The Dysfunctional Self-Focus Attributes Scale-7 (DSAS-7): A Machine Learning-based Development of a Shortened Version of the DSAS

초록

The Dysfunctional Self-focus Attributes Scale (DSAS) is a 15-item tool designed to measure dysfunctional self-focus. However, its length can be burdensome in clinical practice. In this study, we aimed to develop a data-driven, shortened version of the DSAS that accurately predicts the total DSAS score. We collected a sample of about 1,000 responses and employed exploratory factor analysis (EFA) as well as confirmatory factor analysis (CFA) to identify the underlying structure of the DSAS. To further refine the scale, we used eXtreme Gradient Boosting (XGBoost) to select the items most predictive of the total DSAS score. Through EFA and CFA, we identified and validated a two-factor structure for the DSAS items. From each factor, we selected key items based on their contribution to predicting the total DSAS score using XGBoost. With the seven key items, we developed a shortened version of the DSAS, the DSAS-7, which performed exceptionally well in predicting the total DSAS score (R-2 = 0.938). Additionally, the DSAS-7 demonstrates robust predictive power across heterogeneous data samples from nurses (R-2 = 0.942), people infected with coronavirus (R-2 = 0.927), and patients with cancer (R-2 = 0.920), as well as a general population sample of 600 adults (R-2 = 0.943). The DSAS-7 offers a concise and efficient tool for assessing dysfunctional self-focus in a clinical context. This study highlights the effectiveness of integrating traditional factor analysis with machine learning techniques to develop shortened versions of questionnaires while maintaining both performance and reliability.

키워드