Data Sources
Primary Database
Chinese Psychological Health Guard Project (CPHG)
Left-behind children T1
n = 179,877
Validation Database 1
CPHG Single-parent children T1
n = 48,128
Validation Database 2
CPHG Left-behind children T2
n = 133,904
Age Validation (CLDS)
Adolescents (15-18)
T1: n = 1,075; T2: n = 742
Age Validation (CLDS)
Young Adults (19-30)
T1: n = 3,207; T2: n = 2,052
Age Validation (CLDS)
Middle-aged Adults (31-65)
T1: n = 15,758; T2: n = 12,697
Age Validation (CLDS)
Older Adults (≥66)
T1: n = 948; T2: n = 955
Primary Data Processing
Training Set (70%)
n = 125,914
Model Development
Testing Set (30%)
n = 53,963
Internal Validation
Feature Selection & Model Development
Regression Model: Recursive Feature Elimination (RFE) with Multiple Linear Regression
Classification Model: Logistic Regression, Random Forest, Support Vector Machine
Evaluation: 10-fold Cross-validation, ROC Analysis, Brier Score, Decision Curve Analysis
Resulting Stratified Screening System
Tier 1: Rapid Screening (4 items)
C18 "I feel sad" + C09 "I think my life has been a failure" +
C06 "I feel depressed" + C19 "I feel that people dislike me"
AUC = 0.982, Sensitivity = 0.945, Specificity = 0.926
Tier 2: Enhanced Assessment (9 items total)
Above 4 items + C12 "I feel unhappy" + C07 "I feel that everything I do is an effort" +
C14 "I feel lonely" + C02 "I don't feel like eating" + C01 "Things bother me"
R² = 0.957 (predicts full CES-D-20 score)
External Validation Results
Cross-sample Validation
Consistent performance across all validation samples:
R² > 0.94, AUC > 0.97, Accuracy > 0.91
Optimal Clinical Utility in Risk Threshold Range (0.3-0.6)