Arthritis and longitudinal trajectories of depressive symptoms and cognition in middle-aged and older adults: a prospective tri-cohort study

Figures

• 

  Fig. 1. Participant selection for analyses of arthritis, depressive symptoms, and cognitive outcomes in prospective ageing cohorts from China, England, and the United States. Legend: Participant selection is shown separately for the CHARLS (Panel A; 2011–2018), the ELSA (Panel B; 2008–2015), and the HRS (Panel C; 2010–2017). The incident depressive-symptom analysis excluded participants with positive depressive-symptom status at baseline and required baseline and follow-up depressive-symptom assessments and valid follow-up time. The incident cognitive-impairment analysis additionally excluded participants meeting the education-stratified cognitive-impairment criterion at baseline or lacking cognitive follow-up or valid cognitive follow-up time

• 

  Fig. 2. Depressive-symptom-free survival according to baseline arthritis status in the China Health and Retirement Longitudinal Study, 2011–2018. Legend: Kaplan–Meier estimates show the probability of remaining free from incident depressive symptoms according to baseline arthritis status. Incident depressive symptoms were defined using cohort-specific Center for Epidemiologic Studies Depression Scale thresholds: a score of ≥ 10 in the China Health and Retirement Longitudinal Study and ≥ 4 in the English Longitudinal Study of Ageing and the Health and Retirement Study. Vertical marks indicate censored observations. The table below each plot shows the number of participants at risk. P values were obtained using the log-rank test

• 

  Fig. 3. Depressive-symptom-free survival according to baseline arthritis status in the English Longitudinal Study of Ageing, 2008–2015. Legend: Kaplan–Meier estimates show the probability of remaining free from incident depressive symptoms according to baseline arthritis status. Incident depressive symptoms were defined using cohort-specific Center for Epidemiologic Studies Depression Scale thresholds: a score of ≥ 10 in the China Health and Retirement Longitudinal Study and ≥ 4 in the English Longitudinal Study of Ageing and the Health and Retirement Study. Vertical marks indicate censored observations. The table below each plot shows the number of participants at risk. P values were obtained using the log-rank test

• 

  Fig. 4. Depressive-symptom-free survival according to baseline arthritis status in the Health and Retirement Study, 2010–2017. Legend: Kaplan–Meier estimates show the probability of remaining free from incident depressive symptoms according to baseline arthritis status. Incident depressive symptoms were defined using cohort-specific Center for Epidemiologic Studies Depression Scale thresholds: a score of ≥ 10 in the China Health and Retirement Longitudinal Study and ≥ 4 in the English Longitudinal Study of Ageing and the Health and Retirement Study. Vertical marks indicate censored observations. The table below each plot shows the number of participants at risk. P values were obtained using the log-rank test

• 

  Fig. 5. Model-predicted trajectories of Center for Epidemiologic Studies Depression Scale scores according to baseline arthritis status in prospective ageing cohorts from China, England, and the United States. Legend: Trajectories were estimated using linear mixed-effects models including follow-up time, baseline arthritis status, their interaction, age, sex, education, marital status, physical activity, current smoking, alcohol consumption, body mass index, hypertension, and diabetes, with participant-specific random intercepts and slopes. Predictions were generated at cohort-specific median age and body mass index and for a female participant with upper-secondary or vocational education who was married or partnered, physically active, a current non-smoker, reported no alcohol consumption, and had no hypertension or diabetes. Colored solid and dashed lines represent model-predicted scores, and shaded bands represent approximate 95% prediction intervals for individual-level values. Prediction intervals were truncated at the theoretical score boundaries for visual presentation. Panel A: China Health and Retirement Longitudinal Study, 2011–2018, using the 10-item scale (range 0–30). Panel…

• 

  Fig. 6. Cognitive-impairment-free survival according to baseline arthritis status in the China Health and Retirement Longitudinal Study, 2011–2018. Legend: Kaplan–Meier estimates show the probability of remaining free from incident cognitive impairment according to baseline arthritis status among participants without cognitive impairment at baseline. Global cognition was calculated as total word recall (range 0–20) plus temporal orientation (range 0–4). Cognitive impairment was defined as an education-stratified z score of ≤ − 1.5 relative to the cohort- and education-specific baseline distribution. Vertical marks indicate censored observations. The table below each plot shows the number of participants at risk. P values were obtained using the log-rank test

• 

  Fig. 7. Cognitive-impairment-free survival according to baseline arthritis status in the English Longitudinal Study of Ageing, 2008–2015. Legend: Kaplan–Meier estimates show the probability of remaining free from incident cognitive impairment according to baseline arthritis status among participants without cognitive impairment at baseline. Global cognition was calculated as total word recall (range 0–20) plus temporal orientation (range 0–4). Cognitive impairment was defined as an education-stratified z score of ≤ − 1.5 relative to the cohort- and education-specific baseline distribution. Vertical marks indicate censored observations. The table below each plot shows the number of participants at risk. P values were obtained using the log-rank test

• 

  Fig. 8. Cognitive-impairment-free survival according to baseline arthritis status in the Health and Retirement Study, 2010–2017. Legend: Kaplan–Meier estimates show the probability of remaining free from incident cognitive impairment according to baseline arthritis status among participants without cognitive impairment at baseline. Global cognition was calculated as total word recall (range 0–20) plus temporal orientation (range 0–4). Cognitive impairment was defined as an education-stratified z score of ≤ − 1.5 relative to the cohort- and education-specific baseline distribution. Vertical marks indicate censored observations. The table below each plot shows the number of participants at risk. P values were obtained using the log-rank test

• 

  Fig. 9. Model-predicted trajectories of global cognitive scores according to baseline arthritis status in prospective ageing cohorts from China, England, and the United States. Legend: Global cognition was calculated as total word recall (range 0–20) plus temporal orientation (range 0–4), yielding a total score range of 0–24, with higher scores indicating better cognitive performance. Models and the prediction covariate profile were the same as those described for Fig. 5. Colored solid and dashed lines represent model-predicted scores, and shaded bands represent approximate 95% prediction intervals for individual-level values. Prediction intervals were truncated at the theoretical score boundaries for visual presentation. Panel A: China Health and Retirement Longitudinal Study, 2011–2018; Panel B: English Longitudinal Study of Ageing, 2008–2015; Panel C: Health and Retirement Study, 2010–2017. Y-axis ranges differ across panels to show the observed prediction ranges and should not be used for direct comparison of absolute score levels

• 

  Fig. 10. Model-predicted trajectories of word-recall and temporal-orientation scores according to baseline arthritis status in prospective ageing cohorts from China, England, and the United States. Legend: Word recall was calculated as the sum of immediate and delayed recall (range 0–20), and temporal orientation was the number of correct responses for year, month, day of month, and day of week (range 0–4). Higher scores indicate better cognitive performance. Models and the prediction covariate profile were the same as those described for Fig. 5. Colored solid and dashed lines represent model-predicted scores, and shaded bands represent approximate 95% prediction intervals for individual-level values. Prediction intervals were truncated at the theoretical score boundaries for visual presentation. Panel A: word recall in the China Health and Retirement Longitudinal Study; Panel B: temporal orientation in the China Health and Retirement Longitudinal Study; Panel C: word recall in the English Longitudinal Study of Ageing; Panel D: temporal orientation in the English Longitudinal Study of Ageing; Panel E: word recall in the Health and Retirement Study; Panel F: temporal orientation in the…

• 

  Fig. 11. Association between baseline arthritis and incident depressive symptoms across prespecified subgroups in the China Health and Retirement Longitudinal Study, 2011–2018. Legend: Hazard ratios compare participants with arthritis with those without arthritis for incident depressive symptoms. Subgroup-specific Cox proportional hazards models were adjusted for age, sex, education, marital status, physical activity, current smoking, alcohol consumption, body mass index, hypertension, and diabetes, excluding the variable used to define the corresponding subgroup. N and event counts represent the complete-case sample actually used in each subgroup-specific model. Squares indicate hazard ratios and horizontal lines indicate 95% confidence intervals. P values shown for each subgroup variable are P values for interaction obtained from likelihood-ratio tests comparing nested models with and without the arthritis-by-subgroup interaction term. Confidence intervals extending beyond the plotted range are truncated graphically, while the complete numerical confidence intervals are retained in the text column. CI, confidence interval; HR, hazard ratio; N, number of participants

• 

  Fig. 12. Association between baseline arthritis and incident depressive symptoms across prespecified subgroups in the English Longitudinal Study of Ageing, 2008–2015. Legend: Hazard ratios compare participants with arthritis with those without arthritis for incident depressive symptoms. Subgroup-specific Cox proportional hazards models were adjusted for age, sex, education, marital status, physical activity, current smoking, alcohol consumption, body mass index, hypertension, and diabetes, excluding the variable used to define the corresponding subgroup. N and event counts represent the complete-case sample actually used in each subgroup-specific model. Squares indicate hazard ratios and horizontal lines indicate 95% confidence intervals. P values shown for each subgroup variable are P values for interaction obtained from likelihood-ratio tests comparing nested models with and without the arthritis-by-subgroup interaction term. Confidence intervals extending beyond the plotted range are truncated graphically, while the complete numerical confidence intervals are retained in the text column. CI, confidence interval; HR, hazard ratio; N, number of participants