Does child abuse influence the persistence of suicidal ideation in patients with high suicide risk?

ABSTRACT

Background: Child abuse (CA) is a well-established risk factor for suicidal ideation (SI) and suicidal behaviour (SB). However, only few studies have investigated the impact of CA on the persistence of SI over time. Those existing studies have used varying operationalizations to measure the persistence of SI, leading to inconsistent findings.

Objective: The aim of this study is to examine the persistence of SI by integrating prospective ecological momentary assessments (EMAs) with retrospective structured interviews within one clinical sample.

Method: Patients admitted to psychiatric hospitals following an acute suicidal crisis (n = 75) or recent suicide attempt (n = 107) were assessed shortly after admission and then, after discharge, engaged in a 3-week EMA (EMA phase 1), followed by a 6-month EMA period (EMA phase 2). SI was measured using the Self-Injurious Thoughts and Behaviors Interview at baseline and via EMA prompts during both phases. Persistence was indicated by the number of episodes and repeated reports of SI. Multiple regression analyses were conducted to examine the influence of CA on the persistence of SI.

Results: Retrospective interview data showed that CA predicted the persistence of SI over the year before a suicidal crisis or suicide attempt (β = .22, t = 2.67, p = .008). Prospective EMA data indicated that CA was associated with greater persistence of passive SI during EMA phase 1 (β = .20, t = 2.19, p = .030) and with higher persistence of active SI in EMA phase 2 (β = .21, t = 1,99, p = .050).

Conclusions: This study shows that CA is linked to greater persistence of SI across different methods and time points. It also highlights the importance of distinguishing between passive and active SI when examining persistence.

HIGHLIGHTS

• Retrospective data showed that child abuse is associated with a higher persistence of suicidal ideation, but not with suicide plans, over the year before a suicidal crisis or suicide attempt.

• Prospective data showed that child abuse predicts a higher persistence of passive suicidal ideation in the first three weeks after discharge.

• Child abuse predicted a higher persistence of active suicidal ideation regardless of mental disorders over a six months period after discharge.

1. Introduction

Child abuse (CA) is a significant risk factor in the development of mental disorders in adulthood (Xiao et al., 2023). Moreover, empirical evidence suggests that a history of CA is associated with a 2–3-fold increased risk of both suicidal ideation (SI) and suicidal behaviour (SB) (Angelakis et al., 2019). While systematic reviews and meta-analyses have demonstrated a strong association between CA and SI as well as SB (Angelakis et al., 2019; Liu et al., 2017; Xiao et al., 2023; Zatti et al., 2017), most of these studies have focused on single assessments of SI only and thereby tended to overlook how CA may influence the persistence of SI.

The studies that have examined the impact of CA on the persistence of SI in adults have shown inconsistent findings (e.g. Bedi et al., 2011;Kivelä et al., 2019). This could be due to the fact that these studies have employed various approaches to operationalise persistence, including different reference periods and measurement time points such as retrospective assessments (e.g. Cohen et al., 2020) and prospective designs with multiple time points (e.g. Kivelä et al., 2019). In addition, studies have used different data collection methods, including single items (e.g. Cohen et al., 2020), multi item scales (e.g. Kivelä et al., 2019) and clinical interviews (e.g. Bedi et al., 2011; Mortier et al., 2025). When studies use different ways to assess persistent SI it becomes difficult to directly compare their results or draw clear conclusions about the persistence of SI. Additionally, many studies rely on participants’ memories of past experiences, which can be unreliable, especially when people are asked to recall thoughts from weeks, months or even years ago (Gratch et al., 2021). This recall bias can distort our understanding of the impact of CA on persistence of SI.

A cross-sectional online study found that CA was associated with chronic SI, as measured by a single self-report item assessing long-term suicidal thoughts (Cohen et al., 2020). Similarly, Bedi et al. (2011) showed that sexual abuse in childhood was linked to persistent SI and multiple suicide attempts, even after controlling for major depressive disorder (MDD) and posttraumatic stress disorder (PTSD; Bedi et al., 2011). In this study, persistent SI was measured using a single interview question assessing whether SI has ever persisted for at least 7 days in a row. A recent cross-sectional study in a non-clinical sample demonstrated that childhood adversities were associated with the persistence of SI and SB (Mortier et al., 2025). Specifically, emotional abuse and neglect were linked to the persistence of SI, even when controlling for mental disorders, and physical abuse, emotional abuse, and neglect were significant predictors of the persistence of suicide plans (SP). In this study, persistence was operationalised as the ratio of 12-month to lifetime prevalence, calculated separately for SI and SP. However, contrasting these findings, a 9-year prospective study found no association between childhood adversities (including child abuse and neglect) and the persistence of SI, defined as reporting SI at baseline and at two consecutive assessment points after 6 and 9 years (Kivelä et al., 2019).

Previous studies have demonstrated that the prevalence of persistent SI varied among patients with different mental disorders (e.g. Köhler-Forsberg et al., 2017; Madsen et al., 2016; Nock et al., 2018; Rath et al., 2021). Therefore, when exploring the association between CA and persistent SI, adjusting for mental disorders is essential (Bedi et al., 2011; Kivelä et al., 2019; Mortier et al., 2025). Without accounting for this, the observed association may be overstated or actually reflect effects of underlying mental disorders.

Given the inconsistent findings regarding the relationship between CA and persistent SI, this study aims to examine different operationalizations of the persistence of SI by employing different measurement time points and data collection methods in one sample. A key aspect of this study is the use of ecological momentary assessments (EMAs) alongside traditional interview data. EMA captures SI at the moment as it occurs in participants’ natural environments, reducing recall bias associated with retrospective self-reports (Gratch et al., 2021). To our knowledge, this is the first study that used both EMA and interview data to investigate the association of a history of CA and the persistence of SI. We will further investigate the role of mental disorders in the relationship between CA and the persistence of SI. Building on this, our sub-aim is to examine if CA predicts the persistence of SP and suicide intent as next steps after SI according to the ideation to action framework.

Based on the existing literature and the aims of our study, we hypothesise that CA is a significant predictor of a higher persistence of SI even when controlling for mental disorders.

2. Methods

2.1. Procedure

All participants were recruited during their inpatient psychiatric stay. Following inclusion (see Figure 1), participants underwent a comprehensive baseline assessment that included the German version of the Self-Injurious Thoughts and Behaviors Interview (SITBI-G, Fischer et al., 2014). Additionally, they completed several questionnaires including the Childhood Trauma Questionnaire (CTQ, Wingenfeld et al., 2010), being particularly relevant for the present study. The Catalyst App by Metricwire Inc. that was used for the subsequent EMA phases was installed on participants’ devices, allowing them to notify researchers of their upcoming discharge date and to receive prompts during the EMA phases. The study is divided into two EMA phases. In phase 1, participants rate SI, suicide intent and other variables at four daily semi-random-signal contingent prompts from 1 to 3 days pre-discharge through 21–24 days post-discharge. EMA phase 2 extends the assessment of EMA phase 1 over six months, with four prompts on two random days per week. A more detailed description can be found in Spangenberg et al. (2025a) and https://osf.io/axnws.

Figure 1.

Study procedure.

2.2. Measures

A comprehensive description of all study instruments can be found at https://osf.io/axnws. In this section, we will describe only those instruments relevant to the present analyses.

2.2.1. Baseline-Assessment

2.2.1.1. Childhood Trauma Questionnaire (CTQ)

The German version of the Childhood Trauma Questionnaire (CTQ; Bernstein et al., 2003; Wingenfeld et al., 2010) assesses participants’ experiences of CA. Three subscales of this retrospective self-report instrument measure three subtypes of CA: emotional abuse (EA), physical abuse (PA), and sexual abuse (SA). Each subtype is measured using five items rated on a 5-point Likert scale ranging from 1 (never true) to 5 (very often true). Our analysis focused on the total sum score of CA ranging from 15 to 75, as we examined overall exposure rather than differentiating between specific forms. The internal consistency of the total score (α = .92) and the subscales in our sample (EA: α = .91; PA: α = .87; SA: α = .93) was good to excellent and comparable to that reported in previous validation studies (EA: α = .87 to .89; PA: α = .80 to .89; SA: a = .89 to .96), including the original validation and German validation studies (Klinitzke et al., 2012; Wingenfeld et al., 2010). To determine the presence of CA in our sample, we employed the cut-off values for low to moderate severity proposed by Häuser et al. (2011). For the remaining analyses, we employed the total score of CA, as previous studies involving CA survivors have shown high rates of polyvictimization, suggesting that individuals often experienced two or more types of abuse concurrently or sequentially (Finkelhor et al., 2007; Song et al., 2022).

2.2.1.2. German version of the self-injurious thoughts and behaviors interview (SITBI-G)

The SITBI-G (Fischer et al., 2014; Nock et al., 2007) is a structured interview for the assessment of self-injurious thoughts and behaviours. In the current analysis, the SITBI-G was used to operationalise persistent SI (Item 5: ‘How many separate periods [in which you had suicidal thoughts] did you have in the past year?’) and SP (Item 18: ‘How many separate periods [in which you made a suicide plan] did you have in the past year?’) at the baseline assessment. The German version of the SITBI-G demonstrated good interrater reliability, retest reliability, and construct validity (Fischer et al., 2014).

2.2.1.3. Mental disorders

Clinical diagnoses were extracted from medical records at the psychiatric hospitals in the baseline assessment. The following ICD-10 diagnostic categories were represented in the sample: F1: Mental and behavioural disorders due to psychoactive substance use (e.g. alcohol dependence), F2: Schizophrenia, schizotypal and delusional disorders (e.g. paranoid schizophrenia), F3: Mood (affective) disorders (e.g. depression), F4: Neurotic, stress-related and somatoform disorders (e.g. panic disorder, post-traumatic stress disorder), F5: Behavioural syndromes associated with physiological disturbances and physical factors (e.g. anorexia nervosa), F6: Disorders of adult personality and behaviour (e.g. borderline personality disorder), and F9: Behavioural and emotional disorders with onset usually occurring in childhood and adolescence (e.g. attention-deficit/hyperactivity disorder). Mental disorders were operationalised as having at least one diagnosis from the specified diagnostic categories, instead of focusing on specific individual diagnoses due to high comorbidity rates of mental disorders in the population under study and our transdiagnostic approach. These variables were coded dichotomously, with individuals classified as either having at least one diagnosis within the specified categories or having no diagnosis in these categories.

2.2.2. Ecological momentary assessment (EMA)

EMA was used to assess active and passive SI as well as suicide intent. Participants received prompts at semi-random times between 8:00am and 10:00pm, with a minimum two-hour interval between prompts. Each of the four daily EMA prompts included three items assessing SI and suicide intent:

1. ‘At the moment, I would like to be dead’ (passive SI)

2. ‘At the moment, I'm thinking about taking my life’ (active SI)

3. ‘At the moment, I have the intention to take my own life’ (suicide intent)

Participants rated these statements on a scale from 0 (not at all) to 4 (very strong). For each participant, the average intensity of active SI, passive SI, and suicide intent was calculated separately for each EMA phase.

For both EMA phases, we measured the persistence by calculating, for each participant, the number of prompts with ratings above zero relative to the total number of responses, i.e. persistence of SI was operationalised as the proportion of EMA prompts where SI was reported. This was done separately for active SI, passive SI, and suicide intent. Each rating greater than zero represents the presence of SI at the respective time point. The resulting persistence scores range from 0 to 1, which reflects the percentage of all prompts where SI was reported. For example, a score of 0.3 indicates that SI was reported on 30% of prompts.

Previous research has found no evidence that assessing suicidal thoughts and behaviors (STB) through EMA triggers reactivity in STBs (Chen et al., 2024; Kivelä et al., 2022). However, a few recent studies have reported worsening of STB in a small group of participants (Kivelä et al., 2024: 10–18%; Spangenberg et al., 2025a: 5–6%). Consistent with this evidence, we consider the ethical risk of our design to be minimal but acknowledge the importance of ongoing monitoring of participants’ well-being and transparent reporting of potential adverse effects in EMA research with high-risk populations. Therefore, the study’s safety protocol ensured that emergency contact information for suicidal crises was continuously available within the app, with reminders sent to participants daily during EMA phase 1 and weekly during EMA phase 2.

2.3. Sample

This study is based on data from a larger multicenter prospective project investigating the predictive power of three theoretical approaches on the development of STB. The projects’ procedures, main research questions and intended analytical methods were preregistered prior data collection (https://osf.io/axnws). In the present paper, preliminary data of participants that were recruited between March 2022 and February 2024 in eight psychiatric wards at two study sites are analyzed. The study received approval from the responsible ethical review boards (Leipzig no. 382/17-ek and Duisburg-Essen no. EA-PSY19/23/03102023) and conducted in accordance with the Declaration of Helsinki. All participants were informed that the study was observational, that study data would not be actively monitored, and that no clinical intervention would be provided; informed consent was obtained prior to participation. Of 664 eligible patients, n = 184 were included in the study after giving written informed consent (see https://osf.io/axnws for the study flow and inclusion and exclusion criteria). Of these, n = 107 (58.8%) were admitted to a psychiatric hospital due to an acute suicidal crisis and n = 75 (41.2%) due to a recent suicide attempt. Due to the withdrawal of consent, data of one participant were deleted, resulting in a final sample of 183 participants. A total of 12 participants were excluded from analyses because CTQ scores were completely missing, yielding a final analytic sample of 171 participants.

The mean age of the total sample was 37.0 years (SD = 14.0; ranged from 18 to 77 years). Half of the sample (50.9%, n = 87) identified as male, 46.8% (n = 80) as female, and 2.3% (n = 4) as other (non-binary). The lifetime prevalence of at least one suicide attempt was 71.3% (n = 122), with 45.0% (n = 77) reporting multiple suicide attempts (ranging from 2 to 20). An overview of the sample description for the three subsamples (Interview, EMA phase 1, and EMA phase 2) is shown in Table S1.

The most common diagnoses (including comorbidity) according to the International Statistical Classification of Diseases and Related Health Problems (ICD-10; World Health Organization, 1992) were affective disorders (F3: n = 128; 74.9%), mental and behavioural disorders due to psychoactive substance use (F1: n = 59; 34.5%), neurotic, stress-related and somatoform disorders (F4: n = 51, 29.8%), and personality disorders (F6, n = 44, 25.7%). Over half of the inpatients (55.0%, n = 94) had multiple diagnoses.

By February 2024, a total of 137 participants from the included cohort of 171 initiated the EMA phase 1 with an average compliance rate of 48.2% (range 0% to 100%, SD = 28.7), leading to the collection of 6,061 valid EMA surveys. Subsequently, 126 participants proceeded to EMA phase 2, where the average compliance rate was 30.5% (range 0% to 100%, SD = 28.5), yielding 6,387 valid observations. Results from correlation analyses (Table S2) suggest small associations between compliance rates and diagnostic groups (EMA phase 1: F4, EMA phase 2: F3, F4), but no correlation was observed with regard to CA severity, gender, or age.

2.4. Statistical analyses

All statistical analyses were performed using SPSS version 29 (IBM, 2023). A sensitivity analysis was conducted using G*Power 3.1.9.7 (Faul et al., 2007). For the three subsamples (n₁ = 171, n₂ = 137, and n₃ = 126), the analyses showed that in multiple regression models including gender and CA, the study was sufficiently powered (1 – β = .80, α = .05) to detect effects of at least f₁² = .06, f₂² = .07, and f₃² = .08, respectively. When mental disorders were additionally included as predictors, the study retained sufficient power to detect effects of at least f₁² = 08, f₂² = .10, and f₃² = .11 in the three subsamples.

Since participants were allowed to skip CTQ-items if they found it too burdensome, some responses were missing. After excluding 12 participants without CTQ data, the remaining sample had 3.20% missing CTQ data points (n = 82). Little’s MCAR test was statistically significant (p = .041), indicating that data were not missing completely at random. Missing values were handled using multiple imputation with five imputed datasets, generating plausible estimates based on observed data patterns. For scale scores, means were calculated within each imputed dataset and then averaged to generate the final scale score.

Participants’ demographic and clinical characteristics were described with mean, standard deviation (SD), number (n), and percentage (%). Given the gender differences observed in self-reported prevalence rates of CA (Bodkin et al., 2019), we analyzed gender differences in study variables using t-tests and chi-square-tests due to different scale levels of the study variables.

Multiple linear regression analyses were conducted to examine the relationship between CA as a predictor variable and the persistence of SI as well as SP assessed at baseline, and the persistence of (active and passive) SI and suicide intent measured across EMA, with both EMA phases analyzed separately. To assess the additional variance in persistent SI explained by mental disorders beyond CA, the four most prevalent diagnostic categories (F1, F3, F4, and F6) were included in the models. The assessment of multicollinearity was conducted using variance inflation factors (VIF). All VIF values ranged from 1.07 to 1.33 in the interview data and from 1.04 to 1.32 in the EMA data indicating that multicollinearity was not a concern. Due to observed gender differences in self-reported CA prevalence (Bodkin et al., 2019), we included gender as a covariate in all models.

3. Results

3.1. Study sample

One hundred twenty patients (71.3%) reported at least one lifetime suicide attempt at baseline, while the number of attempts varied between 0 and 20 (M = 2.5, SD = 3.6). Regarding CA, our findings indicate that 78.4% (n = 134) of the participants exceeded the cut-off value for slight to moderate EA, 44.4% (n = 76) for PA, and 42.7% (n = 73) for SA. Based on the cut-off values, 131 patients (82.5%) reported experiencing at least one form of abuse on an occasional or more frequent basis. Half of the participants (n = 99, 55.8%) exceeded the slight-to-moderate cut-off threshold in at least two CA types. Female participants reported significantly higher item means of EA and SA as well as a significantly higher CA total score than men (Table 1).

Table 1.

Means, standard deviations, and gender differences of the study variables.

	Total			Women			Men			Gender differences
Child abuse	N	M	SD	n	M	SD	n	M	SD	t
Total score	167	31.3	13.18	80	34.74	14.14	87	27.6	11.3	−3.58***
Emotional abuse	167	14.73	6.63	80	16.48	6.72	87	12.76	5.96	−3.77***
Physical abuse	167	8.98	5.15	80	9.08	5.33	87	8.82	5.09	−0.32
Sexual abuse	167	7.59	4.47	80	9.19	5.58	87	6.03	2.29	−4.72***
Interview
Persistence of SI	163	8.37	29.01	78	7.18	19.17	85	9.47	35.82	0.5
Persistence of SP	162	3.46	13.77	76	5.71	19.67	86	1.47	3.02	−1.86
EMA 1
Prompts with …
Passive SI	129	0.44	0.42	70	0.5	0.42	59	0.38	0.42	−1.6
Active SI	129	0.32	0.34	70	0.4	0.39	59	0.24	0.33	−2.66**
Suicide intent	129	0.25	0.34	70	0.32	0.36	59	0.15	0.28	−3.03**
Aggregated average intensity of …
Passive SI	129	0.92	1.03	70	1.11	1.12	59	0.69	0.87	−2.36*
Active SI	129	0.62	0.84	70	0.82	0.95	59	0.39	0.61	−3.13**
Suicide intent	129	0.4	0.66	70	0.56	0.78	59	0.22	0.43	−3.12**
EMA 2
Prompts with …
Passive SI	96	0.47	0.41	56	0.53	0.42	40	0.39	0.4	−1.63
Active SI	96	0.42	0.4	56	0.5	0.42	40	0.3	0.35	−2.41*
Suicide intent	96	0.31	0.39	56	0.41	0.43	40	0.17	0.28	−3.32**
Aggregated average intensity of …
Passive SI	96	1.08	1.13	56	1.3	1.22	40	0.78	0.92	−2.39*
Active SI	96	0.81	0.98	56	1.07	1.12	40	0.45	0.58	−3.53***
Suicide intent	96	0.53	0.78	56	0.76	0.91	40	0.21	0.34	−4.16***
Diagnoses		%			%			%		φ
F1	165	99.8		24	41.4		34	58.6		−0.09
F3	165	99.8		60	48		65	52		0.03
F4	165	99.8		33	66		17	34		0.25**
F6	165	99.8		31	73.8		11	26.2		0.31***

Table 1 presents an overview of the descriptive statistics, including means and standard deviations, for all study variables. The statistics are reported for the entire sample and stratified by gender. Because of the small group size (n = 4), non-binary participants were excluded from the analyses.

3.2. Interview

Multiple regression analyses demonstrated that CA predicted persistent SI, based on interview data, even when controlling for mental disorders (Table 2), but did not predict persistent SP (Table 3). After accounting for mental disorders, the overall models became non-significant, therefore the findings require cautious interpretation.

Table 2.

Multiple linear regression analyses predicting persistent SI based on interview data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .05
	Intercept	−4.01	5.92		−0.68	.500	−15.70	7.69
	CA	0.49	0.18	0.22	2.67	.008	0.13	0.85
	gender	−5.52	4.70	−0.10	−1.218	.240	−14.76	3.72
2									R2 = .07
	Intercept	1.16	8.47		0.14	.891	−15.56	17.88
	CA	0.48	0.19	0.21	2.59	.011	0.11	0.85
	gender	−3.08	5.02	−0.05	−0.61	.540	−12.99	6.83
	F1	5.42	4.88	0.09	1.11	.269	−4.22	15.05
	F3	−7.54	6.01	−0.11	−1.25	.212	−19.41	4.34
	F4	−8.44	5.64	−0.13	−1.50	.136	−19.57	2.70
	F6	0.11	5.74	0.00	0.02	.985	−11.24	11.45

Note: N = 161 complete cases, Step 1: F(2,158) = 3.71; p = .027, Step 2: F(6,154) = 1.95, p = .077. CA = Child abuse as a total score, SI = suicidal ideation, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Table 3.

Multiple linear regression analyses predicting persistent SP based on interview data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .03
	Intercept	−1.47	2.81		−0.52	.603	−7.01	4.08
	CA	0.11	0.09	0.10	1.22	.223	−0.06	0.28
	gender	3.62	2.23	0.13	1.62	.107	−0.8	8.03
2									R2 = .07
	Intercept	−4.01	4.04		−0.99	.323	−11.99	3.98
	CA	0.12	0.09	0.11	1.33	.185	−0.06	0.29
	gender	3.37	2.39	0.12	1.41	.160	−1.34	8.09
	F1	0.5	2.33	0.02	0.21	.831	−4.10	5.10
	F3	3.1	2.89	0.10	1.07	.285	−2.61	8.80
	F4	−3.64	2.68	−0.12	−1.36	.176	−8.93	1.65
	F6	3.33	2.74	0.10	1.22	.226	−2.08	8.74

Note: N = 161 complete cases, Step 1: F(2,158) = 2.76, p = .067; Step 2: F(6,154) = 1.84, p = .095. CA = Child abuse as a total score, SP = suicide plan, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Exploratory analyses of the different CA subtypes showed that EA and SA were significant predictors of the persistence of SI over the year before a suicidal crisis or suicide attempt, whereas PA was not (Table S3(a)). For persistence of SP, only SA was a significant predictor. No significant effects were found for EA or PA (Table S3(b))

3.3. EMA 1

As shown in Table 4, multiple regression analyses indicated that CA is a significant predictor of persistent passive SI during EMA phase 1, but this effect was no longer significant when controlling for mental disorders. In contrast, no significant predictive effects were observed for persistent active SI (Table 5) or persistent suicide intent (Table 6). After controlling for mental disorders, all overall models were no longer statistically significant, and therefore, these results should be interpreted carefully.

Table 4.

Multiple linear regression analyses predicting persistent passive SI based on EMA 1 data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .06
	Intercept	0.20	0.10		2.04	.043	0.01	0.39
	CA	0.01	0.00	0.20	2.19	.030	0.00	0.01
	gender	0.08	0.07	0.09	1.03	.306	−0.07	0.23
2									R2 = .07
	Intercept	0.21	0.13		1.59	.114	−0.05	0.48
	CA	0.01	0.00	0.18	1.95	.053	0.00	0.01
	gender	0.05	0.08	0.06	0.64	.525	−0.11	0.21
	F1	−0.05	0.08	−0.06	−0.64	.525	−0.20	0.10
	F3	0.02	0.09	0.02	0.17	.865	−0.17	0.20
	F4	−0.03	0.09	−0.03	−0.33	.742	−0.20	0.14
	F6	0.11	0.09	0.12	1.25	.214	−0.06	0.29

Note: N = 129 complete cases, Step 1: F(2,126) = 3.73, p = .027; Step 2: F(6,122) = 1.54, p = .170. CA = Child abuse as a total score, SI = suicidal ideation, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Table 5.

Multiple linear regression analyses predicting persistent active SI based on EMA 1 data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .07
	Intercept	0.12	0.09		1.39	.166	−0.05	0.29
	CA	0.00	0.00	0.14	1.59	.114	0.00	0.01
	gender	0.14	0.07	0.19	2.16	.033	0.01	0.27
2									R2 = .11
	Intercept	0.09	0.12		0.74	.462	−0.15	0.32
	CA	0.00	0.00	0.14	1.52	.132	0.00	0.01
	gender	0.11	0.07	0.14	1.52	.130	−0.03	0.25
	F1	−0.07	0.07	−0.09	−1.04	.302	−0.2	0.06
	F3	0.08	0.08	0.09	1.00	.320	−0.08	0.24
	F4	−0.05	0.08	−0.06	−0.59	.559	−0.2	0.11
	F6	0.12	0.08	0.14	1.55	.124	−0.03	0.27

Note: N = 129 complete cases, Step 1: F(2,126) = 4.75, p = .010; Step 2: F(6,122) = 2.39, p = .032.

CA = Child abuse as a total score, SI = suicidal ideation, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Table 6.

Multiple linear regression analyses predicting persistent suicide intent based on EMA 1 data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .08
	Intercept	0.05	0.08		0.69	.493	−0.10	0.21
	CA	0.00	0.00	0.13	1.51	.134	0.00	0.01
	gender	0.15	0.06	0.22	2.51	.013	0.03	0.27
2									R2 = .10
	Intercept	0.01	0.11		0.13	.900	−0.20	0.22
	CA	0.00	0.00	0.14	1.5	.136	0.00	0.01
	gender	0.13	0.06	0.20	2.08	.040	0.01	0.26
	F1	−0.01	0.06	−0.02	−0.23	.818	−0.14	0.11
	F3	0.06	0.07	0.08	0.79	.429	−0.09	0.21
	F4	−0.04	0.07	−0.06	−0.58	.562	−0.18	0.10
	F6	0.06	0.07	0.08	0.91	.366	−0.08	0.20

Note: N = 129 complete cases, Step 1: F(2,126) = 5.59, p = .005; Step 2: F(2,126) = 2.17,p = .050.

CA = Child abuse as a total score, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Exploratory analyses showed that, EA and SA significantly predicted persistent passive SI in EMA phase 1, whereas PA did not (Table S4(a)). For both persistent active SI and persistent suicide intent, SA was the only significant predictor, with EA and PA showing no significant effects (Table S4(b and c)).

3.4. EMA 2

Results from the multiple regression analyses showed that CA was a significant predictor of persistent active SI during EMA phase 2, also when accounting for mental disorders (Table 7). In contrast, no such predictive effects were found for persistent passive SI (Table 8) or persistent suicide intent (Table 9). Once mental disorders were controlled for, all overall models lost statistical significance, implying that these findings should be interpreted with care.

Table 7.

Multiple linear regression analyses predicting persistent active SI based on EMA 2 data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .09
	Intercept	0.13	0.11		1.22	.226	−0.08	0.34
	CA	0.01	0.00	0.21	1.99	.050	0.00	0.01
	gender	0.14	0.08	0.17	1.68	.095	−0.03	0.31
2									R2 = .14
	Intercept	0.03	0.15		0.22	.829	−0.27	0.33
	CA	0.01	0.00	0.22	2.06	.042	0.00	0.01
	gender	0.07	0.09	0.08	0.74	.458	−0.11	0.25
	F1	−0.13	0.09	−0.15	−1.47	.146	−0.3	0.04
	F3	0.14	0.11	0.14	1.24	.219	−0.08	0.37
	F4	0.01	0.1	0.01	0.1	.918	−0.19	0.21
	F6	0.16	0.1	0.18	1.63	.107	−0.03	0.35

Note: N = 96 complete cases, Step 1: F(2,93) = 4.79, p = .010; Step 2: F(6,89) = 2.40, p = .034.

CA = Child abuse as a total score, SI = suicidal ideation, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Table 8.

Multiple linear regression analyses predicting persistent passive SI based on EMA 2 data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .05
	Intercept	0.26	0.11		2.29	.024	0.03	0.48
	CA	0.00	0.00	0.16	1.49	.139	0.00	0.01
	gender	0.10	0.09	0.12	1.13	.261	−0.08	0.27
2									R2 = .10
	Intercept	0.14	0.16		0.86	.392	−0.18	0.45
	CA	0.01	0.00	0.19	1.69	.095	0.00	0.01
	gender	0.02	0.10	0.03	0.24	.807	−0.17	0.21
	F1	−0.13	0.09	−0.15	−1.49	.139	−0.31	0.04
	F3	0.17	0.12	0.16	1.40	.166	−0.07	0.40
	F4	−0.01	0.11	−0.01	−0.09	.931	−0.22	0.20
	F6	0.16	0.10	0.18	1.55	.124	−0.04	0.36

Note: N = 96 complete cases, Step 1: F(2,93) = 2.46, p = .091; Step 2: F(6,89) = 1.66, p = .141. CA = Child abuse as a total score, SI = suicidal ideation, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

Table 9.

Multiple linear regression analyses predicting persistent suicide intent based on EMA 2 data.

							95% CI
Step		b	SE	β	t	p	LL	UL
1									R2 = .10
	Intercept	0.07	0.10		0.66	.510	−0.14	0.27
	CA	0.00	0.00	0.12	1.16	.247	0.00	0.01
	gender	0.21	0.08	0.27	2.64	.010	0.05	0.37
2									R2 = .13
	Intercept	0.02	0.15		0.12	.904	−0.27	0.31
	CA	0.00	0.00	0.14	1.3	.198	0.00	0.01
	gender	0.17	0.09	0.22	1.92	.059	−0.01	0.35
	F1	−0.07	0.08	−0.08	−0.83	.408	−0.23	0.10
	F3	0.07	0.11	0.07	0.63	.534	−0.15	0.29
	F4	−0.06	0.10	−0.07	−0.64	.525	−0.26	0.13
	F6	0.12	0.09	0.14	1.29	.202	−0.07	0.30

Note: N = 96 complete cases, Step 1: F(2,93) = 5.53, p = .005; Step 2: F(6,89) = 2.24,p = .047.

CA = Child abuse as a total score, F1 = Mental and behavioural disorders due to psychoactive substance use, F3 = Mood (affective) disorders, F4 = Neurotic, stress-related and somatoform disorders, F6 = Disorders of adult personality and behaviour, b = regression coefficient; SE = standard error, CI = confidence interval, LL = lower level, UL = upper level. Significant effects are bold.

In exploratory analyses, EA significantly predicted persistent passive SI in the EMA phase 2, with PA and SA showing no significant effects (Table S5(a)). Both EA and SA significantly predicted persistent active SI, whereas PA did not (Table S5(b)). No CA subtype significantly predicted persistent suicide intent (Table S5(c)).

4. Discussion

To investigate the relationship between CA and the persistence of SI, this study used data from assessments conducted with patients, who were admitted to psychiatric hospitals due to an acute suicidal crisis or following a suicide attempt, both during inpatient treatment and after hospital discharge. By integrating findings from retrospective interview data with prospective EMA data, we examined how CA influence the persistence of SI across different data measurement methods and time periods.

Our findings demonstrate that CA predicted persistence of SI over the year before a suicidal crisis or suicide attempt based on retrospective interview data (Table 4), persistence of passive SI based on EMA phase 1 data (Table 8), and persistence of active SI based on EMA phase 2 data (Table 7). Associations with persistence of SP and suicide intent were not significant.

In line with Cohen et al. (2020) CA was associated with the persistence of SI in the past year. Exploratory analyses of CA subtypes showed that EA and SA were significant predictors of the persistence of SI over the year before a suicidal crisis or suicide attempt, whereas PA was not. These findings partially align with those of Mortier et al. (2025), who reported that EA was linked to the persistence of SI, even when controlling for mental disorders. Similarly, Bedi et al. (2011) found an association between SA and persistent SI, although their definition of persistence was limited to sustained ideation over one week rather than an entire year.

In the present study, CA did not predict the persistence of SP over the year before a suicidal crisis or suicide attempt. When examining the subtypes separately, only SA emerged as a significant predictor of persistence, while no significant effects were found for either EA or PA. This finding stands in contrast to the results reported by Mortier et al. (2025), who identified both PA and EA as significant predictors of persistent SP. One reason for the discrepant results may lie in the characteristics of the samples, as Mortier et al. (2025) examined first year university students, whereas our study examined a high-risk clinical sample. Another important factor is how SP was operationalised: in their study both explicitly planning and more general thoughts on methods of suicide were coded as a SP, while in the current study, we required the report of an actual, concrete plan, which may explain why the patterns of predictors did not fully align.

Within the first three weeks after discharge, CA significantly predicted a higher persistence of passive, but not active SI or suicide intent. This suggests that, directly after the inpatient treatment, the lasting effects of CA are more likely to be associated with passive thoughts of not wanting to live, rather than with concrete active thoughts or the intent to die. However, when controlling for mental disorders, this effect was no longer significant, indicating that the association between CA and the persistence of passive SI might be partly explained by underlying mental disorders. In the first weeks after discharge, potentially protective factors such as clinical stabilisation and increased social support may buffer against the progression to active SI or suicide intent. Nevertheless, passive SI may persist in individuals with a history of CA and appears more resistant to immediate change. A recent study demonstrated that individuals with a history of child maltreatment (CM) are more likely to experience passive SI, and that CM also serves as a significant predictor of passive SI (Olgiati et al., 2025). This study also found that the lack of remission of passive SI after a treatment is associated with factors like lifetime STB and insomnia, which are more prevalent in people with a history of CA (Angelakis et al., 2019; Reffi et al., 2022).

According to the Interpersonal Theory of Suicide (IPTS), thwarted belongingness (TB) and perceived burdensomeness (PB) are proximal drivers of passive SI (Joiner, 2005; van Orden et al., 2010). TB is seen as a psychologically painful state characterised by the frustration of the fundamental need for social connection, reflected in feelings of loneliness and the absence of reciprocal care (van Orden et al., 2010; van Orden et al., 2012). PB is defined as a mental state marked by self-hatred and the belief that one is a burden to others (van Orden et al., 2010, 2012). Various studies have found heterogeneous associations between CA and the two IPTS components TB and PB (e.g. Chalker et al., 2022; Schönfelder et al., 2019; Walker et al., 2024). A recent study showed that chronic victimisation in childhood was associated with negative self – and world-related beliefs, which in turn predicted higher PB and TB (Walker et al., 2024). Specifically, self-beliefs mediated the relationship between CA and PB, while world beliefs mediated the link between CA and TB. These findings imply that changes in basic views of oneself and the world can shape how survivors of CA experience feelings of burden and social connection (Walker et al., 2024), which may also help explain why they show greater persistence of passive SI even shortly after inpatient treatment.

Over a longer period of six to seven months post-discharge, the predictive role of CA shifted, such that CA was associated with a higher persistence of active SI, even when controlling for mental disorders. Neither the persistence of passive SI nor suicide intent was significantly predicted at this later phase. Our findings contrast with, to our knowledge, the only existing prospective study investigating the influence of CA on the persistence of SI, which found no effect of CA on persistent SI (Kivelä et al., 2019). However, this study employed a longer follow-up period of 3–6 years, suggesting that the effects of CA may change, when considering a longer period of time.

The persistence of active SI over an extended period following discharge may be explained by the intensification of psychosocial and emotional stressors that were less apparent or more effectively managed during the inpatient treatment and the initial weeks after discharge. This is particularly relevant given that CA often disrupt the formation of secure attachment bonds (e.g. Ritz et al., 2025; Shahab et al., 2025) and may foster maladaptive (interpersonal) schemas (Aloi et al., 2025; Estévez et al., 2017), which can exacerbate vulnerability to such stressors. These enduring patterns in individuals with a history of CA may lead to difficulties in interpersonal functioning (Aloi et al., 2025) and emotion regulation (Barlow et al., 2017; Racine & Wildes, 2015), which can in turn mediate the link to SI (Roley-Roberts et al., 2023). This may not only increase the likelihood of SI, but also foster the use of SI as a recurring maladaptive strategy to cope with distress. This aligns with findings from a recent study showing that individuals with persistent SI exhibited significantly higher levels of suicide-specific rumination and experienced greater comfort from suicidal thoughts compared to participants with non-persistent SI (Hensel & Teismann, 2025). Moreover, the link can also be understood within the framework of suicide-related metacognitive processes, which involve how individuals perceive and interpret the meaning, control, and role of SI (Forkmann et al., 2023). Such metacognitions may therefore play a role in explaining the persistence of SI, particularly in individuals with a history of CA, although further research is needed to clarify if and how these processes are involved. Overall, it should be noted that although suicide theories often mention child abuse as a risk factor for STB (e.g. IPTS (Joiner, 2005), integrated motivational-volitional model (O'Connor and Kirtley, 2018)), there is no established theory that explicates the specific mechanisms linking child abuse to (persistent) STB.

In summary, this study demonstrated that CA is associated with a higher persistence of SI over the year before a suicidal crisis or suicide attempt, as measured retrospectively through an interview, as well as with a higher persistence of SI assessed prospectively over three weeks and six months after inpatient treatment via EMA. However, differences emerged depending on the operationalisation of SI, particularly between passive and active SI, which in turn also led to variations in the effects of CA across different measurement time points. This finding is consistent with previous research indicating that passive and active ideation represent conceptually distinct types of SI, although they remain highly correlated (Spangenberg et al., 2025b; Wastler et al., 2023). Overall, both retrospective and prospective data collection methods were able to demonstrate the impact of CA on the persistence of SI, while the specific effects varied depending on the operationalisation of SI and the subtype of CA considered. Nevertheless, prospective EMA appears to provide a more accurate assessment of the persistence of SI compared to retrospective interviews. This may be attributed to EMA’s higher temporal resolution (Kleiman et al., 2023) and its capacity to reduce recall bias (Gratch et al., 2021).

Building on these findings, future research should investigate whether a higher persistence of SI leads also to greater risk for suicide attempts, as indicated by previous findings showing an association between persistent SI and multiple attempts among individuals with a history of SA (Bedi et al., 2011). Moreover, examining the role of suicide-related metacognitive processes, particularly as potential coping mechanisms in stressful situations, may further advance our understanding of mechanisms that sustain SI over time. Clinically, while the observed effects of CA on the persistence of SI are small, they may still provide valuable insights into identifying individuals at increased risk for persistent SI. Routine assessment of CA in clinical practice may help clinicians detect risk profiles associated with a higher persistence of SI. However, since the mechanisms underlying the persistence of SI are not yet fully understood and require further investigation – especially in individuals with a history of CA, an elevated risk alone does not warrant specific intervention strategies at this stage.

Several limitations of the present study should be acknowledged. First, mental disorder diagnoses were collected at the beginning of the inpatient treatment, and it is possible that diagnostic status may have changed subsequently, potentially influencing the outcomes. Second, the assessment of the number of episodes in the SITBI-G (Fischer et al., 2014) presents a methodological limitation of the instrument itself, as the absence of a standardised definition of episodes may affect participants’ interpretation of when one episode ends and another begins, which could bias estimates of both duration and frequency of episodes and, in turn, the operationalisation of persistence. In addition, the SITBI-G does not differentiate between passive and active SI and refers to a comprehensive assessment of SI. Third, our findings should be interpreted in light of ongoing conceptual and methodological challenges regarding the definition of the persistence of SI (Spangenberg et al., 2025c). Although our operationalisation reflects overall frequency rather than uninterrupted continuity, it offers a meaningful extension beyond retrospective or low-frequency assessments, which likewise cannot determine if SI occurred consistently. In line with these difficulties, a recent review underscored the lack of a consistent conceptualisation and operationalisation of persistent SI (Spangenberg et al., 2025c), highlighting the need for a standardised definition and therefore greater consistency across studies examining persistent SI, since different operationalizations may capture distinct aspects of SI patterns and could lead to varying influences on observed relationships. By using EMA data, we aimed to contribute to this ongoing discussion and illustrate how high-frequency assessments can provide additional insights into the persistence of SI. Nevertheless, future research should work toward developing a clearer and more standardised definition of persistence of SI. Moreover, the explained variance in the analyses conducted in this study was small. As Franklin et al. (2017) emphasised, predicting suicidality remains a significant challenge, since several factors influence SI, resulting in low variance explained for a single factor. Nevertheless, the patterns of persistence observed in individuals with a history of CA provide insights into the course of SI in this specific sample and potential mediating factors should be further investigated, as they may increase the explained variance. Furthermore, participants could skip CTQ items if they found them too burdensome, leading to some missing data, which may have underestimated the prevalence of CA and potentially biased the results toward lower reported rates. Overestimation is equally plausible if participants with lower CA exposure skipped more items. However, only 6.6% of the sample was excluded due to completely missing CTQ data, suggesting that substantial bias unlikely. Despite this low risk, future replications will clarify the generalizability of these findings. Additionally, the sample consisted of individuals hospitalised for suicide attempt or acute suicidal crisis, which may limit generalizability to lower-risk populations. Finally, the study focuses on persistence of SI and does not capture trajectories or transitions, such as progressions from passive to active SI. Still, findings indicated that both frequency and mean intensity of (passive and active) SI and suicide intent increased between EMA phase 1 and 2.

In conclusion, this study demonstrated that CA is associated with a higher persistence of SI as captured through different methods of data collection and across varying measurement time points. Yet, the findings also revealed that results depend on the operationalisation of SI, underlining the importance of distinguishing between passive and active SI when examining persistence.

Consent to participate

All participants provided written informed consent prior study participation.

Declaration of generative AI

During the preparation of this work the author used Perplexity.ai to improve language and readability of the work. After using this tool, the author reviewed and edited the content as needed and takes full responsibility for the content of the publication.

Funding Statement

This work was supported by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG) grant number FO 784/8-1, GL 818/8-1, SP 1556/5-1.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Ethical considerations

The ethic committees approved the study protocol (University of Duisburg Essen no. EA-PSY19/23/03102023, University of Leipzig: no. 382/17-ek).

Data availability statement

The data that support the findings of this study are available from the corresponding author, JS, upon reasonable request.

Supplemental Material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/20008066.2026.2639279.