Obesity, psychological well-being related measures, and risk of seven non-communicable diseases: evidence from longitudinal studies of UK and US older adults

Background We examined the role of psychological well-being related measures in explaining the associations between obesity and increased risk of non-communicable diseases (NCDs: hypertension, heart disease, stroke, diabetes, arthritis, cancer, and memory-related disease) in older adults. Methods Data were from the English Longitudinal Study of Ageing (ELSA), UK (baseline: Wave 4—2008/2009; n = 8127) and the Health and Retirement Study (HRS), US (baseline: Waves 9 and 10—2008/2010; n = 12,477). Objective body mass index was used to define obesity. A range of psychological well-being related measures (e.g., depressive symptoms, life satisfaction) was available in ELSA (n = 7) and HRS (n = 15), and an index of overall psychological well-being was developed separately in each study. NCDs were from a self-reported doctor diagnosis and/or other assessments (e.g., biomarker data) in both studies; and in ELSA, NCDs from linked hospital admissions data were examined. Longitudinal associations between obesity status, psychological well-being measures, and NCDs were examined using Cox proportional hazard models (individual NCDs) and Poisson regression (a cumulative number of NCDs). Mediation by psychological well-being related measures was assessed using causal mediation analysis. Results Obesity was consistently associated with an increased prospective risk of hypertension, heart disease, diabetes, arthritis, and a cumulative number of NCDs in both ELSA and HRS. Worse overall psychological well-being (index measure) and some individual psychological well-being related measures were associated with an increased prospective risk of heart disease, stroke, arthritis, memory-related disease, and a cumulative number of NCDs across studies. Findings from mediation analyses showed that neither the index of overall psychological well-being nor any individual psychological well-being related measures explained (mediated) why obesity increased the risk of developing NCDs in both studies. Conclusion Obesity and psychological well-being may independently and additively increase the risk of developing NCDs.

stadiometer to the nearest millimetre [1].In HRS, as part of EFTF, a trained interviewer used the Healthometer 830KL scale to measure participants' weight in light clothing without shoes to the nearest half pound.A tape measure was then used to measure standing height without shoes to the nearest quarter inch [2].In both studies, BMI was calculated as weight in kilogram (kg) divided by height in metres squared (m 2 ).
Candidate mediators: Psychological well-being related measures 1. Depressive symptoms An 8-item version of the Centre for Epidemiology Depression Scale (CED-S) was used to quantify depressive symptoms [3].Participants were asked whether they felt negative emotions during the past week (e.g., "did you feel depressed?","did you feel everything you did was an effort") with "yes" and "no" options of answers.All the responses were added together to create a total score ranging from 0 to 8. A higher score indicates greater depressive symptoms.

Enjoyment of life
Enjoyment of life was defined following previous studies [4,5].This was assessed using four questions (e.g., "I enjoy the things that I do," "I enjoy being in the company of others") from the Control, Autonomy, Self-Realization and Pleasure (CASP-19) CASP-19 quality of life questionnaire [6].Responses on a 4-point Likert scale from 0= "often" to 3= "never" were reverse-coded and summed.An enjoyment of life index ranged from 0 to 12 with a higher score indicating greater enjoyment of life.

Eudemonic well-being
Following a previous approach [7], eudemonic well-being was determined by adding together the responses of the remaining 15 items from CASP-19 that were not used to define enjoyment of life (e.g., "How often feels what happens to them is out of their control", "How often feels left out of things").The total score ranged from 0 to 45 with a higher score indicating a greater level of eudemonic well-being.

Life satisfaction
Life satisfaction was assessed using the Satisfaction with Life Scale (SWLS) [8].This scale consisted of five 7-point Likert-scale items (e.g., "in most ways my life is close to my ideal", "the conditions of my life are excellent") (1= "strongly disagree" to 7= "strongly agree").The summary score was determined by summing all responses.A higher score on a possible range of 1-to-35 for a total score indicated greater life satisfaction.

Loneliness
Following a previous study [9], loneliness was measured using a three-item loneliness scale [10], adopted from the 20-item Revised UCLA loneliness scale [11].The following questions were asked to the participants, "How often do you feel: you lack companionship?", "isolated from others?", "left out?") with three options of answers, 1= "hardly ever or never," 2= "some of the time," and 3= "often".The summary score on a possible range of 1-to-9 was generated by summing responses of all the items with a higher score indicating a higher loneliness.

Social support
Participants' reports on positive experiences of social support for different types of relationships (spouse or partner, children, other immediate family members, and friends) were used to define social support.Three questions were asked to participants (e.g., "How much do they really understand the way you feel about things?", "How much can you rely on them if you have a serious problem?") for each relationship with four options of answers available (1= "a lot" to 4= "not at all") (e.g., as in Khondoker, Rafnsson et al [12]).Responses to each question were reverse-coded to make a higher value indicate more positive experience and then averaged for each relationship.The final summary score which represents overall social support was generated by re-averaging the social support scores from all the reported relationships.

Social strain
Social strain was defined in the same manner as social support.Social strain indicated negative social experiences received from different relationships as were assessed for social support.Three items on a 4-point-Likert scale were used to evaluate social strain (e.g., "How much do they criticise you?", "How much do they let you down when you are counting on them?") from each relationship (1= "a lot" to 4= "not at all") (e.g., as in Khondoker, Rafnsson et al [12]).Responses were reversecoded to indicate more negative experiences for a greater value and then averaged for each relationship.An overall social strain score ranging from 1 to 4 was calculated by re-averaging the average social strain scores from all reported relationships.

Positive affect
Thirteen items from the Positive and Negative Affect scale (PANAS-X) were used to evaluate positive affect [13].Participants were asked how they felt in the past 30 days (e.g., "During the last 30 days, to what degree did you feel: determined?", "enthusiastic?") with five options of responses available (1= "not at all'" to 5= "very much").A summary score was generated by adding together all responses.Greater positive affect was indicated by a higher score on a scale from 1 to 65.

Negative affect
Negative affect was assessed using the remaining 12 items from PANAS-X [13].Participants were asked the extent to which they felt negative emotions in the last 30 days (e.g., "During the last 30 days, to what degree did you feel: afraid?", "upset?").As with positive affect, participants' answers on a 5-point-Likert scale (1= "not at all'" to 5= "very much") were summed.A total score ranged from 1 to 60 with a higher score indicating greater negative affect.

Purpose in life
A 7-item domain of purpose in life from the Ryff Measures of Psychological Wellbeing [14,15] was used to evaluate purpose in life.Participants' responses to 6-point-Likert-scale items (e.g., "I enjoy making plans for the future and working to make them a reality.","I am an active person in carrying out the plans I set for myself.")(1= "strongly disagree" to 6= "strongly agree") were added together to create a total score ranging from 1 to 42.A higher score indicated greater purpose in life.

Anxiety
Five items from Beck Anxiety Inventory (BAI) were used to quantify anxiety [16].Participants responded to what degree they felt the following situations in the past week (e.g., "I had fear of the worst happening", "I was nervous") with four possible responses (1= "never" to 4= "most of the time").
A total score was generated by adding together all responses.Greater anxiety was indicated by a higher score on a scale from 1 to 20.

Hopelessness
Two items adapted from Beck, Weissman et al [17] and two from Everson, Kaplan et al [18] (e.g., "I feel it is impossible for me to reach the goals that I would like to strive for.","I don't expect to get what I really want.") on a 6-point Likert scale (1= "strongly disagree" to 6= "strongly agree") were used to evaluate hopelessness.All the responses were summed, resulting in a summary score ranging from 1 to 24 with a higher score indicating greater hopelessness.

Optimism
Optimism was measured using three items (e.g., "I'm always optimistic about my future.","In uncertain times, I usually expect the best.") on a 6-point Likert scale (1= "strongly disagree" to 6= "strongly agree") from the Life Orientation Test-Revised (LOT-R) [19].A summary score ranging from 1 to 18 was computed by totalling the items' responses.A higher score indicated greater optimism.

Pessimism
As with optimism, the remaining three items from the LOT-R (e.g., "If something can go wrong for me it will.","I hardly ever expect things to go my way.") on a 6-point Likert scale (1= "strongly disagree" to 6= "strongly agree") were used to assess pessimism [19] A total score ranging from 1 to 18 was created by adding all the responses.A higher score indicated greater pessimism.

Cynical hostility
The Cook-Medley Hostility Inventory was used to evaluate cynical hostility.Participants' responses to five 6-point-Likert-scale questions (e.g., "Most people dislike putting themselves out to help other people.","Most people will use somewhat unfair means to gain profit or an advantage rather than lose it.")(1= "strongly disagree" to 6= "strongly agree") were added to create a total score [20,21].
A higher score on a possible range of 1-to-30 for a total score indicated greater cynical hostility.

Personal constraints
Participants were asked five items regarding perceptions of constraints over their life circumstances (e.g., "I often feel helpless in dealing with the problems of life.","Other people determine most of what I can and cannot do.").Their answers on a 6-point Likert scale (1= "strongly disagree" to 6= "strongly agree") were added to generate a total score [22,23].A higher level of perceived constraints is indicated by a higher total score on a scale from 1 to 30.

Mastery
Five items on a 6-point Likert scale (1= "strongly disagree" to 6= "strongly agree") regarding beliefs to perform specific behaviours (e.g., "I can do just about anything I really set my mind to.", "When I really want to do something, I usually find a way to succeed at it.")were used to assess the level of perceived mastery [22,23].Participants' responses to all the items were totalled to create a summary score ranging from 1 to 30.A higher score indicates greater perceived mastery.

Main analyses
Causal mediation analysis using the 'med4way' command in STATA version 17 [24] was used to examine psychological well-being related measures as a potential explanation of the association between obesity and risk of developing NCDs.This mediation analysis allows a survival outcome with binary or continuous independent and mediator variables.Separate single mediation models were developed to examine mediation by an index of overall psychological well-being and some psychological well-being related measures in the association between obesity and an NCD, controlling for sociodemographic covariates.Participants who developed an NCD before and up to baseline were excluded from mediation analysis for that NCD.This causal mediation analysis decomposed the total effect into four components: (1) controlled direct effect ("not explained by either the mediator or exposure-mediator interaction; explained only by the exposure"), (2) reference interaction ("explained only by the interaction between the exposure and the mediator"), (3) mediated interaction ("explained by both interaction and mediation or mediated interaction"), and (4) pure indirect effect ("explained only by mediation") [24,25].Both proportions due to mediated interaction and pure indirect effect together constitute the overall proportion due to mediation or the proportion mediated [25].
Mediation analysis using med4way fitted two regression models of the exposure on the mediator and the exposure on the outcome adjusted for and in interaction with the potential mediator.We selected linear regression for a model of obesity on psychological well-being related measure.To estimate the effect of obesity on NCDs, an accelerated failure time (AFT) model with a Weibull distribution was used (e.g., as in Karlsson, Zhan et al [26]) because Cox regression might provide biased estimates in mediation analyses of non-rare outcomes [27,28].As psychological well-being related measures were on a continuous scale, mediation at mean levels of the mediator was tested (e.g., as in Karlsson, Zhan et al [26]).
Across ELSA and HRS, we found that some sociodemographic characteristics (e.g., being older, male, non-White, not currently married, in lower household wealth quintiles) were associated with being excluded from the analysis due to missing observations.To address potential selection bias, we used inverse probability weighting approach [29,30].First, we included sociodemographic covariates in a logistic regression model to estimate the probability of being retained at follow-up or included in the analytic sample.New sample weights were then calculated as the inverse probability of being retained and combined with baseline weights.We applied these combined weights in all the analyses to compensate for the differences in participants' i) non-response characteristics at baseline and ii) characteristics associated with being retained at follow-up.

Additional analyses
We examined whether obesity at baseline was associated with the cumulative number of NCDs (min=0, max=7) reported across the follow-up period, to assess overall disease burden.We assessed this association by controlling the number of NCDs at baseline, and therefore, participants with one or more NCDs at baseline were not excluded to reduce further sample loss.As the outcome was a count variable, Poisson regression was used for this analysis.If we found evidence of this association, we would further examine mediation by psychological well-being related measures using med4way with a count outcome.
Findings were presented as incidence-rate ratio (IRR) along with 95% confidence intervals (CI) and pvalues.
We conducted robustness comparative analysis for ELSA by linking ELSA data with the Admitted Patient Care (APC) data from the National Health Service (NHS) Hospital Episode Statistics (HES) (e.g., as in Bu, Zaninotto et al [31]).This is because the list of NCDs available in ELSA is based on self-report and not exhaustive.Therefore, APC data can provide more accurate information on the diagnosis of some NCDs than self-reports in ELSA.We attempted to replicate the main analysis using information on NCDs reported in APC data.For the analysis predicting an NCD, we excluded participants with that condition at baseline, either self-reported in ELSA or identified from a diagnosis reported in APC data (e.g., as in Bu, Zaninotto et al [31]).Cox-proportional hazard regression models were then used to examine the association between obesity and each NCD from APC data, and causal mediation analysis was used to examine mediation by psychological well-being related measures.
We also conducted additional analyses that were not pre-registered.Psychological well-being related measures (e.g., depressive symptoms) have been shown to be associated with weight changes [32] and this may increase risk of developing NCDs [33].Therefore, we examined the extent to which weight changes between waves (between Waves 4 and 6 in ELSA; and between Waves 9-10 and 11-12 in HRS) explained the prospective associations between psychological well-being related measures at baseline and NCDs (as opposed that psychological well-being related measures explained the associations between obesity and NCDs).For these analyses, the follow-up period for each NCD began from Wave 6 for ELSA and Waves 11-12 for HRS.

Self-reported diagnosis
"Has a doctor ever told you that you have a memory-related disease?"

-OR -Interview
The total scores from responses to 27point modified Telephone Interview for Cognitive Status was used to classify participants into probable dementia (a score of 0-6) [57][58][59].

-OR -
The average scores from responses to the short-form Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) questionnaire (16 questions) was used to classify participants into probable dementia (an average score of ≥ 3.5 on a scale from 1 to 5) [12,60].

-OR -
The average scores from responses to the short-form Informant Questionnaire on Cognitive Decline in the Elderly (IQCODE) questionnaire (16 questions) was used to classify participants into probable dementia (an average score of ≥ 3.5 on a scale from 1 to 5) [12,60].
Compared to a pre-registered study protocol (https://doi.org/10.17605/OSF.IO/JRQAP), we also included an additional assessment of dementia using the short-form IQCODE questionnaire in this study.

Information on planned protocol analyses not reported in the present study
We followed our pre-registered study protocol (https://doi.org/10.17605/OSF.IO/JRQAP) in conducting the analysis.All planned protocol analyses were conducted and reported in the present study, except for additional analyses of metabolic status: i) the associations between obesity and NCD stratified analyses by metabolically health status and ii) metabolically healthy status as an independent variable predicting NCDs.We plan to conduct and then report the analyses focusing on metabolic status in a separate research report given the large number of analyses and different focus of research questions.

Table A1 .
Obesity-related NCDs and supporting studies

Table A2 .
Descriptions of obesity-related NCDs in ELSA and HRS

Table B1 .
Longitudinal associations between Class II & III obesity (vs.normal weight) and the incidence of NCDs in ELSA *p<0.05; **p<0.01;***p<0.001n=analytical sample size (excluding participants with the outcome before and at baseline and missing observations); HR=hazard ratio; CI=confidence interval Model 1: The association between obesity and an NCD was adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).Comparisons between overweight vs. normal weight and between Class I obesity vs. normal weight are not presented.

Table B2 .
Longitudinal associations between psychological well-being related measures and the incidence of NCDs in ELSA *p<0.05; **p<0.01;***p<0.001n=analytical sample size (excluding participants with the outcome before and at baseline and missing observations); HR=hazard ratio; CI=confidence interval Model 1: The association between psychological well-being related measure and an NCD was adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for obesity status.

Table B3 .
Cross-sectional associations between obesity and psychological well-being related measures in ELSA aComparisons between overweight vs. normal weight are not presented.b Comparisons between overweight vs. normal weight and Class I obesity vs. normal weight are not presented.All the associations were adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).

Table B4 .
Mediation by psychological well-being related measures in the longitudinal associations between obesity and the incidence of NCDs in ELSA n=analytical sample size (excluding participants with the outcome before and at baseline and missing observations); Estimate=the overall proportion due to mediation or the proportion mediated; CI=confidence interval All the mediation analyses were adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.As only a binary or continuous independent variable is allowed, causal mediation analysis excluded participants with overweight and Class I obesity for the comparison between Class II & III obesity vs. normal weight, resulting in a smaller analytical sample size compared to previous regression analyses that allowed an ordinal independent variable to be fitted.See Tables 1 (main document), B1, B2, and B3 to determine whether preconditions for successful mediation were met.

Table B5 .
Longitudinal associations between obesity and the cumulative incidence of NCDs in ELSA Comparisons between overweight vs. normal weight and Class I obesity vs. normal weight are not presented.Model 1: The association between psychological well-being related measure and a cumulative number of NCDs at follow-up was adjusted for the number of NCDs at baseline, age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).
a Comparisons between overweight vs. normal weight are not presented.b

Table B6 .
Longitudinal associations between psychological well-being related measures and the cumulative incidence of NCDs in ELSA The association between psychological well-being related measure and a cumulative number of NCDs at follow-up was adjusted for the number of NCDs at baseline, age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for obesity status.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).

Table B7 .
Longitudinal associations between psychological well-being related measures and changes in BMI (between Waves 4 and 6) in ELSA Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).

Table B8 .
Longitudinal associations between changes in BMI (between Waves 4 and 6) and the incidence of NCDs in ELSA

Table B9 .
Longitudinal associations between obesity (vs.normal weight) and the incidence of NCDs reported from healthcare administrative data in ELSA The association between obesity and an NCD was adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).Comparisons between overweight vs. normal weight for each NCD are not presented.

Table B10 .
Longitudinal associations between Class II & III obesity (vs.normal weight) and the incidence of NCDs reported from healthcare administrative data in ELSA The association between obesity and an NCD was adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.
Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).Comparisons between overweight vs. normal weight and between Class I obesity vs. normal weight are not presented.

Table B11 .
Longitudinal associations between an index of overall psychological well-being and the incidence of NCDs reported from healthcare administrative data in ELSA ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for obesity status.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).Comparisons between overweight vs. normal weight for each NCD are not presented.

Table B12 .
Mediation by an index of overall psychological well-being in the longitudinal associations between obesity and the incidence of NCDs reported from healthcare administrative data in ELSA As only a binary or continuous independent variable is allowed, causal mediation analysis excluded participants with overweight for the comparison between obesity vs. normal weight, and participants with overweight and Class I obesity for the comparison between Class II & III obesity vs. normal weight, resulting in a smaller analytical samplesize compared to previous regression analyses that allowed an ordinal independent variable to be fitted.

Table B13 .
Longitudinal associations between psychological well-being related measures and the incidence of NCDs reported from healthcare administrative data in ELSA *p<0.05; **p<0.01;***p<0.001n=analytical sample size (excluding participants with the outcome before and at baseline and missing observations); HR=hazard ratio; CI=confidence interval Model 1: The association between psychological well-being related measure and an NCD was adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for obesity status.

Table B14 .
Mediation by psychological well-being related measures in the longitudinal associations between obesity and the incidence of NCDs reported from healthcare administrative data in ELSA

B9, B10, and B13 to determine whether preconditions for successful mediation were met.Table B15 .
Longitudinal associations between obesity and the cumulative incidence of NCDs reported from healthcare administrative data in ELSA Comparisons between overweight vs. normal weight and Class I obesity vs. normal weight are not presented.Model 1: The association between psychological well-being related measure and a cumulative number of NCDs at follow-up was adjusted for the number of NCDs at baseline, age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, and wealth index.Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness). b

Table B16 .
Longitudinal associations between psychological well-being related measures and the cumulative incidence of NCDs reported from healthcare administrative data in ELSA

Table B17 .
Longitudinal associations between changes in BMI (between Waves 4 and 6) and the incidence of NCDs reported from healthcare administrative data in ELSA

Table B18 .
Interaction analyses between obesity and an index of overall psychological well-being in predicting NCDs in ELSA All the interaction analyses were adjusted for age, sex, ethnicity, marital status, paid employment status, completion of degree qualification level, wealth index, obesity, and index of overall psychological well-being related measures.Model 1: for obesity * index of overall psychological well-being related measures (comparisons between overweight vs. normal weight are not presented) Model 2: for Class II & III obesity * index of overall psychological well-being related measures (comparisons between overweight vs. normal weight and between Class I obesity vs. normal weight are not presented).Findings from the Health and Retirement Study (HRS) Index of overall psychological well-being was developed by re-standardizing the average standardized scores of five psychological well-being related measures (depressive symptoms, enjoyment of life, eudemonic well-being, life satisfaction, and loneliness).

Table C1 .
Longitudinal associations between Class II & III obesity (vs.normal weight) and the incidence of NCDs in HRS The association between obesity and an NCD was adjusted for age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of 10 psychological well-being related measures (depressive symptoms, life satisfaction, loneliness, positive affect, negative affect, purpose in life, anxiety, hopelessness, pessimism, and personal constraint).Comparisons between overweight vs. normal weight and between Class I obesity vs. normal weight are not presented.

Table C2 .
Longitudinal associations between psychological well-being related measures and the incidence of NCDs in HRS The association between psychological well-being related measure and an NCD was adjusted for age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.Model 2: Model 1 with an additional adjustment for obesity status.

Table C3 .
Cross-sectional associations between obesity and psychological well-being related measures in HRS a Comparisons between overweight vs. normal weight are not presented.bComparisonsbetween overweight vs. normal weight and Class I obesity vs. normal weight are not presented.All the associations were adjusted for age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of 10 psychological well-being related measures (depressive symptoms, life satisfaction, loneliness, positive affect, negative affect, purpose in life, anxiety, hopelessness, pessimism, and personal constraint).

Table C4 .
Mediation by psychological well-being related measures in the longitudinal associations between obesity and the incidence of NCDs in HRS Estimate=the overall proportion due to mediation or the proportion mediated; CI=confidence interval; NA=not applicable (e.g., no significant differences in standardised scores of loneliness, purpose in life, and hopelessness between participants with obesity vs. normal weight at p<0.01 -See TableC3) All the mediation analyses were adjusted for age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.As only a binary or continuous independent variable is allowed, causal mediation analysis excluded participants with overweight for the comparison between obesity vs. normal weight, and participants with overweight and Class I obesity for the comparison between Class II & III obesity vs. normal weight, resulting in a smaller analytical sample size compared to previous regression analyses that allowed an ordinal independent variable to be fitted.See Tables 4 (

main document), C1, C2, and C3 to determine whether preconditions for successful mediation were met.Table C5 .
Longitudinal associations between obesity and the cumulative incidence of NCDs in HRS Comparisons between overweight vs. normal weight and Class I obesity vs. normal weight are not presented.Model 1: The association between psychological well-being related measure and a cumulative number of NCDs at follow-up was adjusted for the number of NCDs at baseline, age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.Model 2: Model 1 with an additional adjustment for an index of overall psychological well-being.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of 10 psychological well-being related measures (depressive symptoms, life satisfaction, loneliness, positive affect, negative affect, purpose in life, anxiety, hopelessness, pessimism, and personal constraint). b

Table C6 .
Longitudinal associations between psychological well-being related measures and the cumulative incidence of NCDs in HRS The association between psychological well-being related measure and a cumulative number of NCDs at follow-up was adjusted for the number of NCDs at baseline, age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.Model 2: Model 1 with an additional adjustment for obesity status.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of 10 psychological well-being related measures (depressive symptoms, life satisfaction, loneliness, positive affect, negative affect, purpose in life, anxiety, hopelessness, pessimism, and personal constraint).

Table C7 .
Mediation by psychological well-being related measures in the longitudinal association between obesity and the cumulative incidence of NCDs in HRS Estimate=the overall proportion due to mediation or the proportion mediated; CI=confidence interval; NA=not applicable (e.g., no significant differences in standardised scores of hopelessness and index of overall psychological well-being between participants with obesity vs. normal weight at p<0.01 -See TableC3) All the mediation analyses were adjusted for the number of NCDs at baseline, age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.Index of overall psychological well-being was developed by re-standardizing the average standardized scores of 10 psychological well-being related measures (depressive symptoms, life satisfaction, loneliness, positive affect, negative affect, purpose in life, anxiety, hopelessness, pessimism, and personal constraint).As only a binary or continuous independent variable is allowed, causal mediation analysis excluded participants with overweight and Class I obesity for the comparison between Class II & III obesity vs. normal weight, resulting in a smaller analytical sample size compared to previous regression analyses that allowed an ordinal independent variable to be fitted.See

Tables C3, C5, and C6 to determine whether preconditions for successful mediation were met.Table C9 .
Longitudinal associations between changes in BMI (between Waves 9-10 and 11-12) and the incidence of NCDs in HRS (excluding participants with the outcome before and up to Waves 11-12 and missing observations); HR=hazard ratio; CI=confidence interval All the associations were adjusted for age, sex, ethnicity, marital status, employment status, the number of years in education, and wealth index.

Table C10 .
Interaction analyses between obesity and an index of overall psychological well-being in predicting NCDs in HRS (excluding participants with the outcome before and at baseline and missing observations); All the interaction analyses were adjusted for age, sex, ethnicity, marital status, employment status, the number of years in education, wealth index, obesity, and index of overall psychological well-being related measures.Model 1: for obesity * index of overall psychological well-being related measures (comparisons between overweight vs. normal weight are not presented) Model 2: for Class II & III obesity * index of overall psychological well-being related measures (comparisons between overweight vs. normal weight and between Class I obesity vs. normal weight are not presented).Index of overall psychological well-being was developed by re-standardizing the average standardized scores of 10 psychological well-being related measures (depressive symptoms, life satisfaction, loneliness, positive affect, negative affect, purpose in life, anxiety, hopelessness, pessimism, and personal constraint).