|
Affiliation
|
Centre for Big Data Research in Health (CBDRH), UNSW Sydney, Sydney, NSW, Australia. |
|
Abstract
|
PURPOSE: Records of antidepressant dispensings are often used as a surrogate measure of depression. However, as antidepressants are frequently prescribed for indications other than depression, this is likely to result in misclassification. This study aimed to develop a predictive algorithm that identifies patients using antidepressants for the treatment of depression.
METHODS: Pharmaceutical Benefits Scheme (PBS) and Medicare Benefits Schedule (MBS) claims data were linked to follow-up questionnaires (completed in 2012-2013) for participants of the 45 and Up Study-a cohort study of residents of New South Wales, Australia, aged 45 years and older. The sample composed participants who were dispensed an antidepressant in the 30 days prior to questionnaire completion (n = 3162). An algorithm based on patient characteristics, pharmaceutical dispensings, and claims for mental health services was built using group-lasso interaction network (glinternet), with self-reported receipt of treatment for depression as the outcome. The predictive performance of the algorithm was assessed via bootstrap resampling.
RESULTS: The algorithm composes 15 main effects and 11 interactions, with type of antidepressant dispensed and claims for mental health services the strongest predictors. The ability of the algorithm to discriminate between antidepressant users with and without depression was 0.73. At a predicted probability cut-off of 0.6, specificity was 93.8% and sensitivity was 23.6%.
CONCLUSIONS: Using this algorithm with a high probability cut-off yields high specificity and facilitates the exclusion of individuals using antidepressants for indications other than depression, thereby mitigating the risk of confounding by indication when evaluating the outcomes of antidepressant use.
© 2019 John Wiley & Sons, Ltd.
Language: en |