Optimizing the assessment of suicide attempters with a decision tree

Abstract

BackgroundOptimizing psychiatric assessments could help to standardize the use of structured instruments in clinical practice. In recent years, several research groups have applied Computerized Adaptive Tests (CATs) to simplify assessments in depression, anxiety and also suicidal behaviors. We aimed to construct a shortened test to classify suicide attempters using a decision tree methodology that allows the integration of relevant clinical information, namely the history of past suicide attempts, in the construction of the test.MethodsThe sample was composed of 902 adult participants in three subsamples: first-time suicide attempters, psychiatric inpatients that never attempted suicide and healthy controls. The performance of a decision tree built using the items of a previously developed scale for suicidal risk was examined. The history of past suicide attempts was used to separate patients in the decision tree. The data was randomly divided in a training set and a test set. The test set, that contained 25% of the data, was used to determine the accuracy of the decision tree. Twenty-five cross-validations of this set up were conducted.ResultsThe first four items of the decision tree classified correctly 81.4% of the patients.ConclusionAs a result of a methodology based on decision trees that, contrary to CATs, can incorporate relevant information in building the test we were able to create a shortened test capable of separating suicidal and non-suicidal patients. Using all the information that is available improves the precision and utility of instruments adapted for psychiatric assessments.Disclosure of interestThe authors have not supplied their declaration of competing interest.