2.1 Participants

A sample of 47 adult out-patients formed the depressed pain group. They were recruited from outpatient pain clinics and general practices in London, UK. To be eligible for the study, participants had to read and write English fluently enough to be able to accurately complete the stem completion task, answer interview questions and fill in questionnaires. The inclusion criteria for depressed pain patients were musculoskeletal pain (e.g. lower back, neck, shoulder etc.) persistent for more than 3 months. Patients were only included if they rated their current level of pain, and the level of pain that they had experienced in the past few months as 3 or above on an 11-point Numerical Rating Scale (NRS), where 0=no pain and 10=extremely painful [22]. General practitioners and clinicians excluded patients with “red flags” [23], progressive disorders such as cancer, significant learning disabilities, psychotic symptoms, dementia-type syndromes, addiction problems or bipolar disorders, as well as people under the age of 18 and pregnant women. A criterion of ≥8 on the HADS Depression Scale was established for inclusion in the depressed pain group [24].

A non-clinical control sample was also tested. This comprised a diverse group of 42 healthy control participants recruited through a large population survey [25], which was based on records from several general practices. Control participants were screened twice: first, they were screened by the General Health Questionnaire in the initial survey study by Parsons et al. [25], and second, screening questions were asked by the first author at the beginning of the appointment. Control participants have been matched for socio-economic status, age and gender and were tested in general practices. Control participants were included only if they had a pain-free history, were free of severe and disabling mental and physical illness and were not receiving medication or counselling for these disorders. Potential control participants were excluded if they scored higher than 2 on the NRS, above 7 on the HADS depression subscale or SCID criteria were present. Participants in both groups were tested sequentially on recruitment into the study until the target numbers had been met.

Diagnoses of current major depression were established with standard structured clinical interviews – the Structured Clinical Interview for DSM-IV (SCID; [26]). A trained clinical psychologist conducted all interviews and scored all interviews. The psychologist was blind to questionnaire scores. A depression symptom severity score was calculated as the sum of DSM-IV depression symptoms in the SCID (coded as 1=absent, 2=subthreshold and 3=present).

2.2 Procedure

Patients were recruited on the premises of the pain clinics, where the general practitioners and clinicians in respective clinics and surgeries informed patients about the study and handed them invitations to participate. All patients attending consultations with the participating general practitioners and clinicians, who fit in the inclusion criteria, were approached. All participants provided information about demographic and clinically relevant descriptive data (gender, age, education level, pain duration, pain severity etc.). Of those who left their details with the researcher, only 7.2% did not take part due to difficulties in attending the appointment. All participants provided informed consent. The University Ethics Committee and London Research Ethics Committee (LREC) approved this study. After consenting to the study, participants were informed that the research was about how people think when they have pain. To avoid possible priming from the HADS, the SCP was always presented first. Each participant completed the measures in a single assessment session.

2.3 Measures

2.4 Development of consensus agreed coding rules and development of coding manual

Theory-driven coding criteria were generated by a panel of experts working in the field of chronic pain (n=4) and pain-related co-morbidities and the coding criteria were agreed upon in consensus meetings, where ambiguous sentence completions were discussed and ambivalent ratings were resolved. All experts had previously conducted research using the Sentence Completion Task in pain populations [12], [13], some had previously conducted systematic reviews and meta-analyses of psychological risk-factors and psychological treatments for chronic pain [43], [44], [45] and all were thoroughly familiar with the field. The author blind-coded all the sentence completions using the manual developed for this purpose. The first part of the manual contains guidelines on defining the valence and content of completions and the second provides a glossary of responses developed from previous research in the area of pain and depression [12], [46].

2.5 Statistical analyses

The following statistical analyses were conducted using the Statistical Package for Social Sciences (SPSS for Windows, version 25.0). Chi-square analyses and independent t-tests were used to compare the two groups (depressed pain group versus healthy control group). Where t-tests have been used to compare the groups, in the majority of cases in the following analyses, the t value is reported assuming equal variances in the two groups. For some of the findings, Levene’s test for equality of variances was significant, indicating that the group variances were not equivalent. In these cases, the t value is reported without assuming equal variances, as indicated in the text. The internal consistency of the SCP for negative completions, negative health completions and negative pain completions was tested using the Kuder-Richardson coefficient (KR-20; [47]). Inter-rater agreement was assessed by inta-class correlations (ICC). ICC values were computed between raters’ totals for each valence and content judgements on the selected participants [48]. The content and construct validity of the SCP (e.g. relationships between SCP negatives, HADS depression scores and number of symptoms on the SCID) was investigated by Pearson correlation coefficients. Correlations were interpreted according to the definitions provided by Tabachnick and Fidell ([45]; ≤0.30=weak correlation; ≤0.60=moderate correlation; ≤0.80=strong correlation. A p-value of 0.05 was set as the critical level at or below which the results would be considered statistically significant.

In the absence of previous studies using a sentence completion task in a chronic pain population with two groups (depressed pain versus controls), estimates of effect size were used from studies comparing clinically depressed patients (without pain) with non-depressed controls. In both of the available studies [18], [21] the effect size for differences in the proportion of negative completions ranged from 2.6 for a 15 item scale to 2.8 for a 48 item scale i.e. very large effect sizes. A power calculation for the key one-way comparison between groups (Online calculator [http://psych.wisc.edu/henriques/power.html ] gives a power=0.99 (α=0.05) for a sample size of 20 per group. Similarly, Clark-Carter [49] suggests a minimum sample size of 20 participants per group to demonstrate a medium to large effect size in design with a statistical power of 80% and α of 0.05. We therefore opted for approximately 40 participants per group in order to maintain sufficient power, given that participants will be excluded a posterior if they did not fulfil inclusion criteria.

3.1 Subject characteristics

The average age of participants was 45.46 years (SD=12.92) and ranged between 20 and 74 years. The reported duration of pain in the depressed pain group was 80.9 months (SD=104.37), an equivalent of 6.75 years. The majority of depressed pain patients complained of back pain (91.5%), 6.4% of joint pain in arms, hands, legs and 2.1% of other musculoskeletal pain as primary complaint.

Table 2 reports the descriptive statistics for the depressed pain group and the control participants. The median for pain duration was 92.0 (minimum: 56.2 months and maximum: 98.5 months). No differences were found with regard to gender [Χ²(1)=5.07, n.s.] or age [t(87)=−1.92, n.s.], but there was a significant effect for education [Χ²(1)=50.86, p=0.002], showing that more control participants had obtained a degree compared to the participants in the depressed pain group. Table 2 also summarises the mean scores on pain severity, pain interference and pain disability as measured by the PDI for the depressed pain group. These measures have only been filled in by the depressed pain patients; comparisons between groups were therefore not possible.

3.2 Item analyses

The internal consistency of the SCP for negative completions, negative health completions and negative pain completions was tested using the KR-20. A coefficient of r₂₀=0.75 was observed for negative completions, r₂₀=0.68 for negative health, and r₂₀=0.70 for negative pain completions, indicating adequate internal consistency, as all coefficients were well within acceptable conventional limits. By convention, it has been suggested that a lenient cut-off of 0.60 is common in exploratory research and internal consistency should be at least 0.70 or higher for an adequate scale [50], [51], [52].

Table 3 reports the proportion of participants who made a negative completion on each item. These were computed separately for each group, and the difference in proportions between groups was used to estimate each item’s discriminative capacity. Maximum possible discrimination was +1.00, which would be observed for any item that elicited only negatives from the depressed pain group, and no negatives from the control group. The most discriminating items therefore had the largest positive differences, and negative differences were observed if the control group produced more negatives than the depressed pain group. Inspection of the discrimination measures, as shown in Table 3, revealed large item variability (range −0.02 to +0.55). The positive-verb and negative-verb items were generally less discriminating than those with neutral verbs or simple nouns. Overall, the positive-verb items (Item No. 5; 8; 19; 22; for instance “I care…”) and the future related items (Item No. 4; 13; 18; 27; for instance “In the future I…”) revealed a floor effect of negative thinking; even in the depressed pain group negative completions were relatively uncommon. Conversely, the negative-verb items (Item No. 7; 14; 17; 24; for instance “I worry…”) revealed a ceiling effect of negative thinking; even in the control group negative completions were common. This high level of item variability suggests that between-group differences are not simply attributable to a global negative response bias in the depressed pain group, meaning the tendency to produce negative completions across all items irrespective of content. On the other hand, the specific content of each item had considerable impact on its tendency to elicit negative thinking not only from the depressed pain patients.

3.3 Reliability of coding

Reliability of coding was assessed by inter-rater agreement. An independent rater coded 22.5% of the total sample (n=20), composed of 10 randomly selected data from depressed pain patients, and 10 randomly selected control participants (560 stem completions in total). The author and the independent rater were blind to group status and HADS scores while coding the completed sentences. Intra-class correlations (ICC) were computed between the raters’ totals for each valency and content judgements on the selected participants [48]. ICC values were computed before and after the establishment of the coding manual. When raters coded the responses without the use of explicit coding rules, established by the coding manual, ICC values were (the content was summed over all 28 items): pain (ICC=0.82); health (ICC=0.74); other (ICC=0.79). The ICC coefficient for negative completions for each stem type was: self-positive (ICC=0.53); self-negative (ICC=0.49); self-neutral (ICC=0.68); self-neutral negative (ICC=0.58), world (ICC=0.55); future (ICC=0.51); past (ICC=0.67).

After the use of the coding manual, the ICC values were considerably higher: pain (ICC=1.00), health (ICC=0.99), other (ICC=0.99). The ICC coefficient for negative completions for each stem type was: self-positive (ICC=1.00); self-negative (ICC=0.92); self-neutral (ICC=0.91); self-neutral negative (ICC=0.94), world (ICC=0.98); future (ICC=0.99); past (ICC=0.95).

3.4 Validity

Descriptive statistics for the main measures are shown in Table 4. As would be expected, the depressed pain group scored significantly higher on the HADS Depression Scale [t(87)=13.97, p<0.001], and the HADS Anxiety Scale [t(87)=8.17, p<0.001]. As predicted for content validity, the number of SCP negative statements was significantly greater in the depressed pain group [t(82)=7.02, p<0.001], and this group also produced significantly fewer positive statements [t(82)=−6.08, p<0.001]. There was no significant difference between the groups for neutral statements [t(82)=−2.00, p=0.049]. As expected, the number of negative pain completions [t(83)=4.48, p<0.001] and negative health completions [t(83)=3.76, p<0.001] was significantly greater in the depressed pain group. In the depressed pain group, the Pearson correlation between SCP negatives and the HADS Depression score was r=0.60 (p<0.001), indicating good construct validity. Also in the depressed pain group, the Pearson correlation between SCP negatives and the number of symptoms on the SCID was r=0.56 (p<0.001).

4.1 Limitations and recommendations for future research

There are several limitations for the current study. Firstly, participants were recruited through convenience sampling, and might have had an interest in pain, mood or cognition, which might represent a selection bias and would lead to the problem of generalizability to other populations. Second, the classification of healthy participants and a clinical group with chronic pain and possibly depression might be problematic, as it was not possible to build further groups in order to distinguish between patients with chronic pain and depression and patients with chronic pain but without depression. Future studies should employ this enhanced classification as the two conditions are likely to co-vary. Further, data for the present study were obtained by means of self-report and the disadvantages linked with self-report data are well documented and may include a social desirability bias. Although the wide variability demonstrated in the negative responses would argue against the presence of a social desirability bias.

It is believed that the specificity of the SCP is enhanced because there are no pre-defined negative statements in the SCP (only sentence stems) in contrast to questionnaire measures [21]. In the SCP, participants cannot endorse negatives they would not have produced otherwise and the present results support the assumption that negative sentence completions reflect actual thought-patterns, and are not an artefact of negative response bias. It is however possible that the respective context and response biases might influence the way some respondents make the completions to the SCP; similar to questionnaire measures there might be social circumstances that would enhance or reduce the generation of the negative stem completions, depending of the favourable outcome for the individual [55], [56]. However, it could be argued that the critical test of this would be the relevance of the responses to the process of case-formulation [57]. Although it is unlikely that the SCP will be able to uncover all relevant problems or beliefs in chronic pain patients, the SCP responses can be useful in generating hypotheses about some of them. Future studies should focus on what proportion of problems and beliefs will be identified by the SCP, how much this will vary across cases, and how accurately or reliably clinicians will agree about target problems and beliefs. Furthermore, future research should extend this measure in order to capture other domains and content categories such as anxious thoughts or self-denigration in order to test the hypothesis of an absence of self-denigration in depressed pain patients in contrast to clinically depressed patients without pain [12].

The SCP elicits negative thoughts that are known to accompany depressed mood in chronic pain, and it is plausible that their content is influenced partly by current events, memories, and ruminations (which will fluctuate), and partly by underlying assumptions, beliefs or schema (which will be more stable). There is no data for test-retest reliability available yet, thus it is not known at present how reliably the SCP is able to capture negative thoughts in the same individuals tested for a second time, for instance, a week after the initial testing.

A further caveat in the interpretation of the study is linked to the assessment of depression in addition to the HADS: the SCID interview was not audio-taped, due to restrictions in the ethical approval, thus it was not possible to double-rate all interviews with a second rater blind to the patient scores and the group a patient was assigned to. Future studies employing the SCP should test the inter-rater agreement of the SCID for co-morbid depressive disorders in chronic pain. Furthermore, it is important to explore the relationship between the level of anxiety and negative pain or health-related content, given that one might argue that negative health content is a function of concurrent anxiety rather than depression. Finally, the relationship between negative health-related responses and somatic representations of depression as assessed, for instance by the Beck Depression Inventory, should be the focus of further investigation. The following areas for future investigations are recommended: (a) With regard to cognitive specificity, future studies should investigate (similarly to the SCP in depression) whether SCP elicits negative thinking specific to depression in chronic pain, not normal negative thinking or negatives associated with emotional disorders (e.g. anxiety disorder) or chronic physical conditions (e.g. diabetes); (b) further studies should test the responsiveness, thus the ability to detect changes, of the SCP, before and after treatment; (c) with regard to sensitivity, future research is needed to develop thresholds for caseness by investigating the number of negative pain/negative health completions, which are needed to classify depressed from non-depressed cognition in chronic pain; and (d) the continuation of the process of validation in different settings and samples of chronic patients (e.g. rheumatoid arthritis, fibromyalgia) would be advisable, as the nature and content of depressive thinking might vary between different chronic pain conditions.

4.2 Theoretical and clinical implications

We propose that the SCP would be a useful measure of cognitive change and would expect that valence change would be closely related to mood as a function of treatment outcome in chronic pain, similarly to the responsiveness of the SCD, which has been demonstrated in the context of clinical depression [21]. Future research will have to test whether the SCP is able to detect cognitive change in chronic pain, for instance after a pain management programme.

While anxiety and depression in pain have a long research tradition, positive outlook represents a relatively new, but promising area, which focuses on emotional well-being and pleasure [58], [59]. At the moment there is only limited research on this aspect and more research is needed to understand the nature and characteristics of positive outlook in the context of chronic pain [60], [61]. In particular, it would be helpful to continue the process of validation by comparing the SCP responses with additional criterion variables, for instance with the Depression, Anxiety and Positive Outlook Scale (DAPOS; [60]) to be able to compare SCP responses with the sub-scale of positive outlook. Further research should also deploy other more generic measures of positive and negative affect, such as the Positive and Negative Affect Schedule (PANAS; [62]) as well as investigate the relationships between cognitive content and emotion regulation in mood disorders in the context of chronic pain [63], [64], [65].