Takemi Program in International Health
The Takemi Program in International Health seeks to improve health and health systems around the world by welcoming mid-career health professionals and scholars to the Harvard T. H. Chan School of Public Health to conduct path breaking research and develop their leadership skills.
665 Huntington Avenue, Bldg. 1, Room 1210
Boston, MA 021151, USA
Identification of the key factors predicting depression dynamics based on cohorts with diverse socio-economic conditions in Asia
Jie Jane Zhao,*1,2 Yuanqing Xia,1 Tiffany Leung,1 Laura D. Kubzansky,2 and Liming Liang2
*Takemi Fellow, Harvard T.H. Chan School of Public Health
¹ School of Public Health, The University of Hong Kong
² Harvard T.H. Chan School of Public Health, Boston, MA
Background
Depression affects millions of people worldwide and has become an urgent public health issue. Early detection and early treatment of depression will be of great value for lowering the risk of depression.
Research Gap
- Previous prediction models were mainly built in western settings, which may not be generalized to other settings with different culture.
- Depression is an evolving condition, previous prediction models used depression score measured at a single time point.
Aim
This study aims to identify and compare the key predictors for depression dynamics using cohorts with diverse socio-economic conditions in Asia.
Methods
Data sources:
- China Health and Retirement Longitudinal Study (CHARLS)
- The Indonesian Family Life Survey (IFLS)
Statistical analysis:
- Use two approaches (1) including common predictors of two cohorts; (2) including all predictors available in each cohort
- Using CatBoost to select key factors for depression score change (categorized to 0: no change/get better; 1: get worse) and compare the selected predictors from approach 1.
Results
- We used area under the curve (AUC) to assess performance.
- Based on 34 common predictors in two cohorts, the AUC is 0.64 in CHARLS and 0.58 in IFLS.
- Using all available predictors (53 in CHARLS, 66 in IFLS), the AUC is 0.75 in CHARLS and 0.58 in IFLS.
- Comparing the top ten important factors selected in each dataset, we find that only four variables (40%) are shared between two cohorts, and differ in ranking of importance.
- In CHARLS, sleep quality, satisfaction with life and memory were additionally selected in approach 2, which improved AUC.
Table 1: Top 10 important factors by dataset (ranked by importance score)
| Rank | CHARLS | IFLS |
|---|---|---|
| 1 | Activities of daily living | Activities of daily living |
| 2 | Self-rated health | Cognitive function |
| 3 | Age | Education |
| 4 | Cognitive function | Birth place |
| 5 | Transfer money to parents | Age |
| 6 | Father’s education | Number of living children |
| 7 | Transfer money to children | Mother alive |
| 8 | BMI | BMI |
| 9 | Smoke | Self-rated financial status |
| 10 | Childhood self-rated financial status | Mother’s education |
Discussion
- Some factors such as activities of daily living, self-rated health, cognitive function, BMI and financial status are key factors predicting depression dynamics in different settings, despite their importance is different in different settings.
- We should consider local settings when considering how to identify the vulnerable groups.
- Including measures of sleep quality, satisfaction with life and memory will be helpful for predicting depression dynamics.
References
- Handajani YS, et al. Clin Pract Epidemiol Ment Health. 2022.
- Shen et al. BMC Psychiatry. 2024.
- Li et al. Front. Psychol. 2024.
- Tama TD, Astutik E, Reuwpassa JO. Yale J Biol Med. 2021.
Contact
Jie Jane Zhao
janezhao@hku.hk
janezhao@hsph.harvard.edu
Acknowledgement
HKU and Harvard T.H. Chan Takemi Program