From PHQ-2 to PHQ-2W: Data-driven identification of depressed mood and fatigue for optimized weighted depression...

This study aims to optimize depression screening tools through a data-driven approach, identifying the most predictive core item combination from the PHQ-9 scale to construct a new simplified depression screening tool.

Using 11 international datasets, we employed RFECV to select the most predictive item combination from the PHQ-9. Logistic regression models were developed and externally validated across multiple independent datasets, systematically comparing the newly developed PHQ-2W (items 2 + 4) with the traditional PHQ-2 (items 1 + 2).

Data analysis revealed that the combination of depressed mood (item 2) and fatigue (item 4) demonstrated the strongest predictive power, explaining approximately 80 % of the variation in PHQ-9 total scores. With a PHQ-9 total score cut-off of ≥15, the PHQ-2W model achieved an AUC of 0.9590, sensitivity of 0.9308, and specificity of 0.8533, with a cut-off of ≥10, the model achieved an AUC of 0.9545, sensitivity of 0.8999, and specificity of 0.8632. External validation showed that compared to the traditional PHQ-2, PHQ-2W performed better across multiple metrics, with average improvements of 1.9-3.0 percentage points in AUC and 3.0 percentage points in accuracy. Notably, with a PHQ-9 ≥ 10 cut-off, PHQ-2W demonstrated a substantial 17 percentage point improvement in sensitivity (0.8714 vs. 0.6980).

Depressed mood and fatigue, as core symptoms of depression, together form the affective-somatic core network of depression. Through weight allocation and flexible threshold adjustment, PHQ-2W combines efficiency with adaptability, offering a new approach and practical direction for early identification and intervention of depression.