Accuracy of the Adaptive Computerized COPD Exacerbation Self-management Support (ACCESS) application to support patients’ exacerbation self-management. Preliminary results

Background: Early recognition and prompt treatment may reduce exacerbations and prevent complications in patients with COPD. However, many patients find it difficult to follow the instructions of written exacerbation action plans. To support patients more adequately we developed (i) a Bayesian network model to predict exacerbation risk and (ii) an expert-based model to provide exacerbation treatment advice. We integrated both models in an application (ACCESS) for smartphone and tablet. Aim: To assess the validity of ACCESS, i.e. the validity of the Bayesian network model to predict exacerbation risk, and to examine the relationship between the risk prediction and the expert-based treatment advice. Methods: In a 3-month prospective cohort study, 54 patients with COPD recorded exacerbation-related data in diaries. Diagnostic test characteristics were calculated to establish the validity of ACCESS, using exacerbations diagnosed by chest physicians as gold standard. The relationship between the advice to take prompt action and the risk predictions of ACCESS was estimated using a multilevel binary logistic regression analysis. Results: Sensitivity and specificity of detecting an exacerbation by ACCESS were 86.6% (95%CI 68.4-95.6) and 82.4% (95%CI 68.3-91.1), respectively. Area under the Curve (AUC) was 0.88. The odds ratio of providing the advice to take prompt action when an exacerbation was detected by ACCESS was 8.3 (95%CI 6.3-10.8). Conclusion: The validity of ACCESS to predict exacerbations appears to be high. The advice to take prompt action was strongly related to the risk of an exacerbation predicted by ACCESS.