Persistently high and fluctuating trajectories of total and somatic depressive symptoms increase diabetes risk: Two prospective cohort studies

Abstract

BACKGROUND

Depression is a significant risk factor for diabetes, particularly type 2 diabetes. However, depressive symptoms differ from clinical depression. Previous research has not fully considered the relationship between the trajectory of depressive symptoms and the risk of developing diabetes over time.

AIM

To investigate the association between depressive symptoms, their trajectories, and the risk of developing diabetes in two prospective cohort studies.

METHODS

In the first phase we analyzed the association between depressive symptoms and the risk of developing diabetes separately using the Health and Retirement Study (HRS). Depressive symptom trajectories were assessed by examining changes in depressive symptoms at baseline and again 8 years later. We then identified specific depressive symptom trajectories that increased the risk of diabetes in the second phase. Finally, we confirmed the association between depressive symptoms and their trajectories with diabetes risk using the English Longitudinal Study of Ageing (ELSA) as a validation study. Depressive symptom trajectories were categorized into five states based on changes in the modified 8-item Center for Epidemiological Studies-Depression scores: Persistently high; increasing; fluctuating; decreasing; and persistently low. Diabetes mellitus was defined as self-reported, physician-diagnosed diabetes. Cox proportional hazards models were used to assess hazard ratios (HR) and 95% confidence intervals (CI), adjusting for potential confounders.

RESULTS

In the first phase a total of 27658 participants were included (HRS: 18633, ELSA: 9025), among whom 6582 had depressive symptoms (HRS: 4547, ELSA: 2035), 6407 had somatic depressive symptoms (HRS: 4414, ELSA: 1993), and 26415 had cognitive-affective depressive symptoms (HRS: 17755, ELSA: 8660). We found that overall depressive symptoms (HRS: HR = 1.14, 95%CI: 1.07-1.22; ELSA: HR = 1.18, 95%CI: 1.03-1.34) and somatic depressive symptoms (HRS: HR = 1.14, 95%CI: 1.07-1.22; ELSA: HR = 1.25, 95%CI: 1.10-1.42) increased the risk of diabetes, while cognitive depressive symptoms were not associated with diabetes risk. Over an 8-year follow-up we identified 19729 trajectories of overall, somatic, and cognitive-affective depressive symptoms (HRS: 13918, ELSA: 5811). In the second phase we found that persistently high (HRS: HR = 1.22, 95%CI: 1.06-1.40, ELSA: HR = 1.54, 95%CI: 1.16-2.05 in total and HRS: HR = 1.24, 95%CI: 1.07-1.43, ELSA: HR = 1.79, 95%CI: 1.36-2.35 in somatic) and fluctuating (HRS: HR = 1.09, 95%CI: 1.01-1.17, ELSA: HR = 1.33, 95%CI: 1.14-1.55 in total and HRS: HR = 1.10, 95%CI: 1.02-1.18, ELSA: HR = 1.31, 95%CI: 1.13-1.53 in somatic) trajectories of overall and somatic depressive symptoms increased the risk of diabetes, while increasing trajectories may also raise diabetes risk. However, decreasing trajectories were not associated with diabetes risk. Cognitive-affective depressive symptoms showed no association with diabetes risk regardless of trajectory changes. Sensitivity analyses confirmed the reliability of the findings.

CONCLUSION

Persistently high and fluctuating trajectories of overall and somatic depressive symptoms increased the risk of diabetes, while decreasing trajectories were not associated with diabetes risk. In contrast trajectories of cognitive-affective depressive symptoms show no relationship with diabetes risk. Focusing on depressive symptom trajectories, particularly those of somatic depressive symptoms, represented a viable strategy for future diabetes prevention.

Key Words: Depressive symptom; Trajectories; Diabetes; Cohort study; Epidemiology

Core Tip: Overall and somatic depressive symptoms increased the risk of developing diabetes, while cognitive-affective depressive symptoms did not. Furthermore, persistently high and fluctuating trajectories of overall and somatic depressive symptoms were associated with an increased risk of diabetes, whereas trajectories of cognitive-affective depressive symptoms were not.

INTRODUCTION

Diabetes is one of the leading causes of death and disability worldwide. As of 2021 there were 529 million people with diabetes globally, and the disease ranked high in terms of years lived with disability[1-3]. By 2050 it is projected that there will be over 1.31 billion people with diabetes, leading to a significant increase in age-standardized prevalence[1]. This will place a heavy economic burden on patients and their families. Early detection and management of diabetes can greatly prevent and even reverse its onset. This is a crucial strategy for addressing the significant challenge of preventing and reducing the impact of diabetes in the future[1].

Depressive symptoms differ from clinical depression. Depressive symptoms are one of the early significant risk factors for diabetes. These symptoms represent the early stages of depression and are characterized by low mood, reduced interest, sleep disturbances, and other somatic and cognitive-emotional depressive symptoms. Previous studies have confirmed that clinically significant depression can increase the risk of developing diabetes by 65%[4]. Meta-analyses of longitudinal studies have found that depression can increase the risk of type 2 diabetes by 40%-60%[5,6]. However, depressive symptoms can increase the risk of diabetes even before they develop into full-blown depression, and this effect is independent of various confounding factors[7-9]. In individuals over the age of 50, depressive symptoms are associated with an increased risk of developing diabetes [hazard ratio (HR) = 1.62, 95% confidence interval (CI): 1.15-2.29][10]. Currently, it remains unclear how somatic and cognitive-emotional depressive symptoms are related to diabetes risk. The impact of different trajectories of total, cognitive-emotional, and somatic depressive symptoms on diabetes risk is also not well understood. Revealing the relationship between these symptoms and their trajectories and the risk of diabetes will make a significant contribution to the efficient prevention of diabetes in the future.

Building on the aforementioned background, this study aimed to explore the relationship between total, somatic, and cognitive-emotional depressive symptoms and their trajectories with the risk of diabetes through two prospective cohort studies. We aimed to reveal the causal relationship between depressive symptoms and their trajectories and the risk of diabetes. Subsequent monitoring of changes in depressive symptoms will allow for timely intervention in their trajectories, thereby reducing the risk of diabetes. This approach will facilitate early prevention of diabetes, significantly reducing the years of disability and economic losses caused by the disease (Figure 1).

Figure 1 An overview of the study design and results. In the first phase of this study, a prospective cohort analysis was conducted to examine the association between total, somatic, and cognitive-affective depressive symptoms and the risk of diabetes onset. It was found that both total and somatic depressive symptoms increased the risk of diabetes. In the second phase the trajectory of depressive symptoms was determined from baseline to the second study, followed by a follow-up of diabetes occurrence. The English Longitudinal Study of Ageing cohort validated the aforementioned findings, thereby confirming the reproducibility of our study. The results showed that persistently high and fluctuating total and somatic depressive symptom trajectories promoted the development of diabetes. This study provided a new scientific basis for subsequent targeted interventions based on depressive symptoms and their trajectories to reduce the risk of diabetes. This figure was created by BioRender.com. HRS: Health and Retirement Study; ELSA: English Longitudinal Study of Ageing; HR: Hazard ratio; CI: Confidence interval; USA: United States of America; UK: United Kingdom.

MATERIALS AND METHODS

Study design and population

The Health and Retirement Study (HRS) (as the discovery study) and the English Longitudinal Study of Ageing (ELSA) cohorts (as the validation study) were used. They are prospective, nationally representative cohorts of middle-aged and older adults (aged 50 and above) in the United States and the United Kingdom, respectively[11-14]. As previously reference described the two-stage design involving two mutually validating cohorts was employed. In the HRS we used Wave 4 from 1998 as the baseline data, following up for 6 years (Wave 5 in 2000 and Wave 6 in 2002) until Wave 7 in 2004 (second study) to analyze the trajectory of depressive symptoms. From Wave 7 (second study) to Wave 12 (end study), we examined the association between depressive symptom trajectories and diabetes events. For ELSA (as the validation study), Wave 3 in 2006 was chosen as the baseline, with follow-up for 6 years (Wave 4 in 2008 and Wave 5 in 2010) until Wave 6 in 2012 (second study), serving as the exposure phase to analyze depressive symptom trajectories. From Wave 6 to Wave 9 (end study), diabetes events were observed to validate the association between depressive symptom trajectories and diabetes events found in the HRS. Additionally, we explored the relationship between total, somatic, and cognitive-emotional depressive symptoms and diabetes events using data from Wave 4 to Wave 12 in the HRS and from Wave 3 to Wave 9 in ELSA (Figures 1 and 2).

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

The investigation was conducted in two phases in HRS (discovery study) and ELSA (validation study). In the first phase we analyzed the association between total, somatic, and cognitive-emotional depressive symptoms and the risk of diabetes from baseline to the end study. In the second phase we examined the changes in depressive symptom trajectories over an 8-year follow-up period from the first phase (baseline to the second study). Finally, we analyzed the association between depressive symptom trajectories identified in the second phase (from the second study to the end of the study) and diabetes events occurring at the end of the follow-up.

The inclusion criteria for this study were as follows: (1) Age of 50 years or older; (2) Had complete depressive symptom information for three or more of the four follow-up visits in the exposure trajectory; and (3) Had complete diabetes information during the follow-up period. Patients who developed diabetes during the depressive symptom trajectory follow-up period were excluded. The cohorts used in this study, HRS and ELSA, were conducted in strict accordance with the principles of the Declaration of Helsinki. The HRS cohort received approval from the Institute for Social Research and the Survey Research Center at the University of Michigan, while the ELSA cohort was approved by the London Multicentre Research Ethics Committee. All participants provided written informed consent, either personally or through a legal guardian. Therefore, the repetitive ethical approval and informed consent were not conducted again in this study.

Depressive symptoms and their trajectory assessment

In this study depressive symptoms and their trajectories were assessed using the validated modified 8-item Center for Epidemiological Studies-Depression (CESD) scale[15,16]. Data are collected by asking participants in the HRS and ELSA cohorts about their experiences with each item of the modified 8-item CESD over the past week. The total score was calculated by summing the number of “yes” responses for each of the eight items, with two positive items being reverse-scored, resulting in a score range of 0 to 8. The total depressive symptoms were evaluated using a CESD score of 3 or higher, as reported in previous literature[14,17,18]. The two subtypes of depressive symptoms, cognitive-emotional and somatic depressive symptoms[19], were identified through core emotional symptoms such as “feeling depressed”, “feeling lonely”, and “feeling sad” and key somatic symptoms such as “restless sleep” in the modified 8-item scale[14]. The assessment criteria for cognitive-emotional and somatic depressive symptoms were set at a CESD score of 2 or higher[14,20].

The evaluation of depression symptom trajectories was based on determining the relationship between depression symptoms and diabetes risk and used methods from previous studies. The trajectories of depressive symptoms (persistently low, decreasing, fluctuating, increasing, and persistently high) were assessed based on the changes from baseline to the second study as described earlier[14,17]. The depression symptom trajectory model was previously reported and used in multiple studies[14,17].

A persistently low trajectory was defined as having low depressive symptoms (not meeting the diagnostic criteria for depression) at any of the four time points from baseline to the second study. A decreasing trajectory was characterized by elevated depressive symptoms at the first time point followed by a decrease at the subsequent three time points or by elevated symptoms at the first two time points and a decrease at the next two time points. An increasing trajectory was identified when symptoms did not increase at the first time point but increased at all subsequent time points or when they did not increase at the first two time points but increased at the later time points. Persistently high was defined as having elevated depressive symptoms at all four time points. A fluctuating trajectory was defined as a fluctuation in the trajectory of depressive symptoms, such as an increase followed by a decrease or a decrease followed by an increase, but the change was different from the previous four trajectories (Supplementary Table 1). Both somatic and cognitive-emotional depressive symptom trajectories were distinguished using this method. In subsequent analyses participants with a persistently low depressive symptom trajectory were considered the reference group for other trajectories.

Diabetes outcomes

In the HRS and ELSA cohorts, diabetes diagnosis during follow-up was based on self-reported and physician-diagnosed events with participants asked, “Has a doctor ever told you that you have diabetes?”. A diabetes event was recorded if reported during follow-up, and subsequent follow-ups would reconfirm the presence of diabetes. Only participants that confirmed diabetes in two or more follow-ups were considered as diabetic. Furthermore, the diabetes diagnosis was also verified using multiple criteria, including the use of anti-diabetic medications (such as insulin and oral hypoglycemic agents), fasting blood glucose levels ≥ 7.0 mmol/L, and glycated hemoglobin levels ≥ 6.5%. The self-reported health status in the HRS and ELSA data were matched with medical records, ensuring good reliability[14,21].

Covariates

Based on previous studies reporting potential confounding factors affecting diabetes risk, we identified three categories of covariates. First, demographic variables included age (50 years or older), sex (male or female), education level (less than high school, high school, college or above, other), and marital status (married/partnered, separated/divorced/widowed, single). Health behaviors and conditions included alcohol consumption (ever drank and never drank), smoking status (ever smoked and never smoked), and body mass index (BMI). Lastly, self-reported and physician-diagnosed conditions such as hypertension and heart disease were considered. For hypertension the diagnosis was confirmed by a combination of antihypertensive medication use, measured systolic and diastolic blood pressure readings, and self-reported physician-diagnosed hypertension history. Using these covariates in subsequent analyses reduced the impact of confounding factors, thereby enhancing the robustness of the study results.

Statistical analysis

At the baseline stage we statistically described the baseline information of the study population based on the presence or absence of total, cognitive-emotional, and somatic depressive symptoms. At the second study stage, we statistically described the baseline demographic information of the study population based on different trajectories of depressive symptoms. For both the baseline and second study stages, continuous variables were represented by means or medians, while categorical variables were expressed as frequencies or percentages.

We used the Cox proportional hazards model to assess the HRs and 95%CIs for the association between different trajectories of depressive symptoms and the risk of developing diabetes, analyzing how the relative risk changes over time. The event time was defined as the duration from baseline to the first occurrence of diabetes or the end of follow-up, which concluded when each participant completed their last survey. We designated the persistently low depressive symptom trajectory as the reference group and fitted three Cox regression models to evaluate the association between depressive symptom trajectories and diabetes risk. Model 1 was an unadjusted crude model that directly assessed the relationship between depressive symptom trajectories and diabetes risk. Model 2 adjusted for demographic factors such as age, sex, education level, and marital status, while Model 3 further adjusted for health behaviors like smoking and alcohol consumption as well as health conditions such as BMI, hypertension, and heart disease in addition to the factors in Model 2. The assumptions of these models were verified using Schoenfeld residuals. Missing data for covariates were imputed using the “mice” package in R, with a maximum of 50 iterations. The first dataset after five imputations was used as the complete dataset for subsequent analyses. For sensitivity analysis, we conducted detailed stratified analyses on data from both the HRS and ELSA cohorts. Additionally, we extended the follow-up of the HRS to Waves 13, 14, and 15 to increase the number of follow-up years. All statistical analyses were performed using R (version 4.4.1). Statistical significance was determined using two-sided tests, with differences considered significant at P < 0.05.

RESULTS

Basic characteristics of the baseline population

As shown in Table 1, the HRS study included a total of 18633 participants among whom there were 4547 patients with overall depressive symptoms, 4414 patients with somatic depressive symptoms, and 17755 patients with cognitive-affective depressive symptoms. The median age of the participants was 66.7 years, with 59.1% (11015) being female. The proportion of participants with a high school education or above was 73.2%, and 12433 (66.7%) were married. Regarding lifestyle, there were 10943 (58.7%) smokers and 9023 (48.4%) drinkers. In terms of health conditions, 8067 (43.3%) had hypertension and 3671 (19.7%) had heart disease. The distribution of these factors among patients with overall, cognitive-affective, and somatic depressive symptoms was relatively close to the overall distribution.

Table 1 Baseline characteristics of the participants.

Source	Variable	Overall	Total depressive symptom			Somatic depressive symptom				Cognitive-affective depressive symptom
			Yes	No	P value	Yes	No	Unknown	P value	Yes	No	Unknown	P value
HRS	Number	18633	4547	14086		4414	14191	28		17755	840	38
	Age, mean (SD)	66.7 (10.2)	67.9 (10.9)	66.3 (9.9)	< 0.001	67.7 (10.7)	66.4 (10.0)	71.5 (13.3)	< 0.001	66.8 (10.2)	65.6 (10.8)	67.6 (11.6)	0.002
	BMI, mean (SD)	26.9 (5.2)	27.4 (6.0)	26.8 (4.9)	< 0.001	27.7 (6.1)	26.7 (4.9)	25.8 (4.7)	< 0.001	26.9 (5.2)	26.9 (5.3)	26.0 (4.6)	0.553
	Sex, n (%)
	Female	11015 (59.1)	3050 (67.1)	7965 (56.5)	< 0.001	2871 (65.0)	8129 (57.3)	15 (53.6)	< 0.001	10533 (59.3)	458 (54.5)	24 (63.2)	0.019
	Male	7618 (40.9)	1497 (32.9)	6121 (43.5)		1543 (35.0)	6062 (42.7)	13 (46.4)		7222 (40.7)	382 (45.5)	14 (36.8)
	Education, n (%)
	Below high school	4990 (26.8)	1884 (41.4)	3106 (22.1)	< 0.001	1833 (41.5)	3144 (22.2)	13 (46.4)	< 0.001	4775 (26.9)	201 (23.9)	14 (36.8)	< 0.001
	High school	6704 (36.0)	1561 (34.3)	5143 (36.5)		1555 (35.2)	5142 (36.2)	7 (25.0)		6416 (36.1)	279 (33.2)	9 (23.7)
	College or above	6937 (37.2)	1101 (24.2)	5836 (41.4)		1026 (23.2)	5903 (41.6)	8 (28.6)		6563 (37.0)	359 (42.7)	15 (39.5)
	Other	2 (0.0)	1 (0.0)	1 (0.0)		0 (0.0)	2 (0.0)	0 (0.0)		1 (0.0)	1 (0.1)	0 (0.0)
	Marital status, n (%)
	Married or partnered	12433 (66.7)	2425 (53.3)	10008 (71.0)	< 0.001	2628 (59.5)	9787 (69.0)	18 (64.3)	< 0.001	11860 (66.8)	553 (65.8)	20 (52.6)	0.005
	Never married	558 (3.0)	151 (3.3)	407 (2.9)		139 (3.1)	418 (2.9)	1 (3.6)		520 (2.9)	36 (4.3)	2 (5.3)
	Separated/divorced/widowed	5620 (30.2)	1963 (43.2)	3657 (26.0)		1641 (37.2)	3970 (28.0)	9 (32.1)		5357 (30.2)	247 (29.4)	16 (42.1)
	Unknown	22 (0.1)	8 (0.2)	14 (0.1)		6 (0.1)	16 (0.1)	0 (0.0)		18 (0.1)	4 (0.5)	0 (0.0)
	Smoking status, n (%)
	Ever smokers	10943 (58.7)	2745 (60.4)	8198 (58.2)	0.028	2687 (60.9)	8241 (58.1)	15 (53.6)	0.021	10398 (58.6)	527 (62.7)	18 (47.4)	0.024
	Never smokers	7567 (40.6)	1770 (38.9)	5797 (41.2)		1698 (38.5)	5856 (41.3)	13 (46.4)		7243 (40.8)	304 (36.2)	20 (52.6)
	Unknown	123 (0.7)	32 (0.7)	91 (0.6)		29 (0.7)	94 (0.7)	0 (0.0)		114 (0.6)	9 (1.1)	0 (0.0)
	Drinking status, n (%)
	Ever drinkers	9023 (48.4)	1713 (37.7)	7310 (51.9)	< 0.001	1575 (35.7)	7438 (52.4)	10 (35.7)	< 0.001	8561 (48.2)	442 (52.6)	20 (52.6)	0.163
	Never drinkers	9609 (51.6)	2833 (62.3)	6776 (48.1)		2838 (64.3)	6753 (47.6)	18 (64.3)		9193 (51.8)	398 (47.4)	18 (47.4)
	Unknown	1 (0.0)	1 (0.0)	0 (0.0)		1 (0.0)	0 (0.0)	0 (0.0)		1 (0.0)	0 (0.0)	0 (0.0)
	Hypertension, n (%)
	Yes	8067 (43.3)	2316 (50.9)	5751 (40.8)	< 0.001	2328 (52.7)	5730 (40.4)	9 (32.1)	< 0.001	7696 (43.3)	355 (42.3)	16 (42.1)	0.816
	No	10566 (56.7)	2231 (49.1)	8335 (59.2)		2086 (47.3)	8461 (59.6)	19 (67.9)		10059 (56.7)	485 (57.7)	22 (57.9)
	Heart problem, n (%)
	Yes	3671 (19.7)	1251 (27.5)	2420 (17.2)	< 0.001	1304 (29.5)	2361 (16.6)	6 (21.4)	< 0.001	3480 (19.6)	182 (21.7)	9 (23.7)	0.28
	No	14962 (80.3)	3296 (72.5)	11666 (82.8)		3110 (70.5)	11830 (83.4)	22 (78.6)		14275 (80.4)	658 (78.3)	29 (76.3)
ELSA
	Number	9052	2035	7017		1993	7045	14		8660	344	48
	Age, mean (SD)	65.1 (10.4)	66.5 (11.2)	64.7 (10.1)	< 0.001	66.6 (11.2)	64.7 (10.1)	73.3 (9.7)	< 0.001	65.1 (10.4)	64.6 (11.5)	69.8 (12.2)	0.006
	BMI, mean (SD)	28.0 (4.9)	28.4 (5.4)	27.8 (4.7)	< 0.001	28.9 (5.5)	27.7 (4.6)	27.7 (5.2)	< 0.001	28.0 (4.9)	28.1 (5.1)	26.9 (4.2)	0.428
	Sex, n (%)
	Female	4985 (55.1)	1319 (64.8)	3666 (52.2)	< 0.001	1268 (63.6)	3710 (52.7)	7 (50.0)	< 0.001	4784 (55.2)	169 (49.1)	32 (66.7)	0.022
	Male	4067 (44.9)	716 (35.2)	3351 (47.8)		725 (36.4)	3335 (47.3)	7 (50.0)		3876 (44.8)	175 (50.9)	16 (33.3)
	Education, n (%)
	Below high school	3453 (38.1)	1049 (51.5)	2404 (34.3)	< 0.001	1044 (52.4)	2401 (34.1)	8 (57.1)	< 0.001	3322 (38.4)	109 (31.7)	22 (45.8)	0.047
	High school	1683 (18.6)	336 (16.5)	1347 (19.2)		320 (16.1)	1362 (19.3)	1 (7.1)		1611 (18.6)	67 (19.5)	5 (10.4)
	College or above	3180 (35.1)	485 (23.8)	2695 (38.4)		461 (23.1)	2716 (38.6)	3 (21.4)		3019 (34.9)	145 (42.2)	16 (33.3)
	Other	736 (8.1)	165 (8.1)	571 (8.1)		168 (8.4)	566 (8.0)	2 (14.3)		708 (8.2)	23 (6.7)	5 (10.4)
	Marital status, n (%)
	Married or partnered	5994 (66.2)	1009 (49.6)	4985 (71.0)	< 0.001	1072 (53.8)	4914 (69.8)	8 (57.1)	< 0.001	5768 (66.6)	204 (59.3)	22 (45.8)	< 0.001
	Never married	453 (5.0)	136 (6.7)	317 (4.5)		117 (5.9)	333 (4.7)	3 (21.4)		412 (4.8)	37 (10.8)	4 (8.3)
	Separated/divorced/widowed	2304 (25.5)	842 (41.4)	1462 (20.8)		754 (37.8)	1548 (22.0)	2 (14.3)		2190 (25.3)	94 (27.3)	20 (41.7)
	NA	301 (3.3)	48 (2.4)	253 (3.6)		50 (2.5)	250 (3.5)	1 (7.1)		290 (3.3)	9 (2.6)	2 (4.2)
	Smoking status, n (%)
	Ever smokers	5608 (62.0)	1365 (67.1)	4243 (60.5)	< 0.001	1342 (67.3)	4257 (60.4)	9 (64.3)	< 0.001	5365 (62.0)	213 (61.9)	30 (62.5)	0.996
	Never smokers	3440 (38.0)	670 (32.9)	2770 (39.5)		651 (32.7)	2784 (39.5)	5 (35.7)		3291 (38.0)	131 (38.1)	18 (37.5)
	NA	4 (0.0)	0 (0.0)	4 (0.1)		0 (0.0)	4 (0.1)	0 (0.0)		4 (0.0)	0 (0.0)	0 (0.0)
	Drinking status, n (%)
	Ever drinkers	6722 (74.3)	1266 (62.2)	5456 (77.8)	< 0.001	1261 (63.3)	5457 (77.5)	4 (28.6)	< 0.001	6451 (74.5)	248 (72.1)	23 (47.9)	< 0.001
	Never drinkers	848 (9.4)	294 (14.4)	554 (7.9)		301 (15.1)	545 (7.7)	2 (14.3)		798 (9.2)	45 (13.1)	5 (10.4)
	NA	1482 (16.4)	475 (23.3)	1007 (14.4)		431 (21.6)	1043 (14.8)	8 (57.1)		1411 (16.3)	51 (14.8)	20 (41.7)
	Hypertension, n (%)
	Yes	3758 (41.5)	961 (47.2)	2797 (39.9)	< 0.001	993 (49.8)	2757 (39.1)	8 (57.1)	< 0.001	3599 (41.6)	141 (41.0)	18 (37.5)	0.833
	No	5294 (58.5)	1074 (52.8)	4220 (60.1)		1000 (50.2)	4288 (60.9)	6 (42.9)		5061 (58.4)	203 (59.0)	30 (62.5)
	Heart problem, n (%)
	Yes	1508 (16.7)	468 (23.0)	1040 (14.8)	< 0.001	489 (24.5)	1015 (14.4)	4 (28.6)	< 0.001	1439 (16.6)	56 (16.3)	13 (27.1)	0.149
	No	7544 (83.3)	1567 (77.0)	5977 (85.2)		1504 (75.5)	6030 (85.6)	10 (71.4)		7221 (83.4)	288 (83.7)	35 (72.9)

HRS: Health and Retirement Study; ELSA: English Longitudinal Study of Ageing; BMI: Body mass index; SD: Standard deviation; NA: Not available.

In the ELSA cohort a total of 9052 participants were included among whom there were 2035 patients with overall depressive symptoms, 1993 patients with somatic depressive symptoms, and 8660 patients with cognitive-affective depressive symptoms. The median age of all participants was 65.1 years, with a median BMI of 28 kg/m², and 55.1% were female. Over 50% of the participants had a high school education or above. There were 5608 (62.0%) smokers and 6722 (74.3%) drinkers. In terms of other diseases, 3758 (41.5%) had hypertension and 1508 (16.7%) had heart disease. In patients with overall and somatic depressive symptoms, the proportions of common diseases such as hypertension and heart disease as well as lifestyle habits like smoking and drinking were significantly higher than those in the control population.

Overall and somatic depressive symptoms increased the risk of developing diabetes, while cognitive-affective depressive symptoms did not increase the risk of diabetes

As presented in Figure 3A in both the HRS and ELSA cohorts, we found that overall (HRS: HR = 1.14, 95%CI: 1.07-1.22; ELSA: HR = 1.14, 95%CI: 1.07-1.22) and somatic (HRS: HR = 1.18, 95%CI: 1.03-1.34, ELSA: HR = 1.25, 95%CI: 1.10-1.42) depressive symptoms increased the risk of diabetes. However, there was no association between cognitive-affective depressive symptoms (HRS: HR = 0.94, 95%CI: 0.82-1.08, ELSA: HR = 1.03, 95%CI: 0.77-1.38) and the risk of diabetes.

Figure 3 Forest plots. A: The associations between total, somatic, and cognitive-affective depressive symptoms and the risk of diabetes in the Health and Retirement Study (HRS) and English Longitudinal Study of Ageing (ELSA) cohorts; B: The associations between trajectories of total, somatic, and cognitive-affective depressive symptoms and diabetes in the Health and Retirement Study cohort; C: The associations between trajectories of total, somatic depressive symptoms and diabetes in the English Longitudinal Study of Ageing cohort. HRS: Health and Retirement Study; ELSA: English Longitudinal Study of Ageing; HR: Hazard ratio; CI: Confidence interval.

Baseline information for the second study

As shown in Table 2 in the HRS cohort, we followed up with a total of 13918 participants who had depressive symptom trajectories (682 with persistently high symptoms, 7252 with persistently low symptoms, 215 with decreasing symptoms, 5465 with fluctuating symptoms, and 304 with increasing symptoms). The average median age of the participants was 70.7 years, with 62.2% being female; 77.7% of the participants had a high school education or above, and 59.4% were married. In terms of lifestyle habits, there were 7455 smokers and 5439 non-smokers as well as 6030 drinkers and 6955 non-drinkers. Regarding general health conditions, 49.9% of the participants had hypertension, and 22.1% had heart disease.

Table 2 Basic characteristics of total depressive symptom trajectories in the second study.

Source	Variable	Overall	Persistently high	Persistently low	Decreasing	Fluctuating	Increasing	P value
HRS	Number	13918	682	7252	215	5465	304
	BMI, mean (SD)	70.7 (9.7)	70.7 (9.7)	69.9 (9.0)	71.1 (10.2)	71.7 (10.5)	71.9 (9.8)	< 0.001
	Age, mean (SD)	27.3 (5.6)	29.0 (7.5)	27.0 (5.1)	27.8 (6.9)	27.5 (5.9)	27.3 (6.3)	< 0.001
	Sex, n (%)
	Female	8654 (62.2)	519 (76.1)	4127 (56.9)	170 (79.1)	3629 (66.4)	209 (68.8)	< 0.001
	Male	5264 (37.8)	163 (23.9)	3125 (43.1)	45 (20.9)	1836 (33.6)	95 (31.2)
	Education, n (%)
	Below high school	3113 (22.4)	311 (45.6)	1105 (15.2)	72 (33.5)	1540 (28.2)	85 (28.0)	< 0.001
	College or above	5630 (40.5)	133 (19.5)	3480 (48.0)	49 (22.8)	1862 (34.1)	106 (34.9)
	High school	5175 (37.2)	238 (34.9)	2667 (36.8)	94 (43.7)	2063 (37.7)	113 (37.2)
	Marital status, n (%)
	Married or partnered	8268 (59.4)	277 (40.6)	5253 (72.4)	99 (46.0)	2465 (45.1)	174 (57.2)	< 0.001
	Never married	373 (2.7)	23 (3.4)	199 (2.7)	10 (4.7)	135 (2.5)	6 (2.0)
	Separated/divorced/widowed	4338 (31.2)	381 (55.9)	1797 (24.8)	106 (49.3)	1931 (35.3)	123 (40.5)
	Unknown	939 (6.7)	1 (0.1)	3 (0.0)	0 (0.0)	934 (17.1)	1 (0.3)
	Smoking status, n (%)
	Ever smokers	7455 (53.6)	416 (61.0)	4072 (56.2)	128 (59.5)	2657 (48.6)	182 (59.9)	< 0.001
	Never smokers	5439 (39.1)	266 (39.0)	3130 (43.2)	85 (39.5)	1837 (33.6)	121 (39.8)
	Unknown	1024 (7.4)	0 (0.0)	50 (0.7)	2 (0.9)	971 (17.8)	1 (0.3)
	Drinking status, n (%)
	Ever drinkers	6030 (43.3)	189 (27.7)	3852 (53.1)	72 (33.5)	1807 (33.1)	110 (36.2)	< 0.001
	Never drinkers	6955 (50.0)	493 (72.3)	3399 (46.9)	143 (66.5)	2726 (49.9)	194 (63.8)
	Unknown	933 (6.7)	0 (0.0)	1 (0.0)	0 (0.0)	932 (17.1)	0 (0.0)
	Hypertension, n (%)
	Yes	6944 (49.9)	454 (66.6)	3605 (49.7)	122 (56.7)	2583 (47.3)	180 (59.2)	< 0.001
	No	6042 (43.4)	228 (33.4)	3647 (50.3)	93 (43.3)	1950 (35.7)	124 (40.8)
	Unknown	932 (6.7)	0 (0.0)	0 (0.0)	0 (0.0)	932 (17.1)	0 (0.0)
	Heart problem, n (%)
	Yes	3076 (22.1)	257 (37.7)	1439 (19.8)	63 (29.3)	1228 (22.5)	89 (29.3)	< 0.001
	No	9910 (71.2)	425 (62.3)	5813 (80.2)	152 (70.7)	3305 (60.5)	215 (70.7)
	Unknown	932 (6.7)	0 (0.0)	0 (0.0)	0 (0.0)	932 (17.1)	0 (0.0)
ELSA
	Number	5811	263	3439	100	1877	132
	BMI, mean (SD)	28.2 (5.2)	29.2 (5.9)	27.9 (4.8)	29.2 (5.6)	28.5 (5.6)	28.4 (5.5)	< 0.001
	Age, mean (SD)	69.2 (9.3)	69.9 (10.6)	68.7 (8.8)	70.3 (9.7)	69.7 (9.9)	72.2 (9.9)	< 0.001
	Sex, n (%)
	Female	3318 (57.1)	181 (68.8)	1749 (50.9)	72 (72.0)	1242 (66.2)	74 (56.1)	< 0.001
	Male	2493 (42.9)	82 (31.2)	1690 (49.1)	28 (28.0)	635 (33.8)	58 (43.9)
	Education, n (%)
	Below high school	1861 (32.0)	129 (49.0)	920 (26.8)	56 (56.0)	705 (37.6)	51 (38.6)	< 0.001
	College or above	2309 (39.7)	71 (27.0)	1534 (44.6)	17 (17.0)	642 (34.2)	45 (34.1)
	High school	1184 (20.4)	39 (14.8)	726 (21.1)	18 (18.0)	376 (20.0)	25 (18.9)
	Other	457 (7.9)	24 (9.1)	259 (7.5)	9 (9.0)	154 (8.2)	11 (8.3)
	Marital status, n (%)
	Married or partnered	3834 (66.0)	111 (42.2)	2534 (73.7)	44 (44.0)	1061 (56.5)	84 (63.6)	< 0.001
	Never married	265 (4.6)	23 (8.7)	141 (4.1)	3 (3.0)	94 (5.0)	4 (3.0)
	Separated/divorced/widowed	1542 (26.5)	126 (47.9)	653 (19.0)	52 (52.0)	668 (35.6)	43 (32.6)
	Unknown	170 (2.9)	3 (1.1)	111 (3.2)	1 (1.0)	54 (2.9)	1 (0.8)
	Smoking status, n (%)
	Ever smokers	3684 (63.4)	193 (73.4)	2118 (61.6)	73 (73.0)	1207 (64.3)	93 (70.5)	< 0.001
	Never smokers	2127 (36.6)	70 (26.6)	1321 (38.4)	27 (27.0)	670 (35.7)	39 (29.5)
	Drinking status, n (%)
	Ever drinkers	4471 (76.9)	147 (55.9)	2859 (83.1)	67 (67.0)	1297 (69.1)	101 (76.5)	< 0.001
	Never drinkers	703 (12.1)	76 (28.9)	318 (9.2)	21 (21.0)	275 (14.7)	13 (9.8)
	Unknown	637 (11.0)	40 (15.2)	262 (7.6)	12 (12.0)	305 (16.2)	18 (13.6)
	Hypertension, n (%)
	Yes	2691 (46.3)	150 (57.0)	1491 (43.4)	61 (61.0)	924 (49.2)	65 (49.2)	< 0.001
	No	3120 (53.7)	113 (43.0)	1948 (56.6)	39 (39.0)	953 (50.8)	67 (50.8)
	Heart problem, n (%)
	Yes	1209 (20.8)	89 (33.8)	615 (17.9)	21 (21.0)	449 (23.9)	35 (26.5)	< 0.001
	No	4602 (79.2)	174 (66.2)	2824 (82.1)	79 (79.0)	1428 (76.1)	97 (73.5)

HRS: Health and Retirement Study; ELSA: English Longitudinal Study of Ageing; BMI: Body mass index; SD: Standard deviation.

In the ELSA cohort 5811 participants met the diagnostic criteria for depressive symptom trajectories (263 with persistently high symptoms, 3439 with persistently low symptoms, 100 with decreasing symptoms, 1877 with fluctuating symptoms, and 132 with increasing symptoms). The median age of the participants was 69.2 years, with a median BMI of 28.2 kg/m². There were 3318 female and 2493 male participants, and the distribution of lifestyle habits and general health conditions was similar to that in the HRS cohort. The baseline information for somatic and cognitive-affective depressive symptom trajectories is presented in Supplementary Tables 2 and 3.

Persistently high and fluctuating trajectories of overall and somatic depressive symptoms increased the risk of developing diabetes, while trajectories of cognitive-affective depressive symptoms were not associated with the risk of diabetes

In the HRS cohort compared with the persistently low trajectory, we found that the persistently high (Model 1: HR = 1.90, 95%CI: 1.67-2.17; Model 2: HR = 1.73, 95%CI: 1.51-1.98; Model 3: HR = 1.22, 95%CI: 1.06-1.40) and fluctuating (Model 1: HR = 1.08, 95%CI: 1.01-1.16; Model 2: HR = 1.23, 95%CI: 1.14-1.33; Model 3: HR = 1.09, 95%CI: 1.01-1.17) depressive symptom trajectories increased the risk of developing diabetes. After adjusting for hypertension, heart disease, and BMI in Model 3, the increasing and decreasing symptom trajectories no longer had statistical significance (Figure 3B). Similar to overall depressive symptoms in the somatic depressive symptom trajectories, persistently high (Model 1: HR = 2.26, 95%CI: 1.98-2.59; Model 2: HR = 1.98, 95%CI: 1.72-2.28; Model 3: HR = 1.24, 95%CI: 1.07-1.43) and fluctuating (Model 1: HR = 1.17, 95%CI: 1.09-1.26; Model 2: HR = 1.31, 95%CI: 1.22-1.41; Model 3: HR = 1.10, 95%CI: 1.02-1.18) trajectories increased the risk of diabetes, while increasing or decreasing trajectories were not associated with the risk of diabetes. Like the impact of cognitive-affective depressive symptoms on the risk of diabetes, their trajectories, including persistently high, increasing, decreasing, and fluctuating, did not affect the risk of diabetes.

We conducted further validation in the ELSA cohort (Figure 3C). In the overall depressive symptom trajectories, persistently high (Model 1: HR = 1.97, 95%CI: 1.50-2.59; Model 2: HR = 1.91, 95%CI: 1.44-2.53; Model 3: HR = 1.54, 95%CI: 1.16-2.05) and fluctuating (Model 1: HR = 1.46, 95%CI: 1.26-1.69; Model 2: HR = 1.48, 95%CI: 1.27-1.72; Model 3: HR = 1.33, 95%CI: 1.14-1.55) trajectories similarly increased the risk of developing diabetes, while increasing and decreasing trajectories did not affect the risk of diabetes. In the somatic depressive symptom trajectories, we also found that persistently high (Model 1: HR = 2.42, 95%CI: 1.86-3.16; Model 2: HR = 2.31, 95%CI: 1.77-3.03; Model 3: HR = 1.79, 95%CI: 1.36-2.35) and fluctuating (Model 1: HR = 1.55, 95%CI: 1.34-1.80; Model 2: HR = 1.55, 95%CI: 1.34-1.81; Model 3: HR = 1.31, 95%CI: 1.13-1.53) trajectories increased the risk of diabetes. Additionally, we discovered that the increasing somatic depressive symptom trajectory (Model 1: HR = 1.93, 95%CI: 1.19-3.13; Model 2: HR = 1.75, 95%CI: 1.07-2.85; Model 3: HR = 1.67, 95%CI: 1.03-2.73) also increased the risk of diabetes, but the decreasing trajectory was not associated with the risk of diabetes. Due to the relatively small number of individuals with persistently low cognitive-affective depressive symptoms, we were unable to draw reliable conclusions about the association between their trajectories and the risk of diabetes.

Sensitivity analyses confirmed the reliability of the study results

In terms of sensitivity analysis, we extended the follow-up of the HRS cohort to waves 13, 14, and 15 (Supplementary Table 4) and found that persistently high and fluctuating trajectories of overall and somatic depressive symptoms similarly increased the risk of diabetes, while other trajectories were not associated with the risk of diabetes. Stratified analysis (Supplementary Tables 4-10) of overall depressive symptom trajectories revealed that all trajectories might affect the risk of diabetes in females. In stratified analyses by smoking, drinking, hypertension, and heart disease, the results were similar to the overall findings, indicating that persistently high and fluctuating trajectories increased the likelihood of developing diabetes. The stratified analysis results for somatic depressive symptom trajectories were close to the overall results, while the stratified analysis results for cognitive-affective depressive symptom trajectories had little impact on the risk of diabetes. In the ELSA cohort different genders, education levels, marital statuses, smoking statuses, and drinking statuses all showed that persistently high and fluctuating trajectories of overall and somatic depressive symptoms increased the risk of diabetes. In summary the results of the stratified analyses confirmed the association between the identified trajectories and the risk of diabetes. All R code for comprehensive analysis is provided in Supplementary material.

DISCUSSION

In our study we found that overall and somatic depressive symptoms increased diabetes risk, while cognitive-affective symptoms did not. Persistently high and fluctuating trajectories of overall depressive symptoms were linked to diabetes risk, but increasing and decreasing trajectories were not. For somatic depressive symptoms persistently high, fluctuating, and increasing trajectories may raise diabetes risk, while decreasing trajectories did not. Cognitive-affective depressive symptom trajectories showed no statistical link to diabetes. Sensitivity analyses confirmed our the robustness of the results.

Overall and somatic depressive symptoms, especially their persistently high and fluctuating trajectories, increased diabetes risk. Prior population-based studies have reported similar links between depressive symptoms and diabetes onset[7,22,29]. Individuals with these symptoms may not adhere to dietary advice and physical activity recommendations, increasing obesity risk, a key diabetes factor[23,24,28-30]. Antidepressant use can also cause weight gain, worsening obesity[31]. Moreover, depressive symptoms may lead to unhealthy behaviors like smoking[23-25,28,30]. Even after adjusting for smoking and drinking, the association remained statistically significant. Depressive symptoms may also activate the hypothalamic-pituitary-adrenal axis and sympathetic-adrenal system, releasing inflammatory factors like IL-6 and CRP, which can promote diabetes development[32-34].

Persistently high and fluctuating trajectories of overall and somatic depressive symptoms may influence diabetes occurrence by altering behavior patterns. They may also trigger stress responses that disrupt the sympathetic-parasympathetic balance, increasing adrenaline and glucocorticoid release[33,34]. Adrenaline can decrease insulin secretion and increase glucagon secretion, raising blood glucose levels[35]. Glucocorticoids can cause glycogenolysis and gluconeogenesis, also leading to elevated blood glucose[36].

The inconsistent association between the increased depressive symptom trajectory and diabetes risk in Model 3 across the HRS and ELSA cohorts may stem from several factors: Sample characteristic differences; immunosenescence/metabolic pathways; healthcare system glycemic control strategies; and cultural reporting differences. In the elderly HRS cohort, accelerated immunosenescence may override depression-specific metabolic pathways. Evidence shows spatial heterogeneity in hepatic lipid metabolism and cholesterol synthesis pathway activation during aging[37]. Also, increased adipose tissue CRTC2 Levels and upregulated HSP90B1 in β cells in older individuals may alter metabolic pathways and impair insulin secretion[37]. Additionally, differences in healthcare ecosystems may affect the detection of subclinical depression-related disorders[38,39]. The more aggressive glycemic control in the HRS cohort may mask metabolic dysregulation linked to depression[38,39], while the stricter diagnostic thresholds of the ELSA cohort may enhance detection sensitivity[40-44]. Culturally, varying perceptions of health and aging may influence the reporting of somatic depressive symptoms[40-44]. For example, in the HRS cohort (Americans), symptoms like fatigue and weight changes may be attributed to aging rather than depression, potentially diluting biological signals[40-44].

The link between rising depressive symptom trajectories and diabetes risk may stem from the hypothalamic-pituitary-adrenal (HPA) axis and sympathetic-adrenal system activation[45]. Chronic HPA axis activation causes hypercortisolism, disrupting glucose homeostasis via glucocorticoid receptor signaling[45]. High baseline cortisol levels boost visceral fat and insulin resistance, fostering a diabetes-prone metabolic environment[46-48]. Notably, HPA axis dysfunction often precedes depressive symptoms and glycemic decline. Low-grade inflammation, driven by cytokines like CRP, IL-1β, and MCP-1, biologically connects depression and diabetes[49]. These markers disrupt insulin signaling, impair glucose uptake, and trigger neuroinflammation, leading to depressive traits[49]. Adipokine imbalance exacerbates this cycle by linking adipose tissue inflammation with central nervous system dysfunction. Prospective studies showed insulin resistance was key in the depression-diabetes link[50]. Conversely, it worsened depressive symptoms by hindering hippocampal neurogenesis and monoaminergic neurotransmission[51,52]. Poor lifestyle habits in those with somatic depressive symptoms contributed to obesity and metabolic issues, upping diabetes risk[45]. Sleep problems, common in depression, also harmed glucose metabolism and insulin sensitivity, paving the way for diabetes[45].

Somatic depressive symptoms directly affected metabolic functions and health behaviors, increasing diabetes risk. Chronic stress and inflammation from these symptoms can harm insulin sensitivity and glucose metabolism. In contrast cognitive-affective symptoms may follow different pathways, being more linked to psychological factors like diabetes distress or emotional burden. Research showed that females with diabetes had higher cognitive-affective symptom scores than males, possibly due to the psychological impact of diabetes complications or burdens[53,54].

Our study found stronger links between depressive symptoms and diabetes risk in females, those with lower education, and individuals with comorbid hypertension and heart disease. Potential mechanisms include hormonal differences like estrogen-mediated insulin resistance in females[55,56], health literacy variations across educational groups influencing self-care and preventive adherence[57], and the synergistic effects of cardiac diseases (e.g., endothelial dysfunction and the HPA axis dysregulation tied to depressive symptoms)[58].

This study used two prospective cohort studies to verify the link between depressive symptom trajectories (overall, somatic, and cognitive-affective) of overall and somatic depressive symptoms boosted diabetes risk, offering new insights for diabetes prevention, especially in community settings. Our research, based on nationally representative cohorts from the United States and the United Kingdom, had a large sample size with good representativeness. Unlike previous studies, we analyzed the association between somatic and cognitive-affective depressive symptom trajectories and diabetes risk. Targeted management of different depressive symptoms could be an effective way to prevent diabetes.

Depressive symptom trajectories, like persistently high and fluctuating, were defined using prior CESD models. The persistently high trajectory with CESD scores ≥ 3 suggests likely progression to clinical depression. The fluctuating trajectory, marked by unstable scores, implies episodic worsening and increased clinical depression susceptibility. Both represent subclinical depressive states predicting future diabetes risk. Tracking CESD scores in these groups helps community healthcare systems target high-risk populations with early interventions, curbing diabetes incidence.

Our study had several limitations. First, our study population primarily consisted of individuals over 50 years from the United States and United Kingdom. Therefore, results may not generalize to other age groups. Depressive symptoms in older adults often present as somatic complaints rather than affective symptoms, which can complicate diagnosis and interaction with chronic diseases like diabetes. Aging is associated with physiological changes such as chronic inflammation, mitochondrial dysfunction, and neuroendocrine alterations (e.g., HPA axis dysregulation), which may exacerbate both depressive symptoms and diabetes progression. These pathways may operate differently in younger populations in whom psychosocial and behavioral factors might dominate. Future research should explore these differences in younger populations.

Second, while we used self-reported questionnaires that may introduce recall bias, we cross-verified the data with medical records to reduce errors. Future research could benefit from objective measures like blood glucose levels. Additionally, we did not adjust for some variables like socioeconomic status, substance use, genetic factors, and physical activity due to incomplete documentation. These unadjusted variables may affect the results and should be considered when interpreting the findings. Moreover, the small sample size of cognitive-affective depressive symptoms in the ELSA cohort may limit the generalizability of our results. Caution is advised when interpreting and extrapolating the findings.

We also acknowledge that our trajectory classification method, though validated, may benefit from alternative approaches in future research to enhance generalizability. The 8-item CESD score used in this study differs from some previous literature, but our results still hold clinical value. Lastly, measurement errors and residual confounding factors may be present. We used longitudinal trajectory modeling to reduce measurement errors, but unmeasured factors like dietary patterns and environmental exposures remain limitations. Future research could adopt new approaches like causal inference methods to address these issues.

CONCLUSION

Overall and somatic depressive symptoms increased the risk of diabetes. The risk of diabetes varied with changes in depressive symptoms with individuals in a decreasing state being less affected. Individuals with increasing, fluctuating, or persistently high somatic depressive symptoms may have a higher risk of developing diabetes in the future. Therefore, promoting screening and identification of groups with fluctuating, increasing, or persistently high depressive symptoms and intervening is key to identifying and preventing high-risk diabetes groups.