The PCL-5 has been widely used since its original English version was released in 2013 by Weathers et al. (2013a), with several adaptations following in other languages for both research and clinical purposes. It is based on the diagnostic principles of PTSD in the DSM-5. It consists of 20 items divided into four clusters (B–E) according to the DSM-5 criteria of PTSD: B (items 1–5)—persistent re-experience of the traumatic event; C (items 6–7)—avoidance of trauma-related stimuli; D (items 8–14)—negative changes in cognition and feelings after the trauma; and E (items 15–20)—trauma-related arousal and hyperreactivity. Each item is scored from 0 to 4 based on the participant's response, with the options ranging from “not at all” to “a little bit,” “moderately,” “quite a bit,” and “extremely.” The total possible score ranges from 0 to 80 (Weathers et al., 2013a).

PCL-5 is self-administered and can be finished in well under 20 min, thus being quite resource saving, which is particularly important in acute crises when time and personnel are limited. It has proven to be a reliable instrument with a high internal consistency (α = 0.94; Blevins et al., 2015). Recommended cut-offs range from 30 to 34 (Weathers et al., 2013a; Blevins et al., 2015). However, cut-offs as well as sensitivity and specificity vary strongly depending on the investigated population. Geier et al. (2018) report an astonishingly high sensitivity of 91% and a specificity of 97% for a cut-off of 34 in a US sample of patients following non-intentional traumatic injury. In an Arabic/Kurdish adaptation of PCL-5 employed amongst displaced persons in Kurdistan, a very low cut-off of 23 yielded a sensitivity of 82% and a specificity of 70% (Ibrahim et al., 2018). Boyd et al. (2021) discuss in their study on Canadian psychiatric outpatients two different cut-offs: 33, resulting in a highly sensitive (94%) but unspecific (49%) test; and a cut-off of 45 for optimal ROC with a sensitivity of 82% and specificity of 69%. Thus, PCL-5 compares very well in terms of validity and psychometric properties to other self-administered instruments such as the Primary Care PTSD Screen for DSM-5 (PC-PTSD-5; sensitivity 0.9, specificity 0.8 vs. CAPS-5 in a veteran-sample; Bovin et al., 2021) and clinician-administered instruments such as the PTSD Symptom Scale Interview for DSM-5 (PSSI-5; sensitivity 0.82, specificity 0.71 vs. CAPS-5; Foa et al., 2016). More diagnostic instruments for PSTD diagnosis exist outside of the DSM-5 system. The International Trauma Questionnaire (ITQ, Cloitre et al., 2018) is a self-administered questionnaire for PTSD and complex PTSD according to ICD-11 criteria. Its internal consistently is with α = 0.88 slightly lower than that of PCL-5, but still good. A validated Russian version exists (Padun et al., 2022), but to our knowledge no validated Ukrainian versions, although a DeepL-translated version curated by a bilingual academic is available (OxCADAT Resources). An elder, cross-culturally validated instrument with very high internal validity is the Harvard Trauma Questionnaire (HTQ; Mollica et al., 1992). However, it is outdated being based on DSM-IV and—to the best of our knowledge—lacks a validated Ukrainian adaptation.

Application of PCL-5 should be preceded by the assessment of the trauma A-criterion of PTSD according to DSM-5. The Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013b), a 17-item self-report scale to evaluate individuals' prior life events, fulfills this purpose very well. However, its application remains relatively uncommon, which accounts for the limited availability of validated versions of the LEC-5 in languages other than English and a few others.

Despite its proven usefulness, several caveats of PCL-5 need to be acknowledged. As stated above, cut-offs and psychometric properties of the PCL-5 are highly heterogeneous depending on the underlying population. Thus, results must be interpreted with cautious consideration of the sociodemographics and disease burden of the investigated sample. Secondly, PTBS A-criterion must be fulfilled based on a separate diagnostic instrument or clinical exploration. Disregarding this step will lead to increased false-positive rates through the risk of assessing the general severity of the psychological burden, possibly unrelated to a PTSD diagnosis. Thirdly, most of PCL-5 items assess symptoms such as disturbances in sleep, intrusive thoughts, negative changes in cognition and emotional state, vegetative symptoms and hyperarousal, which can also occur in depression and anxiety disorders. Our preliminary unpublished work indicates a very strong correlation between PCL-5 scores and scores in Beck's Depression Inventory (BDI-II) and Beck's Anxiety Inventory (BAI) emphasizing the strong symptomatic overlap. This inevitably leads to reduced specificity of PCL-5 in clinical populations with a high burden of psychiatric disease, as reported by Boyd et al. (2021), which may result in high false positive PTSD diagnoses.

Fourthly, it is not stated in the PCL-5 that the questions refer to a specific traumatic event. When answering the questions, the subject might refer to one or several unrelated life events which often do not fulfill the DSM-5 trauma criterion (all items 1–8) or to their general state which is not related to a specific stressful event (all items 9–20). Thus, one cannot avoid a thorough clinical exploration after the testing in which the clinician must find out which life events the subject refers to and what their answers actually mean. Fifthly, as in every self-administered Likert-type-scale, the individual scores are highly subjective and are strongly affected by the subject's tendencies to aggravation, dissimulation, cognitive distortions, motivation and (mis-)understanding of the items, possible leading to highly variable scores in individuals with an objectively similar syndrome.

Taken together, the PCL-5 is a very useful screening tool for PTSD and is advisable to be included in the diagnostic process, but the PTSD diagnosis should not solely rely on its result. A thorough assessment of the psychopathological findings, clinical exploration of the symptoms, the context, the individual medical (and trauma event) history, the linguistic and social background and needs to be performed; differential diagnoses must be considered.

The adaptation of psychological assessment instruments, particularly in cross-cultural studies, typically involves a systematic forward-translation and back-translation process as a standard procedure for clinical and research tools (Forkus et al., 2023; Gadeberg et al., 2017; Thompson et al., 2024). Following the translation phase, the validation process is conducted in three stages: a pilot study to evaluate the instrument's linguistic properties, a main study involving administration to the target population, and test-retest reliability assessments. However, the reliability of such instruments can be compromised if the translation process is not rigorously overseen. Specifically, the lack of oversight by expert committees specialized may lead to deviations in translation quality, thereby affecting the validity and reliability of the instrument (Klotz et al., 2023; Leitgöb et al., 2023). A major issue of translations lies the lack of cross-cultural awareness and knowledge of the psychological assessment process.

After its development, the PCL-5 has been validated in several languages including Portuguese, French, Spanish and Arabic as well as in different populations including healthy controls, war veterans, PTSD patients and refugees (Carvalho et al., 2020; Haghish et al., 2024; Ibrahim et al., 2018; Rivest-Beauregard et al., 2022). To our knowledge, there are no other versions with psychometric validation.

We identified several Ukrainian versions of the PCL-5 used internationally. In the previous study of the first author on Ukrainian refugees in Portugal (Figueiredo et al., 2024), the Ukrainian PCL-5 version was adopted from a version provided by the Oxford Health NHS Foundation Trust, based on (Weathers et al., 2013a). While this version has limited references available, it was deemed more reliable in terms of translation and validity, as partially corroborated by Bezsheiko (2016). However, Bezsheiko also notes that the website providing access to the Ukrainian version of the PCL-5 is no longer available. Adding to this inconsistency, studies employing Ukrainian adaptations of the English original (Weathers et al., 2013a) rarely acknowledge source-related limitations, further complicating the assessment of their reliability.

This highlights a significant gap in the documentation regarding the adaptations made by researchers and clinicians to the PCL-5 instrument (Kovács et al., 2023; Lushchak et al., 2024; Zasiekina et al., 2023). Addressing this gap, the present study aims to compare two Ukrainian adaptations and one Russian of the PCL-5 to assess PTSD among Ukrainian refugees in two different hosting countries: the Ukrainian version developed by Bezsheiko (2016), clinically applied to Ukrainian refugees in Germany, and the Ukrainian version by Figueiredo et al. (2024) administered in research context with Ukrainian refugees in Portugal. Additionally, the study will also examine the Russian version of the instrument available online (Los Angeles County, Department of Mental Health, 2019) considering the second author utilized both Ukrainian and Russian versions to accommodate for potential language preferences of Ukrainian refugees in Germany.

Based on our experience in the validation of psychometric instruments, it is important to emphasize the linguistic variability inherent to the Ukrainian language, which parallels the linguistic variants observed in languages like Portuguese and French (Behr and Shishido, 2016; Borsa et al., 2012. Conversely, when authors of adapted versions assume that a specific language translation and adaptation is suitable for all speakers of that language, this assumption is critically flawed because populations vary based on the regions in which they live, despite sharing a common language (Ashbaugh et al., 2016; Bundgaard and Brøgger, 2019; Hoffman et al., 2022). Professionals working in institutional care settings, particularly those involved in early-stage refugee screening, often lack awareness of this linguistic complexity, which can affect the accuracy and cultural appropriateness of the assessments they conduct (Boettcher et al., 2021; Brücker et al., 2023; Wylie et al., 2018). It is imperative to consider the distinct Ukrainian sociocultural and linguistic context, as variations in language and dialectal differences may substantially impact comprehension and the accuracy of responses to standardized measures (Rizzi et al., 2022; Shibuya et al., 2024).

The phenomenon of code-switching, commonly observed among Ukrainian refugees, warrants particular attention, as it reflects broader linguistic adaptations that may have implications for other bilingual or multilingual populations undergoing similar displacement experiences (Bundgaard and Brøgger, 2019; Urbanaviče et al., 2024; Wylie et al., 2018). Code-switching should be analyzed in the context of insights derived from Ukrainian history, as well as the linguistic and cultural variations between the eastern and western regions of the country. Furthermore, the issue of item differential functioning is often overlooked by researchers and scholars utilizing screening instruments for migrant populations, despite some exceptions (Bockhop et al., 2022; Orlando and Marshall, 2002).

Ukraine is a country with 48 million people and spanning over 1,300 km from east to west and nearly 900 km from north to south with Ukrainian being its official constitutional language. Seventy eight percentage of the population are ethnical Ukrainians, 17% are Russians according to the 2001 Ukraine Census (State Statistics Committee of Ukraine, 2003). Ukraine's language landscape is highly diverse across regions and underwent significant changes over time. In 1989, 64% of the population spoke Ukrainian, 33% Russian as their native language, 3% were bilingual in the 1989 USSR Census (Goskomstat SSSR, 1991). The role of Yiddish severely declined from 14% in the early soviet to 4% in the late soviet era due to decreasing Jewish population (Grenoble, 2003). Following the fall of the USSR and Ukraine's independence, Russian native speakers slightly decreased to 30% (State Statistics Committee of Ukraine, 2003). Full-scale invasion by Russia in 2022 induced a major shift in the language preference of Ukrainians. The preference for Russian as the first language dropped to 13% in a nationally representative survey with ~2,000 respondents in 2024. Interestingly, the Ukrainian speaking population stayed at 63%, while the amount of respondents who considered themselves bilingual increased to 24% (Kulyk, 2024).

Historically, the eastern regions have been under strong influence of the Russian language and culture, the western regions have been influenced by Poland, Moldova and Romania, the northern regions by Poland and Belarus, and the south was affected by Turks and Tatars (Cheskin and Kachuyevski, 2019). Thus, Russian language is especially prevalent (50% or more) in the eastern and regions of Ukraine such as Luhansk, Donetsk, Kharkiv, Kherson and Crimea, while the vast majority of the population in the west-Ukrainian regions such as Lviv, Vinnytsia and Khmelnytskyi prefer Ukrainian (State Statistics Committee of Ukraine, 2003; Kudriavtseva, 2021). In the central regions of Ukraine such as Cherkasy, many people speak so-called Surzhyk, a Russo-Ukrainian pidgin (Bilaniuk, 2004; Hentschel and Palinska, 2022). Furthermory, there are differences in the Ukrainian language, divided into northern, south-eastern and south-western dialects (Medynska et al., 2023). The first two groups can be located within Ukrainian-Belorussian and Ukrainian-Russian continua, the third group comprises of Volhynian–Podolian, Galician, Bukovynian and Transcarpathian dialects with Polish, Slovak, Hungarian and Romanian influences.

In summary, firstly, an east-west special gradient, and secondly, a temporal gradient from Russian to Ukrainian is apparent. Several Ukrainian dialects add to the complexity (Figueiredo and Silva, 2013; Figueiredo, 2019). Also the investigator needs to keep in mind that language always (Figueiredo and Silva, 2013; Figueiredo, 2019)—and even more so after 2022—has been a highly politicized matter in Ukraine, with Ukrainian being associated with expression of ethnic identity, national independence and a sense of cultural belonging (Palinska, 2023). Since Ukrainian currently undergoes rapid changes, with quickly developing military-related loanwords and neologisms (Del Gaudio, 2024), regular linguistic evaluations of any psychometric instrument will be required.

Initially, a back-translation of the two Ukrainian and the Russian versions of the PCL-5 into English was conducted using the DeepL translation program (DeepL SE, 2025). Subsequently, corrections were made by a native Ukrainian bilingual fluent in both Ukrainian and Russian, who ensured precise word-for-word accuracy. A bilingual psychologist proficient in Ukrainian and Russian refined these corrections. Please refer to in the Supplementary Table 1 for item-wise back-translation and the original version.

The analysis focused on word choices, semantic nuances, and inconsistencies identified across the translations as compared to the original English original. The emphasis was placed on preserving the original meaning inherent in the construct being assessed, ensuring that the translated items align with the conceptual framework of the PCL-5.

Supplementary Table 1 provides a detailed summary analysis of each PCL-5 item across the three translated versions, and with comparisons to the original version developed by Weathers et al. (2013a). At first glance, different terminology for the traumatic event becomes very expressive. In the original version (OV), as well as in the first Ukrainian version (UV1), it is referred to as “stressful experience”. In the second Ukrainian version (UV2) it is referred to it as a “stressful event/situation”, while in the Russian version (RV) it is named as “experienced stress”. These translations result in subtle semantic differences. A stressful experience, and even more so a stressful event/situation, implies one concrete, specific event, while “experienced stress” can be understood as a less acute and rather prolonged process, such as chronic life stress over several months and years. This can lead to profound misinterpretations of all items referring to this stress.

Differences in sentence structure also catch the eye. In the OV, all items are phrased as questions, omitting the verb like “are you” or “do you have”. The translated versions are less consistent in their structure. UV1 follows the original but items 16 and 17 step out of line. In the UV2 the items are not phrased as questions except 19 and 20. A part of the items is phrased addressing the 2nd person (“you”). In the RV, items are phrased as questions, but again, a part is phrased addressing the 2nd person. These structural inconsistencies within the scale can deteriorate its readability and understanding.

There are different translations for words like “upset” and “distressed”, or “anxiety”, “fear” and “horror” or “alertness” and “vigilance” or “detached”, “distant” and “separated” which are similar but have individual subtle semantic differences. In item 3 in the UV1 the explanatory sentence within the brackets is omitted, which may negatively impact the understanding of this item. In item 7 of the UV1, “conversations”, “actions” and “situations” are omitted in the description within the brackets. In item 9, the OV states “the world is completely dangerous”, which also occurs word by word in UV2. Item 9 in the RV includes “the world is very dangerous”, while the UV1 contains the statement “the world is a dangerous place”. These are semantically quite different statements and comprise gradations from “complete” to “very” to a simple “dangerous”, which will likely result in different scores for this item. In item 12, the RV translated “activities” as a “job” which obviously has a different meaning and may result in completely different scores. The translations of item 18 differ a lot between the versions. While it says “feeling jumpy or easily startled” in the OV, the translations range from “feelings of constant tension” to “anxiety or fearfulness” to “nervousness or anxiety”. These are different emotions, which are once again highly likely to result in different scores.

Our comparison, with no precedents in this field, revealed that none of the 20 PCL-5 items across the three versions aligned entirely with the English original, compromising the integrity of the construct being assessed. Furthermore, the nuances of emotional expression were not consistently captured, showing lexical and semantic differences inherent to the translations. These discrepancies may undermine the reliability of emotional assessment, potentially affecting the individuals' ability to accurately convey their experiences (Shih, 2024; Zaid et al., 2021). Table 1 illustrates a summary of observed discrepancies in a color-dot scheme.

The 20 items of the PCL-5 are designed to assess symptoms such as anxiety, intrusive memories, and abnormal daily behaviors. However, the observed linguistic and semantic variances particularly affect the wording associated with specific emotions, including the expression of emotional responses tied to the memory of traumatic experiences.

In sum, relatively high semantic incongruence regarding conceptual meaning was observed across all versions as compared to the original English original, posing a risk to the validity of the assessment and underscoring the necessity for meticulous translation and cultural adaptation processes.

We adapted four established text similarity metrics from machine translation evaluation to assess the degree of correspondence between original and back-translated psychometric items. Each metric addresses a different aspect of textual overlap:

Calculations were performed in customized Python scripts written with assistance of ChatGPT (OpenAI, 2025) using NumPy (Harris et al., 2020), Matplotlib (Hunter, 2007), Natural Language Toolkit (Loper and Bird, 2002), pandas (McKinney, 2010) and scikit-learn (Pedregosa et al., 2011) libraries.

For all metrics, scores range from zero (no similarity) to one (identical). In our context, ≥0.95 indicates near-identical form, 0.85–0.94 high similarity with minor wording differences, 0.70–0.84 moderate similarity potentially affecting nuance, and < 0.70 low similarity warranting careful review for conceptual equivalence.

Statistical analysis was performed in Python. Similarity metrics and expert scores across the three back-translated versions were analyzed using one-factor ANOVA. Spearman's rank correlation was used for analysis of correlation between similarity metrics and expert scores. All values are presented as mean ± standard error of the mean if not stated otherwise.

Two experts scored similarity of each back-translated item to its original on a scale from one to three; with a score of one being the low similarity indicating significant semantic or structural discrepancy; two indicating medium similarity with some differences but closely matching the English original; and a score of three indicating high similarity with minor or no differences.

Please refer to Table 2 for item-wise expert ratings and automated similarity metrics. In the expert rating, translated versions scored between 2.05 ± 0.18 (UV1) to 2.35 ± 0.17 (RV) and 2.40 ± 0.15 (UV2), indicating mediocre similarity with possible semantic or structural discrepancies to the original. Albeit ANOVA revealed no statistical difference in the expert ratings between these translations [F_{(2, 57)} = 1.27, p = 0.29], low scores in domain E of PCL-5 strike the eye especially in UV1 (1.67) and UV2 (2.00), while the highest scores prevail in domain B, ranging from 2.40 in UV1 to 2.80 to UV2 and in domain C (a score of 3.00 in both, OV2 and RV).

Damerau–Levenshtein, Levenshtein and METEOR scores all ranged between 0.55 and 0.62, while Soft Cosine scores ranged between 0.74 and 0.75. These metrics indicate low to moderate similarity to the original, with a relatively high likelihood for loss of nuance and semantic differences. These scores did not differ significantly amongst translations [F_{(2, 57)} = 0.18, p = 0.84; F_{(2, 57)} = 0.04, p = 0.96; F_{(2, 57)} = 0.18, p = 0.84; F_{(2, 57)} = 0.03, p = 0.97, respectively]. Similarly to expert ratings, lowest scores were found in domain E (0.31–0.59 in the first three methods and 0.60–0.67 in Soft Cosine]. Apparently, translation of items regarding trauma-related arousal and hyperreactivity poses a linguistic challenge.

Then we calculated Spearman's rank correlation (ρ) between expert ratings and semantic discrepancy scores (Table 3). For all three back-translations, we found a strong correlation between expert ratings and Damerau–Levenshtein and Levenshtein scores (ρ between 0.59 and 0.75, p < 0.01 to p < 0.001). For METEOR and Soft Cosine the correlation with expert ratings was weak and statistically insignificant (ρ between 0.17 and 0.34, p > 0.1). Thus, Damerau–Levenshtein and Levenshtein, but not METEOR nor Soft Cosine, seem suitable for semantic validation of translations of psychological instruments.