You are here

  • Home
  • Archive
  • Volume 13, Issue 7
  • Estimating causal effects of physical disability and number of comorbid chronic diseases on risk of depressive symptoms in an elderly Chinese population: a machine learning analysis of cross-sectional baseline data from the China longitudinal ageing soci…

Article Text

Article menu

Download PDFPDF

Health informatics
Original research
Estimating causal effects of physical disability and number of comorbid chronic diseases on risk of depressive symptoms in an elderly Chinese population: a machine learning analysis of cross-sectional baseline data from the China longitudinal ageing social survey
  1. Zhenjie Wang1,
  2. Hanmo Yang2,
  3. Chenxi Sun3,
  4. Shenda Hong4
  1. 1 Institute of Population Research, Peking University, Beijing, People's Republic of China
  2. 2 T. H. Chan School of Public Health, Harvard University, Boston, Massachusetts, USA
  3. 3 School of Intelligence Science and Technology and the Key Laboratory of Machine Perception (Ministry of Education), Peking University, Beijing, People's Republic of China
  4. 4 National Institute of Health Data Science, Peking University, Beijing, People's Republic of China
  1. Correspondence to Dr Zhenjie Wang; zhenjie.wang{at}pku.edu.cn; Dr Shenda Hong; hongshenda{at}pku.edu.cn

Abstract

Objective This study aimed to explore the causal effects of physical disability and number of comorbid chronic diseases on depressive symptoms in an elderly Chinese population.

Design, setting and analysis Cross-sectional, baseline data were obtained from the China Longitudinal Ageing Social Survey, a stratified, multistage, probabilistic sampling survey conducted in 2014 that covers 28 of 31 provincial areas in China. The causal effects of physical disability and number of comorbid chronic diseases on depressive symptoms were analysed using the conditional average treatment effect method of machine learning. The causal effects model’s adjustment was made for age, gender, residence, marital status, educational level, ethnicity, wealth quantile and other factors.

Outcome Assessment of the causal effects of physical disability and number of comorbid chronic diseases on depressive symptoms.

Participants 7496 subjects who were 60 years of age or older and who answered the questions on depressive symptoms and other independent variables of interest in a survey conducted in 2014 were included in this study.

Results Physical disability and number of comorbid chronic diseases had causal effects on depressive symptoms. Among the subjects who had one or more functional limitations, the probability of depressive symptoms increased by 22% (95% CI 19% to 24%). For the subjects who had one chronic disease and those who had two or more chronic diseases, the possibility of depressive symptoms increased by 13% (95% CI 10% to 15%) and 20% (95% CI 18% to 22%), respectively.

Conclusion This study provides evidence that the presence of one or more functional limitations affects the occurrence of depressive symptoms among elderly people. The findings of our study are of value in developing programmes that are designed to identify elderly individuals who have physical disabilities or comorbid chronic diseases to provide early intervention.

  • Health informatics
  • mental health
  • causal effects
  • older population
  • Chinese

Data availability statement

Data are available on reasonable request. The datasets generated and/or analysed during the current study are available atin the China Longitudinal Ageing Social Survey (CLASS) repository, (http://class.ruc.edu.cn/index.php?r=index/index&hl=en).

https://creativecommons.org/licenses/by/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution 4.0 Unported (CC BY 4.0) license, which permits others to copy, redistribute, remix, transform and build upon this work for any purpose, provided the original work is properly cited, a link to the licence is given, and indication of whether changes were made. See: https://creativecommons.org/licenses/by/4.0/.

https://doi-org.ezproxy.u-pec.fr/10.1136/bmjopen-2022-069298

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    STRENGTHS AND LIMITATIONS OF THIS STUDY

    • The current study used a large, representative population-based sample covering 28 of 31 provincial areas in China.

    • The study estimated causal effects using the conditional average treatment effect method.

    • We were unable to evaluate the effects of lifestyle factors since information on these factors was not collected in the China Longitudinal Ageing Social Survey.

    • The findings of this cross-sectional study established a cause-and-effect relationship based on the use of a machine learning algorithm; these must be considered in further research.

    Introduction

    Currently, depressive symptoms (DSs) are one of the most prevalent mental disorders worldwide.1 The characteristics of DS include symptoms of sadness, depressed mood and loss of interest.1 The presence of DS is usually associated with low quality of life, cancer, chronic diseases, suicide and other conditions.2–5 DS can also place a heavy burden on families, communities and health services worldwide.6 7 The prevalence of DS varies widely and ranges from 1% to 16% among middle-aged and elderly populations.8–10 In China, the prevalence of DS among elderly people has increased rapidly from 3.9% to 17% in the past two decades.11–13

    Many studies have provided evidence that specific demographic variables are associated with DS risk,14 although the associations are inconsistent.10 13 An inverse or U-shaped association between age and DS has been found in many studies.15–17 Demographic status and level of education were also found to be associated with DS risk.18–20 However, some studies reported no association of DS with demographic status or level of education.14 17 21

    Machine learning (ML) models can handle complex relationships between variables and can capture non-linearities and interactions that may be missed by traditional statistical models. Currently, researcher can evaluate the treatment effect though the average effect treatment effect.22 Conditional average treatment effect (CATE) estimation is a specific ML technique used in the context of causal inference. CATE aims to estimate the effect of a treatment on an outcome for a specific subpopulation or individual, rather than the average effect across the entire population. Very few studies have used ML to obtain evidence regarding the causal effects of physical disability or the number of comorbid chronic diseases on DS risk in the elderly Chinese population.

    In this study, we used ML to explore the CATE of physical disability and number of comorbid chronic diseases on DS using data from the Chinese Longitudinal Ageing Social Survey (CLASS).

    Methods

    Data source

    We drew the sample used in the current study from the cross-sectional baseline data the CLASS, which was conducted in 2014.23 The CLASS was collected by the National Survey Research Center, Renmin University. The CLASS used a stratified, multistage, probabilistic sampling method to select a nationally representative sample covering 28 of 31 provincial areas in China. A total of 11 511 older adults were surveyed. In this study, the sample comprised 7496 subjects 60 years of age or older who answered the questions on DS and other independent variables of interest collected in 2014. All the participants were interviewed face to face by trained interviewers.

    Measurement of variables

    Depressive symptoms

    DSs were assessed using the nine-item Centre for Epidemiological Studies Depression Scale (CES-D). This scale includes three items designed to assess positive feelings and two items each designed to assess negative emotions, somatic symptoms and sense of marginalisation. The nine-item CES-D is reliable and valid for detecting nonpsychotic mental disorders among older Chinese adults.24 For each item, a score of 0 (rarely or none of the time), 1 (some of the time) or 2 (most of the time) was assigned; thus, the total score ranged from 0 to 18. Since the coding of the positive effect items was reversed, a higher score indicates a higher level of DSs. In the current study, on the nine-item scale, the total possible score was 18 (9 items multiplied by 2, the highest response). A standardised cut-off score of 4.8 for the nine-item form of the CES-D was established using Kohout’s formula.25 In this study, the internal Cronbach’s alpha for the nine items was 0.75.

    Physical disabilities

    Activities of daily living (ADLs) were assessed using the Barthel Index. For each individual, the self-reported number of comorbid chronic diseases that were based on self-reported previous medical record, including health problems such as hypertension, diabetes, heart disease, renal disease, liver disease, stroke, tuberculosis, arthritis and respiratory diseases, was categorised as ‘0’, ‘1’ or ‘≥2’.

    Confounding variables

    Confounding variables, which were significantly associated with DSs,26 under consideration were gender (male, female), age, marital status (married, widowed/divorced/unmarried), ethnicity (Han, others), residence (rural, urban), educational level (junior high school and above, primary school, never attended school) and living arrangements (lives alone, lives with others). Income was categorised into five levels using quintiles of household income (yuan) (Q1: ≤3000; Q2: >3000 and ≤10 000; Q3: >10 000 and ≤24 000; Q4: >24 000 and ≤36 000; Q5: >36 000).

    Estimation of causal effects

    Causal analysis aims to determine the causal relationships between variables (treatments) and outcomes. Causal relationships are essentially different from correlations in that causal relationships mainly solve the problem of ‘why’. In this work, we used causal ML, an open-source Python software provided by Uber, to perform causal analysis using ML algorithms based on recent research.27 Causal ML provides a standard interface that allows users to estimate the average treatment effect (ATE) based on experimental or observational data.

    To estimate the causal relationship between the variable of interest (known as treatment indicator E) and the outcome Y, we performed a two-step analysis: (1) we built a ML model and (2) we estimated the ATE based on causal ML. First, we built a random forest classifier (RFC) and used it to predict DS using all the variables except the treatment indicator (denoted X).28 The RFC contains 100 decision trees and uses the Gini purity index to judge the segmentation criteria of a tree node. The maximum depth of a single tree was set to unlimited so that each tree would be segmented until each leaf node of the tree model contained only one sample. The RFC output was 1 if the individual had DS and 0 otherwise. The RFC model was built using a scikit-learn ML framework.29

    We then set the above trained RFC as the base learner for causal ML and estimated the treatment effect based on the algorithm described here.30 The algorithm, which uses a single base learner from the ML model and has the treatment indicator as a feature, operates in two stages. In the first stage, the algorithm estimates the average outcomes after specific treatments using the trained RFC model, where the outcome is calculated as Embedded Image . In the second stage, the ATE was calculated as Embedded Image . We also used SHAPley Additive exPlanations, a game theory approach, to explain the output of the causal analysis results modelled by the ML models.31 We analysed the data using Python V.3.7.14 based on the anaconda3 development platform, causalml V.0.10.0, sklearn V.0.23.2, pandas V.1.0.1 and numpy V.1.21.0.

    Survey conduct and consent

    The interviewer-administered questionnaire was prepared in the local language (Chinese). Consent to participation in the study was received from each participant prior to data collection, and the data collectors were trained to provide any requested information or clarification at any time during the interview. Only participants who wished to continue participating in the study after providing informed consent were included in the study.

    Patient and public involvement

    None.

    Results

    Descriptive statistics

    The characteristics of subjects with and without DSs are summarised in table 1. The overall average CES-D score was 4.56 (SD: 3.56). The prevalence of DS in the Chinese population over 60 years of age was 43%. In the current study, male subjects, urban residents, people living with others and people of Han ethnicity accounted for the majority of elderly Chinese individuals with DS.

    View this table:
    Table 1

    Subjects’ characteristics by their depressive symptom status according to the 9-item Centre for Epidemiological Studies Depression Scale among the elderly population in China

    Estimated causal effects

    The estimated causal effects of physical disability and chronic diseases on DSs are presented in table 2. Physical disability and the number of comorbid chronic diseases had causal effects on DS. Among the subjects who had one or more functional limitations, those who had one chronic disease, and those who had two or more chronic diseases, the risk probability of DS increased by 22%, 13% and 20%, respectively.

    View this table:
    Table 2

    Estimated causal effects of physical disability and chronic disease on depressive symptoms

    The effects of the mechanism direction of physical disability and number of comorbid chronic diseases on DS are presented in figures 1 and 2 in descending order of importance In descending order of importance among subjects who had one or more functional limitations, the variables are as follows: educational level, age, number of comorbid chronic diseases, wealth quantile, marital status, living arrangement, residence, gender and ethnicity. In descending order of importance among subjects who had only one chronic disease, the variables are as follows: wealth quantile, age, educational level, residence, gender, marital status, physical disability, living arrangement and ethnicity. In descending order of importance among subjects who had one or more functional limitations, the variables are as follows: wealth quantile, age, residence, educational level, gender, marital status, living arrangement, physical disability and ethnicity. Age, residence, gender and physical disability contributed to DS risk among subjects who had one or more chronic diseases.

    Figure 1
    Figure 1

    The importance sort and mechanism direction of physical disability with depressive symptoms by using explainable machine learning predictions. The importance of variables (Mean of SHAP value with 3 decimal places: risk factor (RF): mean of SHAP value >0; protective factor (PF): mean of SHAP value <0; no effect (NE): mean of SHAP value=0) by descending were education level (NE), age groups (RF), number of comorbid chronic disease (PF), wealth quantile (NE), marital status (PF), living arrangement (NE), residence (RF), gender (RF) and ethnicity (PF). SHAP, SHAPley Additive exPlanations.

    Figure 2
    Figure 2

    The importance sort and mechanism direction of number of comorbid chronic disease with depressive symptoms by using explainable machine learning predictions. (A) Number of comorbid chronic disease: 1 (Ref: 0). The importance of variables (mean of SHAP value with 3 decimal places: risk factor (RF): mean of SHAP value >0; protective factor (PF): mean of SHAP value <0; no effect (NE): mean of SHAP value=0) by descending were wealth quantile (NE), age groups (RF), education level (RF), residence (RF), gender (RF), marital status (PF), physical disability (RF), living arrangement (NE) and ethnicity (PR). (B) Number of comorbid chronic disease: ≥2 (Ref: 0). The importance of variables (mean of SHAP value with 3 decimal places: risk factor (RF): mean of SHAP value >0; protective factor (PF): mean of SHAP value <0; no effect (NE): mean of SHAP value=0) by descending were wealth quantile (PF), age groups (RF), residence (RF), education level (NE), gender (RF), marital status (NF), living arrangement (NE), physical disability (NF) and ethnicity (PF). SHAP, SHAPley Additive exPlanations.

    Discussion

    Overall, physical disability and the number of comorbid chronic diseases present had causal effects on DS risk. Subjects who had one or more functional limitations had a 22% increased risk of DS. Subjects who had one chronic disease and those who had two or more chronic diseases showed increases of 13% and 20%, respectively, in their DS risk.

    In the current study, we found causal effects of physical disability and comorbid chronic diseases on DS. A previous study suggested that some health variables, such as poor self-rated health and disability, are associated with DS, indicating that health status tends to have similar effects on mental health regardless of social context.32 Physical disability is considered an inability or a diminished capacity to perform basic self-care activities. Physical disability lessens an individual’s ability to interact with the physical and social environment, hence increasing the risk of DS.33 Increased disability in ADLs and instrumental ADLs have been proven to be connected with DSs.34 35 Failure to perform ADLs has been positively associated with DSs in elderly people, and physical disability accounts for incidences of DSs in elderly individuals.36 37 Furthermore, the finding that physical disability leads to an increase in DSs in geriatric individuals over time is supported by longitudinal studies.38Given the influence of physical disability and DSs on the well-being of elderly individuals, it is urgent to find ways to address these issues. The results presented here demonstrate that impairment of ADLs has a significant effect on DSs in elderly people, consistent with the findings of previous studies that showed a relationship between limitations on activity and psychosocial problems.33 34

    The number of chronic diseases present in an individual was found to be associated with DS risk, consistent with previous studies. These associations could be explained by several factors, including unhealthy lifestyle, shared genetic influences and inflammation related to disease.39 Previous studies have suggested that experiencing pain and physical disability could affect mood in elderly individuals.20 Another explanation is that subjects with DSs might be more sedentary than those without DSs; such a lifestyle could cause increased bone resorption and decreased bone formation.40 41

    Causal inference using ML methods such as CATE can be a powerful tool for analysing observational data. They model the treatment effect using regression or decision trees. ML can learn causal effects because it can identify and account for confounding factors, which are variables that may affect both the treatment and outcome variables. By controlling for confounding factors, ML algorithms can estimate the causal effect of the treatment variable on the outcome variable, even in the absence of a randomised controlled trial. Specifically, CATE aims to estimate the effect of a treatment on an outcome for a specific subpopulation or individual, rather than the average effect across the entire population. This method can help identify treatment effects for different subgroups of the population and enable personalised treatment recommendations.

    Strengths and limitations

    Our study has many strengths, including a large sample size, a population-based design and adjustment for a wide range of socioeconomic characteristics. Another noticeable strength is that the measurement of all physical illnesses took place prior to the CES-D measurement, minimising the risk of reverse causation. However, our study also has several limitations that should be taken into account by future researchers. First, the physical conditions information were self-reported, which should also be considered for future research. Moreover, although we used ML to estimate the causal effects of the factors on DS, these findings should be interpreted cautiously in view of the cross-sectional nature of the data used in the present analysis. We will be able to extend the current study to determine causality when longitudinal data are available. Although the current data were collected in 2014, they are available on the web. Because all the information was collected by trained interviewers, the quality of the survey was reliable. This point should also be considered with caution by future researchers. It is important to note that the CLASS does not provide information on lifestyle factors (ie, weight, height, smoking, alcohol consumption and other factors), some of which have been suggested as risk factors for depression.42–44

    Conclusion

    In conclusion, in this large population-based study in an elderly Chinese population, physical disability and the number of comorbid chronic diseases were found to increase DS risk. This study provides evidence of the effect of one or more functional limitations on the occurrence of DSs among elderly people. This is helpful in planning preventive measures and in improving knowledge of how to treat DSs. The findings made in our study are valuable for the development of prevention programmes designed to identify elderly individuals who have physical disabilities or a number of comorbid chronic diseases and provide early intervention.

    Data availability statement

    Data are available on reasonable request. The datasets generated and/or analysed during the current study are available atin the China Longitudinal Ageing Social Survey (CLASS) repository, (http://class.ruc.edu.cn/index.php?r=index/index&hl=en).

    Ethics statements

    Patient consent for publication

    Ethics approval

    This study involves human participants but all procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the 1964 Declaration of Helsink and its later amendments or comparable ethical standards. The surveys were conducted within the legal framework governed by statistical law in China. This article does not contain any studies with animals performed by any of the authors. Participants gave informed consent to participate in the study before taking part.

    Acknowledgments

    We would like to thank the Institute of Gerontology and National Survey Research Center at Renmin University of China for providing the CLASS data.

    References

    1. ↵
      2. Whiteford HA ,
      3. Ferrari AJ ,
      4. Degenhardt L , et al
      . The global burden of mental, neurological and substance use disorders: an analysis from the global burden of disease study 2010. PLoS One 2015;10:e0116820. doi:10.1371/journal.pone.0116820
    2. ↵
      2. Dalton SO ,
      3. Laursen TM ,
      4. Ross L , et al
      . Risk for hospitalization with depression after a cancer diagnosis: a nationwide, population-based study of cancer patients in Denmark from 1973 to 2003. J Clin Oncol 2009;27:1440–5. doi:10.1200/JCO.2008.20.5526
      OpenUrlAbstract/FREE Full Text
    3. ↵
      2. Hsu Y-T ,
      3. Liao C-C ,
      4. Chang S-N , et al
      . Increased risk of depression in patients with parkinson disease: a nationwide cohort study. Am J Geriatr Psychiatry 2015;23:934–40. doi:10.1016/j.jagp.2014.10.011
      OpenUrl
    4. ↵
      2. Ayerbe L ,
      3. Ayis S ,
      4. Wolfe CDA , et al
      . Natural history, predictors and outcomes of depression after stroke: systematic review and meta-analysis. Br J Psychiatry 2013;202:14–21. doi:10.1192/bjp.bp.111.107664
      OpenUrlAbstract/FREE Full Text
    5. ↵
      2. Cuijpers P ,
      3. Vogelzangs N ,
      4. Twisk J , et al
      . Comprehensive meta-analysis of excess mortality in depression in the general community versus patients with specific illnesses. Am J Psychiatry 2014;171:453–62. doi:10.1176/appi.ajp.2013.13030325
      OpenUrlCrossRefPubMed
    6. ↵
      1. World Health Organization
      . Depression–a global public health concern. 2012. Available: http://hesp- news.org/2012/10/05/depression-a-global public-health-concern/ [Accessed 29 Nov 2018].
    7. ↵
      2. Walker ER ,
      3. McGee RE ,
      4. Druss BG
      . Mortality in mental disorders and global disease burden implications: a systematic review and meta-analysis. JAMA Psychiatry 2015;72:334–41. doi:10.1001/jamapsychiatry.2014.2502
      OpenUrl
    8. ↵
      2. Wilhelm K ,
      3. Mitchell P ,
      4. Slade T , et al
      . Prevalence and correlates of DSM-IV major depression in an Australian national survey. J Affect Disord 2003;75:155–62. doi:10.1016/s0165-0327(02)00040-x
      OpenUrlCrossRefPubMedWeb of Science
    9. ↵
      2. Waraich P ,
      3. Goldner EM ,
      4. Somers JM , et al
      . Prevalence and incidence studies of mood disorders: a systematic review of the literature. Can J Psychiatry 2004;49:124–38. doi:10.1177/070674370404900208
      OpenUrlCrossRefPubMed
    10. ↵
      2. Djernes JK
      . Prevalence and predictors of depression in populations of elderly: a review. Acta Psychiatr Scand 2006;113:372–87. doi:10.1111/j.1600-0447.2006.00770.x
      OpenUrlCrossRefPubMedWeb of Science
    11. ↵
      2. Chen R ,
      3. Copeland JR ,
      4. Wei L
      . A meta-analysis of epidemiological studies in depression of older people in the people's Republic of China. Int J Geriatr Psychiatry 1999;14:821–30. doi:10.1002/(sici)1099-1166(199910)14:10<821::aid-gps21>3.0.co;2-0
      OpenUrlCrossRefPubMed
    12. ↵
      2. Li D ,
      3. Zhang D ,
      4. Shao J , et al
      . A meta-analysis of the prevalence of depressive symptoms in Chinese older adults. Arch Gerontol Geriatr 2014;58:1–9. doi:10.1016/j.archger.2013.07.016
      OpenUrlCrossRefPubMed
    13. ↵
      2. Zhang L ,
      3. Xu Y ,
      4. Nie H , et al
      . The prevalence of depressive symptoms among the older in China: a meta-analysis. Int J Geriatr Psychiatry 2012;27:900–6. doi:10.1002/gps.2821
      OpenUrlCrossRefPubMed
    14. ↵
      2. Gao S ,
      3. Jin Y ,
      4. Unverzagt FW , et al
      . Correlates of depressive symptoms in rural elderly Chinese. Int J Geriatr Psychiatry 2009;24:1358–66. doi:10.1002/gps.2271
      OpenUrlCrossRefPubMed
    15. ↵
      2. Kessler RC ,
      3. Birnbaum HG ,
      4. Shahly V , et al
      . Age differences in the prevalence and co-morbidity of DSM-IV major depressive episodes: results from the WHO world mental health survey initiative. Depress Anxiety 2010;27:351–64. doi:10.1002/da.20634
      OpenUrlCrossRefPubMedWeb of Science
    16. ↵
      2. Pan A ,
      3. Franco OH ,
      4. Wang Y-F , et al
      . Prevalence and geographic disparity of depressive symptoms among middle-aged and elderly in China. J Affect Disord 2008;105:167–75. doi:10.1016/j.jad.2007.05.003
      OpenUrlCrossRefPubMedWeb of Science
    17. ↵
      2. Yunming L ,
      3. Changsheng C ,
      4. Haibo T , et al
      . Prevalence and risk factors for depression in older people in Xi'an China: a community-based study. Int J Geriatr Psychiatry 2012;27:31–9. doi:10.1002/gps.2685
      OpenUrlPubMed
    18. ↵
      2. Lorant V ,
      3. Croux C ,
      4. Weich S , et al
      . Depression and socio-economic risk factors: 7-year longitudinal population study. Br J Psychiatry 2007;190:293–8. doi:10.1192/bjp.bp.105.020040
      OpenUrlAbstract/FREE Full Text
    19. ↵
      2. Van de Velde S ,
      3. Bracke P ,
      4. Levecque K
      . Gender differences in depression in 23 European countries. Cross-national variation in the gender gap in depression. Soc Sci Med 2010;71:305–13. doi:10.1016/j.socscimed.2010.03.035
      OpenUrlCrossRefPubMed
    20. ↵
      2. Zhou X ,
      3. Bi B ,
      4. Zheng L , et al
      . The prevalence and risk factors for depression symptoms in a rural Chinese sample population. PLoS ONE 2014;9:e99692. doi:10.1371/journal.pone.0099692
    21. ↵
      2. van der Wurff FB ,
      3. Beekman ATF ,
      4. Dijkshoorn H , et al
      . Prevalence and risk-factors for depression in elderly Turkish and Moroccan migrants in the Netherlands. J Affect Disord 2004;83:33–41. doi:10.1016/j.jad.2004.04.009
      OpenUrlCrossRefPubMedWeb of Science
    22. ↵
      2. Pearl J
      . Causality. In: Causality: Models, Reasoning, and Inference. Cambridge: Cambridge University Press, 2009. doi:10.1017/CBO9780511803161
    23. ↵
      The Chinese longitudinal ageing social study. n.d. Available: http://class.ruc.edu.cn/English/Home.htm
    24. ↵
      2. Silverstein M ,
      3. Cong Z ,
      4. Li S
      . Intergenerational transfers and living arrangements of older people in rural China: consequences for psychological wellbeing. J Gerontol B Psychol Sci Soc Sci 2006;61:S256–66. doi:10.1093/geronb/61.5.s256
      OpenUrlCrossRefPubMedWeb of Science
    25. ↵
      2. Kohout FJ ,
      3. Berkman LF ,
      4. Evans DA , et al
      . Two shorter forms of the CES-D depression symptoms index. J Aging Health 1993;5:179–93. doi:10.1177/089826439300500202
      OpenUrlCrossRefPubMed
    26. ↵
      2. Wang Z ,
      3. Yang H ,
      4. Guo Z , et al
      . Socio-demographic characteristics and co-occurrence of depressive symptoms with chronic diseases among older adults in China: the China longitudinal ageing social survey. BMC Psychiatry 2019;19:310. doi:10.1186/s12888-019-2305-2
    27. ↵
      2. Chen H ,
      3. Harinen T ,
      4. Lee JY , et al
      . CausalML: python package for causal machine learning. ArXiv 2020. doi:10.48550/arXiv.2002.11631
    28. ↵
      2. Breiman L
      . "Random forests". Mach Learn 2001;45:5–32. doi:10.1023/A:1010933404324
      OpenUrlCrossRefPubMedWeb of Science
    29. ↵
      2. Pedregosa F ,
      3. Varoquaux G ,
      4. Gramfort A , et al
      . Scikit-learn: machine learning in python. J Mach Learn Res 2012;12:2825–30.
      OpenUrlCrossRefPubMed
    30. ↵
      2. Künzel SR ,
      3. Sekhon JS ,
      4. Bickel PJ , et al
      . Meta-learners for estimating heterogeneous treatment effects using machine learning. Proc Natl Acad Sci U S A 2019;116:4156–65. doi:10.1073/pnas.1804597116
      OpenUrlAbstract/FREE Full Text
    31. ↵
      2. Lundberg S ,
      3. Lee SI
      . A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing System. 2017: 4768–77.
    32. ↵
      2. Kino S ,
      3. Jang S-N ,
      4. Kawachi I
      . Healthy but unhappy? Cross-national comparative analysis of depressive symptoms in Japanese vs. Korean elders. Arch Gerontol Geriatr 2021;95:104426. doi:10.1016/j.archger.2021.104426
      OpenUrl
    33. ↵
      2. Bruce ML
      . Depression and disability in late life: directions for future research. Am J Geriatr Psychiatry 2001;9:102–12. doi:10.1097/00019442-200105000-00003
      OpenUrlCrossRefPubMedWeb of Science
    34. ↵
      2. Bowen ME ,
      3. Ruch A
      . Depressive symptoms and disability risk among older White and Latino adults by nativity status. J Aging Health 2015;27:1286–305. doi:10.1177/0898264315580121
      OpenUrlCrossRefPubMed
    35. ↵
      2. Hybels CF ,
      3. Pieper CF ,
      4. Blazer DG
      . The complex relationship between depressive symptoms and functional limitations in community-dwelling older adults: the impact of subthreshold depression. Psychol Med 2009;39:1677–88. doi:10.1017/S0033291709005650
      OpenUrlCrossRefPubMed
    36. ↵
      2. Newsom JT ,
      3. Schulz R
      . Social support as a mediator in the relation between functional status and quality of life in older adults. Psychol Aging 1996;11:34–44. doi:10.1037/0882-7974.11.1.34
      OpenUrlCrossRefPubMedWeb of Science
    37. ↵
      2. Kennedy GJ ,
      3. Kelman HR ,
      4. Thomas C
      . The emergence of depressive symptoms in late life: the importance of declining health and increasing disability. J Community Health 1990;15:93–104. doi:10.1007/BF01321314
      OpenUrlCrossRefPubMed
    38. ↵
      2. Lin I-F ,
      3. Wu H-S
      . Does informal care attenuate the cycle of ADL/IADL disability and depressive symptoms in late life? J Gerontol B Psychol Sci Soc Sci 2011;66:585–94. doi:10.1093/geronb/gbr060
      OpenUrlPubMedWeb of Science
    39. ↵
      2. Jia LN ,
      3. Zhuang HL ,
      4. Wang XY , et al
      . Depression status of the elderly and its relationship with quality of life in community of Fuzhou city. Chinese Journal of Geriatrics 2011;30:970–3. doi:10.3760/cma.j.issn.0254-9026.2011.11.024
      OpenUrl
    40. ↵
      2. Bonanno GA ,
      3. Westphal M ,
      4. Mancini AD
      . Resilience in the face of loss and potential trauma. Annu Rev Clin Psychol 2011;7:511–35. doi:10.1146/annurev-clinpsy-032210-104526
      OpenUrlCrossRefPubMed
    41. ↵
      2. Assil SM ,
      3. Zeidan ZA
      . Prevalence of depression and associated factors among elderly Sudanese: a household survey in Khartoum state. East Mediterr Health J 2013;19:435–40. doi:10.26719/2013.19.5.435
      OpenUrl
    42. ↵
      2. Wiltink J ,
      3. Michal M ,
      4. Wild PS , et al
      . Associations between depression and different measures of obesity (BMI, WC, WhtR, WHR). BMC Psychiatry 2013;13:223. doi:10.1186/1471-244X-13-223
    43. ↵
      2. Lopresti AL ,
      3. Hood SD ,
      4. Drummond PD
      . A review of lifestyle factors that contribute to important pathways associated with major depression: diet, sleep and exercise. J Affect Disord 2013;148:12–27. doi:10.1016/j.jad.2013.01.014
      OpenUrlCrossRefPubMed
    44. ↵
      2. Lo A
      . Lifestyle factors on depression. In: Pachana N , ed. Encyclopedia of Geropsychology. Springer, 2015. doi:10.1007/978-981-287-080-3

    Footnotes

    • Contributors Conceptualisation: ZW; methodology: SH; software: ZW and SH; writing: ZW; writing—review: ZW, SH, HY and CS; writing—editing: HY and CS.

    • Funding This work was supported by the National Social Science Fund of China (No. 20BRK020), the National Natural Science Foundation of China (No. 62102008) and the Economic & Social Research Council (ESRC) (No. ES/P011055/1).

    • Disclaimer The funding agencies did not participate in the research or review any details of this study, and the authors are independent of the funder

    • Competing interests None declared.

    • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

    • Provenance and peer review Not commissioned; externally peer reviewed.