Nathalie Gault1,2,3, Johann Castañeda-Sanabria4,3, Sylvie Guillo4,2,3, Stéphanie Foulon1, Florence Tubach1,4,2,3. 1. APHP, Département d'Epidémiologie et Recherche Clinique, Hôpital Bichat, 75018, Paris, France. 2. Univ Paris Diderot, 75018, Paris, France. 3. INSERM UMR 1123 ECEVE, 75010, Paris, France. 4. APHP, Centre de Pharmacoépidémiologie, 75018, Paris, France.
Abstract
PURPOSE: The self-controlled (SC) designs offer the advantage of inherently controlling for time-invariant confounders, which may not be collected in electronic healthcare databases. To be valid, the design requires that some characteristics must be present. Recommendations for their use have been published in 2012, which encourage their implementation when these key characteristics are valid. We aimed at describing the potentially missed opportunities for the use of SC in pharmacoepidemiological studies on electronic healthcare databases. METHODS: We searched Medline for the articles published in two different 6-month periods before and after the 2012 recommendations. The potentially opportunity for the use of SC design is based on major required characteristics (abrupt onset event, rare or recurrent event, and acute or intermittent exposure). RESULTS: In total, 94 papers in 2011 were analysed and 106 in 2014. SC designs were rarely used (2% in 2011 and 1% in 2014). A potentially missed opportunity for SC use was found in 16% of papers in 2011 and 15% in 2014. CONCLUSIONS: We found that SC designs were underused and identified situations in which they could have been applied, but were not. Their use should be encouraged.
PURPOSE: The self-controlled (SC) designs offer the advantage of inherently controlling for time-invariant confounders, which may not be collected in electronic healthcare databases. To be valid, the design requires that some characteristics must be present. Recommendations for their use have been published in 2012, which encourage their implementation when these key characteristics are valid. We aimed at describing the potentially missed opportunities for the use of SC in pharmacoepidemiological studies on electronic healthcare databases. METHODS: We searched Medline for the articles published in two different 6-month periods before and after the 2012 recommendations. The potentially opportunity for the use of SC design is based on major required characteristics (abrupt onset event, rare or recurrent event, and acute or intermittent exposure). RESULTS: In total, 94 papers in 2011 were analysed and 106 in 2014. SC designs were rarely used (2% in 2011 and 1% in 2014). A potentially missed opportunity for SC use was found in 16% of papers in 2011 and 15% in 2014. CONCLUSIONS: We found that SC designs were underused and identified situations in which they could have been applied, but were not. Their use should be encouraged.