contact indicating the birth of a child had to be no more than 14 weeks after oseltamivir prescription or influenza diagnosis). Details on ICPC- and ATC-codes used to classify high-risk patients have been published elsewhere [10].
Additional patient characteristics comprised gender, month of prescribing, total number of contacts with the practice in 2009, and the total number of prescriptions in 2009 besides oseltamivir. Other underlying chronic health conditions were based on a selection of diseases with a high prevalence, a long-term course and a serious illness, as used in the National Public Health Compass (http://www.nationaalkompas.nl). This selection is based on the list of chronic conditions from the Australian Family Medicine Research Centre (http://www.fmrc.org.au) and adapted from ICPC-2 to ICPC-1. Practice characteristics included practice type, dispensing practice, and level of urbanisation and geographic region of the practice location.
Statistical analysis
In the group of patients with a diagnosis of influenza, we evaluated the proportion of patients whom had been prescribed oseltamivir and whom should have been prescribed it. Univariate logistic regression analyses were performed on the association between the prescription of oseltamivir and a recommendation for a prescription, as well as other patients’ and practices’ characteristics. Multivariable regression analyses were performed to assess potential determinants independently associated with the prescription of oseltamivir. Likewise, we evaluated whether oseltamivir was prescribed according to the guidelines among all patients receiving oseltamivir. Because of the recommendation to prescribe oseltamivir to both very young (under two years of age) and old (60 years or above) patients, we added age-squared in the multivariable models to take into account the u-shaped association when using the continuous age variable. Only variables that significantly improved the model fit, based on Likelihood-ratio tests (p-value < 0.05), were included. Since patients were clustered within practices, we used multilevel logistic regression. All analyses were performed in Stata version 11.2 (StataCorp LP, College Station, TX, USA).