Stress and mental health in older adulthood: testing linear dose–response and steeling effects of early-life adversity Authors: Myriam V. Thoma, Jan Höltge, Simona-Viktoria Balsiger, Florence Bernays, Shauna L. Rohner Published: 2026 ABSTRACT Although mental health in later life has gained increasing attention, research on how early-life adversity (ELA) affects older adults remains limited. This study examined whether ELA exposure relates to mental health via a linear dose–response effect or a curvilinear association as proposed by the Steeling Effect Theory. ELA and past-year mental health disorders were assessed in N = 211 older adults (M = 69.8 years; 46.4% female), including a high-risk group with substantial ELA (RG; n = 100) and a lower-risk control group (CG; n = 111). Linear and curvilinear associations were tested using negative binomial generalized linear models. Overall, 59.2% of participants met criteria for at least one past-year mental health disorder (RG: 68.0%; CG: 51.4%). Higher levels of ELA were associated with an increased number of mental health disorders (1.5% per unit increase), supporting a linear dose–response effect. No evidence was found for a curvilinear (steeling) effect. Findings underscore the long-term impact of early-life stress on mental health in older adulthood. In this sample, a linear model better characterized the association between ELA and past-year mental health disorder count than a curvilinear model. These findings are consistent with the importance of preventing and reducing ELA as part of broader efforts to support mental health across the life course.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s10433-026-00928-7. Introduction The global increase in the proportion of older adults—projected to rise from 12% of the world’s population aged 60 + in 2015 to 22% by 2050 (World Health Organization [WHO], 2025a)—highlights the need to deepen the understanding of mental health heterogeneity in later life. Despite rapid demographic changes, research on the etiological factors that contribute to both vulnerability and resilience in older adulthood remains limited. Compounding this issue, mental health conditions in older adults are frequently underrecognized and undertreated, with stigma often preventing individuals from seeking the help they need (WHO, 2025b). These challenges underscore the need for more comprehensive research to identify key risk and protective factors and to better characterize pathways to vulnerability and resilience in later life. Stress exposure is a key factor in life-course models of psychopathology and resilience. Early-life adversity (ELA) is particularly important in this regard because adverse experiences during sensitive developmental periods may shape stress responsivity, coping, and later mental health. Empirical work has consistently linked ELA with elevated risk for adverse mental health outcomes (e.g., Björkenstam et al. 2016; Hillberg et al. 2011; Kessler et al. 2010). In addition, longitudinal evidence suggests that adverse childhood environments are associated with heightened emotional reactivity to stress, which in turn predicts the later onset of mood and anxiety disorders (McLaughlin et al. 2010b). Adult stressful life events have also been associated with past-year mental health disorders, with evidence that childhood adversity may heighten vulnerability to later stress exposure (McLaughlin et al. 2010a). Later life is shaped by life-course stress processes, in which earlier social and experiential conditions may influence later exposure to stressors and health-related vulnerabilities (Pearlin et al. 2005). In parallel, older adulthood is marked by a substantial burden of age-related chronic conditions, multimorbidity, disability, and mental and neurological disorders (Prince et al. 2015). These life-course and health-related processes may contribute to heterogeneity in mental health in older adulthood and underscore the relevance of examining stress–mental health associations in the context of population ageing. For Switzerland, national projections indicate that the population aged 65 + will increase from about 1.8 million currently to about 2.7 million by 2055, and its share will rise from just under one-fifth to more than one-quarter of the total population (Federal Statistical Office 2025). Two theoretical accounts are particularly relevant for understanding how ELA may relate to mental health outcomes in later life: linear dose–response accounts and the Steeling Effect Theory. Dose–response accounts Dose–response accounts assume that increasing stressor exposure is associated with increasingly adverse outcomes. From this perspective, individuals exposed to higher levels of ELA would be expected to show worse mental health in later life than individuals exposed to lower levels of ELA or no ELA. This assumption is consistent with cumulative stress and allostatic load models, according to which repeated or chronic stress exposure can accumulate over time and increase biological and psychological vulnerability (Felitti et al. 1998; McEwen 1998). It is also consistent with life-course perspectives on cumulative disadvantage, which suggest that early adversity can initiate or amplify later inequalities in health and functioning (Dannefer 2003; Ferraro and Shippee 2009; Pearlin et al. 2005). Applied to mental health, a dose–response account predicts a positive linear association between ELA and later psychopathology: higher adversity should be associated with a higher likelihood or greater burden of mental health problems. Although dose–response associations between childhood adversity and mental health have been well documented in younger and mixed-age samples, evidence in older adulthood remains comparatively limited (e.g., Lee et al. 2017; Raposo et al. 2014). Steeling effect theory An alternative perspective is offered by the Steeling Effect Theory (Liu 2015; Rutter 2012), which describes the conditions under which stress can potentially lead to higher adaptive functioning, resilience, and beneficial outcomes. This theory proposes a curvilinear association between stress and health outcomes, suggesting that comparatively favorable outcomes may occur among individuals with ‘some’ or ‘moderate’ stress exposure, although no precise definition or operationalization of what constitutes ‘some’ or ‘moderate’ stress currently exists (see Höltge et al. 2018). Similar to dose–response accounts, the Steeling Effect Theory expects worse outcomes among individuals exposed to extreme stress (Liu 2015; Rutter 2012). However, it differs from dose–response accounts in its assumption that moderate stress exposure may, under some conditions, foster adaptive learning, coping, or stress resistance. Rather than assuming uniformly increasing vulnerability, the Steeling Effect Theory suggests that individuals become ‘steeled’, desensitized, or strengthened in response to successfully coping with manageable stress exposure (Seery et al. 2010a, b). This process is assumed to increase adaptability and resistance to future stress and may contribute to better health and well-being in the long term (Liu 2015; Rutter 2012). Thus, the Steeling Effect Theory provides an alternative model for understanding mental health heterogeneity in older adulthood, particularly when associations between adversity and outcomes deviate from a strictly linear dose–response pattern. Empirical evidence on dose–response and steeling in later life Empirical support for steeling-consistent patterns has been reported across different age groups and outcome domains, although operationalizations of adversity and outcomes vary substantially (for a review see Höltge et al. 2018). Steeling-consistent findings typically follow a curvilinear or buffering pattern in which moderate prior adversity, or accumulated experience with stressors, is associated with comparatively more adaptive outcomes than low/no adversity or high adversity. For example, in laboratory stress paradigms, moderate early-life stress has been associated with lower cortisol activity or reactivity during acute stress compared with low or high adversity (Gunnar et al. 2009; Hagan et al. 2014). In adult community samples, ‘some’ lifetime adversity has been related to lower distress and functional impairment and higher life satisfaction relative to no adversity or high adversity (Seery et al. 2010a, b), and similarly to less functional impairment, less chronic pain, and lower healthcare utilization (Seery et al. 2010a, b). In older adulthood, the empirical picture is limited and mixed. Höltge et al. (2019) reported inverse U-shaped associations between ELA and both individual quality of life and mental health in older adults, with the most favorable outcomes observed at moderate levels of ELA; mental health mediated the association between ELA and individual quality of life, whereas no curvilinear association was found for physical health. Kok et al. (2021), using 16-year longitudinal data from the Longitudinal Aging Study Amsterdam, focused on accumulated negative life events rather than ELA alone. They found that cumulative and proximate life events were independently associated with more depressive symptoms, supporting cumulative burden. At the same time, the detrimental effect of proximate life events on depressive symptoms became weaker as cumulative life events increased, suggesting a steeling effect; the authors also reported that higher mastery and lower neuroticism were unexpectedly associated with weaker steeling effects (Kok et al. 2021). Zhang (2023), in a cross-sectional study of older adults in China, found that elder abuse and adverse childhood experiences (ACEs) were each associated with higher depressive symptoms, but that moderate exposure to ACEs (one or two ACEs) attenuated the association between elder abuse and depressive symptoms, particularly for financial exploitation; exposure to three or more ACEs did not show this buffering pattern (Zhang 2023). Other recent findings in older populations support linear or dose–response interpretations rather than steeling-consistent patterns. Using data from the same longitudinal cohort as the present study, Thoma et al. (2024a, b) examined ACEs as a brief cumulative indicator of ELA, recent stressful life experiences over 21 months, and broader health outcomes, including physical illnesses and self-reported mental and physical health. That study supported linear rather than curvilinear stress–health relationships for both ELA and recent stress, and found that ELA moderated the association between recent stress and physical illnesses (Thoma et al. 2024a, b). Zong et al. (2025), using longitudinal data from the Survey of Health, Ageing and Retirement in Europe from adults aged 50 + in 25 European countries and Israel during the early and middle phases of the COVID-19 pandemic, found dose–response associations between childhood adversity and adverse mental health outcomes, including depression, loneliness, anxiety, and trouble sleeping. Additional moderation analyses indicated that these associations varied across outcomes by COVID-19-related adversity, COVID-19 stringency measures, living alone, and age (Zong et al. 2025). Remaining gaps and the present study Taken together, prior findings suggest that associations between ELA, later stress exposure, and mental health in older adulthood cannot be assumed to follow a single pattern. Some studies support steeling-consistent curvilinear or buffering effects in specific contexts (Höltge et al. 2019; Kok et al. 2021; Zhang 2023), whereas others support linear dose–response patterns (Thoma et al. 2024a, b; Zong et al. 2025). Importantly, much of this evidence has relied on self-report indicators of mental health, broad health outcomes, or moderation designs involving later-life stressors. It remains unclear whether proposed steeling-consistent patterns extend to clinically assessed mental disorders as outcomes in later life. The present study therefore extends prior work in two ways. First, although it uses data from the same broader longitudinal cohort as Thoma et al. (2024a, b), it addresses a distinct research question and uses different operationalizations. Whereas Thoma et al. (2024a, b) used ACEs as a brief cumulative ELA indicator and examined broader health outcomes, including self-reported mental health, the present manuscript uses the Childhood Trauma Questionnaire (CTQ; Bernstein & Fink 1998; Gast et al. 2001; Klinitzke et al. 2012) to capture childhood maltreatment severity and focuses on clinically assessed past-year DSM-5 mental health disorders assessed via structured diagnostic interview. Second, the present study explicitly compares linear dose–response and curvilinear, steeling-consistent functional forms for the association between ELA and clinically assessed mental health disorders in older adulthood. Study aims and hypotheses The present study addresses two key questions. First, to what extent is ELA linearly associated with the number of past-year mental health disorders in older adulthood? Second, is there evidence for a curvilinear association, as proposed by the Steeling Effect Theory, in which moderate levels of ELA are associated with fewer mental health disorders than minimal or high levels of ELA? We tested two competing predictions regarding the functional form of the stress–mental health association. First, in line with dose–response accounts, we expected higher levels of ELA to be associated with a greater number of past-year mental health disorders in older adults (H1). Second, in line with the Steeling Effect Theory, we tested the alternative prediction of a curvilinear relationship, such that moderate levels of ELA would be associated with fewer past-year mental health disorders, whereas both minimal and high exposure would be linked to worse outcomes (H2). Overall, the present study aims to advance understanding of how ELA relates to mental health in older adulthood by testing competing functional forms, namely linear dose–response versus curvilinear, steeling-consistent associations. Methods This study was conducted as part of the National Research Program 76 ‘Welfare and Coercion – Past, Present, and Future’, commissioned by the Swiss Federal Council in 2017 and managed by the Swiss National Science Foundation (http://www.nrp76.ch/en). The program broadly examined Swiss compulsory social measures and placements (CSMP), using various perspectives, including historical and psychological approaches. In Switzerland, CSMP refer to welfare-related coercive interventions used in the last century (up to 1981), whereby local authorities could remove minors from their families and place them in out-of-home care settings such as foster families, children’s homes/residential institutions, poor houses, or other facilities. These measures were often implemented in response to perceived social norm violations and frequently occurred under conditions that later survivors have described as highly adverse, including neglect, maltreatment, and, in some cases, child labor (e.g., “Verdingkinder”; Knüsel et al. 2024; Leuenberger and Seglias 2008). The present longitudinal study investigated “Differential aging trajectories in high-risk individuals with past experiences of early adversity” as one of 27 research proposals selected for the NRP76 program. The ethics committee of the Faculty of Arts and Social Sciences at the University of Zurich granted approval for the study’s original protocol (ID: 19.4.3), as well as its amended version (ID: 20.12.24), which included adjustments due to the COVID-19 pandemic (e.g., a three-month delay in follow-up assessments and the integration of protective measures). All participants provided written informed consent prior to participation. The findings from this project have been published in previous works (see Eising et al. 2021; Maercker et al. 2022; Pfluger et al. 2021a, b; Thoma et al. 2021a; Thoma et al. 2022; Thoma et al. 2020; Thoma et al. 2024a, b). Participants The study sample consisted of native Swiss German speakers aged 50 years or older and included two distinct groups: A risk group (RG) comprising individuals who had experienced CSMP in Switzerland for at least one year, and a control group (CG) of individuals with no history of CSMP; those affected by CSMP are considered a risk group because these enforced out-of-home placements were frequently associated with heightened exposure to early-life adversity, including neglect, maltreatment, and, in some cases, child labor (Knüsel et al. 2024; Kuhlman et al. 2013; Leuenberger and Seglias 2008). Participants were recruited between July and December 2019. Risk group participants were primarily recruited via a contact list provided by the Swiss Federal Office of Justice (2020). This list was compiled in the context of individuals’ applications for a “solidarity contribution” related to their CSMP experiences; during that process, applicants could indicate whether they were willing to be contacted for research purpose. Additional recruitment strategies included participant referrals and reaching out to publicly known figures who had previously disclosed their CSMP experiences. CG participants were recruited via public advertisements and flyers, senior networks, care homes, contacts of the study team, and through the University Research Priority Program ‘Dynamics of Healthy Aging’ (https://www.dynage.uzh.ch/en.html). Study design Individuals interested in participating contacted the study team for further details and underwent an initial screening to confirm eligibility. Inclusion criteria were age ≥ 50 years and Swiss German as native language; additional group-specific criteria were a history of CSMP in Switzerland for at least one year for the RG, and no history of CSMP for the control group. Eligible participants were invited to two baseline assessments (A1 and A2), scheduled within one week of each other (each up to two hours), conducted between July and December 2019 either at the university’s department of psychology or at participants’ homes. Prior to A1, participants received an information packet and preliminary questionnaires; written informed consent was obtained at the beginning of A1. RG participants first provided information about their CSMP experiences, after which both RG and CG participants completed structured clinical interviews to assess mental health disorders. Participants received additional questionnaires to complete at home (i.e., Home-Package I) and return at A2. Upon completing A2, which included the assessment of ELA (CTQ; see Measures), participants were compensated with 240 Swiss Francs (approximately $250) and provided with emergency contacts and psychological support resources. After baseline, participants were contacted by phone every three months over 21 months (October 2019 to July 2021) for brief interviews. Follow-up assessments (A3 and A4) were conducted from April to October 2021. For the present manuscript, A1 data were used for socio-demographics, A2 data for ELA, and A3 data for past-year DSM-5 mental health disorders (see Measures). A complete overview of all measures included in the broader NRP76 longitudinal project assessment battery is provided in Supplementary Table S1, including assessment wave/time point and whether each measure was used in the present analyses. Upon completing A4, participants received an additional reimbursement of 240 Swiss Francs and were again provided with emergency contacts and support resources. Measures Standardized self-report measures were administered in German and printed in larger fonts to facilitate readability. If necessary, interviewers assisted the participants in the case of reading or writing difficulties. The broader project included additional measures beyond the scope of this paper; accordingly, we report here only the instruments used in the current analyses (see prior cohort publications for details on the wider assessment battery). Socio-demographics A self-report questionnaire assessed basic socio-demographic information, i.e., age and gender (female, male, other) at baseline (A1). Mental health disorders Past-year mental health disorders, as classified by the DSM-5, were assessed during A3 using the German adaptation of the structured clinical interview for mental health diagnostics (DIPS; Margraf et al. 2017a, b). The study considered a range of mental health conditions, including affective disorders (e.g., bipolar disorders, dysthymia, major depressive disorder), anxiety disorders (e.g., separation anxiety, panic disorder, agoraphobia, social anxiety disorder, specific phobias, generalized anxiety disorder), obsessive–compulsive disorders, trauma- and stressor-related disorders (e.g., acute stress disorder, posttraumatic stress disorder), somatic stress disorders (e.g., somatic symptom disorder, illness anxiety disorder), and psychotic symptoms. To facilitate analysis, the total number of mental health disorders reported within the past year was calculated (total possible range = 0–20, total realized range = 0–10 [RG = 0–10, CG = 0–7]), with a higher score indicating a higher number of past-year mental health disorders. Early-life adversity (ELA) The German version of the Childhood Trauma Questionnaire (CTQ; Bernstein & Fink 1998; Gast et al. 2001; Klinitzke et al. 2012) was used at A2 to retrospectively assess trauma, maltreatment, and neglect in childhood and/or adolescence. The CTQ is comprised of 25 items across five subscales (emotional, physical, and sexual abuse, and emotional and physical neglect), as well as three validity items. The items are rated on a five-point Likert scale (1 = “never true” to 5 = “very often true”). Scores range from 5 to 25 for each subscale, with higher scores indicative of more trauma, maltreatment, and neglect in childhood and/or adolescence. In the current study, the CTQ showed good internal consistency across the subscales: Emotional abuse (α = 0.84), physical abuse (α = 0.84), sexual abuse (α = 0.96), emotional neglect (α = 0.89), and physical neglect (α = 0.80). The CTQ sum score was computed from all items (total possible range = 25–125, total realized range = 38–120 [RG = 42–120, CG = range = 38–116]), excluding the three validity items. The measure demonstrated a good internal consistency (α = 0.84). The CTQ data have previously been published in the application of different research questions, sample sizes, and group compositions (Pfluger et al. 2021; Thoma et al. 2021b). Data analysis RStudio version 4.1.3 (RStudio Team 2022) and the “missRanger” package were used for data imputation (trees = 5000; Mayer 2021). All other statistical analyses were performed using SPSS version 28.0 (IBM Corp 2021). Little’s test (Little 1988) indicated that missing values were missing completely at random (MCAR; χ2 (119, N = 255) = 120.61, p = 0.44). Missing values were imputed for cases with less than 30% missings on the main study variables (i.e., CTQ, DIPS); cases with a higher missing rate were excluded from the analysis (Wolf & Best, 2010). The percentage of missing data per variable in the final analysis was as follows: CTQ = 9.48%, DIPS = 5.2%, gender = 0%, age = 0%. Dropout analyses were conducted by comparing completers and dropouts using independent-samples t-tests (for continuous variables: age and ELA) and χ2 tests (for categorical variables: gender, group membership [RG vs. CG]). No multivariate outliers were identified using Mahalanobis distance (p > 0.01). Sample characteristics were computed using Pearson’s Chi-square test for categorical variables, independent sample t-tests for variables with normal distribution, and Mann–Whitney U tests for variables with non-normal distribution. To test the study hypotheses, the RG and CG were combined into one sample to provide the largest possible variation of adversity (as well as outcomes). This enabled the investigation of the Steeling Effect models, which require a broad range of adversity levels, including low exposure. Descriptive statistics were used to examine group differences between the RG and CG on the study variables. To test hypothesis 1 (linear, dose–response assumption) and hypothesis 2 (curvilinear assumption), the count variable ‘number of past-year mental health disorders’ was examined. It showed a highly skewed distribution of 2.03 across both groups (RG = 1.67, CG = 2.16). ELA showed a mild skew of 0.63 (RG = 0.07, CG = 1.86). Following best practices for addressing this overdispersion (e.g., Lee and Conway 2022), generalized linear models that estimated a negative binomial distribution based on a log link function were compared to a generalized linear model characterized by a Poisson regression, including the linear predictor and control variables (age and gender). The negative binomial model best fit the data: AICPoisson = 631.57, AICnegativebinomial = 616.75. Furthermore, the assumption of Poisson regression that the mean and the variance should be equal was violated, supporting the use of negative binomial models. Accordingly, a negative binomial generalized linear model was used to examine the relationships between ELA and mental health disorders. The Incidence Rate Ratio (IRR), an exponentiated coefficient analogous in form to an odds ratio but applied to count outcomes (e.g., Lee and Conway 2022) was used for interpretation and represents the proportional change in the expected count of mental health disorders associated with a one-unit change in the predictor. All models were constructed using the same procedure. First, the effects of the covariates (i.e., age, gender) on the outcome variable were examined. Second, the main predictor was added (linear model). Third, the quadratic term of the main predictor was added to investigate the curvilinear relationships (curvilinear model). Model fit was determined using the Akaike Information Criterion (AIC; Akaike 1974), with lower AIC values indicating a better fit. In the case that the curvilinear model provided a better fit than the linear model, a two-lines test was conducted to validate the curvilinear effect. This was necessary because quadratic regressions can yield false positive results when estimating non-linear effects (Simonsohn 2018). The two-lines test fits two regression lines for low versus high values of the predictor (ELA) around an empirically estimated breakpoint. A curvilinear effect is considered valid only if both regression lines are significant and have opposite directions. Additionally, exploratory analyses were conducted using the negative binomial generalized linear models for each ELA subscale to investigate how the different child maltreatment types were associated with past-year mental health disorders. Results Sample characteristics While N = 255 participants were initially included at A1, three participants (all from the RG) dropped out after A1, and an additional 25 participants (RG: n = 19; CG: n = 6) dropped out after A2. Reasons for dropout included death (RG: n = 6; CG: n = 2), worsening health conditions (RG: n = 7; CG: n = 1), excessive demand or being afraid of the testing situation (RG: n = 2, CG: n = 2), problems in recontacting the participant (RG: n = 3), and unknown or other reasons (RG: n = 4, CG: n = 1). Participants who dropped out were significantly older (t(234) = − 3.16, p = 0.002) and more likely to belong to the RG (χ2(1) = 10.07, p = 0.002) than the analytical sample. No significant differences were found for gender (χ2(1) = 0.00, p = 0.984) and ELA (t(225) = −0.78, p = 0.437). All dropout participants had more than 70% missing values across key study variables. In addition, 16 participants were excluded from the current analysis due to more than 30% missing data on at least one of the main scales (i.e. CTQ, DIPS). Reasons for high item nonresponse were: Did not fill in the corresponding questionnaire (RG: n = 2; CG: n = 3), difficulty answering the questions (RG: n = 1), and unknown (RG: n = 5; CG: n = 5). The final data set included 211 participants (Mage = 69.8 years, SDage = 10.8, 46.4% female). Table 1 presents data on sample characteristics, early-life adversity, and mental health disorders. Significant group differences were found for all types of child maltreatment: physical abuse (RG: M = 12.4; CG: M = 6.7), sexual abuse (RG: M = 10.8; CG: M = 6.6), emotional abuse (RG: M = 14.5, CG: M = 10.1), physical neglect (RG: M = 15.1, CG: M = 7.7), and emotional neglect (RG: M = 20.4; CG: M = 13.5). Table 1Sample characteristics: Socio-demographics, early-life adversity, and number of mental health disordersTotal sample (N = 211)RG (n = 100)CG (n = 111)RG vs. CGSocio-demographics:pAge (years, M (SD);age range = 50–95)69.8 (10.8)69.1 (11.8)70.4 (9.8)t(193.058) = 0.890.377Gender (female; %)46.442.050.5χ2(1, n = 211) = 1.510.219CTQ: M (SD)pSum score58.1 (22.0)73.2 (17.8)44.6 (15.7)t(209) = −12.39**Physical abuse9.4 (5.0)12.4 (5.1)6.7 (3.0)t(158.424) = −9.72**1Sexual abuse8.6 (5.8)10.8 (6.9)6.6 (3.6)t(147.021) = −5.44**1Emotional abuse12.2 (5.9)14.5 (5.6)10.1 (5.3)t(209) = −5.88**1Physical neglect11.2 (5.2)15.1 (4.4)7.7 (3.1)t(175.631) = −13.93**1Emotional neglect16.8 (6.1)20.4 (4.4)13.5 (5.5)t(205.376) = −10.24**1Number of mental health disorders:pSum score M (SD)1.3 (1.7)1.7 (2.0)0.9 (1.3)z = − 3.42**CG, control group; CTQ, childhood trauma questionnaire; M, mean; RG, risk group; SD, standard deviation; χ2, Pearson’s chi-squared test; t, two-sided t-tests comparing RG values with those of the CG; p, p-value; *, p <.01; **, p <.0011The subscales were additionally tested with the Mann–Whitney U test due to non-normal distribution in the data. Results did not change. Table 1Sample characteristics: Socio-demographics, early-life adversity, and number of mental health disordersTotal sample (N = 211)RG (n = 100)CG (n = 111)RG vs. CGSocio-demographics:pAge (years, M (SD);age range = 50–95)69.8 (10.8)69.1 (11.8)70.4 (9.8)t(193.058) = 0.890.377Gender (female; %)46.442.050.5χ2(1, n = 211) = 1.510.219CTQ: M (SD)pSum score58.1 (22.0)73.2 (17.8)44.6 (15.7)t(209) = −12.39**Physical abuse9.4 (5.0)12.4 (5.1)6.7 (3.0)t(158.424) = −9.72**1Sexual abuse8.6 (5.8)10.8 (6.9)6.6 (3.6)t(147.021) = −5.44**1Emotional abuse12.2 (5.9)14.5 (5.6)10.1 (5.3)t(209) = −5.88**1Physical neglect11.2 (5.2)15.1 (4.4)7.7 (3.1)t(175.631) = −13.93**1Emotional neglect16.8 (6.1)20.4 (4.4)13.5 (5.5)t(205.376) = −10.24**1Number of mental health disorders:pSum score M (SD)1.3 (1.7)1.7 (2.0)0.9 (1.3)z = − 3.42**CG, control group; CTQ, childhood trauma questionnaire; M, mean; RG, risk group; SD, standard deviation; χ2, Pearson’s chi-squared test; t, two-sided t-tests comparing RG values with those of the CG; p, p-value; *, p <.01; **, p <.0011The subscales were additionally tested with the Mann–Whitney U test due to non-normal distribution in the data. Results did not change. Sample characteristics: Socio-demographics, early-life adversity, and number of mental health disorders Age (years, M (SD); age range = 50–95) CG, control group; CTQ, childhood trauma questionnaire; M, mean; RG, risk group; SD, standard deviation; χ2, Pearson’s chi-squared test; t, two-sided t-tests comparing RG values with those of the CG; p, p-value; *, p <.01; **, p <.001 1The subscales were additionally tested with the Mann–Whitney U test due to non-normal distribution in the data. Results did not change. Mental health disorders Table 2 depicts the prevalence rates of past-year DSM-5 mental health disorders for the total sample, and separately for the RG and CG. On average, participants across the total sample reported 1.3 (SD = 1.7) past-year mental health disorders (RG: M = 1.7, SD = 2.0; CG: M = 0.9, SD = 1.3), with a significant difference between the groups (z = − 3.42, p < 0.001). From the total sample, 59.2% of participants presented with a past-year mental health disorder (RG: 68%; CG: 51.4%). The most prevalent mental health problems in the RG were insomnia (RG: 19.0%; CG: 9.9%), somatic stress disorder (RG: 16.0%; CG: 9.9%), and hallucinations (RG: 15.0%; CG: 4.5%). Significant group differences were found for hallucinations (χ2(1) = 6.75, p = 0.01) and hearing voices (χ2(1) = 5.69, p = 0.02). Table 2Mental health disorders (diagnosis) and group comparisonsTotal sample (N = 211)RG (n = 100)CG (n = 111)RG vs. CGPast-year diagnosis; n (%)χ2pAnxiety disordersSeparation anxiety10 (4.7)4 (4.0)6 (5.4)0.230.7511Panic disorder/attacks3 (1.4)3 (3.0)0 (0)3.380.1051Agoraphobia19 (9.0)13 (13.0)6 (5.4)3.700.054Social phobia9 (4.3)4 (4.0)5 (4.5)0.0311Specific phobia22 (10.4)11 (11.0)11 (9.9)0.070.796Generalized anxiety disorder10 (4.7)7 (7.0)3 (2.7)2.150.1971Bipolar disordersBipolar manic0 (0)0 (0)0 (0)––Bipolar hypomanic3 (1.4)1 (1)2 (1.8)0.2411Depressive disordersMajor depression16 (7.6)9 (9.0)7 (6.3)0.550.460Obsessive–compulsive disordersCompulsive thoughts0 (0)0 (0)0 (0)––Compulsive actions1 (0.5)0 (0)1 (0.9)0.9111Trauma and stress related disordersPosttraumatic Stress Disorder12 (5.7)7 (7.0)5 (4.5)0.610.434Somatic stress disordersSomatic disorder27 (12.8)16 (16.0)11 (9.9)1.750.186Hypochondria0 (0)0 (0)0 (0)––Sleep–wake disordersInsomnia30 (14.2)19 (19.0)11 (9.9)3.560.059Hypersomnia4 (1.9)2 (2.0)2 (1.8)0.0111Psychotic symptomsHallucinations20 (9.5)15 (15.0)5 (4.5)6.75**Hearing voices5 (2.4)5 (5.0)0 (0)5.69*1Delusions16 (7.6)11 (11.0)5 (4.5)3.170.075Disordered speech7 (3.3)5 (5.0)2 (1.8)1.680.2601CG, control group; PTSD, posttraumatic stress disorder; RG, risk group; χ2, Pearson’s chi-squared test; p, p-value; *, p <.05; **, p <.011 = Fisher’s exact test Table 2Mental health disorders (diagnosis) and group comparisonsTotal sample (N = 211)RG (n = 100)CG (n = 111)RG vs. CGPast-year diagnosis; n (%)χ2pAnxiety disordersSeparation anxiety10 (4.7)4 (4.0)6 (5.4)0.230.7511Panic disorder/attacks3 (1.4)3 (3.0)0 (0)3.380.1051Agoraphobia19 (9.0)13 (13.0)6 (5.4)3.700.054Social phobia9 (4.3)4 (4.0)5 (4.5)0.0311Specific phobia22 (10.4)11 (11.0)11 (9.9)0.070.796Generalized anxiety disorder10 (4.7)7 (7.0)3 (2.7)2.150.1971Bipolar disordersBipolar manic0 (0)0 (0)0 (0)––Bipolar hypomanic3 (1.4)1 (1)2 (1.8)0.2411Depressive disordersMajor depression16 (7.6)9 (9.0)7 (6.3)0.550.460Obsessive–compulsive disordersCompulsive thoughts0 (0)0 (0)0 (0)––Compulsive actions1 (0.5)0 (0)1 (0.9)0.9111Trauma and stress related disordersPosttraumatic Stress Disorder12 (5.7)7 (7.0)5 (4.5)0.610.434Somatic stress disordersSomatic disorder27 (12.8)16 (16.0)11 (9.9)1.750.186Hypochondria0 (0)0 (0)0 (0)––Sleep–wake disordersInsomnia30 (14.2)19 (19.0)11 (9.9)3.560.059Hypersomnia4 (1.9)2 (2.0)2 (1.8)0.0111Psychotic symptomsHallucinations20 (9.5)15 (15.0)5 (4.5)6.75**Hearing voices5 (2.4)5 (5.0)0 (0)5.69*1Delusions16 (7.6)11 (11.0)5 (4.5)3.170.075Disordered speech7 (3.3)5 (5.0)2 (1.8)1.680.2601CG, control group; PTSD, posttraumatic stress disorder; RG, risk group; χ2, Pearson’s chi-squared test; p, p-value; *, p <.05; **, p <.011 = Fisher’s exact test Mental health disorders (diagnosis) and group comparisons CG, control group; PTSD, posttraumatic stress disorder; RG, risk group; χ2, Pearson’s chi-squared test; p, p-value; *, p <.05; **, p <.01 1 = Fisher’s exact test Testing the linear, dose–response effects (hypothesis 1) Bivariate correlations examined the associations between the assessed variables (see Supplementary Material, Appendix A and Table S2). To test hypothesis 1, a negative binomial generalized linear model was applied (see Table 3). Results indicated that each one-point increase in ELA was associated with a 1.5% increase in the expected number of past-year mental health disorders (IRR = 1.015, p < 0.001), consistent with a linear dose–response relationship. Table 3Negative binomial generalized linear model with mental health as a log link function of linear and curvilinear ELALinear modelCurvilinear modelBSE BIRRBSE BIRRDV: DIPSConstant1.940**0.55736.9601.4760.85024.375Gender0.0030.15301.0030.0070.15281.007Age − 0.040**0.00700.961 − 0.039**0.00710.962ELA0.015**0.00351.0150.0300.02041.030ELA20.0000.00021.000AIC616.746618.220N, 211. AIC, Akaike information criterion, B, unstandardized regression coefficient; DIPS, diagnostisches Interview bei psychischen Störungen (number of past-year mental health disorders); DV, dependent variable; ELA, linear term of early-life adversity; ELA2, quadratic term of early-life adversity; IRR, incidence rate ratio; SE B, standard errors. *, p <.01; **, p <.001 Table 3Negative binomial generalized linear model with mental health as a log link function of linear and curvilinear ELALinear modelCurvilinear modelBSE BIRRBSE BIRRDV: DIPSConstant1.940**0.55736.9601.4760.85024.375Gender0.0030.15301.0030.0070.15281.007Age − 0.040**0.00700.961 − 0.039**0.00710.962ELA0.015**0.00351.0150.0300.02041.030ELA20.0000.00021.000AIC616.746618.220N, 211. AIC, Akaike information criterion, B, unstandardized regression coefficient; DIPS, diagnostisches Interview bei psychischen Störungen (number of past-year mental health disorders); DV, dependent variable; ELA, linear term of early-life adversity; ELA2, quadratic term of early-life adversity; IRR, incidence rate ratio; SE B, standard errors. *, p <.01; **, p <.001 Negative binomial generalized linear model with mental health as a log link function of linear and curvilinear ELA N, 211. AIC, Akaike information criterion, B, unstandardized regression coefficient; DIPS, diagnostisches Interview bei psychischen Störungen (number of past-year mental health disorders); DV, dependent variable; ELA, linear term of early-life adversity; ELA2, quadratic term of early-life adversity; IRR, incidence rate ratio; SE B, standard errors. *, p <.01; **, p <.001 To illustrate the practical significance of the 1.015 IRR, we compared predicted disorder counts across the possible range of the CTQ scale (25–125). A participant at the low end of the scale (Score = 25) would have a predicted rate of 0.69 disorders, whereas a participant at the high end (Score = 125) would have a predicted rate of 3.06 disorders. This represents a 4.43-fold increase in the burden of mental health disorders across the spectrum of measured stressors. These predicted differences illustrate that the apparently small per-unit IRR may translate into meaningful differences in the expected number of past-year mental health disorders across the adversity range. Exploratory analyses with the CTQ subscales (i.e., physical, sexual, emotional abuse, physical, emotional neglect) also showed positive linear, dose–response relationships between all ELA types and mental health disorders (see Supplementary Material, Appendix B). Testing the curvilinear effect (hypothesis 2) To test hypothesis 2, negative binomial generalized linear models were applied. No evidence was found for a curvilinear relationship between ELA and past-year mental health disorders (see Table 3). Discussion The overarching aim of the current study was to extend the literature on the etiological factors linked to mental health heterogeneity in older adulthood. Specifically, this study focused on the stress-outcome relationship by examining past-year mental health disorders in older adults with varied levels of ELA. The tested associations between CTQ-assessed ELA and past-year mental health disorder count were better characterized by a linear dose–response pattern than by a curvilinear, steeling-consistent pattern. With regard to hypothesis 1 and the stress-outcome relationship, the study findings were in support of linear, dose–response relationships. Specifically, the linear associations between total ELA and each ELA type and past-year mental health disorders in older adulthood, corroborate and extend previous research with younger age groups (e.g., Björkenstam et al. 2016; Rossi et al. 2021). As such, the findings indicate that with higher levels of ELA, worse mental health outcomes may be expected in older adulthood. The increased risk for psychopathology in response to ELA may be due to the sensitive nature of early-life phases with respect to physical and psychosocial development (Heim and Nemeroff 2001; Lupien et al. 2009). For example, research on minors exposed to ELA has shown a heightened risk for biopsychosocial maladaptation, such as altered sympathetic nervous system stress reactivity in adulthood (e.g., Fergusson et al. 2013; Kuras et al. 2017). Moreover, extreme, repeated, or chronic stress may become excessively demanding or overwhelming, hindering the effective application of adequate stress coping mechanisms (Liu 2015; Pfluger et al. 2021). This, in turn, can heighten vulnerability to subsequent stress experiences and increase the likelihood of the development of psychopathology (Starr et al. 2014). The linear, dose–response patterns demonstrated in the present study reinforce this notion that the accumulation of stress and adversity across the lifespan can increase the risk of mental health disorders in later life. Regarding hypothesis 2 and the Steeling Effect Theory, the results of the current study did not support a steeling-consistent curvilinear association of ELA with mental health disorders. This contrasts with findings from the survey study by Höltge et al. (2019), which observed a steeling effect on mental health in older adults exposed to moderate levels of ELA. However, Höltge et al. (2019) used a self-report screening instrument, whereas the current study employed a structured clinical interview. While clinical interviews offer greater diagnostic specificity, symptom-based measures may be more sensitive to detecting steeling-consistent patterns (e.g., symptoms rather than diagnoses; Seery et al. 2010a, b). Furthermore, although both studies assessed ELA using the CTQ, the CTQ captures comparatively severe stressors that may be less likely to yield steeling effects (Höltge et al. 2018; Liu 2015; Rutter 2012). Moreover, a central assumption of the Steeling Effect Theory is that potential benefits of moderate stress exposure depend on successful coping/mastery and subsequent adaptive learning. Because we did not assess how participants coped with or appraised their stressors (e.g., coping efficacy, perceived mastery, or growth-related responses), we could not examine whether “moderate” exposure was accompanied by successful adaptation. Thus, the absence of a curvilinear pattern may partly reflect unmeasured heterogeneity in coping processes, which should be addressed in future studies. Future studies should therefore use integrative adversity assessments that capture both severe maltreatment and non-maltreatment adversities, as well as key characteristics such as timing, chronicity, and perceived impact, to better distinguish conditions under which steeling-consistent patterns might emerge. Several additional limitations should be noted. First, ELA was assessed retrospectively using the CTQ and may therefore be affected by recall and reporting biases. Second, although the study used longitudinal cohort data, the present analyses are observational and do not allow causal conclusions about the effects of ELA on later mental health. Third, the sample was enriched for individuals with a history of CSMP and consisted of native Swiss German-speaking older adults; therefore, findings may not generalize to all older populations or to other sociocultural contexts. Fourth, dropout analyses indicated that participants who dropped out were older and more likely to belong to the risk group, which may have affected the representativeness of the analytical sample. Finally, the outcome aggregated heterogeneous mental health disorders into a single count, thereby weighting different diagnostic categories equally and capturing diagnostic breadth rather than severity, impairment, or clinical complexity. Despite these limitations, a key strength of this study lies in its comprehensive assessment of the most prevalent mental health disorders (excluding dementia) in an older adult sample with varying levels of ELA exposure. Additionally, this study uniquely integrates multiple stress models within a single research framework, allowing for a comparative evaluation of their assumptions. Older adulthood remains an underrepresented population, not only in research on stress-related effects, but also in the broader field of mental health studies. Further investigation into the relationship between stress and mental health in older adulthood is essential, given the substantial burden of mental ill-health (Prince et al. 2015), the importance of stress in the etiology and pathogenesis of mental health disorders (Kessler et al. 2010), the frequent under-identification of mental illness in older adulthood (WHO, 2025b), and ongoing population ageing worldwide (WHO, 2025a). This study underscores the importance of considering early-life adversity when addressing mental health in older adulthood. Across models, evidence supported a linear dose–response association between ELA and past-year mental health disorders, whereas support for a steeling-consistent curvilinear pattern was limited in this sample. Future research should further delineate the conditions and mechanisms under which steeling effects may emerge in later life. Supplementary Information Below is the link to the electronic supplementary material.Supplementary file1 (DOCX 19 kb) Supplementary file1 (DOCX 19 kb)