Published on 15 September 2026
Complex PTSD prevalence: what can be concluded from 167 studies?
In brief
A systematic review with meta-analysis brings together 167 studies and 138,681 participants to estimate the prevalence of complex post-traumatic stress disorder (CPTSD), the entity introduced by ICD-11. The global pooled prevalence, all samples taken together, is 6.2% (95% CI 3.7 to 10.3%). In trauma-exposed samples alone it reaches 12.4% (95% CI 7.7 to 19.3%). It is highest in clinical samples (44.7%), in survivors of domestic violence or sexual abuse (40.0%) and in military samples (36.4%), and lowest in emergency services (7.4%). The decisive point is not the average, it is the spread: in the detailed table of the supplementary material, the prevalences reported study by study run from 0 to 93%. At that scale of variation, a pooled prevalence describes the range of what has been observed from one study to the next; it does not supply an expected value transposable to a given caseload. No gender difference in prevalence was observed, which is not the same as a demonstration that none exists. What the study supports robustly is simpler, and it is enough to change a consultation: in trauma-exposed populations, and still more in care settings, CPTSD is common, and not looking for it amounts to not seeing it.
The context
ICD-11 separates two diagnoses where everyday practice often handles only one. PTSD in the strict sense combines re-experiencing, avoidance and a sense of current threat. CPTSD adds three dimensions known as disturbances in self-organization: affective dysregulation, a negative and persistent self-concept, and lasting difficulties in interpersonal relationships. Those three dimensions are not adjectives of severity, they are criteria.
DSM-5-TR does not retain CPTSD as a distinct diagnosis. A clinician who reasons with DSM and uses a self-report questionnaire built on DSM therefore has no item exploring those three dimensions. The question this meta-analysis puts is one of frequency: if CPTSD accounts for only a fringe of post-traumatic presentations, the distinction stays academic; if it accounts for a substantial share, the failure to explore it becomes a case-finding problem. This study addresses neither the timetable on which a given health system moves to ICD-11, nor the availability of validated translations of the instruments used.
The study at a glance
| Item | Content |
|---|---|
| Population | Trauma-exposed samples and potentially trauma-exposed samples, all ages, across a large number of countries |
| DetailThe authors separate samples actually exposed to trauma from potentially exposed samples, and run the publication bias analyses separately on those two sets. The geographical reach is wide: the sample table lists dozens of countries, and two of the included works are themselves multinational, one covering 39 countries and the other 52. | |
| Measurement | CPTSD prevalence measured with validated ICD-11 assessments |
| DetailThe authors list among their limitations the lack of clinician-administered instruments. A prevalence measured by questionnaire is not a prevalence of diagnoses made at interview. | |
| Literature search | Five databases searched up to 31 January 2025, peer-reviewed articles |
| DetailWeb of Science, Scopus, PsycNET, Psych and Behavioural Sciences Collection and PubMed. The supplementary material documents two search waves, on 21 November 2023 and then on 31 January 2025, and English-language limiters on three of the five databases. | |
| Outcome | Pooled prevalence of CPTSD, overall and by population subgroup |
| DetailRandom-effects model, meta package in R. Prevalences are pooled after a Freeman-Tukey (double arcsine) transformation, as the axes of the funnel plots in the supplementary material indicate. Gender-specific analyses were conducted where prevalence was reported separately for women and men. | |
| Design and numbers | Systematic review with random-effects meta-analysis: 167 studies, 138,681 participants |
| DetailThe quality of the included studies was assessed with the Joanna Briggs Institute (JBI) checklist for prevalence studies. No study was excluded on that basis. | |
Quality control
A prevalence meta-analysis has no randomisation, no masking and no comparator arm: the RoB 2 tool does not apply to it. Two things can be checked here: the conduct of the review itself, and the quality of the studies it aggregates, assessed with the JBI checklist.
| Domain | Judgement |
|---|---|
| Search strategy | Five databases, two waves, full search string |
| FindingThe search string is published database by database, with the number of records returned at each of the two dates. That is the expected level of transparency, and it is met. | |
| Scope of the corpus | English, peer-reviewed journals |
| FindingThe search is restricted to articles published in peer-reviewed journals, with English-language limiters on several databases. Both restrictions are defensible in the name of standardisation, but they remove from the corpus work often conducted in under-represented populations. The direction in which they move the pooled prevalence cannot be determined here. | |
| Quality of the included studies | JBI applied, no exclusion |
| FindingThe JBI tool is the one recommended for prevalence studies. The authors state that no study was set aside after this assessment: the checklist therefore describes the corpus, it does not filter it. | |
| Heterogeneity of the samples | Samples not comparable with one another |
| FindingThe table of characteristics gathers under a single estimate nationally representative samples, students recruited online, inpatients, asylum seekers and veterans. This is the structural limitation of the whole edifice: the pooled prevalence aggregates populations for which nothing guarantees that a single quantity is being estimated. | |
| Definition of trauma | Inconsistent definitions, limitation acknowledged |
| FindingThe authors themselves list among their limitations the inconsistency of trauma definitions from one study to the next, together with the lack of clinician-administered instruments. The detailed table confirms the diversity of the exposure assessment tools. | |
| Publication bias | Clearly asymmetric funnel plots |
| FindingThe four funnel plots in the supplementary material show marked asymmetry on both sets of samples, exposed and potentially exposed. A trim-and-fill procedure was applied: it imputes a large number of missing studies on the low-prevalence side, which points towards an overestimation of the pooled prevalences by the published corpus. The size of that correction does not appear in the material consulted. | |
| Funding and declared interests | Public and charitable doctoral scholarships |
| FindingThe first author reports financial support from Australian Rotary Health and the Rotary Club of Victor Harbor in the form of a PhD scholarship; another author is supported by an Australian Government Research Training Program stipend. The authors declare no other financial interests, relationships, patents or activities that could be perceived to influence the submitted work. No industry funding is declared. | |
The findings
| Result | Reported value |
|---|---|
| Global pooled estimate | 6.2% (95% CI 3.7 to 10.3%) |
| ReadingThis is the estimate across all samples taken together, not a general population prevalence. The distinction matters: the authors indicate that prevalence is lower in the general community and among emergency and health workers than in exposed populations. | |
| Trauma-exposed samples | 12.4% (95% CI 7.7 to 19.3%) |
| ReadingThis is the estimate closest to a clinical question: among people whose trauma exposure is established, about one in eight meets the ICD-11 criteria for CPTSD, with a confidence interval that is already wide. | |
| Clinical samples | 44.7% |
| ReadingThe highest subgroup. What counts as a clinical population varies from one study to another, which contributes to the observed spread. No confidence interval accompanies this figure in the material consulted. | |
| Domestic violence or sexual abuse | 40.0% |
| ReadingThe second highest subgroup. Consistent with the theoretical construction of the diagnosis, which targets interpersonal and repeated trauma, without this work being able to establish a causal relationship. | |
| Military samples | 36.4% |
| ReadingThe authors put forward, as explanatory hypotheses, the intensity of the exposure, limited support and cultural factors. These are interpretations, not measured results. | |
| Emergency services | 7.4% |
| ReadingThe lowest subgroup. The authors’ hypothesis, selection at recruitment and support systems, is likewise an interpretation. A healthy worker effect cannot be ruled out on this kind of data. | |
| Spread between samples | Prevalences reported from 0 to 93% depending on the sample |
| ReadingRead off the detailed table of the supplementary material. The highest values come from samples made up of people already diagnosed with PTSD, the lowest from community samples or controls. A new study, run tomorrow in a new setting, could produce almost any value in that range. | |
| Gender | No difference in prevalence observed |
| ReadingThe authors note that this finding contrasts with the typically higher rates of PTSD in women. Not observing a difference is not demonstrating that it is nil: the distinction is not rhetorical, it forbids writing that men and women are equally affected. | |
| Publication bias | Asymmetric funnel plots, trim-and-fill procedure applied |
| ReadingThe figures in the supplementary material show a clear asymmetry and a large number of studies imputed on the low-prevalence side. The practical conclusion is that the pooled estimates are more likely inflated than understated. | |
Critical appraisal
| Domain | Judgement |
|---|---|
| Spread and meaning of the pooled estimate | Major limitation |
| FindingWith individual prevalences running from 0 to 93%, the pooled prevalence stops being a point estimate and becomes a description of the spread. This is not a footnote reservation: it governs the reading of every figure in the paper. The subgroups proposed, by population type, explain part of that spread without resolving it. | |
| Recruitment of the samples | Populations not comparable |
| FindingNationally representative samples, students, patients, refugees, veterans, emergency workers: the corpus mixes recruitment methods whose representativeness is not of the same order. The estimates meant to hold for the general community are the most vulnerable to this limitation. | |
| Measurement of the diagnosis | No clinician interview |
| FindingA limitation the authors acknowledge. Aggregating prevalences obtained with different instruments and different thresholds, without a diagnostic interview, is a source of spread in itself. How much of that spread is attributable to the instrument alone is not established by this work. | |
| Absence of a gender difference | Observation, not demonstration |
| FindingNo difference was observed, and the supplementary table shows that several studies do not report prevalence by gender. In a corpus this dispersed, a real but moderate effect can stay undetectable. The result to retain is the absence of a conclusion, not a demonstrated equality. | |
| Publication bias | Marked asymmetry |
| FindingTwo elements compound each other: a clear asymmetry of the funnel plots, corrected by trim-and-fill in the direction of overestimation, and a corpus restricted to peer-reviewed journals and largely to the English language. | |
| Transposition to a given health system | Not addressed by the study |
| FindingNo country-specific result is isolated here. The study addresses neither the actual use of ICD-11 in a given health system, nor the existence of validated translations of the instruments used. Both points have to be established elsewhere, and not inferred from this work. | |
| Proportionality of the conclusions | Measured conclusions |
| FindingA point to the authors’ credit: the limitations are stated in the abstract itself, “inconsistent trauma definitions and lack of clinician-administered instruments”, and the conclusion does not go beyond what the data allow. The risk of hardening lies downstream, in the reuse of a single subgroup figure. | |
Level of evidence
Where confidence is high, and where it is not
Confidence is high on three points. First, the existence of a considerable corpus: 167 studies and 138,681 participants, assembled by a search documented database by database and at two dates. Second, the qualitative finding that CPTSD is common, and not marginal, in trauma-exposed populations and more so still in samples recruited in care settings. Third, the fact that this finding recurs across very different cultural contexts, including in two multinational works that between them cover dozens of countries.
Confidence is low on everything that is a number. None of the pooled prevalences can be used as an expected value in a given population, and saying so is not methodological over-caution: it follows directly from a spread running from 0 to 93% from one sample to the next, and from a publication asymmetry that works in the direction of inflation. Confidence is also low on the comparison between genders, whose result is inconclusive, on the comparison between measurement instruments, which is not what this work is about, and on any projection to a particular national setting, which is not addressed. The design of a systematic review with meta-analysis qualifies how the evidence is organised, not the solidity of each estimate that comes out of it.
The colleague test
What an experienced colleague would say if you put this study to them in two minutes, between two consultations.
“ The figure that stops me is 44.7% in clinical samples. But I look at the detail right next to it, studies running from 0 to 93%, and I tell myself that this average gathers samples that have nothing to do with each other, students recruited online and displaced populations. So I do not turn it into a rate for my own patients, that would be dishonest. What I keep is something else: when I see someone with post-traumatic stress, I now ask the questions about emotion regulation, self-image and relationships. Before, I did it when it crossed my mind. And I cannot say whether it is more frequent in women, this study does not settle that. ”
What this means in practice: the usable result is not the number, it is the shift of clinical attention. The question is no longer only whether there is a trauma and re-experiencing, but what is happening to affect regulation, self-image and the bond with others.
What you can do with this
- Explore the three disturbances in self-organization systematically in a patient exposed to trauma: affective dysregulation, persistent negative self-concept, lasting relational difficulties. This is the best supported implication of the work, and it costs three questions.
- Know that the instrument shapes what you find. The PCL-5 is built on DSM-5 criteria and does not explore those three dimensions; instruments aligned with ICD-11 do cover them, although the factor structure of the International Trauma Questionnaire has limitations of its own. The study does not establish which of these strategies improves patient outcomes, and it says nothing about the availability of a validated translation in a given language.
- What can be said to a patient: that after certain traumas, interpersonal and repeated ones in particular, the difficulties often go beyond nightmares and re-experiencing, and reach the way one sees oneself and enters into relationships; that this is not a personality trait, and that it is described in the international classification.
- What cannot be said: “about one patient in two like you has CPTSD”. The 44.7% in clinical samples is an average across settings too different for it to serve as an individual prognosis.
- Do not conclude anything about gender. Faced with a woman or a man, this study licenses neither raising nor lowering suspicion according to sex: no difference was observed, and that is not enough to conclude equality.
Frequently asked questions
Almost one patient in two in clinical populations: can I transpose that figure to my own caseload?
No. The 44.7% aggregates clinical samples whose definition varies from one study to another, and whose individual prevalences, in the detailed table, range from 0 to 93%. The figure shows that the phenomenon is common in care settings, it does not give a probability applicable to a patient or to a caseload.
Does CPTSD appear in DSM-5-TR?
No. CPTSD is an ICD-11 entity. DSM-5-TR does not retain it as a distinct diagnosis. That is why a clinician working with DSM-derived tools may never meet the question.
Should the PCL-5 be dropped in favour of an ICD-11 aligned instrument?
The study does not compare these two case-finding strategies and does not measure their effect on management. It establishes only that the symptoms explored by ICD-11 instruments and not by the PCL-5 are common in trauma-exposed populations. Adding a tool is a matter of clinical judgement, and the question of a validated translation is not addressed by this work.
Does the absence of a gender difference mean that CPTSD affects men as much as women?
No. No difference was observed, which is not the same as a difference measured as nil. Several included studies do not report prevalence by gender, and the spread between studies is such that a real but moderate difference can stay undetectable. Absence of evidence of a difference is not evidence of absence of a difference.
Do estimates this dispersed not make the meta-analysis useless?
They change what it is for. Such a spread forbids reading the pooled prevalence as a point estimate, but it does not cancel the information: the work documents the size of the corpus, the presence of the phenomenon on every continent, and the ranking of populations, from emergency services at 7.4% to clinical samples at 44.7%. It is a map, not a measurement.
Annotated bibliography
Huynh PA, Kindred R, Perrins K, de Boer K, Miles S, Bates G, Nedeljkovic M (2025). Prevalence of Complex Post-Traumatic Stress Disorder (CPTSD): A Systematic Review and Meta-Analysis. Psychiatry Research, 351, 116586. DOI 10.1016/j.psychres.2025.116586 · PMID 40570696. Source study analysed here. What it brings: a broad aggregation of the literature on CPTSD prevalence, with a useful distinction between exposed and potentially exposed samples, a ranking of populations, and conclusions proportionate to the data. Its limitations: a spread of individual estimates running from 0 to 93% that deprives the pooled estimates of point value, recruitment methods that are not comparable, marked asymmetry of the funnel plots, a corpus restricted to peer-reviewed journals and largely English-language, and measurements resting on questionnaires rather than diagnostic interviews.
International Classification of Diseases, eleventh revision (ICD-11), World Health Organization. What it brings: the operational definition of CPTSD, with its three disturbances in self-organization, on which every measurement in this meta-analysis rests. Its limitations: it says nothing about actual adoption in a given health system, nor about the availability of instruments validated in a given language. A nosographical reference point, outside the scope of the study.
Joanna Briggs Institute checklist for prevalence studies. What it brings: the quality assessment tool used here, suited to prevalence studies, where the risk of bias tools designed for trials do not apply. Its limitations: in this work it describes the corpus without filtering it, since no study was excluded on its basis.
What was consulted. References verified on 12 August 2026 against the published version (publisher record, abstract and highlights) and against its 87-page supplementary material. The full text of the article could not be consulted: values appearing neither in the published abstract nor in the supplementary material are not carried here. DOI, volume, pagination, funding and declaration of interests were taken from the PubMed record, whose title and author list match the document read exactly. This analysis underwent an independent double reading.