We searched for articles published in the English language after January 1995 and listed in the PsycINFO, MEDLINE, and PubMed databases. January 1995 is when the first studies based on the DSM IV diagnosis handbook could have emerged. The search was last updated in November 2016. We used a number of textword combinations related to CP, such as “chronic pain,” “migraine,” “chronic daily headache,” “fibromyalgia,” “widespread pain,” “musculoskeletal pain,” “rheumatoid pain,” “chronic back pain,” and “chronic spinal pain” combined with textword for posttraumatic stress “Stress Disorders,” “post-traumatic/,” and “PTSD.” In addition, all the reference lists of the studies included were hand searched for further relevant publications. The search string for the search in the Embase, MEDLINE, and PsycINFO databases is attached as Supplementary material.

After eliminating all duplicate studies, two reviewers independently examined the titles and abstracts of all the extracted articles. We retrieved the full texts of articles that were considered, by at least one reviewer, to meet the inclusion criteria. The inclusion criteria were studies published in English reporting the point prevalence (within 1 month) of PTSD based on a structured assessment according to the PTSD criteria of the DSM IV or the International Statistical Classification of Diseases and Related Health Problems (10th Edition; ICD-10). Exclusion criteria were military-only samples, samples recruited from a mental health-care setting, studies using non-structured assessments (such as informal clinical interviews) or assessments that did not represent sufficiently the diagnostic criteria for PTSD. Migraine was considered a CP disorder, based on the assumption that it is a chronic disorder, despite being characterized by episodic presentation of symptoms (24). After retrieving the full texts, the reviewers extensively scrutinized them to confirm the eligibility of the studies based on the above criteria. Any disagreements relating to inclusion eligibility were resolved through discussions.

The reviewers (Ajmal Hussain and Johan Siqveland) extracted information about sample size, number of participants with PTSD, pain location, the setting of recruitment, the method of PTSD assessment, and nationality. Both reviewers agreed on the final data extraction, ensuring accuracy, and disagreements concerning data selection were resolved via discussion between them.

We assessed the risk of bias using an 8-item scale (range 0–8; see Supplementary Material) adapted for this study from the EBMH Notebook list for evaluating risk of bias in prevalence studies (25). This scale comprises various criteria for evaluating sources of bias, such as the quality of the diagnostic assessments and sample size. We gave the studies included one point for each criterion met, with higher scores, therefore, indicating a lower risk of bias. Studies that scored a seven or eight were classified as having a low risk of bias. Those scoring between three and six, and those scoring between 0 and 2, were of medium and high risk, respectively.

We hypothesized that the prevalence of PTSD would vary according to pain location, assessment method, and population, so we performed three analyses of subgroups. In the first subgroup analysis, studies included were grouped according to pain location: (a) headaches, including migraines, tension-type headaches, chronic daily headaches, and other headaches; (b) back pain, including pain originating from the spine and chronic lower back pain; (c) CWP, according to the American College of Rheumatology criteria of widespread pain or FM syndrome; and (d) other CP conditions including peripheral neuropathic pain or mixed CP. In the second subgroup analysis, we divided studies according to the population studied: (a) participants recruited from clinical settings and (b) participants from non-clinical settings. Finally, in the third subgroup analysis, we grouped the studies according to the assessment method used: (a) self-reporting of assessed PTSD symptoms and (b) a structured clinical interview for assessing a PTSD diagnosis.

As we had expected there to be significant heterogeneity across the studies included, we used a random-effects model. We analyzed the prevalence of PTSD using a binomial-normal random-effects meta-regression model (26) in the “metafor” R package (27). A random-effects model allows for explicit modeling of the heterogeneity of the parameters investigated in each study, assuming that the true logit transformed prevalence rates are normally distributed among the studies. We calculated the 95% confidence interval (95% CI) for all estimates. Heterogeneity was assessed and reported using the I² statistic, which describes the variance between studies as a proportion of the total variance. The I² statistic varies between 0 and 100%, with 0% being interpreted as no heterogeneity, 25% as low heterogeneity, 50% as medium heterogeneity, and 75% and above as a high degree of heterogeneity (28).

We categorized the studies according to pain location, population, and assessment, including them in the model as categorically coded predictors in three separate sub-analyses. Finally, we estimated the regression coefficients and transformed them into prevalence rates for easier interpretation and reporting.

After excluding duplicates, the literature search yielded 1,589 records. In the initial screening of abstracts, 1,473 articles were excluded, as they did not meet the inclusion criteria. After reading the full texts, we excluded 95 studies the reasons for this presented in Figure 1.

The prevalence of PTSD in the studies included is given in Table 1. A forest plot depicting the prevalence of PTSD in the included studies is presented in Figure 2.

Twenty-one articles with a total of 6,750 participants fulfilled the inclusion criteria and were included in the analysis. These 21 studies varied in sample size between 60 patients with migraine (38) and 1,478 patients with orofacial pain (30). Six of the studies included participants with widespread pain, seven studies included participants with headaches, and two studies included participants with back pain. Six studies included participants with other pain conditions or mixed pain. Three structured or semi-structured diagnostic interviews to diagnose PTSD were used in 10 of the studies, with four questionnaires used in 11 studies. All studies made diagnostic categorization according to the DSM IV criteria for PTSD.

The pooled prevalence across all studies was 9.7% (95% CI 5.2–17.1). The lowest prevalence was reported in a population-based study of persons with lower back pain (n = 110), where none of the participants had PTSD (32). In contrast, the highest prevalence was reported in a study of patients with FM (n = 77), where 57% of participants had PTSD (29).

The prevalence rates of PTSD with 95% CIs and heterogeneity statistics according to the location of the pain, the study population, and the assessment method are presented in Table 2.

The seven studies of persons suffering with headache (n = 1,555) indicated a PTSD prevalence ranging from 2.0 to 28.3%, with a pooled estimate of 11.2% (95% CI 5.1–22.8%). The six studies of persons with CWP (n = 2,189) indicated a PTSD prevalence from 4.8 to 57.0%, with a pooled estimate of 20.5% (95% CI 9.5–39.0%). The two studies on persons with back pain (n = 675) demonstrated PTSD prevalence rates of 0 and 0.6%, respectively, with a pooled estimate of 0.3% (95% CI 0.0–2.4%). The six studies including other pain conditions or mixed pain conditions reported PTSD prevalences from 1.6 to 29.2%, with a pooled estimate of 9.1% (95% CI 3.8–20.5%).

The 16 studies of clinical samples (n = 4,719), reported PTSD prevalences from 0.6 to 57%, pooled estimate 11.7% (95% CI 6.0–21.5%). The five studies with non-clinical samples (n = 2,059), reported PTSD prevalences from 0 to 25.7%, pooled estimate 5.1% (95% CI 1.4–17.2%).

The 11 studies where PTSD was assessed with a structured clinical interview (n = 1,790) indicated PTSD prevalence rates from 0.6 to 57%, pooled estimate 4.5% (95% CI 2.1–9.3%). The 10 studies that included self-reported assessments of PTSD (n = 4,988) reported a PTSD prevalence from 4.8 to 45.8%; pooled estimate 20.4% (95% CI 10.6–35.5%).

The total heterogeneity in the included studies was very high (I² = 98.6). To understand further this large variation in the included we also calculated heterogeneity estimates in the subgroup analysis based on location of pain, population, and assessment method. However, this did not reduce the heterogeneity substantially (I² 94.2–98.6). Because PTSD is more common in women than it is in men (47), we also analyzed the studies with gender as a co-variate. However, we found that this did not reduce heterogeneity either (I² = 98.3).

In Table 1, scores for the risk of bias are presented. Of the 21 studies included here, three had a low risk of bias and 18 had a medium risk. The lack of comparison between responders and non-responders, unclear exclusion criteria, and self-reported PTSD symptoms were the most common causes for the higher levels of risk of bias.

We applied strict inclusion criteria in terms of only including diagnoses according to the latest diagnostic manuals for mental disorders, which had been available for long enough for studies to have emerged. A later version of the DSM manual for psychiatric diagnosis, DSM 5, was published during the course of this work; however, none of the studies included here involved making diagnostic evaluations according to this newer manual. Despite the attempts to make the inclusion strict to limit heterogeneity between the studies, the meta-analysis showed that the PTSD prevalence varied widely in that there very large amounts of unexplained variance across the studies. Therefore, the pooled mean prevalence has to be interpreted cautiously. The groups included in this review were also heterogeneous in terms of type of pain conditions and degree of clear somatic origin due to the inclusion criteria in the primary studies. The difficulties with finding homogenous associations between PTSD and pain is further hindered by our lack of understanding of the causal relationship between the conditions and more research and model development on this area is clearly needed.

Few of the studies included here gave an adequate description of the recruitment process or rate of non-participation. However, this represents the most common risk of bias in these studies, and so the participants included may not be fully representative of the CP population. In addition, the number of studies, and hence the participants, included was limited. Even for the most extensively studied pain condition, headaches, only seven studies including 1,555 participants were included. Only articles written in English were included, but as the prevalence of PTSD varies between different countries, adding publications in other languages might have altered our findings. Finally, as some studies included here relied on self-reported CP, not all participants in these studies may have met the formal criteria for a CP disorder.

The identification of mental disorders in persons seeking treatment for CP may be important for several reasons. Pain treatment is often long term and requires good adherence from the patient. A diagnosis of PTSD is related to poor adherence and, therefore, it is important to assess whether this exists alongside the CP (54). The current review indicates that PTSD is prevalent in persons with CP, and particularly among persons seeking treatment for pain, and in those with widespread pain. Clinicians treating CP should be aware of the increased risk of comorbidity with PTSD in their patients, and particularly in patients with widespread pain. In addition to the risk of non-adherence, PTSD is related to an increased risk of substance use disorder (55), so the possibility for dependence needs to be weighed against the benefit for the patient before starting treatment with opioids. Interventions with proven effectiveness in the treatment of both PTSD and CP [such as cognitive behavioral therapy (56, 57), somatic experiencing (58), or antidepressants (59, 60)] may be considered as the first treatment in persons with comorbid pain and PTSD.