Longitudinal trajectories of somatic and cognitive-affective depressive symptoms influence stroke risk across different populations: Three prospective cohort studies

Abstract

BACKGROUND

Depressive symptoms differ from clinical depression. However, the relationship between depressive symptom trajectories and stroke risk across diverse geographic regions remains unclear.

AIM

To address the gap in the existing understanding of the relationship between depressive symptom trajectories and stroke risk, the current study utilized three representative cohorts.

METHODS

In this study, we used three representative cohorts from Asia, Europe, and the Americas: China Health and Retirement Longitudinal Study (CHARLS), English Longitudinal Study of Ageing (ELSA), and Health and Retirement Study (HRS). Depressive symptoms were assessed using the 8-item Center for Epidemiological Studies Depression scale and categorized into somatic and cognitive-affective subtypes. The trajectories of depressive symptoms were monitored over four surveys starting from baseline and classified into five distinct states: persistently low, decreasing, fluctuating, increasing, and consistently high. Self-reported physician diagnoses were used to evaluate the subsequent stroke events. Hazard ratios (HRs) and 95% confidence intervals (95%CIs) were computed using Cox proportional-risk models adjusted for potential confounding factors.

RESULTS

A total of 7990 participants from CHARLS (females: 52.3%, mean age: 63.4 years), 5642 participants from ELSA (females: 56.2%, mean age: 63.7 years), and 12260 participants from HRS (females: 61.4%, mean age: 64.7 years) participated in this study. The median follow-up periods were 5 years for CHARLS, 8 years for ELSA, and 10 years for HRS. In comparison with the persistently low trajectory, consistently high and fluctuating trajectories of total depressive symptoms increased the risk of stroke in all three cohorts (CHARLS: HR = 1.80, 95%CI: 1.36-2.38; ELSA: HR = 1.50, 95%CI: 1.02-2.21; HRS: HR = 1.45, 95%CI: 1.29-1.62 for consistently high; CHARLS: HR = 1.47, 95%CI: 1.14-1.90; ELSA: HR = 1.44, 95%CI: 1.17-1.77; HRS: HR = 1.26, 95%CI: 1.13-1.41 for fluctuating). Increasing trajectories enhanced the risk in the European cohort (ELSA: HR = 1.71, 95%CI: 1.06-2.74), while decreasing trajectories did not increase stroke risk in any cohort. For somatic depressive symptoms, consistently high and fluctuating trajectories increased the risk of stroke across all cohorts (CHARLS: HR = 2.16, 95%CI: 1.67-2.79; ELSA: HR = 1.94, 95%CI: 1.34-2.81; HRS: HR = 1.79, 95%CI: 1.49-2.15 for consistently high; CHARLS: HR = 1.35, 95%CI: 1.20-1.62; ELSA: HR = 1.56, 95%CI: 1.27-1.92; HRS: HR = 1.33, 95%CI: 1.20-1.46 for fluctuating). Increasing trajectories only increased the risk in the European cohort (ELSA: HR = 1.95, 95%CI: 1.11-3.43), while decreasing trajectories did not increase stroke risk in the European and American cohorts. For cognitive-affective depressive symptoms, consistently high and fluctuating trajectories increased the risk in the Asian and European cohorts (CHARLS: HR = 2.06, 95%CI: 1.52-2.81; ELSA: HR = 1.25, 95%CI: 1.02-1.54 for consistently high; CHARLS: HR = 1.63, 95%CI: 1.23-2.16; ELSA: HR = 1.58, 95%CI: 1.11-2.24 for fluctuating). Increasing trajectories increased the risk only in the American cohort (HRS: HR = 14.67, 95%CI: 1.87-114.91).

CONCLUSION

Consistently high and fluctuating trajectories of total and somatic depressive symptoms were associated with an increased risk for stroke across all populations. Consistently high, fluctuating, and increasing trajectories of cognitive-affective symptoms pose a risk for certain populations. These findings highlight the importance of targeted interventions for managing depressive symptoms as potential strategies for stroke prevention, particularly in regions where specific symptom trajectories are prevalent.

Key Words: Stroke; Depressive symptom; Cohort study; Epidemiology; Middle-aged and elderly population

Core Tip: Consistently high and fluctuating trajectories of total and somatic depressive symptoms are associated with an increased risk of stroke across all populations. Consistently high, fluctuating, and increasing trajectories of cognitive-affective symptoms pose a risk for certain populations. These findings highlight the importance of targeted interventions to manage depressive symptoms as a potential strategy for stroke prevention, particularly in regions where specific symptom trajectories are more prevalent.

INTRODUCTION

Stroke is the third-leading cause of age-standardized mortality worldwide after ischemic heart disease and coronavirus disease 2019 and has the highest age-standardized death rate among neurological diseases, posing a severe threat to the health of people worldwide[1,2]. With the progression of population aging and the extension of life expectancy, the overall social and economic impact of stroke is expected to increase[3]. Therefore, developing targeted prevention and treatment strategies on the basis of risk factors for stroke holds much value for reducing the incidence of stroke[4]. Many common risk factors for stroke have been identified, including diabetes and hypertension. However, the risk factors for stroke have not been fully elucidated. Moreover, stroke risk may vary across different populations. Therefore, employing representative cohorts to identify specific stroke risk factors in particular groups would be of great value for targeted stroke prevention.

Depressive symptoms are increasingly being recognized as a key area of focus in the prevention of cardiovascular and cerebrovascular diseases. Unlike clinical depression, depressive symptoms are a set of states that may be related to depression but do not meet the full diagnostic criteria for clinical depression[5]. Several prospective studies have investigated the association between depressive symptoms and the risk of stroke and yielded inconsistent results[6-9]; among these, some studies have failed to confirm that depression increases the risk of stroke[1-3], while others have found a positive correlation between depression and stroke risk[4]. Importantly, these studies primarily explored depressive symptoms at a single time point and did not consider the trajectory of depressive symptoms over time, which may not fully elucidate the role of depressive symptoms in stroke risk[5]. Moreover, depressive symptoms are dynamic, with differences in their changes over time appearing as sustained high, fluctuating, increasing, decreasing, and sustained low patterns[5,6]. However, studies on the relationship between these trajectory patterns and stroke risk are lacking. One study found that trajectories of consistently high, increasing, and fluctuating depressive symptoms increased the risk of stroke in the American population; however, this study did not further investigate the relationship between the trajectories of depressive symptom subgroups, such as somatic depressive symptoms and cognitive-affective symptoms, and stroke risk in different populations[6]. In addition, a prospective study explored the trajectories and associated risk factors of depressive symptoms in the 10 years following stroke, focusing mainly on changes in depressive symptom trajectories after stroke[7].

Prospective studies exploring the trajectories of depressive symptoms and the risk of stroke are still scarce, and further exploration of depressive symptom trajectories in subgroups and their associations with stroke risk has not yet been conducted. Therefore, to clarify the association between overall, somatic, and cognitive-affective depressive symptom trajectories and the risk of stroke in different populations, we utilized three regionally representative prospective cohorts, namely, the China Health and Retirement Longitudinal Study (CHARLS; Asia), the English Longitudinal Study of Ageing (ELSA; Europe), and the Health and Retirement Study (HRS; America), for an 8-year cohort observation to ascertain the trajectories of depressive symptoms and conducted assessments of the risk of stroke events to elucidate the link between somatic and cognitive-affective depressive symptom trajectories and stroke occurrence (Figure 1).

Figure 1 An overview of the study design and results. This study established the trajectories of total, somatic, and cognitive depressive symptoms through an 8-year follow-up of three prospective cohort studies. Subsequently, it explored the relationship between depressive symptom trajectories and the risk of stroke occurrence. The three geographically representative cohorts confirmed that persistently high and fluctuating depressive symptoms increase the risk of stroke, while a reduction in depressive symptoms does not. The increased risk of stroke associated with a sustained increase in depressive symptoms was more pronounced in European and American populations, while further research is needed to confirm this in Asian populations. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study.

MATERIALS AND METHODS

The three regionally representative prospective cohort studies, CHARLS, ELSA, and HRS, all strictly adhered to the ethical principles of the Declaration of Helsinki. CHARLS was approved by the Biomedical Ethics Committee of Peking University, with all participants signing informed consent forms. ELSA received approval from the multi-centre research ethics committee in London, and obtained informed consent from all participants or their families during the study. HRS received ethical approval from the Institute for Social Research and the Survey Research Center at the University of Michigan, with all participants providing their signed informed consent.

Study design and population

This study utilized data from CHARLS, ELSA, and HRS, which are regionally representative prospective cohort studies conducted in China, the United Kingdom, and the United States, respectively[8-10]. The study employed data from the first wave of CHARLS (wave 1, 2011, T1), the third wave of ELSA (wave 3, 2006, T1), and the fourth wave of HRS (wave 4, 1998, T1) as the baseline for the initial research. Subsequently, follow-up assessments were conducted using data from three additional waves: The fourth wave of CHARLS (wave 4, 2018, T4), the sixth wave of ELSA (wave 6, 2012, T4), and the seventh wave of HRS (wave 7, 2004, T4), which were considered the second round of surveys. Data from the four waves, spanning from baseline to the second survey, were used to assess the patterns of depressive symptom trajectories. Additional follow-up surveys were conducted until the end of the study period, namely the fifth wave of the CHARLS (wave 5, 2020, T5), the ninth wave of the ELSA (wave 9, 2018, T5), and the twelfth wave of the HRS (wave 12, 2014, T5), to reveal the incidence of endpoint events.

Figure 2 illustrates the selection process for the study population. The study population initially consisted of 59791 participants, including 18605 from CHARLS, 19802 from ELSA, and 21384 from HRS. Of these, 14762 participants were excluded from the study because of missing age information or because they did not meet the age inclusion criteria. In addition, 11135 participants were excluded because they had missing depressive symptom information in the initial wave or had experienced a stroke at baseline. Consequently, 33827 participants met the inclusion criteria and were included in the baseline depression symptom trajectory study. However, 7935 participants were excluded due to missing depressive symptom or stroke information during follow-up until the second study. Therefore, a final total of 25892 participants were included in the second study, including 7990 participants from CHARLS, 5642 participants from ELSA, and 12260 participants from the HRS.

Figure 2 Flowchart of the study population selection process. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study.

Assessment of total, somatic, and cognitive-affective depressive symptom trajectories

The 8-item modified version of the Center for Epidemiological Studies Depression (CES-D) scale, which has been validated for reliability in assessing depressive symptoms among older adults, was used to evaluate depressive symptoms[11,12]. At baseline and in the second study, each participant was asked whether they had experienced each of the eight symptoms in the past week. Responses to each of the 8 items, with “yes” or “no” answers coded as 1 and 0 respectively (with reverse scoring for positive items), were summed to yield a total score ranging from 0 to 8. A score of 3 or higher is considered to indicate clinically significant depressive symptoms; thus, a threshold of 3 was chosen to distinguish overall depressive symptoms[6,13]. On the basis of the clinical presentation of depressive symptoms, two subtypes were distinguished: Somatic and cognitive-affective. Somatic symptoms mainly include “poor sleep”, “feeling that everything is an effort”, and “inability to continue my life”, reflecting the somatic manifestations of depressive symptoms in older adult populations[14]. Cognitive-affective depressive symptoms primarily include “feeling depressed”, “feeling lonely”, and “feeling sad”, mainly reflecting the core emotional disturbances associated with depressive symptoms[14]. The two subtypes of depressive symptoms, somatic and cognitive-affective, were validated through confirmatory factor analysis, and the two-factor model of depressive symptoms showed sufficient discriminant validity[15]. A CES-D score of 2 or higher in both the somatic and cognitive-affective subgroups indicated significant symptoms of clinical value[16]. Patients clinically diagnosed with depression were excluded from the study.

Depressive symptom trajectories were assessed using the latent transition analysis (LTA) approach. Depressive symptom trajectories, categorized as persistently low, decreasing, fluctuating, increasing, and persistently high, were evaluated on the basis of CES-D score variations across CHARLS (waves 1-4, T1-T4), ELSA (waves 3-6, T1-T4), and HRS (waves 4-7, T1-T4)[5,6] (Supplementary Figures 1 and 2). A persistently low trajectory was identified when depressive symptoms did not escalate at any of the four time points. A decreasing trajectory was marked by high depressive symptoms at the initial time point, followed by reductions in the next three time points or an elevation at the first two time points with a subsequent decrease in the last two. An increasing trajectory was indicated by elevated depressive symptoms at the first time point that continued to rise at all subsequent time points, or showed no initial elevation with a rise at the final two time points. A persistently high trajectory was characterized by sustained high depressive symptoms across all four time points. Fluctuating trajectories included all patterns that did not conform to the aforementioned classifications. This categorization method has also been applied to somatic and cognitive-affective depressive symptom trajectories. For further analysis, participants exhibiting a persistently low trajectory served as the reference group, against which other depressive symptom trajectories were compared.

Stroke outcomes

In the 5^th wave of the CHARLS, the 7^th-9^th waves of the ELSA, and the 8^th-12^th waves of the HRS, the occurrence of first-ever stroke events was assessed on the basis of self-reported physician diagnoses during the follow-up period. In each wave of these three representative prospective cohorts, participants or their family members (for those unable to communicate directly) were asked, “Has a doctor ever told you that you have had a stroke?” If a participant or a proxy family member reported a diagnosis of stroke during the follow-up period, it was considered a stroke event. In the subsequent follow-up wave, the reported stroke status from the previous wave was reconfirmed, and any discrepancies were corrected. Reports of transient ischemic attacks were not coded as strokes and were therefore excluded[17]. Previous data from CHARLS, ELSA, and HRS have demonstrated that self-reported health conditions are highly consistent with medical record data and strongly associated with clinical validation studies[17,18].

Covariates

Drawing on previous research on the risk factors for stroke and the related variables in prospective cohorts, we identified covariates in three key areas. First, self-reported demographic and socioeconomic variables at baseline included age (continuous), sex (male or female), highest level of education (below high school, high school graduate/general educational development, college or 4-year college and above, and others), and marital status (married/partnered, separated/divorced/widowed, unmarried). Second, the health behavior and health status variables at baseline included alcohol consumption status (ever consumed alcohol, never consumed) and smoking status (ever smoked, never smoked). Finally, the self-reported physician-diagnosed health conditions at baseline were hypertension (yes/no), diabetes (yes/no), and heart disease (yes/no). Adjusting for these covariates is crucial for controlling confounding factors when evaluating the influence of depressive symptom trajectories on the risk of stroke.

Statistical analysis

We conducted statistical evaluations for both the baseline and study populations to identify the trajectories of depressive symptoms. Continuous variables were expressed as mean ± SD or median (interquartile range), while categorical variables are represented as frequencies (percentages). Health data missing during follow-up were imputed using data from the previous wave, and multiple imputations by chained equations were employed to fill in the remaining missing values.

To analyze the association between the trajectories of depressive symptoms and the risk of stroke occurrence, we used Cox proportional-hazard regression to calculate hazard ratios (HRs) and their 95% confidence intervals (95%CIs). The time to event was defined as the duration from the start to the first occurrence of stroke or the end of follow-up, with the first occurrence being the primary reference. Follow-up ended with the last completed survey for each participant. Three models were fitted for Cox regression, using participants with persistently low symptoms as the reference. Model 1 was unadjusted, with the depressive symptom trajectory as the sole variable. Model 2 was further adjusted for potential demographic confounders such as age, sex, education level, and marital status. Model 3 expanded the adjustments from model 2 to include health behaviors (smoking, alcohol consumption) and health conditions (hypertension, diabetes, and heart disease)[5,19]. Schoenfeld residuals were used to verify the proportional-hazard assumption of the Cox regression model. The “mice” package in R was employed for multiple imputation by chained equations to handle missing data in covariates, performing five rounds of imputation to generate five datasets, with a maximum of 50 iterations per dataset. All statistical analyses were conducted using R software (version 4.4.2). Two-sided tests were used for statistical testing, and differences were considered statistically significant at P < 0.05.

Sensitivity analysis

To enhance the reliability of our findings, we expanded the stroke event endpoint in the HRS cohort to include waves 13, 14, and 15, thereby extending the duration of follow-up. This allowed us to evaluate the stability of the link between depressive symptom trajectories and the risk of stroke. Additionally, we incorporated body mass index (BMI) as a covariate in Model 3 for both the HRS and ELSA cohorts to mitigate the influence of BMI as a potential confounder. We refined our criteria for identifying depressive symptom trajectories by excluding individuals who reported only extreme symptoms at the conclusion of the trajectory assessment. This ensured a more consistent representation of depressive symptomatology throughout the study period. Furthermore, for a score of 3, we required a minimum change of two points across consecutive evaluations to classify participants as having high depressive symptoms. Constrained by CHARLS data, we did not exclude stroke events that occurred during the trajectories of depressive symptoms. To ensure the robustness of our research findings, we explored the association between depressive symptom trajectories and stroke risk after excluding stroke events that occurred during the depressive symptom trajectories in both the ELSA and HRS. To minimize the influence of health behavioral pathways on our results, we adjusted for health status and behaviors in the second study. Finally, to maintain the integrity of our analysis, we focused on participants with complete data on exposure and outcomes, thus circumventing biases that might arise from incomplete follow-up data.

RESULTS

Baseline characteristics of the initial and second study populations

After applying the inclusion and exclusion criteria, we enrolled 18605 individuals from the CHARLS cohort (average age, 63.4 years; female patients, 52.3%), 19802 participants from the ELSA cohort (average age, 64.5 years; female patients, 54.5%), and 21384 participants from the HRS cohort (average age, 66.5 years; female patients, 58.1%) in our initial research. The detailed baseline characteristics of the participants are presented in Supplementary Table 1. Female participants slightly outnumbered male participants in each cohort, and approximately 70% of the participants were married or had partners. The cohorts showed similar health profiles, with the incidence of hypertension, diabetes, and heart disease being 40%, 10%, and 20%, respectively. Regarding educational attainment, 87% of the CHARLS participants had less than high school education, while over 50% of ELSA and HRS participants had completed high school or higher education. Additionally, the HRS and ELSA cohorts had a considerably higher percentage of smokers and drinkers than the CHARLS cohort.

From an initial pool of 59791 individuals, 25964 were excluded due to incomplete records, lack of initial depression data, or baseline stroke, yielding 33827 individuals for the subsequent analysis. After an 8-year follow-up, 7935 participants were excluded due to missing depression information or stroke events, leaving 25892 participants for the second study. Within the CHARLS cohort, 659 out of 7990 participants experienced their first stroke, while 286 new stroke cases emerged among 5642 participants in the ELSA cohort over 8 years, and 2032 new stroke patients were identified among 12895 participants in the HRS cohort over a decade. The baseline characteristics for the second study, as detailed in Supplementary Table 1, were similar to those in the initial study, with an average age of approximately 65 years (CHARLS: 63.4 years, ELSA: 63.7 years, HRS: 64.7 years), a majority of female participants (CHARLS: 56.2%, ELSA: 57.0%, HRS: 61.4%), and common health issues such as hypertension (CHARLS: 42.0%, ELSA: 38.5%, HRS: 39.8%), diabetes (CHARLS: 14.0%, ELSA: 7.7%, HRS: 10.9%), and heart disease (CHARLS: 22.4%, ELSA: 13.2%, HRS: 15.1%). The cohorts showed variations in smoking rates (CHARLS: 42.9%, ELSA: 60.3%, HRS: 57.7%), alcohol consumption rates (CHARLS: 33.3%, ELSA: 79.9%, HRS: 52.0%), and low educational levels (CHARLS: 63.1%, ELSA: 32.6%, HRS: 21.9%).

Determination of the five types of depression symptom trajectories

Figure 3 and Tables 1 and 2 present the shifts in participant numbers across total, somatic, and cognitive-affective depressive symptom trajectories, as measured by the CES-D scale, within the three prospective cohorts after an 8-year period of observation. In the CHARLS cohort, which encompassed 7990 individuals, the distribution of total depressive symptom trajectories was as follows: Persistently high (1649), increasing (335), fluctuating (4361), decreasing (321), and persistently low (1324). The somatic symptom trajectories were less varied, with participants classified as persistently low (2600), fluctuating (4685), or persistently high (705). Cognitive-affective trajectories exhibited a broader range, including persistently low (1634), persistently high (1144), decreasing (238), fluctuating (4734), and increasing (240) trajectories. Tables 2 and 3 provide a similar breakdown for the ELSA and HRS cohorts, detailing the variations in depressive symptom trajectories.

Figure 3 Sankey diagram of participants, waves, and trajectories in three prospective cohorts. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study.

Table 1 Changes in depressive symptoms during the four waves of follow-up in three prospective cohort studies.

Type	CES-D score	CHARLS				ELSA				HRS
		Wave 1	Wave 2	Wave 3	Wave 4	Wave 3	Wave 4	Wave 5	Wave 6	Wave 4	Wave 5	Wave 6	Wave 7
Depressive_symptom_trajectory	0	1431	1586	1616	1183	2537	2797	2479	2774	30	27	18	20
Depressive_symptom_trajectory	1	1696	2024	1697	1301	1419	1273	1346	1213	199	206	143	194
Depressive_symptom_trajectory	2	1552	1606	1290	994	597	502	655	578	5437	5493	5813	5865
Depressive_symptom_trajectory	3	1448	1273	1074	890	367	356	407	404	3112	3030	2765	2698
Depressive_symptom_trajectory	4	1529	1172	1035	922	227	250	242	244	1632	1567	1529	1519
Depressive_symptom_trajectory	5	1469	1009	866	927	169	194	177	163	965	997	986	964
Depressive_symptom_trajectory	6	1061	738	704	837	131	133	155	109	578	572	634	608
Depressive_symptom_trajectory	7	826	495	552	716	120	92	106	93	241	276	280	283
Depressive_symptom_trajectory	8	249	193	221	220	75	45	75	64	66	92	92	109
Somatic_trajectory	0	3295	3590	3213	2314	2894	3257	2855	3160	6230	6444	6774	6839
Somatic_trajectory	1	3562	3337	2967	2448	1685	1426	1612	1436	3566	3454	3010	2975
Somatic_trajectory	2	2941	2263	1894	1949	600	549	689	605	1676	1551	1570	1537
Somatic_trajectory	3	1463	906	981	1279	460	402	479	433	788	811	906	909
Cognitive_affective_trajectory	0	2329	2557	2521	1969	34	28	38	54	91	63	57	73
Cognitive_affective_trajectory	1	2233	2537	1953	1517	161	176	165	180	434	439	352	393
Cognitive_affective_trajectory	2	2099	1836	1533	1257	4240	4229	4163	4267	8778	8593	8673	8651
Cognitive_affective_trajectory	3	2374	1608	1500	1509	815	825	855	786	1901	1956	1973	1929
Cognitive_affective_trajectory	4	1771	1179	1132	1355	307	290	313	257	810	892	900	913
Cognitive_affective_trajectory	5	455	379	416	383	66	65	77	61	246	317	305	301

The China Health and Retirement Longitudinal Study cohort we used wave 1 as the baseline for the cohort with 4 waves of follow-up (follow-up to wave 4); the English Longitudinal Study of Ageing cohort used wave 3 as the baseline for the cohort with 4 waves of follow-up (follow-up to wave 6); and the Health and Retirement Study cohort used Wave 4 as the baseline for the cohort with 4 waves of follow-up (follow-up to wave 7). CES-D: Center for Epidemiological Studies Depression; CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study.

Table 2 Depressive symptom trajectory information determined from the four waves of follow-up data in three prospective cohort studies.

Type	Term	CHARLS	ELSA	HRS
Overall		7990	5642	12260
Depressive_symptom_trajectory	Consistently low	1324	3394	2405
Depressive_symptom_trajectory	Consistently high	1649	260	7037
Depressive_symptom_trajectory	Decreasing	321	100	36
Depressive_symptom_trajectory	Fluctuating	4361	1751	2688
Depressive_symptom_trajectory	Increasing	335	137	94
Somatic_trajectory	Consistently low	2600	3418	548
Somatic_trajectory	Consistently high	705	251	7180
Somatic_trajectory	Decreasing	0	47	106
Somatic_trajectory	Fluctuating	4685	1838	4256
Somatic_trajectory	Increasing	0	88	167
Cognitive_affective_trajectory	Consistently low	1634	3001	10595
Cognitive_affective_trajectory	Consistently high	1144	2177	13
Cognitive_affective_trajectory	Decreasing	238	93	217
Cognitive_affective_trajectory	Fluctuating	4734	340	1422
Cognitive_affective_trajectory	Increasing	240	31	13

CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study.

Table 3 Hazard ratios between trajectories of total depressive symptoms and the risk of stroke occurrence.

Type	Term	Source	Model	HR	Lower_CI	Upper_CI	SE	Beta	P value
Depressive_symptom_trajectory	Consistently low	Reference
Depressive_symptom_trajectory	Consistently high	CHARLS	1	2.06	1.57	2.71	0.14	5.19	2.08E-07
Depressive_symptom_trajectory	Consistently high	CHARLS	2	2.22	1.68	2.93	0.14	5.61	2.01E-08
Depressive_symptom_trajectory	Consistently high	CHARLS	3	1.80	1.36	2.38	0.14	4.08	4.60E-05
Depressive_symptom_trajectory	Decreasing	CHARLS	1	1.27	0.79	2.05	0.24	0.98	3.27E-01
Depressive_symptom_trajectory	Decreasing	CHARLS	2	1.33	0.82	2.15	0.24	1.17	2.44E-01
Depressive_symptom_trajectory	Decreasing	CHARLS	3	1.12	0.69	1.81	0.25	0.46	6.46E-01
Depressive_symptom_trajectory	Fluctuating	CHARLS	1	1.54	1.20	1.98	0.13	3.34	8.25E-04
Depressive_symptom_trajectory	Fluctuating	CHARLS	2	1.60	1.24	2.06	0.13	3.62	2.95E-04
Depressive_symptom_trajectory	Fluctuating	CHARLS	3	1.47	1.14	1.90	0.13	2.95	3.21E-03
Depressive_symptom_trajectory	Increasing	CHARLS	1	1.39	0.88	2.19	0.23	1.41	1.58E-01
Depressive_symptom_trajectory	Increasing	CHARLS	2	1.43	0.91	2.26	0.23	1.55	1.21E-01
Depressive_symptom_trajectory	Increasing	CHARLS	3	1.26	0.80	1.99	0.23	0.99	3.20E-01
Depressive_symptom_trajectory	Consistently low	Reference
Depressive_symptom_trajectory	Consistently high	ELSA	1	1.96	1.35	2.86	0.19	3.51	4.40E-04
Depressive_symptom_trajectory	Consistently high	ELSA	2	1.70	1.16	2.49	0.20	2.70	7.03E-03
Depressive_symptom_trajectory	Consistently high	ELSA	3	1.50	1.02	2.21	0.20	2.05	4.09E-02
Depressive_symptom_trajectory	Decreasing	ELSA	1	1.26	0.62	2.55	0.36	0.64	5.20E-01
Depressive_symptom_trajectory	Decreasing	ELSA	2	1.00	0.49	2.04	0.36	0.00	9.97E-01
Depressive_symptom_trajectory	Decreasing	ELSA	3	0.88	0.43	1.80	0.36	-0.35	7.25E-01
Depressive_symptom_trajectory	Fluctuating	ELSA	1	1.61	1.32	1.97	0.10	4.69	2.68E-06
Depressive_symptom_trajectory	Fluctuating	ELSA	2	1.51	1.23	1.85	0.10	3.91	9.31E-05
Depressive_symptom_trajectory	Fluctuating	ELSA	3	1.44	1.17	1.77	0.11	3.44	5.90E-04
Depressive_symptom_trajectory	Increasing	ELSA	1	2.28	1.42	3.64	0.24	3.44	5.91E-04
Depressive_symptom_trajectory	Increasing	ELSA	2	1.75	1.09	2.81	0.24	2.33	1.98E-02
Depressive_symptom_trajectory	Increasing	ELSA	3	1.71	1.06	2.74	0.24	2.22	2.63E-02
Depressive_symptom_trajectory	Consistently low	Reference
Depressive_symptom_trajectory	Consistently high	HRS	1	1.77	1.59	1.97	0.05	10.49	9.91E-26
Depressive_symptom_trajectory	Consistently high	HRS	2	1.66	1.49	1.86	0.06	8.92	4.82E-19
Depressive_symptom_trajectory	Consistently high	HRS	3	1.45	1.29	1.62	0.06	6.34	2.36E-10
Depressive_symptom_trajectory	Decreasing	HRS	1	0.86	0.32	2.29	0.50	-0.30	7.62E-01
Depressive_symptom_trajectory	Decreasing	HRS	2	0.75	0.28	2.00	0.50	-0.58	5.61E-01
Depressive_symptom_trajectory	Decreasing	HRS	3	0.67	0.25	1.80	0.50	-0.79	4.28E-01
Depressive_symptom_trajectory	Fluctuating	HRS	1	1.34	1.20	1.50	0.06	5.20	2.04E-07
Depressive_symptom_trajectory	Fluctuating	HRS	2	1.33	1.19	1.49	0.06	4.97	6.58E-07
Depressive_symptom_trajectory	Fluctuating	HRS	3	1.26	1.13	1.41	0.06	4.05	5.19E-05
Depressive_symptom_trajectory	Increasing	HRS	1	1.62	1.03	2.55	0.23	2.07	3.82E-02
Depressive_symptom_trajectory	Increasing	HRS	2	1.39	0.88	2.19	0.23	1.42	1.54E-01
Depressive_symptom_trajectory	Increasing	HRS	3	1.31	0.83	2.06	0.23	1.16	2.45E-01

Model 1 explores the unadjusted risk ratios of depressive symptom trajectories with stroke risk. Model 2 adjusts for age, gender, marital status, and education level in addition to the depressive symptom trajectories and stroke risk ratios. Model 3 further adjusts for smoking, alcohol consumption, hypertension, diabetes, and heart problems in addition to age, gender, marital status, education level, and the risk ratios of depressive symptom trajectories with stroke risk. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study; HR: Hazard ratio; CI: Confidence interval.

Persistently high and fluctuating trajectories of total depressive symptoms increased the incidence of stroke across all populations, while an increasing trajectory increased the risk of stroke in some populations

As depicted in Figure 4 and Table 3, within the CHARLS cohort, when using the persistently low trajectory of total depressive symptoms as the reference point, both the persistently high (model 1: HR = 2.06, 95%CI: 1.57-2.71; model 2: HR = 2.22, 95%CI: 1.68-2.93; model 3: HR = 1.80, 95%CI: 1.36-2.38) and fluctuating (model 1: HR = 1.54, 95%CI: 1.20-1.98; model 2: HR = 1.60, 95%CI: 1.24-2.06; model 3: HR = 1.47, 95%CI: 1.14-1.90) trajectories were associated with an increased incidence of stroke. In the ELSA cohort, the risk of stroke was found to increase with trajectories that were persistently high (model 1: HR = 1.96, 95%CI: 1.35-2.86; model 2: HR = 1.70, 95%CI: 1.16-2.49; model 3: HR = 1.50, 95%CI: 1.02-2.21), increasing (model 1: HR = 2.28, 95%CI: 1.42-3.64; model 2: HR = 1.75, 95%CI: 1.09-2.81; model 3: HR = 1.71, 95%CI: 1.06-2.74), and fluctuating (model 1: HR = 1.61, 95%CI: 1.32-1.97; model 2: HR = 1.51, 95%CI: 1.23-1.85; model 3: HR = 1.44, 95%CI: 1.17-1.77). Similarly, in the HRS cohort, the risk of stroke was elevated for individuals with persistently high (model 1: HR = 1.77, 95%CI: 1.59-1.97; model 2: HR = 1.66, 95%CI: 1.49-1.86; model 3: HR = 1.45, 95%CI: 1.29-1.62), increasing (model 1: HR = 1.62, 95%CI: 1.03-2.55; model 2: HR = 1.39, 95%CI: 0.88-2.19; model 3: HR = 1.31, 95%CI: 0.83-2.06), and fluctuating (model 1: HR = 1.34, 95%CI: 1.20-1.50; model 2: HR = 1.33, 95%CI: 1.19-1.49; model 3: HR = 1.26, 95%CI: 1.13-1.41) trajectories of total depressive symptoms. However, a decreasing trajectory of depressive symptoms did not increase the risk of stroke across the CHARLS, ELSA, and HRS cohorts. An increasing trajectory of depressive symptoms was associated with an increased risk of stroke in European and American populations. However, this was not observed in Asian populations.

Figure 4 Persistently high and fluctuating trajectories somatic depressive symptoms increased the risk of stroke onset across all populations. A: Relationship between total depressive symptom trajectory and stroke risk in representative cohort studies from Asia (China Health and Retirement Longitudinal Study), Europe (English Longitudinal Study of Ageing), and Americas (Health and Retirement Study) regions; B: Relationship between different total, somatic and cognitive-affective depressive symptom trajectories and stroke risk. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study.

Persistently high and fluctuating trajectories somatic depressive symptoms increased the risk of stroke onset across all populations

As depicted in Figure 4 and Tables 4 and 5, our analysis revealed that within the CHARLS cohort, somatic depressive symptom trajectories showed only three states: Persistently high, persistently low, and fluctuating. When using the persistently low trajectory as the reference, both the persistently high (model 1: HR = 2.53, 95%CI: 1.97-3.24; model 2: HR = 2.65, 95%CI: 2.05-3.42; model 3: HR = 2.16, 95%CI: 1.67-2.79) and fluctuating (model 1: HR = 1.43, 95%CI: 1.19-1.72; model 2: HR = 1.47, 95%CI: 1.22-1.77; model 3: HR = 1.35, 95%CI: 1.12-1.62) trajectories were associated with an increased risk of stroke. In the ELSA cohort, with the persistently low trajectory as the reference, the persistently high (model 1: HR = 2.49, 95%CI: 1.74-3.56; model 2: HR = 2.18, 95%CI: 1.51-3.15; model 3: HR = 1.94, 95%CI: 1.34-2.81) and fluctuating (model 1: HR = 1.84, 95%CI: 1.51-2.25; model 2: HR = 1.67, 95%CI: 1.37-2.05; model 3: HR = 1.56, 95%CI: 1.27-1.92) trajectories also increased the incidence of stroke. However, the increasing trajectory raised the risk of stroke in model 1 (HR = 2.60, 95%CI: 1.49-4.56), model 2 (HR = 2.02, 95%CI: 1.15-3.54), and model 3 (HR = 1.95, 95%CI: 1.11-3.43). In the HRS cohort, the persistently high (model 1: HR = 2.25, 95%CI: 1.89-2.68; model 2: HR = 2.22, 95%CI: 1.85-2.66; model 3: HR = 1.79, 95%CI: 1.49-2.15) and fluctuating (model 1: HR = 1.54, 95%CI: 1.40-1.69; model 2: HR = 1.48, 95%CI: 1.34-1.63; model 3: HR = 1.33, 95%CI: 1.20-1.46) trajectories were associated with an increased risk of stroke, while the increasing and decreasing trajectories were not related to stroke risk. In summary, all three prospective cohorts showed that somatic depressive symptom trajectories that are persistently high and fluctuating increase the risk of stroke in all populations, while increasing trajectories increase the risk of stroke in the European population; however, decreasing trajectories were not related to stroke risk (Figure 4A).

Table 4 Hazard ratios between trajectories of somatic depressive symptoms and the risk of stroke occurrence.

Type	Term	Source	Model	HR	Lower_CI	Upper_CI	SE	Beta	P value
Somatic_trajectory	Consistently low	Reference
Somatic_trajectory	Consistently high	CHARLS	1	2.53	1.97	3.24	0.13	7.31	2.76E-13
Somatic_trajectory	Consistently high	CHARLS	2	2.65	2.05	3.42	0.13	7.50	6.62E-14
Somatic_trajectory	Consistently high	CHARLS	3	2.16	1.67	2.79	0.13	5.83	5.57E-09
Somatic_trajectory	Fluctuating	CHARLS	1	1.43	1.19	1.72	0.09	3.79	1.54E-04
Somatic_trajectory	Fluctuating	CHARLS	2	1.47	1.22	1.77	0.10	4.03	5.56E-05
Somatic_trajectory	Fluctuating	CHARLS	3	1.35	1.12	1.62	0.10	3.10	1.92E-03
Somatic_trajectory	Consistently low	Reference
Somatic_trajectory	Consistently high	ELSA	1	2.49	1.74	3.56	0.18	4.98	6.42E-07
Somatic_trajectory	Consistently high	ELSA	2	2.18	1.51	3.15	0.19	4.18	2.95E-05
Somatic_trajectory	Consistently high	ELSA	3	1.94	1.34	2.81	0.19	3.51	4.40E-04
Somatic_trajectory	Decreasing	ELSA	1	2.25	1.00	5.08	0.41	1.96	4.97E-02
Somatic_trajectory	Decreasing	ELSA	2	1.81	0.80	4.10	0.42	1.42	1.57E-01
Somatic_trajectory	Decreasing	ELSA	3	1.64	0.72	3.72	0.42	1.17	2.41E-01
Somatic_trajectory	Fluctuating	ELSA	1	1.84	1.51	2.25	0.10	6.07	1.31E-09
Somatic_trajectory	Fluctuating	ELSA	2	1.67	1.37	2.05	0.10	4.97	6.79E-07
Somatic_trajectory	Fluctuating	ELSA	3	1.56	1.27	1.92	0.10	4.26	2.08E-05
Somatic_trajectory	Increasing	ELSA	1	2.60	1.49	4.56	0.29	3.34	8.29E-04
Somatic_trajectory	Increasing	ELSA	2	2.02	1.15	3.54	0.29	2.44	1.47E-02
Somatic_trajectory	Increasing	ELSA	3	1.95	1.11	3.43	0.29	2.33	2.00E-02
Somatic_trajectory	Consistently low	Reference
Somatic_trajectory	Consistently high	HRS	1	2.25	1.89	2.68	0.09	9.09	9.78E-20
Somatic_trajectory	Consistently high	HRS	2	2.22	1.85	2.66	0.09	8.69	3.56E-18
Somatic_trajectory	Consistently high	HRS	3	1.79	1.49	2.15	0.09	6.22	4.84E-10
Somatic_trajectory	Decreasing	HRS	1	1.44	0.92	2.27	0.23	1.58	1.13E-01
Somatic_trajectory	Decreasing	HRS	2	1.46	0.93	2.30	0.23	1.63	1.03E-01
Somatic_trajectory	Decreasing	HRS	3	1.27	0.80	2.00	0.23	1.02	3.06E-01
Somatic_trajectory	Fluctuating	HRS	1	1.54	1.40	1.69	0.05	8.99	2.37E-19
Somatic_trajectory	Fluctuating	HRS	2	1.48	1.34	1.63	0.05	8.02	1.04E-15
Somatic_trajectory	Fluctuating	HRS	3	1.33	1.20	1.46	0.05	5.64	1.69E-08
Somatic_trajectory	Increasing	HRS	1	1.26	0.86	1.85	0.20	1.19	2.36E-01
Somatic_trajectory	Increasing	HRS	2	1.15	0.79	1.69	0.20	0.72	4.69E-01
Somatic_trajectory	Increasing	HRS	3	1.02	0.70	1.50	0.20	0.12	9.05E-01

Model 1 explores the unadjusted risk ratios of depressive symptom trajectories with stroke risk. Model 2 adjusts for age, gender, marital status, and education level in addition to the depressive symptom trajectories and stroke risk ratios. Model 3 further adjusts for smoking, alcohol consumption, hypertension, diabetes, and heart problems in addition to age, gender, marital status, education level, and the risk ratios of depressive symptom trajectories with stroke risk. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study; HR: Hazard ratio; CI: Confidence interval.

Table 5 Hazard ratios between trajectories of cognitive depressive symptoms and the risk of stroke occurrence.

Type	Term	Source	Model	HR	Lower_CI	Upper_CI	SE	Beta	P value
Cognitive_affective_trajectory	Consistently low	Reference
Cognitive_affective_trajectory	Consistently high	CHARLS	1	2.31	1.71	3.11	0.15	5.46	4.82E-08
Cognitive_affective_trajectory	Consistently high	CHARLS	2	2.53	1.87	3.43	0.16	5.98	2.23E-09
Cognitive_affective_trajectory	Consistently high	CHARLS	3	2.06	1.52	2.81	0.16	4.63	3.73E-06
Cognitive_affective_trajectory	Decreasing	CHARLS	1	1.39	0.80	2.42	0.28	1.16	2.47E-01
Cognitive_affective_trajectory	Decreasing	CHARLS	2	1.45	0.83	2.54	0.28	1.32	1.87E-01
Cognitive_affective_trajectory	Decreasing	CHARLS	3	1.26	0.72	2.20	0.28	0.81	4.16E-01
Cognitive_affective_trajectory	Fluctuating	CHARLS	1	1.70	1.29	2.25	0.14	3.73	1.95E-04
Cognitive_affective_trajectory	Fluctuating	CHARLS	2	1.76	1.33	2.33	0.14	3.95	7.83E-05
Cognitive_affective_trajectory	Fluctuating	CHARLS	3	1.63	1.23	2.16	0.14	3.38	7.22E-04
Cognitive_affective_trajectory	Increasing	CHARLS	1	1.74	1.04	2.89	0.26	2.12	3.41E-02
Cognitive_affective_trajectory	Increasing	CHARLS	2	1.84	1.10	3.07	0.26	2.34	1.94E-02
Cognitive_affective_trajectory	Increasing	CHARLS	3	1.53	0.92	2.55	0.26	1.62	1.05E-01
Cognitive_affective_trajectory	Consistently low	Reference
Cognitive_affective_trajectory	Consistently high	ELSA	1	1.44	1.18	1.75	0.10	3.62	2.95E-04
Cognitive_affective_trajectory	Consistently high	ELSA	2	1.28	1.05	1.57	0.10	2.40	1.63E-02
Cognitive_affective_trajectory	Consistently high	ELSA	3	1.25	1.02	1.54	0.10	2.16	3.07E-02
Cognitive_affective_trajectory	Decreasing	ELSA	1	1.67	0.89	3.15	0.32	1.58	1.13E-01
Cognitive_affective_trajectory	Decreasing	ELSA	2	1.51	0.80	2.86	0.33	1.27	2.06E-01
Cognitive_affective_trajectory	Decreasing	ELSA	3	1.47	0.77	2.78	0.33	1.17	2.41E-01
Cognitive_affective_trajectory	Fluctuating	ELSA	1	1.76	1.24	2.49	0.18	3.18	1.48E-03
Cognitive_affective_trajectory	Fluctuating	ELSA	2	1.63	1.15	2.32	0.18	2.72	6.60E-03
Cognitive_affective_trajectory	Fluctuating	ELSA	3	1.58	1.11	2.24	0.18	2.52	1.16E-02
Cognitive_affective_trajectory	Increasing	ELSA	1	1.53	0.49	4.80	0.58	0.74	4.62E-01
Cognitive_affective_trajectory	Increasing	ELSA	2	1.49	0.48	4.68	0.58	0.69	4.93E-01
Cognitive_affective_trajectory	Increasing	ELSA	3	1.43	0.46	4.49	0.58	0.61	5.40E-01
Cognitive_affective_trajectory	Consistently low	Reference
Cognitive_affective_trajectory	Consistently high	HRS	1	2.15	0.30	15.29	1.00	0.77	4.44E-01
Cognitive_affective_trajectory	Consistently high	HRS	2	2.20	0.31	15.71	1.00	0.79	4.30E-01
Cognitive_affective_trajectory	Consistently high	HRS	3	2.04	0.29	14.55	1.00	0.71	4.76E-01
Cognitive_affective_trajectory	Decreasing	HRS	1	2.89	0.40	21.00	1.01	1.05	2.93E-01
Cognitive_affective_trajectory	Decreasing	HRS	2	2.82	0.39	20.51	1.01	1.02	3.07E-01
Cognitive_affective_trajectory	Decreasing	HRS	3	2.54	0.35	18.50	1.01	0.92	3.57E-01
Cognitive_affective_trajectory	Fluctuating	HRS	1	2.53	0.36	18.05	1.00	0.93	3.54E-01
Cognitive_affective_trajectory	Fluctuating	HRS	2	2.77	0.39	19.81	1.00	1.02	3.10E-01
Cognitive_affective_trajectory	Fluctuating	HRS	3	2.56	0.36	18.28	1.00	0.94	3.49E-01
Cognitive_affective_trajectory	Increasing	HRS	1	16.76	2.15	130.96	1.05	2.69	7.19E-03
Cognitive_affective_trajectory	Increasing	HRS	2	15.92	2.03	124.70	1.05	2.63	8.42E-03
Cognitive_affective_trajectory	Increasing	HRS	3	14.67	1.87	114.91	1.05	2.56	1.05E-02

Model 1 explores the unadjusted risk ratios of depressive symptom trajectories with stroke risk. Model 2 adjusts for age, gender, marital status, and education level in addition to the depressive symptom trajectories and stroke risk ratios. Model 3 further adjusts for smoking, alcohol consumption, hypertension, diabetes, and heart problems in addition to age, gender, marital status, education level, and the risk ratios of depressive symptom trajectories with stroke risk. CHARLS: China Health and Retirement Longitudinal Study; ELSA: English Longitudinal Study of Ageing; HRS: Health and Retirement Study; HR: Hazard ratio; CI: Confidence interval.

Persistently high and fluctuating trajectories of cognitive-affective symptoms increased the risk of stroke in Asian and European populations, but the increasing trajectory increased the risk of stroke in American populations

In the CHARLS cohort, which is representative of the Asian population, cognitive-affective depressive symptom trajectories that were persistently high (model 1: HR = 2.31, 95%CI: 1.71-3.11; model 2: HR = 2.53, 95%CI: 1.87-3.43; model 3: HR = 2.06, 95%CI: 1.52-2.81) or fluctuating (model 1: HR = 1.70, 95%CI: 1.29-2.25; model 2: HR = 1.76, 95%CI: 1.33-2.33; model 3: HR = 1.63, 95%CI: 1.23-2.16) increased the risk of stroke. However, in the increasing trajectory, models 1 (HR = 1.74, 95%CI: 1.04-2.89) and 2 (HR = 1.84, 95%CI: 1.10-3.07) showed an increased risk of stroke, while model 3 (HR = 1.53, 95%CI: 0.92-2.55) did not confirm this relationship. In the ELSA cohort, which was also characterized by persistently high (model 1: HR = 1.44, 95%CI: 1.18-1.75; model 2: HR = 1.28, 95%CI: 1.05-1.57; model 3: HR = 1.25, 95%CI: 1.02-1.54) and fluctuating (model 1: HR = 1.76, 95%CI: 1.24-2.49; model 2: HR = 1.63, 95%CI: 1.15-2.32; model 3: HR = 1.58, 95%CI: 1.11-2.24) trajectories, the risk of stroke was increased among Europeans. Decreasing and increasing trajectories were not associated with stroke occurrence. Conversely, in the HRS cohort, an increasing trajectory of cognitive-affective depressive symptoms promoted stroke occurrence, whereas persistently high and fluctuating trajectories did not increase the risk of stroke in Americans. In summary, persistently high and fluctuating cognitive-affective depressive symptom trajectories elevated the incidence of stroke in Asian and European populations, whereas an increasing trajectory was associated with stroke occurrence in American populations, and a decreasing trajectory did not increase the risk of stroke across Asian, European, and American populations.

Sensitivity analysis confirmed the robustness of the study results

When we conducted stratified analyses based on sex, education, marital status, smoking habits, alcohol use, hypertension, diabetes, and heart disease for the trajectories of total, somatic, and cognitive-affective depressive symptoms, the results remained unaffected (Supplementary Tables 2-10). The associations between overall, cognitive-affective, and somatic depressive symptom trajectories and stroke risk remained consistent with the overall findings across all subgroups. Notably, in the Asian and European populations, only a persistently increasing symptom trajectory was linked to an elevated stroke risk among individuals with comorbid diabetes and cardiac conditions. In contrast, among American populations, the associations aligned with the overall results regardless of diabetes or cardiac comorbidity status. We extended our follow-up of the HRS cohort to waves 13, 14, and 15, and these analyses yielded similar outcomes to our earlier studies (Supplementary Table 11). The results for the CHARLS and ELSA cohorts were also updated using the latest data. Furthermore, incorporating BMI as a covariate in the HRS and ELSA analyses did not significantly change our initial findings (Supplementary Tables 12 and 13). We confirmed the robustness of our results by partially accounting for the dismissal of depressive symptoms (Supplementary Table 14) and excluding all stroke incidences that occurred before the follow-up period in the second study (Supplementary Table 15). The outcomes of the LTA included the Akaike information criterion and Bayesian information criterion (Supplementary Table 16 and Supplementary Figures 1 and 2). On the basis of these indices and trajectory plots, our comprehensive evaluation revealed that five trajectories were optimal for this study. As shown in Supplementary Table 17, we conducted a measurement invariance analysis of CES-D scores within the three datasets. The results indicated that the standard deviation ratios for all three datasets were within the ideal range (0.8-1.2). Specifically, the HRS data showed nearly perfect measurement stability (ratio approximately 1.000), while the ELSA and CHARLS data also demonstrated good measurement stability, with minor fluctuations within the acceptable range. Thus, CES-D measurement invariance in this longitudinal study was stable. This stability ensured that the lack of measurement invariance did not affect our study of the relationship between depressive symptom trajectories and stroke risk.

DISCUSSION

In this study, which included three prospective, regionally representative cohorts, we identified the trajectories of total, somatic, and cognitive-affective depressive symptoms over an 8-year period and explored the association between these depressive symptom trajectories and the risk of stroke incidence. We discovered that in comparison with participants showing persistently low trajectories, those with persistently high and fluctuating trajectories were at an increased risk of stroke, and this result was consistent across all three prospective cohorts. The increasing trajectory was found to increase the risk of stroke among American and European populations, but not among Asian populations. The persistently high and fluctuating trajectories of somatic depressive symptoms elevated the risk of stroke in all three studies. Additionally, for cognitive-affective symptom trajectories, persistently high and fluctuating states increased the risk of stroke among Asian and European populations, whereas an increasing trajectory increased the risk among American populations. Decreasing trajectories of total, somatic, and cognitive-affective depressive symptoms did not affect the risk of stroke. Sensitivity analyses adjusted for potential confounding factors and covariates that could influence the relationship between depressive symptoms and stroke risk yielded robust results, confirming the reliability of our findings.

The trajectory of depressive symptoms may show a state of continuous stability or dynamic change. Prospective cohort studies have elucidated how these trajectories affect the risk of stroke. Subsequently, strategies can be developed to prevent stroke on the basis of the trajectory of depressive symptoms. Our research indicates that consistently high and fluctuating overall depressive symptom trajectories are associated with an increased incidence of stroke, as are increasing trajectories in certain populations. A parallel finding from a prospective American cohort study supports these results, showing that consistently high, fluctuating, and increasing depressive symptom trajectories are associated with stroke occurrence, whereas decreasing trajectories do not affect stroke risk[6]. Additional research combining the HRS and ELSA cohorts revealed that persistently high, fluctuating, and increasing depressive symptom trajectories increased the risk of heart disease[5], which is associated with the mechanisms underlying stroke, further validating our findings. However, in the CHARLS cohort, which represents the Asian population, an increasing trajectory of depressive symptoms was not found to increase stroke risk, possibly because of regional and demographic differences. The lack of statistical significance between decreasing trajectories and stroke risk suggests that individuals who have previously experienced high levels of depressive symptoms may have a reduced risk of stroke if they maintain a reducing symptom trajectory over time. This implies that interventions aimed at depressive symptom trajectories could potentially serve as preventive strategies against stroke risk.

The association between persistently high, fluctuating, and increasing trajectories of depressive symptoms and stroke may be related to the activation of the sympathetic nervous system, dysregulation of the hypothalamic-pituitary-adrenal (HPA) axis, dysfunction of platelet aggregation, and immune inflammation during the exacerbation of depressive symptoms[20]. Depression is positively correlated with the levels of C-reactive protein (CRP), interleukin (IL)-1, and IL-6, which increase the risk of atherosclerosis and may contribute to stroke[21-26]. Both persistently high and fluctuating or increasing trajectories of depressive symptoms have the potential to promoting the release of inflammatory molecules such as CRP and possibly increasing the risk of stroke[27]. Depression is also associated with unhealthy lifestyle factors, such as smoking, lack of exercise, poor diet, and obesity, all of which are potential risk factors for stroke and may contribute to the increased risk associated with worsening depressive symptoms[20,28-31]. Additionally, exacerbation of depressive symptoms can lead to increased use of antidepressant medications, which have a positive correlation with stroke risk and may be influenced by the dosage and duration of medication use[20,32,33]. Furthermore, trajectory-specific mechanisms, wherein the worsening of depressive symptoms can lead to the rapid onset or exacerbation of conditions such as obesity, hypertension, and diabetes, may also increase the risk of stroke[34]. Fluctuating depressive symptoms may also cause pathological changes in the physiological mechanisms related to stroke risk, such as immune dysregulation[35]. However, the specific mechanisms underlying these trajectories require further in-depth investigation.

Additionally, this study found that the three prospective cohorts consistently showed that a trajectory of persistently high and fluctuating somatic depressive symptoms increased the risk of stroke, whereas increasing and decreasing trajectories did not, which may be related to the severity of somatic depressive symptoms. To date, no prospective studies have systematically explored the trajectories of somatic and cognitive-affective depressive symptoms in relation to the risk of stroke. Somatic depressive symptoms, which mainly manifest as sleep disturbances and fatigue, may activate the sympathetic nervous system, increasing blood pressure and heart rate and thereby increasing the risk of stroke[34,36,37]. Moreover, somatic depressive symptoms can affect the HPA axis, influencing cardiovascular health and increasing the risk of stroke[20]. Additionally, they can influence immune regulation and inflammation, accelerating atherosclerosis and further exacerbating the risk of stroke[25,26,38].

Cognitive-affective depressive symptom trajectories showed differences in their association with stroke risk across the three cohorts, with the Asian and European cohorts demonstrating that persistently high and fluctuating trajectories increased stroke risk, and the American cohort showing that an increasing trajectory increased stroke risk. Thus, the effect of cognitive-affective depressive symptom trajectories on stroke risk may be influenced by factors such as region and ethnicity. Ethnocultural differences may influence the strength and direction of the associations between depressive symptom trajectories and stroke risk. For instance, in Asian cultures, individuals may be more inclined to internalize depressive symptoms, which can lead to greater prominence of cognitive-emotional symptoms. This, in turn, may strengthen the association between symptom trajectories and stroke risk. Additionally, certain cultures may have a tendency to express emotional distress through physical symptoms, which can also affect the relationship between different symptom trajectories and stroke risk. Cognitive-affective symptoms, which mainly include low mood and reduced interest, may contribute to the onset of stroke through unhealthy lifestyle factors such as smoking, lack of exercise, and obesity[20,28-30]. Additionally, this relationship may involve changes in brain neurotransmitters, since alterations in the expression of certain neurotransmitters can affect mood regulation and the stress response, thereby influencing stroke risk[39-41].

Our findings hold substantial clinical value. First, the results imply that the trajectory of depressive symptoms should be accounted for in stroke-prevention strategies. Depressive symptom trajectories, including persistently high, fluctuating, and increasing patterns, are crucial for preventing stroke. Persistently high and fluctuating trajectories of somatic depressive symptoms are key targets for intervention. Cognitive-affective depressive symptom trajectories vary in their impact on stroke risk. High and fluctuating trajectories predominantly threaten stroke risk in Asian and European populations, whereas increasing trajectories pose a risk in American populations. These findings are valuable for future targeted stroke-prevention strategies. Moreover, our study found that a decreasing trajectory of depressive symptoms did not increase the risk of stroke; thus, targeted interventions in the trajectories of depressive symptoms is a feasible plan to reduce stroke risk. Additionally, this study was based on three representative prospective cohort studies from different ethnic regions, adhering to strict clinical research designs and including large sample populations for long-term follow-up. The results from the three cohorts showed consistency (generalizability) and differences (population specificity), indicating that targeted measures are required to efficiently intervene for stroke risk associated with different trajectories of depressive symptoms in different populations. Multiple sensitivity analyses were conducted to ensure robustness of the findings. Our research provided a statistical characterization of this phenomenon (persistently high levels and fluctuating trajectories of cognitive-emotional depressive symptoms primarily threaten stroke risk in Asian and European populations, while ascending trajectories primarily threaten stroke risk in the United States). However, research exploration in this area is limited. Therefore, this could be a potential research direction for future cohort studies to further investigate the effects of cognitive-emotional depressive symptom trajectories on stroke risk in different populations, which could significantly contribute to advancing precise stroke prevention.

Our study also had some limitations. First, we excluded participants who lacked CES-D scores and those who experienced new-onset stroke during the exposure period, which limited the generalizability of the results. However, this rigorous standard ensured a causal relationship between depressive symptom trajectories and stroke onset. Additionally, the Asian population included in this study was aged 45 years and older, whereas the American and European populations were aged 50 years and older. Age plays a crucial role in the comparison of the three cohorts. The age ranges of participants in the CHARLS, HRS, and ELSA cohorts were determined using the inclusion criteria set when each cohort was established. Previous comparative studies involving these three cohorts have encountered the same challenge, and a satisfactory solution remains to be identified. However, the average ages of the three cohorts in our study were within a reasonable range. Specifically, as detailed in Supplementary Table 1, the CHARLS, ELSA, and HRS cohorts had average ages of 63.6 years, 64.5 years, and 66.5 years, respectively. Statistical analysis indicated no significant differences in average age among the three cohorts. Thus, the age ranges showed minor differences, and the age trends were comparable across cohorts. Therefore, generalization of our findings to younger populations should be performed with caution. Additionally, due to the limited follow-up data in CHARLS, we did not exclude stroke events that occurred during the trajectories of depressive symptoms. However, we conducted sensitivity analyses by excluding stroke events that occurred during the depressive symptom trajectories from the ELSA and HRS data. The associations between depressive symptom trajectories and stroke risk were consistent with the results of this study, confirming that our results are unaffected and reliable. Moreover, data collection in the three prospective cohort studies used in this research was based on self-reported questionnaires, which may have been subject to recall bias due to the participants' potential inability to remember events accurately. However, we assessed patients with self-reported and newly diagnosed stroke. During the follow-up rounds, the diagnoses were reconfirmed, and the incidence information from cohorts such as ELSA and HRS was cross-checked with medical record systems[42,43], minimizing the impact of recall bias on the results. The strong association between cognitive-affective depressive symptom trajectories and disease risk (HR = 14.67) identified in this study may be related to the relatively small sample size of the increasing cognitive-affective depressive symptom trajectory in the HRS cohort, which could lead to unstable estimates. However, individuals with increasing trajectories were more likely to seek medical care frequently, potentially increasing the detection rate of positive outcomes. Additionally, the lack of antidepressant medication information in the ELSA study limited our ability to account for the influence of antidepressant use, which is a key limitation that may restrict the generalizability of the findings, primarily to associations under natural disease progression. Nevertheless, this study provides scientific evidence for the predictive value of long-term depressive symptom trajectories in untreated states for disease risk, offering insights into the early identification of high-risk populations. Another limitation of this study was the unequal sample sizes in the HRS, CHARLS, and ELSA cohorts, with the HRS cohort being substantially larger. However, because this study focused on stroke events with an incidence of > 5% in all three cohorts, the frequency was sufficient to generate stable estimates. While a larger HRS sample can enhance the precision of 95%CIs, sample size differences alone do not directly introduce bias. Bias typically arises from systematic errors such as measurement or selection bias, and not from sample size disparities. We addressed cross-cohort heterogeneity by standardizing the variable definitions, implementing multivariate adjustments, and employing other relevant methods. To minimize the influence of sample size imbalance on the results, we recommend that future studies combine individual data from multiple cohorts for validation. The three cohorts in our study had different time periods. Our analyses focused on within-cohort patterns rather than cross-cohort comparisons. We also avoided definitive conclusions regarding population differences based solely on these results. Instead, we sought to identify general associations and mechanisms linking depressive symptom trajectories to stroke risk, which may be relevant across populations. Future studies should validate these findings in a more controlled manner across populations, preferably using uniform data-collection protocols and timeframes. Subclinical cerebrovascular lesions can influence depressive symptoms and the risk of stroke. Our analysis may not have fully captured this complex relationship, which is a limitation of this study. To address this, we conducted a sensitivity analysis excluding stroke events within the first 1-2 years of follow-up. This study aimed to reduce reverse causality, where subclinical stroke-related changes may have affect depressive symptoms early in follow-up. The results of the sensitivity analysis aligned with our primary findings, indicating minimal influence of reverse causality. However, we acknowledge that this approach may not entirely rule out bidirectional relationships since subclinical cerebrovascular changes may develop and exert their influence over longer periods. Future studies should use detailed neuroimaging data and adopt long-term longitudinal design. The use of different depression assessment tools such as the CES-D and Patient Health Questionnaire-9 could have also caused measurement bias. To reduce heterogeneity, this study used the CES-D, as reported in the prior literature, across all three cohorts, ensuring methodological consistency with previous studies. Instead of directly comparing the three cohorts, we focused on identifying patterns and trends within each cohort; therefore, our results were unaffected. Nevertheless, future studies should aim to have coordinated assessment protocols to ensure data consistency and comparability across studies. Another limitation of our study was that the study population was primarily from upper-middle-income countries. Additional research is required to apply these findings to low-income groups. Future cohort studies should include populations from lower-middle-income countries to enhance the generalizability of the results. This approach is essential for a more comprehensive understanding of the interaction between depression and stroke risk across racial, ethnic, and socioeconomic groups. In addition, the use of intervals (e.g., biennials) for assessing depressive symptoms may miss short-term fluctuations in severity. This could potentially affect the accuracy of the trajectory classification. However, this did not influence the overall trend of long-term depressive symptom trajectories observed in this study. Future research should consider shortening the assessment intervals. This would better capture the temporal dynamics of depressive symptoms, improve the accuracy of the trajectory classification, and enhance our understanding of the relationship between depressive symptoms and health outcomes. Potential mediators between depressive symptoms and stroke risk were not thoroughly investigated in this study. Due to the constraints of the cohort study data, information on inflammatory factors was not available, precluding a detailed analysis of the mediating effects. Patients with depressive symptoms often have elevated levels of inflammatory factors. These factors may increase the risk of stroke by influencing neurotransmitters and activating the HPA axis, thereby causing vascular endothelial dysfunction. In addition, lack of physical activity can lead to obesity and metabolic syndrome, which may affect the risk of stroke. Future research should delve more deeply into the mediating effects of these factors. A potential limitation of this study is that missing data may have influenced the results. While the missing data in our three prospective cohorts partially satisfied the Missing Completely at Random assumption, we performed multiple imputations on all three cohorts. This step was performed to reduce the effects of missing data on the findings.

Our findings and previous literature highlight the importance of incorporating the identification and management of depressive symptom trajectories into stroke-prevention strategies. Interventions should be tailored to the manifestations of depressive symptoms, focusing on either somatic or cognitive-affective differences in their trajectories. Furthermore, widespread and convenient assessments of depressive symptoms and stroke risk screening, combined with adjustments to treatment plans based on these results, can effectively improve depressive symptoms and correct the trajectories of depressive symptoms, thereby reducing the risk of onset.

CONCLUSION

Fluctuating and persistently high trajectories of total and somatic depressive symptoms can promote the occurrence of stroke events in all populations; fluctuating and persistently high trajectories of cognitive-affective depressive symptoms threatened stroke risk in Asian and European populations; and increasing trajectories of cognitive-affective symptoms increased the risk of stroke in American populations. Additional research is needed to develop precise prevention strategies to delay the progression of depressive symptoms, and targeted interventions are required to correct the trajectories of depressive symptoms and reduce the risk of stroke.

ACKNOWLEDGEMENTS

We thank all workers who contributed to the CHARLS, ELSA, HRS database.