Associations between adverse childhood experiences and progression to incident psychiatric disorders in older adults: A 22-year cohort study

Abstract

The long-term impact of adverse childhood experiences (ACEs) on the development of psychiatric disorders in older adults remains unclear. This study examined associations between ACEs and incident psychiatric disorders in older adults. Data from the 2000-2022 wave of Health and Retirement Study in the USA were analyzed. Time-varying Cox regression and multistate Markov models were applied to explore the impact of ACEs on transitions across five health states: healthy, physical conditions (PC), mental health symptoms (MS), comorbid PC & MS, and psychiatric disorders. Models were adjusted for demographic, behavioral, and disease-related factors. A total of 8628 participants were included, with a mean age of 64.75 years (SD = 8.01). During a mean follow-up duration of 16.8 years, 1429 developed psychiatric disorders (incidence: 9.85 per 1000 person-years). ACEs were significantly associated with higher risk of incident psychiatric disorders in a dose-response relationship. Participants with ACEs had higher transition percentages and intensities from healthy to less healthy states, notably from PC & MS to psychiatric disorders (percentages: 3.7 vs. 3.2%) and from the healthy state to MS (intensities: 0.130 vs. 0.104). They also spent less time in the healthy state and more time in comorbid states, with a 33% higher 22-year cumulative probability of psychiatric disorders (25.3 vs. 19.0%). ACEs appear to have enduring adverse impacts on mental health in later life by accelerating the progression to comorbidity and the development of psychiatric disorders. Early clinical screening and physical-mental health interventions are essential for prevention.

Introduction

With the rapidly aging population worldwide, poor mental health in older adults has become a major public health challenge, contributing substantially to high disability, low quality of life and increased healthcare utilization [1, 2]. Previous research found that approximately 14% of individuals aged 60 and above suffered from psychiatric disorders globally [3]. The 2019 Global Health Estimates indicated that mental health conditions, particularly depression and anxiety, accounted for 10.6% of the total disability-adjusted life years (DALYs) among older adults [1]. Mental health in later life is largely influenced by physical conditions, social environments as well as the cumulative effects of life experiences. Factors, such as exposure to adversity, loss of intrinsic capacity and declining functional ability, can lead to increased psychological distress [1]. Among the early life factors, adverse childhood experiences (ACEs) have emerged as a critical determinant of mental well-being with lasting impact on mental health across the lifespan.

ACEs refer to a range of stressful or adverse conditions and events occurring before age 18 years, which can disrupt a person’s sense of safety, stability, and bonding [4]. Common types of ACEs include child abuse, neglect, or household dysfunction involving substance abuse, mental illness, incarceration, domestic violence, and parental separation or divorce [4, 5]. Additionally, research on ACEs has expanded the scope to other significant stressors such as exposure to natural disasters, extreme economic hardship, bullying, school and community violence, life-threatening illnesses or injuries, and even school grade retention [6, 7]. Exposure to ACEs can increase the risk of a wide range of psychiatric problems, including depression, anxiety, post-traumatic stress disorder and substance use disorders [8–13]. For example, a large birth cohort study found that each additional ACE was associated with a 52% increase in the odds of developing any psychiatric disorder in adulthood, even after adjusting for shared genetic and environmental influences [13]. Neuroimaging research has shown that ACEs can lead to alterations of blood flow, activity, and connectivity in specific brain regions which may potentially mediate the impact of ACEs on the development of psychiatric disorders [14, 15].

Despite increasing research exploring the long-term consequences of ACEs and linking ACEs to depressive symptoms [16, 17], cognitive decline [18] and loneliness [18, 19] in older adults, important gaps remain. Late-onset psychiatric disorders often develop against a background of long-standing psychological vulnerability, chronic physical illness or multimorbidity, with subthreshold psychiatric symptoms and physical conditions that frequently precede clinical psychiatric diagnosis [20, 21]. However, existing studies mainly include overall mental health outcomes using traditional longitudinal or survival models in short follow-up periods, which estimate the risk of a single endpoint (i.e., incident psychiatric disorder), but do not capture how individuals move through intermediate physical and mental health states over time.

In this context, ACEs may not just have a simple link to the overall risk of psychiatric disorders in older age but may affect lifelong health trajectories by accelerating the progression from initially healthy states to physical illness, subclinical mental health symptoms or their co-occurrence, before ultimately developing into diagnosed psychiatric disorders. Conventional Cox models are not able to distinguish between these pathways. However, multistate models address these limitations by explicitly estimating transitions between multiple health states, quantifying the probabilities, intensities and duration of movement between states [22], thereby allowing the examination of whether ACEs influence entry into intermediate states (e.g., physical illness or subclinical mental health symptoms), progression to comorbid states, recovery to healthier states, or subsequent transition to psychiatric disorders.

To address these gaps, we conducted this secondary analysis of 22-year follow-up data from the nationally representative Health and Retirement Study (HRS) to: (1) examine the associations between different ACEs and the onset of psychiatric disorders in later life, incorporating time-varying covariates and repeated assessments; and (2) explore how ACEs relate to the changes across various health states (e.g., physical and mental health symptoms transitioning to psychiatric disorders). We hypothesized that ACEs would be associated with an increased risk of developing psychiatric disorders in older adults after adjusting for time-varying covariates, and individuals with ACEs would experience a higher likelihood of progressing from healthy states to physical and mental health conditions, and ultimately developing psychiatric disorders.

Methods

Study design and samples

This was a cohort study based on the most recent 22-year dataset (2000-2022; waves 5-16) of the HRS. The HRS is a nationally representative longitudinal study that has biennially surveyed U.S. residents aged 50 and older since 1992 (Wave 1) using multi-stage, probability design. The details of the HRS methodology have been described previously [23]. The study is funded by the National Institute on Aging (grant numbers NIA UO1AG009740 and NIA RO1AG073289) and is conducted by the University of Michigan [24]. Our study used the data obtained from the RAND HRS Longitudinal File (including 2000-2020 wave of sociodemographic and health-related data), HRS core data (including ACEs-related data and 2022 wave of all related data) and exit data (including all-cause mortality data). The RAND HRS Longitudinal File, developed by the RAND Center for the Study of Aging, provides systematically cleaned and processed data for all waves up to 2020 wave [25], and therefore the HRS core data from the 2022 wave were obtained directly and harmonized accordingly.

The study cohort included individuals who participated in the 2000 wave (i.e., baseline) and followed up longitudinally up to the 2022 wave. Following previous studies [26, 27], older adults were defined as individuals aged ≥50 years. Participants who fulfilled the inclusion criteria: 1) were aged 50 years or older at the 2000 baseline; 2) were interviewed during at least one follow-up assessment between 2000 and 2022 wave; 3) completed ACEs assessments during the 2008-2012 surveys; and 4) had no prior physician-diagnosed psychiatric disorders at baseline, based on the HRS records of the psychiatric disorder item. Following previous research [28–30], for covariates with missing values related to demographic, health, or lifestyle variables, we applied imputation using last observation carried forward (LOCF) and next observation carried backward (NOCB), as previously recommended [31, 32]. The HRS received ethical approval from the Institutional Review Board of Virginia Commonwealth University (HM20023839). All participants provided informed consent. As a secondary analysis of HRS data, this study was exempt from ethical review.

Measurements

Adverse childhood experiences

Following previous studies [33–35], the presence of ACEs was assessed using nine items extracted from the Leave-Behind Questionnaires (LBQ) [36] that was completed during the 2008-2012 waves of the HRS. We selected items that explicitly referred to events occurring before age 18 years, drawing from three LBQ sections: Lifetime Traumas [37], Quality of Relationship with Mother Early in Life [38], and Lifetime Traumas Before the Age of 18 [37]. From Lifetime Traumas, only items with reported occurrence before age 18 were retained, including natural disasters, combat-related violence, physical assault, life-threatening illness or accident. One item reflecting maternal neglect was included to avoid redundancy, and all four items from Lifetime Traumas Before Age 18 (i.e. grade repetition, trouble with the police, parental substance-use-caused family problems, and physical abuse by parents) were used.

All items were handled as binary variables (1 = “Yes” and 0= “No”). The total scores of ACEs ranged from 0 to 9, with higher scores indicating more severe ACEs. The description and coding of the ACE items are listed in Table S1.

Ascertaining incident psychiatric disorders

Following previous research [39, 40], psychiatric disorders were ascertained using the standardized question at each HRS wave: “Have you ever been diagnosed with emotional, nervous, or psychiatric problem by a doctor?”. This question captured lifetime physician-diagnosed psychiatric disorders up to the time of each interview. Participants who reported any prior psychiatric diagnosis at baseline were excluded. The study outcome was the onset of psychiatric disorders, defined as the first self-reported physician-diagnosed psychiatric disorder occurring during follow-up. Follow-up time was calculated from baseline (2000 wave) to the wave at which psychiatric disorders were first identified, or to censoring due to death, loss to follow-up, or the end of follow-up (2022 wave), whichever occurred first.

Covariates

Sociodemographic data

Sociodemographic variables potentially related to psychiatric disorders included age, sex (male vs. female), race (White/ Caucasian, Black/ African American or other), marital status (married, divorced/separated, widowed, or never married/other), educational level (>12 years vs. ≤12 years), living with others (yes vs. no), vigorous physical activity (≥3 times/ week, yes vs. no), alcohol use (yes vs. no) and cigarette smoking (yes vs. no).

Health-related variables

Chronic diseases included seven self-reported physician-diagnosed major physical diseases recorded in the HRS: diabetes, cancer, heart disease, stroke, hypertension, lung diseases, and arthritis.

The severity of depressive symptoms was evaluated using the 8-item Center for Epidemiologic Studies Depression Scale (CESD-8) [41]. Each item was scored as 1 or 0 (yes or no), with items 4 (“I was happy”) and 6 (“I enjoyed life”) being reverse scored. The total scores ranged from 0 to 8, with higher scores indicating more severe depressive symptoms. A cutoff score of 4 was used to identify individuals having depressive symptoms (depression hereafter) [42]. Previous research has demonstrated that the CESD-8 had similar psychometric properties with the original version [43].

Cognitive function was assessed using the modified version of the Telephone Interview for Cognitive Status (TICS-m) [44]. Following a previous study [45], the total score of TICS-m was calculated by adding up the respective scores of memory, attention and calculation domains, which ranged from 0 to 27, with higher scores indicating better cognitive performance. A cutoff score of 11 was used to identify individuals having cognitive impairment [46, 47].

Insomnia symptoms (insomnia hereafter) was measured using the 4-item Jenkins Sleep Scale (JSS-4) [48]. Each item was rated on a 3-point Likert scale ranging from 1 (“most of the time”) to 3 (“rarely or never”). Item 4 (“feel rested in the morning”) was reverse scored. Higher total scores reflected more severe insomnia symptoms, and a cutoff score of 5 was used to identify those with insomnia symptoms [49]. All covariates were assessed at each HRS biennial wave.

Definition of health states

In this study, five health states were defined to characterize longitudinal transitions in mental and physical health status during follow-up: (1) absence of physical conditions or mental health symptoms (No PC & No MS); (2) presence of physical conditions only (PC; including diabetes, stroke, heart conditions or cancers); (3) presence of mental health symptoms only [MS; depression (CESD-8 score ≥ 4) and (or) cognitive impairment (TICS-m score ≤ 11)]; (4) co-occurrence of both physical conditions and mental health symptoms (PC & MS); and (5) having psychiatric disorders, defined as the absorbing state.

Statistical analyses

Frequencies and percentages as well as means and standard deviations (SDs) were used to describe categorical and continuous variables, respectively. All covariates were compared between different number of ACEs using Chi-square, analysis of variance and Wilcoxon rank sum tests, as appropriate. All statistical analyses were conducted using R (version 4.4.2) [50]. A two-sided P-value < 0.05 was considered statistically significant.

To examine the longitudinal association between ACEs and the risk of psychiatric disorders in older age (PDOA), we used Cox proportional hazards regression models with time-varying covariates [51]. Participants were followed up from baseline (2000 wave) until the occurrence of a psychiatric disorder, death, loss to follow-up, or end of study (2022 wave), whichever came first. Time-varying covariates, such as health-related variables and lifestyle factors, were updated at each wave to account for the changes over time. Three Cox models were constructed, adding covariates progressively. Model 1 adjusted for age, sex and race; Model 2 added marital status, education level, living with others, lifestyle variables (i.e., smoking, alcohol use, and vigorous physical activity) based on Model 1; Model 3 further added major physical diseases (i.e., diabetes, cancer, heart conditions, stroke, hypertension, lung diseases, and arthritis), depression (i.e., total CESD-8 score), insomnia (i.e., total JSS-4 score) and cognitive function (i.e., total TICS-m scores) based on Model 2. Following previous studies [13, 52], ACEs were analyzed in two forms: (1) total number of ACEs (i.e., 0, 1, 2, and 3+ ACEs) as a categorical variable with the 0 ACE group as the reference, and (2) continuous ACE scores (ranging 0-9). Schoenfeld residuals were used to assess the proportional hazards assumption and to ensure no violation was found. Kaplan-Meier survival curves were plotted to examine potential differences in incidence rates across number of ACEs.

We also conducted subgroup analyses to explore whether the associations between ACEs and incident PDOA were moderated by age (50-64 years vs. ≥65 years), sex, marital status, educational level, living with others, alcohol use, cigarette smoking, engagement in vigorous activity, insomnia (JSS-4 score < 5 vs. ≥5), depression (CESD-8 score < 4 vs. ≥4), and cognitive impairment (TICS-m score ≤ 11) between the group with and without ACEs.

To assess the robustness of the associations between ACEs and incident psychiatric disorders, five sensitivity analyses were conducted by: (1) excluding cases with psychiatric disorders during the first two years after the baseline; (2) excluding cases with psychiatric disorder in the first follow-up wave; (3) using the total number of physical diseases as a time-varying covariate in Model 3, instead of adjusting for individual physical diseases; (4) using traditional Cox regression model adjusting only for baseline covariates; (5) performing competing risk analyses using Fine–Gray subdistribution hazard models, given that death may preclude the occurrence of psychiatric disorders. Cumulative incidence function (CIF) curves were constructed to compare the risks of psychiatric disorders and death overtime across different number of ACEs.

Additionally, the multistate Markov model was implemented to capture the dynamic transitions of health status among individuals with ACEs and those without ACEs, focusing particularly on how ACEs associated with transitions across physical and mental health states and ultimately led to psychiatric disorders. The state structure is shown in Fig. 1A. States 1 to 4 were defined as transient states, while state 5 was the absorbing state. Transition intensity parameters (q_ij) represent the instantaneous transition risk from state i to j. These were estimated through maximum likelihood under the assumption of continuous time Markov processes [53]. For instance, q_3->5 reflects the instantaneous hazard of progressing from only having mental health symptoms to a diagnosed psychiatric disorder. Figure 1B illustrates representative individual-level transitions in our sample, showing the heterogeneous patterns of disease progression observed during follow-up. The transition intensity matrix Q based on 5-state model was formulated as follows [53]:

$$Q=\left(\begin{array}{ccccc}-\left(q_{1\to 2}+q_{1\to 3}+q_{1\to 4}+q_{1\to 5}\right)& q_{1\to 2}& q_{1\to 3}& q_{1\to 4}& q_{1\to 5}\\ q_{2\to 1}& -\left(q_{2\to 1}+q_{2\to 3}+q_{2\to 4}+q_{2\to 5}\right)& q_{2\to 3}& q_{2\to 4}& q_{2\to 5}\\ q_{3\to 1}& q_{3\to 2}& -\left(q_{3\to 1}+q_{3\to 2}+q_{3\to 4}+q_{3\to 5}\right)& q_{3\to 4}& q_{3\to 5}\\ q_{4\to 1}& q_{4\to 2}& q_{4\to 3}& -\left(q_{4\to 1}+q_{4\to 2}+q_{4\to 3}+q_{4\to 5}\right)& q_{4\to 5}\\ 0& 0& 0& 0& 0\end{array}\right)$$

Fig. 1. Transitions between transient states and absorbing state (psychiatric disorder), and possible transition paths for the selected subjects with ACEs.

(A) Transitions among multi states in the Markov model: State 1 (No PC & No MS): participants without physical conditions (PC, including diabetes, stroke, heart conditions and cancers) and mental health symptoms (MS, including depressive symptoms and cognitive impairment); State 2 (PC): participants with physical conditions; State 3 (MS): participants with mental health symptoms; State 4 (PC & MS): participants with mental health symptoms and physical conditions; State 5 (absorbing state): participants with psychiatric disorder. The arrows illustrate the possible transitions between the transient status and the absorbing state. Each transition parameter q illustrates the transition intensity and q_i->j is defined as the instantaneous transition risk from state i to j. (B) Possible transition paths for four selected subjects with ACEs. Observations were truncated once the absorbing states appeared. The nodes represent the actual observed states, with solid lines indicating potential transition trajectories observed during the follow-up period. Possible transition paths were selected randomly based on the recorded states. ACEs: adverse childhood experiences.

In this context, the transition probability represents the likelihood of a transition from one state to another within a given time interval. To assess the influence of potential factors on transition dynamics, a series of univariable Markov models was fitted first, each incorporating a single covariate (e.g., ACEs in the total sample, and all other covariates in the ACE-exposed group) independently. Given the limited number of observed transitions for several state changes, multivariable Markov models including multiple covariates simultaneously were not statistically stable and were therefore not fitted. Cumulative transition probabilities over a 22-year horizon were then computed. All multistate models were implemented using the R package msm.

Results

Participant characteristics

Of the 19,580 participants in the 2000 wave, we included 8628 participants who were followed up with 79,477 observations from 2000 to 2022 wave (Figure S1). Baseline sociodemographic and health-related characteristics of participants stratified by ACE number are summarized in Table 1. The mean age of participants was 64.75 years (SD = 8.01), and the majority were female (n = 4936; 57.2%), married (n = 6310; 73.1%), and had lower than high school education (n = 4901; 56.8%). Univariable analyses showed significant differences across ACE number in age, sex, educational level, smoking, suffering from lung diseases, depression score and cognition score (P < 0.05).

Table 1.

Baseline characteristics of the total sample and by different number of ACEs (N = 8628).

Variables	Total (N = 8628)		Number of ACEs								Univariable analyses
			0 (N = 5672, 65.7%)		1 (N = 2160, 25.0%)		2 (N = 608, 7.0%)		≥3 (N = 188, 2.2%)
	N	%	N	%	N	%	N	%	N	%	χ 2	P
Male	3692	42.8	2158	38.0	1072	49.6	350	57.6	112	59.6	169.27	<0.001
Race											5.30	0.506
White/ Caucasian	7331	85.0	4785	84.4	1859	86.1	527	86.7	160	85.1
Black/ African American	1005	11.6	689	12.1	231	10.7	64	10.5	21	11.2
Other	291	3.4	197	3.5	70	3.2	17	2.8	7	3.7
Married	6310	73.1	4120	72.6	1617	74.9	431	70.9	142	75.5	6.10	0.107
Above high school education (>12 years)	3727	43.2	2562	45.2	853	39.5	236	38.8	76	40.4	26.43	<0.001
Living with others	6680	77.4	4354	76.8	1701	78.8	470	77.3	155	82.4	6.31	0.098
Alcohol use	2801	32.5	1831	32.3	722	33.4	177	29.1	71	37.8	6.52	0.089
Smoking	1120	13.0	714	12.6	275	12.7	98	16.1	33	17.6	9.67	0.022
Vigorous physical activity	4499	52.1	2929	51.6	1141	52.8	327	53.8	102	54.3	1.97	0.579
Physical disease
Diabetes	958	11.1	615	10.8	260	12.0	63	10.4	20	10.6	2.68	0.444
Cancer	848	9.8	563	9.9	205	9.5	61	10.0	19	10.1	0.38	0.944
Heart conditions	1298	15.0	818	14.4	361	16.7	90	14.8	29	15.4	6.48	0.091
Stroke	260	3.0	169	3.0	69	3.2	17	2.8	5	2.7	0.44	0.931
Hypertension	3857	44.7	2529	44.6	994	46.0	259	42.6	75	39.9	4.39	0.222
Lung diseases	482	5.6	293	5.2	127	5.9	54	8.9	8	4.3	15.40	0.002
Arthritis	4582	53.1	2979	52.5	1165	53.9	337	55.4	101	53.7	2.72	0.437
	Mean	SD	Mean	SD	Mean	SD	Mean	SD	Mean	SD	F/Z	P
Age (years)	64.75	8.01	65.22	8.11	64.15	7.69	63.22	7.82	62.35	7.67	71.38	<0.001
Depression score (CESD-8)	1.14	1.59	1.07	1.54	1.23	1.66	1.39	1.76	1.35	1.68	34.81*	<0.001
Sleep score (JSS-4)	6.71	2.03	6.69	2.02	6.69	2.01	6.83	2.14	7.06	2.13	7.58*	0.055
Cognition score (TICS-m)	16.08	4.21	16.24	4.20	15.81	4.20	15.56	4.23	16.11	4.34	24.90*	<0.001

Bolded values indicate P < 0.05; ACE: adverse childhood experience; SD: standard deviation; CESD-8: 8-item Center for Epidemiological Studies Depression Scale; JSS-4: 4-item Jenkins Sleep Scale; TICS-m: modified version of the Telephone Interview for Cognitive Status Scale.

^*Analyzed using Wilcoxon rank sum test.

ACEs and the incidence of psychiatric disorders in older adults

Incidence rate of psychiatric disorders in older adults

During a mean follow-up duration of 16.8 years, 1429 (16.6%) participants experienced psychiatric disorders. The overall incidence rate was 9.85 (95%CI: 9.34-10.37) per 1000 person-years. The incidence rate of psychiatric disorders increased with the total number of ACEs: 8.85 per 1000 person-years for participants with no ACE, 11.2, 13.0 and 15.0 per 1000 person-years for those with one, two, and ≥3 ACEs, respectively (Table 2).

Table 2.

The associations between number of ACEs and incident psychiatric disorders in older adults (N = 8628).

Variables	Cases N	Incidence rate, per 1000 person-years	Model 1			Model 2			Model 3
			HR	95%CI	P	HR	95%CI	P	HR	95%CI	P
No. of ACEs
0 ACE	5672	8.85	Reference			Reference			Reference
1 ACE	2160	11.2	1.340	1.188-1.511	<0.001	1.313	1.163-1.482	<0.001	1.185	1.051-1.337	0.006
2 ACEs	608	13.0	1.608	1.336-1.935	<0.001	1.524	1.265-1.835	<0.001	1.279	1.056-1.549	0.012
≥3 ACEs	188	15.0	1.845	1.357-2.508	<0.001	1.785	1.313-2.428	<0.001	1.506	1.128-2.011	0.005
Age (years)			0.993	0.986-1.001	0.075	0.982	0.974-0.989	<0.001	0.970	0.963-0.978	<0.001
Male			0.615	0.549-0.688	<0.001	0.750	0.666- 0.845	<0.001	0.746	0.661-0.842	<0.001
Race
White/ Caucasian			Reference			Reference			Reference
Black/ African American			0.761	0.637-0.908	0.002	0.646	0.538-0.774	<0.001	0.543	0.451-0.653	<0.001
Other			1.097	0.836-1.440	0.503	0.971	0.739- 1.275	0.831	0.841	0.648-1.091	0.191
Married						0.671	0.499-0.902	0.008	0.787	0.580-1.069	0.125
Above high school education (>12 years)						0.960	0.861-1.071	0.465	1.190	1.061-1.335	0.003
Living with others						1.119	0.828-1.513	0.464	1.157	0.849-1.575	0.356
Alcohol use						0.624	0.550-0.709	<0.001	0.703	0.618-0.799	<0.001
Smoking						1.060	0.883-1.272	0.535	0.946	0.786-1.139	0.559
Vigorous physical activity						0.634	0.563-0.714	<0.001	0.813	0.720-0.919	<0.001
Physical disease
Diabetes									0.943	0.833-1.067	0.355
Cancer									0.889	0.775-1.020	0.093
Heart conditions									1.126	1.003-1.265	0.045
Stroke									1.440	1.224-1.694	<0.001
Hypertension									1.100	0.976-1.241	0.120
Lung diseases									1.246	1.075-1.444	0.003
Arthritis									1.295	1.137-1.475	<0.001
Depression score (CESD-8)									1.318	1.287-1.351	<0.001
Sleep score (JSS-4)									1.066	1.038-1.095	<0.001
Cognition score (TICS-m)									0.962	0.949-0.975	<0.001

Bolded values indicate P < 0.05. ACE: adverse childhood experience; HR: Hazard ratio; CI: Confidence interval; CESD-8: 8-item Center for Epidemiological Studies Depression Scale; JSS-4: 4-item Jenkins Sleep Scale; TICS-m: modified version of the Telephone Interview for Cognitive Status Scale.

Model 1 was adjusted for age, sex and race/ethnicity.

Model 2 was adjusted for age, sex, race/ethnicity, marital status, living with others, educational level, vigorous activities, smoking, and alcohol use.

Model 3 was adjusted for the same covariates as model 2 plus history of major physical diseases (diabetes, cancer, heart conditions, stroke, hypertension, lung diseases, and arthritis), total CESD-8 score, total JSS-4 score and total TICS-m score.

The associations between ACEs and incident psychiatric disorders in older adults

Table 2 shows a dose-response relationship between the number of ACE and incident psychiatric disorders in older adults, which persisted across all three models. In the fully adjusted model, participants with two and ≥3 ACEs had 27.9% and 50.6% higher risks of psychiatric disorders, respectively, compared to those with no ACE [adjusted Hazard ratio (aHR): 1.279, 95%CI: 1.056-1.549 vs. aHR: 1.506, 95%CI: 1.128-2.011]. Table S2 shows that the continuous ACE score was also significantly associated with incident psychiatric disorders across three models (Model 3 aHR: 1.142, 95%CI: 1.072-1.216). The Kaplan-Meier survival curve (Figure S2) illustrates significant differences in incidence rates across different number of ACEs (P < 0.001).

Figure 2 shows that the association between ACEs and incident psychiatric disorders was more pronounced among individuals aged 50-64 years (HR: 1.71, 95% CI:1.36-2.15) than those aged ≥65 years (HR: 1.23, 95% CI: 1.09- 1.39; P for interaction: 0.013); among participants with higher educational level (HR: 1.71, 95% CI: 1.45-2.02) than those with lower educational level (HR: 1.12, 95% CI: 0.98-1.29, P for interaction: <0.001); and among those with cognitive decline (HR: 1.49, 95% CI: 1.31-1.70) compared to those without cognitive decline (HR:1.06, 95% CI:0.88-1.27; P for interaction: 0.002).

Fig. 2. The association between ACEs and incident psychiatric disorders in older adults stratified by demographic and clinical correlates.

Graphs show hazard ratios (HRs) and 95% confidence intervals (CIs) for psychiatric disorders in older adults. ACEs were assessed using a binary variable (yes vs. no). ACEs: adverse childhood experiences; HR: Hazard ratio; CI: Confidence interval; CESD-8: 8-item Center for Epidemiological Studies Depression Scale; JSS-4: 4-item Jenkins Sleep Scale.

In sensitive analyses, consistent associations between ACEs and incident psychiatric disorders were observed (Tables S3-S6). The competing risk analyses showed that the association between ACEs and incident psychiatric disorders remained significant when accounting for death as a competing event (Table S7). The CIF curves revealed a higher cumulative risk of psychiatric disorders over time among individuals with greater exposure to ACEs, although death became more prominent as a competing risk after approximately 150 months of follow-up (Figure S3).

Transitions between health states among participants with and without ACEs

State transition percentage and intensity

The number and percentage of observed state transitions during follow-up among participants with ACEs and those without ACEs are summarized in Table S8. Among participants with ACEs, transitions from comorbid physical conditions and mental health symptoms (PC & MS) to psychiatric disorders accounted for the largest proportion (3.7%), followed by transitions from mental health symptoms alone (3.5%). In contrast, among those without ACEs, the proportion transitioning to psychiatric disorders was lower across all states, with the highest transition from the PC & MS state (3.2%). Participants with ACEs were also more likely to transition between vulnerable health states (e.g., from mental health symptoms to PC & MS: 8.2%) compared to those without ACEs (7.5%). Participants with ACEs showed consistently higher transition intensities across most state transitions compared to those without ACEs, e.g., the transitions from No PC & No MS to mental health symptoms (0.130 vs. 0.104) and from physical conditions to PC & MS (0.146 vs. 0.129) (Table S9).

State transition probability and total duration

Figure 3 (left panel) and S4 illustrate the 22-year transition probability curves across five health states, stratified by participants with ACEs and those without ACEs. The probability of remaining in the healthy state (1- > 1, No PC & No MS) declined sharply for both groups, especially in the ACE group. At baseline, the probabilities of state 1 were over 0.8 in both groups, while by the year 2022, the probabilities dropped to 0.153 in the ACE group and 0.196 in the no-ACE group, respectively. The probability of transitioning from the healthy state to more adverse states (e.g., 1- > 2, 1- > 4, and especially 1- > 5) increased over time. Certain transitions, such as recovery to healthy states (e.g., 2- > 1, 3- > 1), remained low and stable in both groups, indicating more likelihood of accelerated health deterioration and less likelihood of recovery.

Fig. 3. Transition probability curves and estimated percentage of total duration for five health states over 22 years.

A Transition probability curves and estimated percentage of total duration for participants with ACEs. B Transition probability curves and estimated percentage of total duration for participants without ACEs.

Notably, the group with ACEs exhibited higher and faster probabilities of progressing from the healthy state or any intermediate states to the absorbing state 5 (psychiatric disorders) over the 22 years; for instance, at 22 years, 1- > 5: ACE 0.253 vs. no ACE 0.190; 2 - > 5: ACE 0.245 vs. no ACE 0.209; 3- > 5: ACE 0.269 vs. no ACE 0.215; 4- > 5: ACE 0.269 vs. no ACE 0.233. We calculated the 22-year cumulative transition probabilities from four transient health states (States 1–4) to incident psychiatric disorders (State 5). The cumulative transition probabilities ranged from 19.0% to 23.3% and from 25.3 to 26.9% across states in the group without ACE and the group with ACEs, respectively. Even among participants initially in a healthy state (State 1), those with ACE exposure had a 33% higher risk of transitioning to psychiatric disorders (25.3 vs. 19.0%) by the year 2022. Furthermore, the group with ACEs exhibited consistent higher probabilities of transitioning between intermediate states (e.g., 1- > 2, 1- > 4, 3- > 4, and 2- > 4), indicating a higher likelihood of progression to early physical conditions, as well as comorbid physical and mental health conditions.

Figure 3 (right panel) shows the standardized percentages of total duration across the five states. Over the 22-year follow-up period, most of the duration was in state 2 (physical conditions), accounting for approximately 48% of the total time across both groups. This was followed by state 4 (PC & MS), comprising around 31%, and state 1 (No PC & No MS), at about 15%. Hence, both total duration of physical conditions and PC & MS prolonged, while those of healthy state (No PC & No MS) shortened. When comparing participants with ACEs and those without ACEs, the group with ACEs spent less time in the healthy state (state 1) (14.5 vs 15.6%) and more time in the comorbid state (state 4) (31.2 vs 30.1%) than their no-ACE counterparts. These trends indicated that ACEs might be associated with earlier onset and longer duration of complex comorbidity prior to the onset of psychiatric disorder.

Univariable associations with state transitions

As shown in Table S10, across the whole sample, univariable Markov models showed that individuals with ACEs tended to have higher transition rates from healthier to less healthy states compared with those without ACEs. Specifically, ACE exposure was associated with higher estimated risks of transition from No PC & No MS to physical conditions (HR = 1.190, 95% CI: 1.086-1.304), mental health problems (HR = 1.253, 95% CI: 1.153-1.362), and to psychiatric disorder (HR = 1.859, 1.439-2.403); and higher risk of transition from physical conditions to PC & MS (HR = 1.138, 95% CI: 1.063-1.218). However, several transition-specific estimates were imprecise with wide confidence intervals, which might be due to the relatively small number of observed transitions for these state changes.

Among participants with ACEs, additional univariable Markov models (Table S11) showed that older age (≥65 years) and male gender were consistently associated with lower transition rates from all states to psychiatric disorders (e.g., No PC & No MS -> Psychiatric disorders, HR: 0.538, 95% CI: 0.362–0.798 for older age; HR: 0.536, 95% CI: 0.359–0.802 for male gender). Vigorous physical activity was associated with lower transition rates from the No PC & No MS state to psychiatric disorders (HR: 0.620, 95% CI: 0.424–0.906). Insomnia (JSS-4 score ≥5) showed an association with higher transition rates from combined physical and mental health symptoms to psychiatric disorders (HR: 2.721, 95% CI: 1.244–5.952). Among physical diseases, lung diseases and arthritis were associated with higher transition rates from No PC & No MS state to psychiatric disorders (HR: 2.592, 95% CI: 1.524–4.356 for lung diseases; HR: 1.530, 95% CI: 1.028–2.275 for arthritis), as well as from physical conditions to psychiatric disorders (HR: 2.274, 95% CI: 1.371-3.773 for lung diseases; HR: 2.846, 95% CI: 1.334-6.072 for arthritis). Higher educational level was associated with higher transition rates from mental health symptoms to psychiatric disorders (HR: 1.864, 95% CI: 1.114–3.120), and from PC & MS to psychiatric disorders (HR: 1.472, 95% CI: 1.005- 2.155).

Discussion

To the best of our knowledge, this was the first cohort study to investigate the associations between ACEs and incident psychiatric disorders in older adults, based on 22 years of follow-up data. We found that ACEs were strongly associated with an increased risk of psychiatric disorders in later life after adjusting for the time-varying confounders regarding lifestyle, physical and mental health conditions. We also identified effect modifications in the relationship between ACEs and incident psychiatric disorders based on stratification by sociodemographic, lifestyle, and health-related factors. Moreover, we explored how ACEs influenced the transition from healthy and intermediate health states to psychiatric disorders and identified risk factors for such transitions.

Our study demonstrated that ACEs were significantly associated with a higher risk of psychiatric disorders in older adults over a 22-year follow-up. Using time-varying Cox regression, we found that the total ACE score and different number of ACEs were all significantly associated with an elevated risk of psychiatric disorders. The risk increased in a dose–response manner, with individuals exposed to three or more ACEs having the highest risk. The competing risk model further validated that these associations were not confounded by the differential risk of mortality across groups. These results are consistent with previous findings that ACEs contribute to poorer cognitive, physical and mental health outcomes in later life [54]. Notably, the assessment of ACEs in this study was different from traditional ACE inventories (e.g., household dysfunction, financial difficulties and sexual abuse) [13, 52]. Instead, we captured a broader set of early-life adversities, including traumatic experiences, grade repetition, trouble with the police, exposure to violence and other major stressors occurring before age 18. Such findings could help quantify the proportion of those exposed that are missed by the lack of assessment of most standard ACE exposures. It is important to clarify that the psychiatric outcome in this study reflected the first self-reported physician diagnosis of any emotional, nervous or psychiatric condition during follow-up. This might represent the culmination of subclinical symptoms, comorbid physical conditions leading to clinical recognition, rather than the onset of a specific defined psychiatric disorder. Therefore, the observed associations should be interpreted as reflecting the risk of first clinical recognition or diagnosis of psychiatric disorder, rather than being etiologically linked to the onset of a specific disorder.

The multistate Markov model used in this study provided dynamic data about the impact of ACEs on mental health trajectories in older adults, which could extend previous longitudinal studies showing that ACEs predict incident mental health problems in later life, often in a dose–response manner [16–19]. First, ACEs might be related to accelerated transitions toward psychiatric disorders in older adults, as ACEs were associated with higher transition frequency and intensity from healthier states to less healthy states (e.g. from No PC & No MS to mental health symptoms, and from PC & MS to psychiatric disorders), and reduced likelihood of recovery back to healthier states (e.g., from mental health symptoms to No PC & No MS), ultimately elevating cumulative risk of psychiatric disorders. Such results are aligned with previous research showing that ACEs accelerated the progression of both physical and mental health conditions, predisposing individuals to complex multimorbidity and psychiatric illness [54].

Second, ACEs were associated with prolonged vulnerability states and reduced time in healthy state. The group with ACEs spent less time in the healthy state (No PC & No MS) and more time in the comorbid PC & MS state over the study period. This pattern indicated that ACEs might be linked not only with the onset of psychiatric disorders but also with a prolonged period of subclinical vulnerability, which overall are likely to exacerbate psychiatric risks. Third, individuals with ACEs showed consistently higher cumulative risk of transitioning from healthy or intermediate states (e.g., physical conditions or mental health symptoms) to having psychiatric disorders. A recent study using multi-state model also found that higher ACE scores (≥4) were associated with an increased probability of forward transitions toward worsening physical health states, such as from robust to pre-frail (HR = 1.37) and from pre-frail to frail (HR = 1.39), as well as a decreased probability of reverting to healthier states (pre-frail to robust, HR = 0.64) [55]. Our findings hence extended previous research by dynamically characterizing how ACEs could influence long-term mental health trajectories in older adults, emphasizing the enduring effects of early life stressors on psychiatric morbidity decades later.

Several potential mechanisms, either alone or in combination, may explain both the predictive associations observed in the time-varying Cox models and the transition-specific patterns identified in the multistate analyses. First, the elevated risk of transitioning from No PC & No MS to mental health symptoms among individuals with ACEs may be due to long-term psychosocial vulnerability, including having increased exposure to stressful life events, higher perceived stress, negative emotionality, and unhealthy behaviors in adulthood [56]. Second, the stronger associations observed for transitions from physical conditions or from PC & MS to psychiatric disorders, and longer duration in PC & MS among older adults with ACE may be due to cumulative biological burden. Chronic stress related to ACEs may induce multisystem dysregulation, including neuroendocrine and inflammatory processes, which may interact with existing physical conditions to amplify the vulnerability to clinically recognized psychiatric disorders [57]. This may partially explain the higher transition intensity toward psychiatric disorders among individuals already in comorbid physical and mental health states. Third, neuroimaging studies indicates that alterations related to ACEs in brain regions critical for emotional regulation and cognitive control, such as the prefrontal cortex and hippocampus, may mediate increased vulnerability to psychiatric disorders, particularly in individuals experiencing cognitive decline [14, 15]. These psychosocial and biological processes can cumulatively generate sustained risk factors, including chronic inflammation, maladaptive coping strategies, and health-risk behaviors, that may reduce recovery from mental health challenges or accelerate transitions to psychiatric disorders.

Subgroup analyses revealed a strong association between ACEs and incident psychiatric disorders among individuals aged 50-64 years and those with higher educational attainment. This aligns with the results of our multi-state model, showing that individuals aged 50-64 years with ACEs had greater risks of transitioning from all health states to psychiatric disorders, while those with a higher education level and ACEs had greater risks of transitioning from mental health symptoms or PC & MS to psychiatric disorders. Individuals aged 50-64 years often experience greater work-related stress, family obligations, and caregiving responsibilities, compared to those aged 65 and older. Such stressors, combined with psychosocial vulnerabilities resulting from ACEs, can contribute to increased emotional distress and risk of psychiatric disorders [56]. Individuals with a higher education level may exhibit greater awareness of mental health symptoms and are more likely to seek help, leading to increased rates of psychiatric disorder diagnosis [58]. Furthermore, despite having socioeconomic advantages, individuals with a higher education level may still experience major psychosocial stressors (e.g., career pressure and social expectations) that can exacerbate the impact of early adversities [59]. Our findings that females with ACEs were associated with increased risks of transitions from all health states to psychiatric disorders is also consistent with the higher prevalence of mental illness observed in females [60]. Biological sensitivities, including hormonal fluctuations and neural differences, combined with gender-specific psychosocial stressors such as caregiving burdens and exposure to gender-based violence, can amplify the impact of early adversities on emotional regulation and psychiatric vulnerability in women [61, 62]. The interaction between ACEs and female-specific factors might contribute to the elevated psychiatric risk across various health trajectories observed in our study.

Furthermore, in the group with ACEs, univariable multi-state analysis identified other factors that were significantly associated with elevated risk of transition to psychiatric disorders. Behavioral factors, such as physical inactivity and insomnia, were linked to increased psychiatric risk, which highlights the role of lifestyle in modulating outcomes of ACEs. Chronic physical conditions, including lung diseases and arthritis, were also associated with higher risk of progression profoundly from healthy (No PC & No MS) or physical condition states to psychiatric disorders, potentially through mechanisms related to systemic inflammation [63] and functional impairment [64]. However, these univariable findings were exploratory, which generated hypotheses for multivariable modeling and mechanistic investigation rather than direct evidence of independent effects. Future studies employing multivariable multistate models or causal inference approaches (e.g., marginal structural models) [65] are needed to distinguish true risk factors from confounded associations.

This study has several clinical implications. First, our findings highlight the importance of preventing and reducing exposure to ACEs at the population level. Multi-level interventions based on socio-ecological models, targeting individual, family, community, and policy levels could provide effective strategies for reducing ACE exposure at its source and decreasing the burden of ACEs [7]. Second, although routine ACE screening has been controversial recently due to its potential harm (e.g. stigma, re-traumatization and resource misallocation without proven interventions) [66], we advocate targeted ACEs screening in high-risk populations (e.g., presenting with physical or mental health symptoms). This screening can facilitate early identification of ACE-related vulnerability, guiding risk-stratified interventions to prevent progression to multimorbidity and psychiatric disorders. Methodological advancements, such as validated brief ACE measures with improved sensitivity, specificity and patient acceptability, are essential to optimize clinical screening [66]. Third, for middle-aged and older adults with ACEs exhibiting transitional physical or mental health states, early integrated interventions during vulnerability windows, such as trauma-informed care (TIC) [67, 68], are needed to enhance their resilience and improve the mental health outcomes. TIC is a framework emphasizing trauma recognition, safety, empowerment and re-traumatization avoidance through integrated physical, mental, and social services [7].

The strengths of this study included a large, nationally representative cohort with longitudinal data spanning over two decades, the use of a comprehensive set of time-varying covariates to reduce potential confounding, and the application of multi-state Markov modeling that enabled a nuanced understanding of dynamic progress to the onset of psychiatric disorders. However, some limitations should be noted. First, ACEs were retrospectively self-reported, which might introduce recall bias [69–71]. However, a prior study has reported on the reasonable validity of adult retrospective reporting of ACEs [72]. Similar to previous research [39, 40], psychiatric disorders were based on self-reported diagnosis by physicians, which might lead to underreporting and cannot distinguish between different types of psychiatric disorders. Second, ACEs were assessed retrospectively in 2008–2012 wave, and study inclusion required survival and completion of the ACE module. This might introduce survivor and attrition selection bias, whereby individuals with more severe early adversity and worse mental health conditions might have been underrepresented. In addition, loss to follow-up over the observation period might have resulted in additional attrition bias if participants with greater ACE exposure or more adverse health trajectories were more likely to drop out, potentially leading to underestimation of transition risks and cumulative incidence. Third, the HRS-based measurement of ACEs did not capture the full range of ACE domains, which might potentially underestimate the true impact of ACEs on late-life psychiatric disorders. Fourth, potential confounding factors that might influence health trajectories [73], such as genetic predisposition, social support, social economic status or later-life stressors, were not included in the analysis. Fifth, missing data in certain time-varying covariates (i.e., CESD-8, JSS-4, and TICS-m) were imputed using LOCF and NOCB methods, which might introduce bias regarding the constancy of data across time. Sixth, due to limited number of transitions between health states, we were unable to perform multivariate Markov modeling incorporating all variables that were significantly associated with transitions in the univariable analyses, thus limiting the adjustment for confounding factors. Finally, although the cohort is representative of the US older adult population, findings might not be generalizable to those in other cultural or socioeconomic contexts.

In conclusion, this long-term cohort study revealed that ACEs were significantly associated with a higher risk of psychiatric disorders as well as accelerated progression to psychiatric disorders in older adults. The association between ACEs and psychiatric disorders in older adults were more pronounced among females, participants aged 50-64 years, and those with higher education level or with cognitive decline. Individuals with ACEs showed faster transitions from healthy and intermediate states to having psychiatric disorders, spent less time in healthy state, and had prolonged periods of multimorbidity prior to onset of psychiatric disorder. Demographic, behavioral, and physical health factors (such as age, sex, education level, physical activity, insomnia, lung diseases, and arthritis) might further modify these risks. Our findings highlighted the enduring impact of early life adversity on mental health trajectories in later life, and further underscored the need for integrated, trauma-informed care to mitigate psychiatric morbidity in older populations.

Author contributions

Study design: Hui-Ying Fan, Mu-Rui Zheng, Qinge Zhang, Chee H. Ng, Yu-Tao Xiang, Gang Wang. Data curation, analysis and interpretation: Hui-Ying Fan, Mu-Rui Zheng, Xiao-Xuan Meng, Qinge Zhang, Chee H. Ng, Yu-Tao Xiang, Zhi-Cheng Du, Zhaohui Su, Teris Cheung, Gabor S. Ungvari. Drafting of the manuscript: Hui-Ying Fan, Mu-Rui Zheng, Yu-Tao Xiang. Critical revision of the manuscript: Chee H. Ng, Gang Wang. Approval of the final version for publication: all co-authors.

Funding Sources

The study was supported by Hainan Provincial Natural Science Foundation of China (No. 826MS0167), Academic Enhancement Support Program of Hainan Medical University (XSTS2026170), Beijing High Level Public Health Technology Talent Construction Project (Discipline Backbone-01-028), the Beijing Municipal Science & Technology Commission (No. Z181100001518005), the Capital’s Funds for Health Improvement and Research (CFH 2024-2-1174) and the University of Macau (MYRG-GRG2023-00141-FHS; CPG2025-00021-FHS).

Data availability

The data that support the findings of this study are available from The University of Michigan Health and Retirement Study Platform at https://hrsdata.isr.umich.edu/data-products/public-survey-data.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-04102-6.