Applying Rasch methodology to examine and enhance precision of the Edinburgh Postnatal Depression Scale

Abstract

BackgroundThe 10-item Edinburgh Postnatal Depression Scale (EPDS) is a widely used depression measure with acceptable psychometric properties, but it uses ordinal scaling that has limited precision for assessment of outcomes in clinical and research settings. This study aimed to apply Rasch methodology to examine and enhance psychometric properties of the EPDS by developing ordinal-to-interval conversion algorithm. MethodsThe Partial Credit Rasch model was implemented using a sample of 621 mothers of infants (birth to 2 years old) who completed the EPDS as a part of a larger online survey. ResultsInitial analysis indicated misfit to the Rasch model attributed to local dependency between individual EPDS items. The best model fit was achieved after combining six locally dependent items into three super-items resulting in non-significant item-trait interaction (χ2(49) = 46.61, p < 0.57), strong reliability (PSI = 0.86), unidimensionality and item invariance across personal factors such as age and mothers' education. This permitted generation of ordinal-to-interval conversion algorithms derived from person estimates of the Rasch model. LimitationsOrdinal-to-interval conversion cannot be applied for individuals with missing data. ConclusionsThe EPDS met expectations of the unidimensional Rasch model after internal modifications, and its precision can be enhanced by using ordinal-to-interval conversion tables published in this article without the need to alter the original scale format. Scores derived from these conversion tables should decrease error and improve conformity with statistical assumptions in both clinical and research use of the EPDS, making monitoring of clinical status and outcomes more accurate.