Moral injury in psychiatric patients with personality and other clinical disorders: development, psychometric properties, and validity of the Moral Injury Events Scale–Civilian Version

ABSTRACT

Background: Moral injury emerges when someone perpetrates, fails to prevent, or witnesses acts that violate their own moral or ethical code. Nash et al. [(2013). Psychometric evaluation of the moral injury events scale. Military Medicine, 178(6), 646–652] developed a short measure, the Moral Injury Events Scale (MIES) to facilitate the empirical study of moral injury in the military. Our study aimed to develop a civilian version of the measure (MIES–CV) and examine its psychometric properties in a sample of psychiatric inpatients .

Methods: In this cross-sectional study, the sample comprised 240 adult patients (71.7% female) with a mean age of 31.57 (SD = 11.69). The most common diagnoses in the sample were anxiety disorders (58.3%), depressive disorders (53.8%), and borderline personality disorder (39.6%). Participants were diagnosed using structured clinical interviews and filled out psychological questionnaires.

Results: Exploratory factor analysis suggested that Nash et al.’s model (Perceived Transgressions, Perceived Betrayals) represents the data well. This two-factor solution showed an excellent fit in the confirmatory factor analysis, as well. Meaningful associations were observed between moral injury and psychopathology dimensions, shame, reflective functioning, well-being, and resilience. The Perceived Betrayals factor was a significant predictor of bipolar disorders, PTSD, paranoid personality disorder, borderline personality disorder, and avoidant personality disorder.

Conclusions: Our study demonstrated that this broad version of the MIES is a valid measure of moral injury that can be applied to psychiatric patients.

HIGHLIGHTS

• The Moral Injury Events Scale–Civilian Version is a reliable and valid instrument.

• The original 2-factor solution (Perceived Transgressions, Perceived Betrayals) yielded a good fit to the data.

• Moral injury’s Perceived Betrayals factor predicted bipolar disorders, PTSD, and three personality disorders (paranoid PD, borderline PD, avoidant PD).

Antecedentes: El daño moral surge cuando alguien perpetra, no logra prevenir o es testigo de actos que violan su propio código moral o ético. Nash et al. (2013) desarrollaron una herramienta breve, la Escala de Eventos de Daño Moral (MIES, por sus siglas en Ingles) para facilitar el estudio empírico del daño moral en el ejército. Nuestro estudio buscó desarrollar una versión civil de la herramienta (MIES-CV) y examinar sus propiedades psicométricas en una muestra de pacientes psiquiátricos hospitalizados.

Métodos: En este estudio transversal, la muestra estuvo compuesta por 240 pacientes adultos (71,7% mujeres) con una edad media de 31,57 (DE = 11,69). Los diagnósticos más frecuentes en la muestra fueron los trastornos de ansiedad (58,3%), trastornos depresivos (53,8%) y trastorno límite de la personalidad (39,6%). Los participantes fueron diagnosticados mediante entrevistas clínicas estructuradas y completaron cuestionarios psicológicos.

Resultados: El análisis factorial exploratorio sugirió que el modelo de Nash et al. (Transgresiones percibidas, Traiciones percibidas) representa bien los datos. Esta solución de dos factores también mostró un excelente ajuste en el análisis factorial confirmatorio. Se observaron asociaciones significativas entre daño moral y dimensiones de psicopatología, vergüenza, funcionamiento reflexivo, bienestar y resiliencia. El factor de Traiciones Percibidas fue un predictor significativo de trastornos bipolares, TEPT, trastorno de personalidad paranoide, trastorno límite de personalidad y trastorno de personalidad por evitación.

Conclusiones: Nuestro estudio demostró que esta versión amplia del MIES es una herramienta válida de daño moral que se puede aplicar a pacientes psiquiátricos.

背景：当某人实施、未能阻止或目睹违反自己的道德或伦理准则的行为时，会出现道德伤害。纳什等（2013）开发了一个简短的测量，即道德伤害事件量表（MIES），以促进军队道德伤害的实证研究。我们的研究旨在开发该测量的平民版（MIES-CV），并在精神病住院患者样本中考查其心理测量特性。

方法：在本横断面研究中，样本包括 240 名平均年龄为 31.57 岁（SD = 11.69）的成年患者（71.7% 为女性）。样本中最常见的诊断是焦虑障碍（58.3%）、抑郁障碍（53.8%）和边缘性人格障碍（39.6%）。参与者通过结构化临床访谈进行诊断并填写心理问卷。

结果：探索性因素分析表明 Nash 等人的模型（感知越界、感知背叛）很好地表征了数据。这种双因素解决方案在验证性因素分析中也表现出极好的拟合。在道德伤害和心理病理学维度、耻辱感、反思功能、幸福感和心理韧性之间观察到有意义的关联。感知背叛因素是双相情感障碍、PTSD、偏执型人格障碍、边缘性人格障碍和回避型人格障碍的重要预测因素。

结论：我们的研究表明，这种广泛版本的 MIES 是道德伤害的有效测量，可应用于精神病患者。

KEYWORDS:

• Moral Injury Events Scale–Civilian Version MIES-CV
• moral injury
• personality disorders
• psychiatric inpatients
• reliability
• validity

PALABRAS CLAVE:

• Escala de Eventos de Daño Moral–Versión Civil MIES-CV
• Daño moral
• Trastornos de personalidad
• Pacientes psiquiátricos hospitalizados
• Fiabilidad
• Validez

关键词:

• MIES-CV
• 道德伤害
• 人格障碍
• 精神病住院患者
• 可靠性
• 有效性

1. Introduction

Moral injury, a term that was originally developed and applied in the context of the military sector, has received a lot of attention recently, across various areas of study. After the initial focus on war veterans’ experiences of having been let down, betrayed, or abandoned by their leaders in combat (Shay, 1994, 2014) the definition of moral injury has been expanded to include experiences of not only of victims but also perpetrators and witnesses (Litz & Kerig, 2019; McEwen et al., 2021) and includes any experiences where ‘a person perpetrates, fails to prevent, bears witness to, or learns about acts that transgress deeply held moral beliefs or expectations’ (Litz & Kerig, 2019, p. 342; Litz et al., 2009). During the last 10–15 years, the concept of moral injury has been applied to civilians’ experiences in a non-military context as well, such as educators (Keefe-Perry, 2018; Levinson, 2015), social workers (Haight et al., 2017), healthcare workers (Murray et al., 2017), refugees (Nickerson et al., 2018), and police officers (Papazoglou & Chopko, 2017). Research has pointed to moral injury being predictive of suicide, depression, and other deleterious mental health outcomes within the traumatized veteran population (Ames et al., 2019; Cameron et al., 2021; Koenig et al., 2019), greater depression symptoms in refugees in Switzerland (Nickerson et al., 2018), lower resilience in veterinarians (Crane et al., 2015), and various psychological symptoms, as well as dehumanization, burnout, depressive symptoms, and PTSD in healthcare workers (Ehman et al., 2023; Testoni et al., 2022).

Overall, the association between moral injury and psychological symptoms has been supported by a growing number of studies with increasingly diverse populations. A recent meta-analysis of 59 empirical studies confirmed statistically significant relationships between all moral injury outcomes and poorer mental health (McEwen et al., 2021). Another recent systematic review of 57 studies came to the same conclusion, finding positive correlations between moral injury and posttraumatic stress symptoms, depression, anxiety, suicide, substance abuse, and other health outcomes like burnout, somatic pain, sleep disturbances, and treatment-seeking behaviours (Hall et al., 2021). Despite the increasing evidence that indicates the strong association between moral injury and psychopathology, the concept has not been studied in a psychiatric population.

1.1. Factor structure and validity of the Moral Injury Events Scale (MIES) and its adapted versions

The Moral Injury Events Scale (MIES; Nash et al., 2013) was the first published measure that made possible the scientific study of moral injury. In the following years, several new instruments have appeared in the literature, e.g. the Moral Injury Symptom Scale–Military Version (MISS–M; Koenig et al., 2018) or the Moral Injury Questionnaire–Military Version (MIQ–M; Currier et al., 2015), however, the MIES remained the dominant measure in the field of studying moral injury.

The studies that used quantitative assessment of moral injury outside the military context so far, have either used self-designed individual items (Ehman et al., 2023; Nickerson et al., 2018), or their own adaptation of the MIES (Andrukonis & Protopopova, 2020; Brennan et al., 2022; Fani et al., 2021; Feinstein et al., 2018; Haight et al., 2017; Khan et al., 2021; Morris et al., 2022; Nickerson et al., 2015; Papazoglou et al., 2019; Testoni et al., 2022; Thomas et al., 2021). From those using an adapted form of MIES, Thomas et al. (2021) and Morris et al. (2022) assessed the psychometric validity of their slightly different scales in civilian and healthcare professional samples, respectively, as we will discuss below.

To date, seven previous studies aimed to evaluate the factor structure of the MIES or its adapted versions in military and non-military samples, resulting in four competing models (Figure S1). Four of these studies focused on military samples. Nash et al. (2013) found a clear two-factor structure (Perceived Transgressions, Perceived Betrayals), whereas Bryan et al.’s (2016) study found an alternative three-factor solution instead (Transgressions–Self, Transgressions–Other, Betrayals). Richardson et al. (2020) attempted to replicate the previously reported two- and three-factor structures, and found that Nash et al.’s two-factor model showed a better fit compared to Bryan et al.’s three-factor model, however, the authors identified a new, theory-informed two-factor model (in which the Transgressions–Other and Betrayal factors of Bryan et al.’s model were merged together to form a new, broader Transgressions–Other factor, while the Transgressions–Self factor remained unchanged), which was superior to both previous models. Finally, Plouffe et al. (2023) found a poor fit in every attempt to find a factor solution with adequate characteristics.

Three further studies involved non-military samples. Papazoglou et al. (2019) found a two-factor structure that was again different from Nash et al.’s and Richardson et al.’s earlier two-factor solutions due to the unintentional omission of item 6. Thomas et al. (2021) made an adapted version of the MIES for civilian use and fitted Bryan et al.’s three-factor model to the data, yielding a good fit. Finally, Morris et al. (2022) replicated Richardson et al.’s new, theory-based two-factor model by using another adapted version of the MIES. Thus, both two and three-factor solutions have received some support, but results are often contradictory.

In most of the above-mentioned studies, the authors also assessed relationships between the MIES and other instruments to further ensure its validity. The MIES was positively associated with negative affectivity, depression, anxiety, anger, shame, guilt, and PTSD symptoms (Bryan et al., 2016; Nash et al., 2013; Papazoglou et al., 2019; Plouffe et al., 2023; Thomas et al., 2021), negatively with perceived social support, positive affectivity, and well-being (Nash et al., 2013; Thomas et al., 2021). To further support the MIES’s validity, the MIES was found to be highly correlated with scores on other moral injury measures (Thomas et al., 2021), demonstrating sufficient criterion validity.

1.2. Critiques of the MIES

Earlier studies made three major critiques regarding the MIES. First, the items of the scale lack specificity (Richardson et al., 2020). For example, it is unclear what ‘things’ refer to that were morally wrong, and what being ‘troubled’ means in the context of potentially morally injurious events. Second, the first six items of the MIES are in fact three question pairs asking about the same experiences using similar wording (e.g. ‘I acted in ways that violated my own moral code or values’ and ‘I am troubled by having acted in ways that violated my own morals or values’), which could result in artificially increased internal consistency estimates (Bryan et al., 2016). Third, including items both about experiencing an event and an internal reaction to the experienced event can also be problematic, since this assumes a strong connection between these two, which may not be the case (Frankfurt & Frazier, 2016).

Thus, despite the various critiques regarding specific aspects of the MIES, overall, it seems to be a reliable and valid measure of moral injury, which has been used in various populations, including recent initial studies on civilian populations as well. However, we know little about the role of moral injury in an especially vulnerable civilian group, in psychiatric patients. Therefore, in order to keep up with the expanding research on moral injury in vulnerable civilian populations, we decided to develop a broad adaptation of the MIES for generalized civilian use and apply the scale to a sample of psychiatric inpatients.

1.3. Aims

We aimed to create a MIES version that can be used in a broad range of populations without restriction to time, context, and perpetrators, and to demonstrate its validity in a specific and very relevant population, in psychiatric patients. As there are no previous results regarding the associations between moral injury and personality disorders, we examined these relationships in an explorative manner.

2. Methods

2.1. Participants and procedure

Participants were patients treated in the inpatient unit of Semmelweis University’s Department of Psychiatry and Psychotherapy in Budapest, Hungary. The patients participated in a complex four-week group schema therapy programme which consisted of two group schema therapy sessions per week, as well as other activities (such as art therapy or relaxation training). Data were collected between February 2020 and January 2022. The sample consisted of 240 patients, aged 18–70 (M = 31.57, SD = 11.69), from which 25% self-identified as male, 71.7% as female, and 3.4% as non-binary. Exclusion criteria were refusal of voluntary participation, acute psychosis, acute intoxication, and organic psychosyndrome.

Diagnoses were based on clinical assessments using the MINI Plus 5.0.0 and SCID–5–PD interviews, as well as self-report questionnaires completed on a secure online platform. All assessments were completed in Hungarian at the beginning of the therapy programme. 59.6% of the patients had at least one personality disorder (PD) diagnosis, specifically 39.6% borderline PD, 17.9% obsessive-compulsive PD, 17.5% avoidant PD, 9.2% paranoid PD, 5.8% narcissistic PD, 4.6% dependent PD, 2.9% histrionic PD, 0.8% schizotypal PD, 0.4% schizoid PD. The most common clinical disorders (CDs, formerly ‘Axis I’ disorders) in the sample were: 58.3% anxiety disorders, 53.8% depressive disorders, 23.8% bipolar disorders, and 14.2% posttraumatic stress disorder (PTSD). 92.5% of the patients had at least one CD diagnosis. Antisocial PD was not assessed due to legal and ethical concerns. Written informed consent was obtained from all patients and the study was approved by the local ethical committee (Scientific and Research Ethics Committee of the Medical Research Council).

2.2. Measures

2.2.1. Moral Injury Events Scale–Civilian Version (MIES–CV)

Moral injury was assessed using the MIES–CV which consists of 9 items and is based on the Moral Injury Events Scale developed by Nash et al. (2013). The instructions and some items were slightly modified to make the instrument able to be used in a civilian context. Specifically, the original instructions (‘Please circle the appropriate number to indicate how much you agree or disagree with each of the following statements regarding your experiences at any time since joining the military’) were changed to ‘Please circle the appropriate number to indicate how much you agree or disagree with each of the following statements’. Item 8 originally read ‘I feel betrayed by fellow service members who I once trusted’ and was changed to ‘I feel betrayed by peers who I once trusted’. Item 9 originally read ‘I feel betrayed by others outside the US military who I once trusted’ and was changed to ‘I feel betrayed by others who I once trusted’. (See the MIES and the MIES–CV in full form in Table S2.)

There are three differences between the MIES–CV and Thomas et al.’s (2021) civilian version, the MIES–C. First, the instructions of the MIES–CV do not include any time reference, while the instructions of the MIES–C ask the participant to think about experiences over the past 6 months. Second, the MIES–CV uses the word ‘peers’ while the MIES–C uses the word ‘friends’ instead of ‘fellow service members’ in item 8. Third, the expression ‘others outside the US military’ in item 9 was changed to ‘others’ in the MIES–CV and to ‘others outside my immediate circle’ in the MIES–C.

Fani et al. (2021) also developed a civilian version of the MIES which is called Moral Injury Exposure and Symptom Scale–Civilian (MIESS–C). The authors did not report the instructions they used, so MIESS–C’s instructions cannot be compared with MIES–C’s instructions. The items of this measure differ from the MIES–CV in four ways: (1) the word ‘people’ was used instead of ‘leaders’ in item 7, (2) item 8 was completely rewritten (‘I am troubled by this betrayal by specific people’) to reflect distress associated with the betrayal identified in item 7, (3) the words ‘the institutions that I am supposed to trust (for example, police, church, schools and governmental workers)’ were used instead of ‘others who I once trusted’, (4) finally, item 10 was added to the scale (‘I am troubled by this betrayal by the institutions that I am supposed to trust’) to reflect distress associated with the betrayal identified in item 9.

The MIES–CV was translated into Hungarian and subsequently back-translated into English before its administration in Hungarian. Participant agreement with each statement is assessed using a 6-point Likert-type scale ranging from 1 = strongly agree to 6 = strongly disagree. In the statistical analyses, we reverse-scored the items so higher scores indicated greater moral injury.

2.2.2. Experience of Shame Scale (ESS)

The ESS (Andrews et al., 2002; Vizin et al., 2016) is a 25-item questionnaire that assesses trait shame (i.e. proneness to experience shame). Example items include: ‘Have you felt ashamed of your ability to do things?’ or ‘Have you felt ashamed of your body or any part of it?’. Responses are coded on a 4-point Likert-type scale (1 = not at all, 4 = very much). The total mean of the ESS (68.16) was considerably higher than the mean obtained from a sample of undergraduates (55.58) (Andrews et al., 2002), suggesting that the patients in our sample were generally more prone to experience shame than undergraduate students. The scale showed high internal consistency (α = .94) in our study.

2.2.3 Symptom Checklist–90–R (SCL–90–R)

The SCL–90–R (Derogatis, 1977; Unoka et al., 2004) is a 90-item self-report questionnaire designed to measure a broad range of psychological problems and symptoms of psychopathology. Each item is rated on a 5-point Likert-type scale from 0 = not at all to 4 = very much. The SCL–90–R comprises 9 subscales and also provides a global psychological distress index, the Global Severity Index (GSI). The SCL–90–R scale scores in this study were generally higher compared to the scores of a normative sample and were similar to the scores of an independent clinical sample (Unoka et al., 2004), suggesting that the patients in this sample generally showed heightened levels of psychopathology. Cronbach’s alphas ranged from .70 to .88 for subscales, whereas Cronbach’s alpha of GSI was .96.

2.2.4. Reflective Functioning Questionnaire–8 (RFQ–8)

The RFQ–8 (Fonagy et al., 2016; Horváth et al., 2023) is a self-report questionnaire that comprises 8 items forming the two scales of certainty (RFQ_C; α = .78) and uncertainty about mental states (RFQ_U; α = .72). Items were rated on a 7-point Likert-type scale from 1 = do not agree at all to 7 = agree completely. To obtain scale scores (RFQ_C, RFQ_U) we used the scoring procedure described in Fonagy et al. (2016).

2.2.5. Connor-Davidson Resilience Scale (CD–RISC)

To measure resilience, we used the 10-item version of the CD–RISC (Campbell-Sills & Stein, 2007; Járai et al., 2015) which is a short form of the original, 25-item CD–RISC developed by Connor and Davidson (2003). The answers are coded on a 5-point Likert-type scale ranging from 0 (not true at all) to 4 (true nearly all the time). The total mean of the CD–RISC in this study was 17.76, which is markedly lower than the mean (27.21) found in Campbell-Sills and Stein’s (2007) study, suggesting that the patients in our study generally had lower level of resilience compared with undergraduates. High internal consistency (α = .88) was found for CD–RISC.

2.2.6. WHO (Five) Well-being Index, 1998 Version (WHO–5)

The WHO–5 (Susánszky et al., 2006; WHO, 1998) is a short and generic, 5-item measure of subjective well-being. The respondents are asked to rate how well each of the statements applies to them when considering the last 2 weeks. The answers had to be indicated on a 6-point Likert-type scale (0 = at no time, 5 = all of the time). The total mean of the WHO–5 in this study was somewhat lower (6.69) than the mean found in a large community sample (7.8) by Susánszky et al. (2006), suggesting that the well-being of the patients was more impaired compared to community people. The WHO–5 exhibited high internal consistency (α = .85).

2.2.7. Mini–International Neuropsychiatric Interview Plus 5.0.0 (MINI Plus 5.0.0)

The MINI Plus 5.0.0 (Sheehan et al., 1998) is a brief, structured diagnostic interview compatible with DSM–IV and ICD–10 criteria. The interview includes 26 modules (labelled with letters from A to Z) and explores current and lifetime diagnoses. Inter-rater reliability was high (κ > .94), and criterion validity compared to the Diagnostic Interview Schedule was good in the Hungarian validation studies (Balázs et al., 1998, 2001). We aimed to reduce the number of former Axis I diagnostic categories we obtained from the MINI interview based on the DSM-IV classification by regrouping them into the following higher order diagnostic groups: Major Depressive Disorder and Dysthymic Disorder into Depressive Disorders; Bipolar I and Bipolar II Disorders into Bipolar Disorders; Panic Disorder, Agoraphobia, Social Phobia, Specific Phobia, and Generalized Anxiety Disorder into Anxiety Disorders; Anorexia Nervosa and Bulimia Nervosa into Eating Disorders; Post traumatic Stress Disorder into Post traumatic Stress Disorder; Obsessive-Compulsive Disorder and Body Dysmorphic Disorder into Obsessive-Compulsive and Related Disorders; Alcohol Dependence/Abuse and other psychoactive substance dependence/abuse disorders into Substance-related and Addictive Disorders; Schizophrenia and Schizoaffective Disorder into Schizophrenia Spectrum and Other Psychotic Disorders; Somatization Disorder and Hypochondriasis into Somatic Symptom and Related Disorders.

2.2.8. Structured Clinical Interview for DSM–5 Personality Disorders (SCID–5–PD)

The SCID–5–PD (First et al., 2015, 2018) is a semi-structured interview designed to assess the 10 DSM–5 personality disorders. The items are rated by a clinician on a 3-point scale: 0 = symptom is absent, 1 = symptom is present but not on a clinically significant level, and 2 = symptom is present on a clinically significant level. Then, the SCID–5–PD can be used to make either categorical personality disorder diagnoses (absent or present) or dimensional diagnoses (by summing 0, 1, and 2 ratings). In the present study, categorical personality disorder diagnoses were made and used.

2.3. Data analysis

All calculations were performed using Mplus 8 (Muthén & Muthén, 1998–2017) and IBM SPSS Statistics 28.0.1 (IBM Corp., 2021).

We examined univariate and multivariate normality by using Cain et al.’s (2017) SPSS macro which is a slightly modified version of DeCarlo’s (1997) SPSS macro. According to Hair et al.’s (2017) recommendations when univariate skewness or kurtosis is greater than +1 or lower than −1 then the distribution can be considered non-normal. Similarly, when Mardia’s test multivariate skewness and/or multivariate kurtosis is significant, then the assumption of multivariate normality is violated.

Inter-item correlations (IIC), item-total correlations (ITC), and Cronbach’s alphas were used for testing internal consistency. We considered IICs between .30 and .70 (Ferketich, 1991) and ITCs above .30 as acceptable. For Cronbach’s alpha, we used the following cut-off values: α ≥ .70: acceptable, α ≥ .80: good, α ≥ .90: excellent (George & Mallery, 2003).

Due to the absence of any prior validation in psychiatric context and mixed results regarding the factor structure, we first applied an exploratory factor analytic (EFA) approach to examine the structural characteristics of the MIES. The optimal number of factors was determined based on Kaiser’s criterion, parallel analysis (PA), minimum average partial (MAP) test, and findings in the previous literature. PA and MAP test were performed using O’Connor’s (2000) SPSS macro. As in Nash et al.’s (2013) paper, principal axis factoring (PAF) method was chosen with an oblique rotation (Promax, κ = 4) because the factors were considered to be correlated.

Then, we used confirmatory factor analysis to confirm the structure found in the EFA. Robust maximum likelihood (MLR) estimation was chosen because it is robust in the presence of normality violations. To assess model fit (Browne & Cudeck, 1992; Hu & Bentler, 1999), we inspected the root mean square error of approximation (RMSEA; ≤ .08), the comparative fit index (CFI; ≥ .90), the Tucker–Lewis Index (TLI; ≥ .90), and the standardized root mean squared residual (SRMR; ≤ .08).

To ensure that the instrument exhibits meaningful relationships with other measures, we examined the associations between the MIES–CV factors and shame, psychopathology dimensions, reflective functioning, resilience, and well-being using Pearson correlations. 95% confidence intervals for correlation coefficients were also computed using an SPSS syntax provided by IBM (https://www.ibm.com/support/pages/confidence-intervals-correlations). Point-biserial correlations and binary logistic regressions were performed to explore if there is a connection between MIES–CV factors and major clinical disorder (formerly ‘Axis I’ disorder) domains and personality disorder (formerly ‘Axis II’ disorder) diagnoses. According to Bonett (2020) ‘a “large” point-biserial correlation is smaller than what might be considered to be a “large” Pearson correlation between two quantitative variables’ (p. 177). Thus, we considered a point-biserial coefficient below .20 a small correlation, a coefficient between .20 and .40 a medium correlation, and .40 or larger a strong correlation.

2.3.1. Missing data

For 13.8% of the participants the SCID–5–PD information and for 7.5% of the participants the MINI Plus 5.0.0 information was missing. Listwise deletion was used to handle missing data.

3. Results

3.1. Exploratory factor analysis

Prior to the factor analysis, Bartlett’s test of sphericity and Kaiser–Meyer–Olkin (KMO) measure of sampling adequacy were used to assess the factorability of the data. Bartlett’s test was significant, χ²(36) = 1137.06, p < .001, and the KMO value was .71 (‘middling’ according to Kaiser, 1974). The results of the tests supported the factorability of the correlation matrix.

Next, the number of factors to be retained was determined. Kaiser’s criterion and parallel analysis suggested extracting 3 factors. In the parallel analysis, the first four observed principal component (PC) eigenvalues were 3.645, 1.886, 1.151, and .858, whereas the mean of random data PC eigenvalues were 1.305, 1.202, 1.125, and 1.055. The original MAP² and the revised MAP⁴ test suggested extracting 2 factors. The minimum average partials for one- to four-factor solutions were .1108, .0995, .1040, .1098, according to the MAP² test, and .0274, .0224, .0266, .0835, according to the MAP⁴ test. Based on these results, we decided to test a two-factor and a three-factor solution.

In the two-factor solution, items 1–6 loaded on the first factor, while items 7–9 loaded on the second factor. All factor loadings were above .30, ranging from .43 to .90, and all communalities were above .20, ranging from .29 to .81. The first factor explained 35.20%, and the second factor explained 17.13% of the total variance. The factors were moderately intercorrelated (r = .31).

In the three-factor solution, items 1–4 loaded on the first factor, items 7–9 loaded on the second factor, while items 5 and 6 formed the third factor (see Table S3). Again, all factor loadings were above .30, ranging from .48 to .92, and the communalities were above .20, ranging from .32 to 88. The first factor explained 36.75%, the second factor explained 18.12%, and the third factor explained 9.71% of the total variance. The correlation was .29 between the first and second factors, .51 between the second and third factors, and .19 between the first and third factors. Considering the large correlation between items 5 and 6 (r = .85, see Table S4) it is plausible to assume that the third factor in the three-factor solution is a so-called ‘tautological factor’ (Eysenck & Eysenck, 1969) or a ‘bloated specific’ (Cattell & Kline, 1977) that emerges when items are characterized by high content similarity, and thus they are largely correlated (Kline, 1986).

Because the two-factor solution replicates the results presented by Nash et al. (2013) and the three-factor solution seems to be implausible, we decided to retain the two-factor solution where the first factor is labelled as Perceived Transgressions and the second factor is labelled as Perceived Betrayals (Table 1). Both factors had good internal consistency, Cronbach’s alphas were .82 for Perceived Transgressions and .84 for Perceived Betrayals.

Table 1. Item-level statistics and exploratory factor analysis of the Moral Injury Events Scale–Civilian Version.

Item	Mean	SD	Skew	Kurt	ITC	F1	F2	h^2
1. I saw things that were morally wrong	2.85	1.76	.70	−.83	.57	.538	.178	.38
2. I am troubled by having witnessed others’ immoral acts	2.63	1.78	.85	−.68	.47	.431	.210	.29
3. I acted in ways that violated my own moral code or values	3.18	1.91	.36	−1.41	.65	.741	−.037	.53
4. I am troubled by having acted in ways that violated my own morals or values	3.03	1.96	.48	−1.37	.67	.756	−.026	.56
5. I violated my own morals by failing to do something that I felt I should have done	2.60	1.62	.92	−.26	.61	.740	−.067	.52
6. I am troubled because I violated my morals by failing to do something that I felt I should have done	2.59	1.65	.95	−.22	.55	.672	−.037	.44
7. I feel betrayed by leaders who I once trusted	3.79	1.80	−.02	−1.47	.58	.111	.590	.40
8. I feel betrayed by peers who I once trusted	3.51	1.90	.13	−1.53	.77	−.063	.904	.79
9. I feel betrayed by others who I once trusted	3.39	1.88	.25	−1.43	.79	−.020	.904	.81

Note: Factor loadings of .30 and above are in bold type. SD = Standard Deviation; Skew = Skewness; Kurt = Kurtosis; ITC = Item-Total Correlation; F1 = Perceived Transgressions; F2 = Perceived Betrayals; h² = Communality.

3.2. Confirmatory factor analysis

For the CFA, robust maximum likelihood (MLR) estimation was chosen because univariate (see Table 1) and multivariate normality (Mardia’s multivariate skewness: b_1,p = 9.09, p < .001, Mardia’s multivariate kurtosis: b_2,p = 116.23, p < .001) were both violated. The model fit of the two-dimensional CFA was poor, χ²(26) = 313.47, p < .001, RMSEA = .22, CFI = .65, TLI = .51, SRMR = .11. Modification indices suggested that placing covariances between error terms for item 1 and 2, item 3 and 4, and item 5 and 6 would increase fit. This decision seemed well-justified because the content of the items of these item pairs shows remarkable similarity, and Nash et al. (2013) also decided to free these error covariances. The modifications resulted in significant improvement: χ²(23) = 28.28, p = .205, RMSEA = .03, CFI = .99, TLI = .99, SRMR = .05 (Figure 1). Based on the cut-off recommendations detailed earlier, the model fit was excellent. The factor loadings in this model ranged from .51 to .93. The two factors were moderately intercorrelated (r = .34).

Figure 1. Confirmatory factor analysis of the Moral Injury Events Scale–Civilian Version.

Confirmatory factor analysis of the Moral Injury Events Scale–Civilian Version. The model has two factors: Perceived Transgressions and Perceived Betrayals. Error covariances were freed between items 1 and 2, items 3 and 4, and items 5 and 6. Factor loadings are ranging from .51 to .93. Factors are moderately intercorrelated (r = .34).

Of note, we also tested the fit of Bryan et al.’s (2016) three-factor model and it yielded an excellent fit as well (χ²(23) = 24.92, p = .354, RMSEA = .02, CFI = 1.00, TLI = 1.00, SRMR = .05) after allowing the error terms for items 5 and 6 to covary. The correlation between Transgressions–Self and Transgressions–Other was high (r = .58).

In sum, our results confirm the two-factor solution described by Nash et al. (2013), and the three-factor solution (i.e. Bryan et al.’s (2016) model) in which the Perceived Transgressions factor is bifurcated into two subfactors based on the perpetrator of the morally injurious acts (i.e. self or others) seems to be adequate, as well. However, because the two-factor solution is a more parsimonious choice we decided to draw on it later in this article.

3.3. Associations with other measures

Table 2 displays the bivariate relations between moral injury and other psychological phenomena. Perceived transgressions showed a significant weak positive correlation with shame and psychopathology dimensions. A significant weak negative association was observable between perceived transgressions and certainty about mental states, and a relationship between perceived transgressions and uncertainty about mental states emerged with a similar magnitude but positive direction.

Table 2. Descriptive statistics and bivariate associations between the Moral Injury Events Scale–Civilian Version and other variables.

		Moral Injury Events Scale–Civilian Version
Measures	M (SD)	Total	Perceived Transgressions	Perceived Betrayals
Experience of Shame Scale	68.16 (16.81)	.34** [.22, .45]	.27** [.15, .38]	.29** [.17, .40]
Symptom Checklist–90–R
Somatization	1.23 (.86)	.27** [.15, .38]	.13* [.00, .25]	.35** [.23, .46]
Obsessive-compulsive	1.90 (.81)	.30** [.18, .41]	.22** [.10, .34]	.28** [.16, .39]
Interpersonal sensitivity	1.92 (.91)	.28** [.16, .39]	.14* [.01, .26]	.36** [.24, .47]
Depression	2.47 (.86)	.30** [.18, .41]	.20** [.08, .32]	.33** [.21, .44]
Anxiety	1.83 (.85)	.28** [.16, .39]	.15* [.02, .27]	.35** [.23, .46]
Hostility	1.23 (.88)	.34** [.22, .45]	.25** [.13, .37]	.33** [.21, .44]
Phobic anxiety	1.46 (1.07)	.28** [.16, .39]	.18** [.05, .30]	.31** [.19, .42]
Paranoid ideation	1.51 (.92)	.37** [.26, .47]	.19** [.06, .31]	.47** [.37, .56]
Psychoticism	1.16 (.66)	.26** [.14, .37]	.23** [.11, .35]	.19** [.06, .31]
Global Severity Index	1.69 (.66)	.37** [.26, .47]	.24** [.12, .36]	.41** [.30, .51]
Reflective Functioning Questionnaire–8
Certainty about mental states	.96 (.77)	−.33** [−.44, −.21]	−.28** [−.39, −.16]	−.25** [−.37, −.13]
Uncertainty about mental states	.93 (.70)	.28** [.16, .39]	.24** [.12, .36]	.20** [.08, .32]
Connor-Davidson Resilience Scale	17.76 (8.48)	−.08 [−.20, .05]	−.02 [−.15, .11]	−.14* [−.26, −.01]
WHO (Five) Well-being Index, 1998 Version	6.69 (5.18)	−.12 [−.24, .01]	−.06 [−.19, .07]	−.16* [−.28, −.03]

Note: 95% confidence intervals of Pearson’s r coefficients are in brackets.

* p < .05, ** p < .01.

Perceived betrayals also showed significant weak associations with shame. Correlations between perceived betrayals and SCL–90–R scales reached a moderate magnitude for 7 of the 9 subscales, whereas the obsessive-compulsive and psychoticism subscales showed only weak associations with perceived betrayals. Regarding reflective functioning, a similar pattern was observable between perceived betrayals and the RFQ–8 subscales as in the case of perceived transgressions. Perceived betrayals showed a significant but weak negative correlation with resilience. The relation between perceived betrayals and subjective well-being also reached the level of significance but the magnitude of r was low.

Next, we examined the relationship between moral injury factors and clinical disorder domains and between moral injury factors and personality disorders. Initially, point-biserial correlations were calculated (Table 3). We found weak but significant associations between perceived transgressions and PTSD, between perceived betrayals and borderline PD, and between perceived betrayals and avoidant PD. Furthermore, the perceived betrayals factor was moderately correlated with paranoid PD, bipolar disorders, and PTSD.

Table 3. Point-biserial correlations and binary logistic regression models predicting clinical disorders and personality disorders (independent variables, italicized) by moral injury factors (dependent variables).

Variables	rpb	B	SE	Wald	p	Exp(B)	95% CI
Bipolar Disorders^a (Nagelkerke’s R^2 = .07)
Perceived Transgressions	.03	−.01	.02	.38	.54	.99	.95–1.03
Perceived Betrayals	.22**	.12	.04	10.45	.00	1.12	1.05–1.20
Constant		−1.98	.60	10.70	.00	.14
Posttraumatic Stress Disorder^a (Nagelkerke’s R^2 = .14)
Perceived Transgressions	.16*	.04	.03	2.13	.14	1.05	.99–1.11
Perceived Betrayals	.26**	.16	.05	10.71	.00	1.17	1.06–1.28
Constant		−4.67	.96	23.71	.00	.01
Paranoid Personality Disorder^b (Nagelkerke’s R^2 = .14)
Perceived Transgressions	.05	−.01	.03	.07	.80	.99	.93–1.06
Perceived Betrayals	.25**	.22	.06	11.27	.00	1.24	1.09–1.41
Constant		−4.46	1.12	15.86	.00	.01
Borderline Personality Disorder^b (Nagelkerke’s R^2 = .05)
Perceived Transgressions	.13	.02	.02	1.41	.24	1.02	.99–1.06
Perceived Betrayals	.18*	.07	.03	4.46	.04	1.07	1.01–1.14
Constant		−1.43	.54	6.96	.01	.24
Avoidant Personality Disorder^b (Nagelkerke’s R^2 = .05)
Perceived Transgressions	−.03	−.03	.02	1.44	.23	.97	.92–1.02
Perceived Betrayals	.16*	.10	.04	6.50	.01	1.11	1.02–1.20
Constant		−1.76	.66	7.11	.01	.17

Note: ^a N = 222, ^b N = 217. r_pb = Point-biserial correlation; B = Unstandardized regression coefficient; SE = Standard error of B; Exp(B) = Exponential value of B; 95% CI = 95% confidence interval of Exp(B).

* p < .05, ** p < .01.

Finally, a series of binary logistic regressions were conducted to test whether the moral injury factors are significant predictors of clinical disorder domains or personality disorders (Table 3). Of note, only those models are reported here in which a significant relationship emerged between at least one predictor and the dependent variable. Perceived betrayals factor was a significant predictor of bipolar disorders, PTSD, paranoid PD, borderline PD, and avoidant PD. The perceived transgressions factor was not a significant predictor of any of the disorders.

4. Discussion

To our knowledge, this is the first study to examine moral injury in a sample of psychiatric inpatients by developing a civilian version of the Moral Injury Events Scale (MIES–CV).

In line with the results of the original validation study (Nash et al., 2013), an exploratory factor analysis (EFA) with a two-factor solution yielded a clear factor structure, adequate factor loadings, and reasonable percentages of explained variance for both factors. However, Kaiser’s criterion and parallel analysis suggested extracting three factors, so we also tested whether a three-factor solution produced a better fit than a two-factor solution. The third factor, which consisted of items 5 and 6, in the three-factor EFA model seemed to be a ‘tautological’ (Eysenck & Eysenck, 1969) or a ‘bloated specific’ (Cattell & Kline, 1977) factor because of the large correlation (r = .85) between items 5 and 6. This result is not surprising if we consider that many critiques (e.g. Bryan et al., 2016) were raised against the wording of the MIES items which are, unfortunately, also applicable to the Hungarian civilian version of the MIES. In light of these findings, we decided to retain the two-factor solution composed of a factor labelled as Perceived Transgressions and a factor labelled as Perceived Betrayals. Furthermore, high internal consistency estimates (αs > .80) were obtained for both scales.

The two-factor structure found by Nash et al. (2013) was confirmed by using exploratory (EFA) and confirmatory factor analysis (CFA) with excellent fit indices. Additionally, an alternative three-factor model proposed by Bryan et al. (2016) was also tested which showed an excellent fit, as well. In sum, both Nash et al.’s (2013) and Bryan et al.’s (2016) models seem to be adequate in a clinical sample. Thus, when the distinction between the perpetrators (i.e. self or others) of the transgressive acts is important, the Perceived Transgressions factor can be split into two subfactors based on the perpetrator, otherwise using the two-factor solution is advised.

When assessing the relationships between the MIES–CV and other scales, we found that its two subscales were associated with other variables in the expected way. Both perceived transgressions and perceived betrayals were significantly associated with shame and psychopathology dimensions. This finding is consistent with the results of earlier studies (e.g. Nash et al., 2013; Nieuwsma et al., 2021; Thomas et al., 2021); however, most of the studies only used the MIES total score in the correlation analyses. Furthermore, only the perceived betrayals factor was significantly correlated with resilience and subjective well-being, meaning that the previously found negative associations between moral injury and resilience (Crane et al., 2015) and moral injury and well-being (Thomas et al., 2021) were only partially replicated. However, in contrast with Ferrajão and Oliveira’s (2014) study in which no significant relationship was found between moral injury and reflective functioning, our results showed that moral injury factors were significantly and positively correlated with difficulties in understanding mental states of self and others (i.e. hypomentalization) as measured by the Reflective Functioning Questionnaire. As Debbané et al. (2022, p. 137) simply stated ‘[m]entalizing is about generating safety, trust, understanding, and collaboration between people.’ People who develop psychopathology and especially BPD, tend to have more childhood adversity (Gunderson et al., 2018) which, in turn, might constitute moral injuries, and also, lead to diminished reflective functioning indeed, since they are confused about the mental states of themselves and others due to being betrayed and traumatized. Lower reflective functioning, in turn, might result in, negatively interpreting situations and also a tendency to expose themselves to situations that eventually hurt them due to a lack of ability to recognize danger and inner discomfort (victimization).

Binary logistic regressions showed that moral injury had significant relationships with bipolar disorders, PTSD, paranoid PD, borderline PD, and avoidant PD; however, only the perceived betrayals component was a significant predictor of the above-mentioned clinical disorder domains and PD diagnoses. According to two recent literature reviews, the association between moral injury and PTSD is present across military and non-military populations (Hall et al., 2021; Koenig et al., 2019). The results regarding personality disorders draw attention to the developmental model of moral injury (Kidwell & Kerig, 2023) which attempts to unite the betrayal trauma theory, the attachment trauma theory, and the concept of perpetration-induced traumatic stress. According to this unified model, moral transgressions or betrayals committed by attachment figures or other trusted persons, peers, and institutions as well as own acts of perpetrations may result in developing maladaptive meaning-making processes which may lead to the development of long-lasting psychopathology (Kidwell & Kerig, 2023). Lastly, the association between moral injury and bipolar disorders is a novel and puzzling finding. A meta-analysis found that adults with bipolar disorder have significantly more severe depressive and manic symptoms if they experienced childhood trauma compared to individuals who also suffer from bipolar disorder but did not report a history of childhood trauma (Agnew-Blais & Danese, 2016). Based on the limited empirical literature, one possible explanation could be that moral injury related to childhood trauma may constitute a vulnerability for the development of certain psychiatric problems, among them, possibly bipolar disorder. Future studies are needed to support the impact of early moral injury on later mental health problems and to establish whether moral injury makes individuals more prone to develop specific psychiatric disorders.

4.1. Limitations and future directions

There are some limitations in the present study, and we should also mention some possible future directions. Our study used a cross-sectional design which does not allow us to infer causality between variables. The high prevalence of personality disorders in the sample may limit the generalizability of the presented findings to other clinical samples. It is certainly a great advantage of the MIES–CV that it contains no limitation regarding time frame (e.g. over the last six months), transgressor (e.g. friends of the respondent), or context (e.g. outside the respondent’s immediate circle). It thus allows for a broad use of the scale. A more recent instrument, the Moral Injury Outcome Scale (MIOS; Litz et al., 2022), published after our study was finished, asks the respondent to think of the worst event that currently bothers them the most and answer further questions regarding the experience. Nevertheless, we think that focusing only on one event limits the breadth of the morally injurious events that could have led to the experience of moral injury and thus can lead to an underestimation of the presence of moral injury. Previous studies and our results made clear that the scale has some methodological and psychometric weaknesses (e.g. similar wording of certain item pairs, which results in large item intercorrelations). For this reason, a fundamental revision in the future seems warranted. Lastly, one must not forget that experiences in the hospital might also continue moral injuries which would be an interesting topic to further investigate in a future study.

5. Conclusions

Our study demonstrated that this broad version of the MIES is a valid measure of moral injury that can be applied to psychiatric patients. Future studies are needed to demonstrate its validity in other specific populations, as well.

Ethical standards

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

Data availability statement

Deidentified data and copies of the supplementary materials can be found in the following Open Science Foundation repository: https://osf.io/vxt7k/.

Disclosure statement

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

Additional information

Funding

The principal investigator Zsolt Szabolcs Unoka was supported by the Hungarian National Research, Development and Innovation Fund (Grant No. NKFI-132546). Erika Evelyn Lévay was supported by the ÚNKP-22-3-II-SE-34 New National Excellence Program of the Ministry for Culture and Innovation from the source of the National Research, Development, and Innovation Fund.