Ten year citation prediction model for systematic reviews using early years citation data

Abstract

AbstractCitation counts are frequently used for assessing the scientific impact of articles. Current approaches for forecasting future citations counts have important limitations. This study aims to analyse and predict the trajectories of citation counts of systematic reviews (SR) based on their citation profiles in the previous years and predict quantiles of future citation counts. We included all SR published between 2010 and 2012 in medical journals indexed in the Web of Science. A longitudinal k-means (KML) clustering approach was applied to identify trajectories of citations counts 10 years after publication, according to the yearly citation count, the proportion of all cites attained in a specific year and the annual variation in citation counts. Finally, we built multinomial logistic regression models aiming to predict in what tercile or quartile of citation counts a SR would be 10 years after publication. Using clustering approaches, we obtained 24 groups of SR. Two groups (7.9% of the articles) had an average of > 200 citations, while two other groups (10.4% of the articles) presented an average of < 10 citations. The model predicting terciles of citation counts attained an accuracy of 72.8% (95%CI = 71.1–74.3%) and a kappa coefficient of 0.59 (95%CI = 0.57–0.62). Prediction of citation quartiles (combining the second and third quartiles into a single group) attained a accuracy of 76.2% (95%CI = 74.7–77.8%) and a kappa coefficient of 0.62 (95%CI = 0.59–0.64). This study provides an approach for predicting of future citations of SR based exclusively on citation counts from the previous years, with the models developed displaying an encouraging accuracy and agreement.