| Summary: | <p>Background: Recurrent major mood episodes and subsyndromal mood instability cause substantial disability in patients with bipolar disorder. Early identification of mood episodes enabling timely mood stabilisation is an important clinical goal. Recent technological advances allow the prospective reporting of mood in real time enabling more accurate, efficient data capture. The complex nature of these data streams in combination with challenge of deriving meaning from missing data mean pose a significant analytic challenge. The signature method is derived from stochastic analysis and has the ability to capture important properties of complex ordered time series data. To explore whether the onset of episodes of mania and depression can be identified using self-reported mood data.</p><p> Objective: To explore whether the onset of episodes of mania and depression can be identified using self-reported mood data.</p><p> Methods: Self-reported mood data were collected from 261 participants with bipolar disorder using the True Colours monitoring system. Manic and depressive episodes were defined as an ASRM score 6 or a QIDS score 10, respectively. The signature method was used to extract features from a rolling window of k-weeks, where k = f4, 6, 8, 12, 20, 50g. Two independent models for prediction of depressive and manic mood episodes were trained using logistic regression with the elastic net regularisation scheme to reduce overfitting. The stability of the generalisation error metrics of the predictive models was estimated by pooling 100 repetitions of 10-fold cross validation.</p><p> Results: The signature method on average accurately predicted 79.2% of precursors to depressive episodes with a sensitivity of 76.9% and specificity of 79.5% (PPV=78.6%, AUC=0.86) and 71.9% of precursors to manic episodes with a sensitivity of 73.3% and specificity 79.2% (PPV=77.1%, AUC=0.83). This was more accurate than predictions (accuracy 77.4% for depression and 69.5% for mania, p-value < 0.001) based up on a model comprises three features: the mean value, variability and a number of missing responses within the window.</p><p> Conclusions: The signature method offers a systematic approach to the analysis of longitudinal self-reported mood data. The accuracies of the signature-based linear models are considerably better than linear models based on manually crafted features and the models have the potential to significantly enhance self-management and clinical care in bipolar disorder.</p>
|
|---|