Non-attendance Prediction Models to Pediatric Outpatient Appointments

Non-attendance to pediatric outpatient appointments is a frequent and relevant public health problem.

Using different approaches it is possible to build non-attendance predictive models and these models can be used to guide strategies aimed at reducing no-shows. However, predictive models have limitations and it is unclear which is the best method to generate them. Regardless of the strategy used to build the predictive model, discrimination, measured as area under the curve, has a ceiling around 0.80. This implies that the models do not have a 100% discrimination capacity for no-show and therefore, in a proportion of cases they will be wrong. This classification error limits all models diagnostic performance and therefore, their application in real life situations. Despite all this, the limitations of predictive models are little explored.

Taking into account the negative effects of non-attendance, the possibility of generating predictive models and using them to guide strategies to reduce non-attendance, we propose to generate non-attendance predictive models for outpatient appointments using traditional logistic regression and machine learning techniques, evaluate their diagnostic performance and finally, identify and characterize the population misclassified by predictive models.