This study is a secondary analysis using real patient data (baseline and immediate post-treatment) from an ongoing multi-center, randomized controlled trial (expected finalization in June 2022) approved by the local ethics committees (University Hospital Ghent and University Hospital Brussels – ref no. B.U.N. 670201835625). The primary purpose of this ongoing trial is to examine the added value of cognitive-behavioral therapy for insomnia (CBT-I) to the current best physical therapy treatment for nCSP with comorbid insomnia. All participants gave informed consent. The full study protocol of the ongoing trial is registered at Clinicaltrials.gov (NCT03482856) and is published elsewhere.25

Adults with nCSP and comorbid insomnia were recruited via flyers in the participating universities, university hospitals (Ghent and Brussels), workplaces and public places, through advertisements, social media, primary care practices, and occupational health services. People had to send an email to an institutional email address specifically created for the original trial to show their interest. The inbox was only accessible by 3 researchers who worked on the original trial. Thereafter, the potential participants received a booklet with the study details and were requested to sign the informed consent and to fill out an online questionnaire which was used to perform a screening based on inclusion and exclusion criteria (Table 1). Potential participants were called by telephone for a verbal screening and to schedule the home-based polysomnography (Alice PDX system, Philips Respironics IncTM) assessment to screen for underlying sleep pathologies.26,27

Because the recruitment of participants for the main trial was still ongoing at the conception of this analysis, a sample size calculation was conducted specifically for estimating the required sample size for answering the research question of this analysis. Sample size calculations were performed with G*Power 3 (Düsseldorf, Germany). The required number of participants was calculated for a linear multiple regression analysis based on a medium effect size (Cohen's f²) of 0.20 and 4 predictors in the final model.28 A total of 65 participants was required allowing for a type I error of .05 and aiming for 80% power.

The experimental and control interventions had a similar structure, the same number of sessions and the same session duration, only the content was different. All participants received 18 sessions of therapy during 14 weeks. The control and experimental groups received pain neuroscience education (PNE) and cognition targeted exercise therapy (CTET) while the experimental group also received additional CBT-I. The first three session for both groups focused on PNE.29-34 The first session was a 1-hour group session addressing the general aspects of PNE with the possibility for patients to ask questions. After this session, all participants received an information leaflet to inform their significant other and as refresher for themselves. Patients were also asked to complete a form to indicate feared activities. The second session was a home-based online module which consists of videos alternating with online questioning. The last PNE session was a 30-minute individual session, addressing the patient's specific questions, translating the content to the patient's daily life, and questioning and discussing the patient's perceptions and goals.

All other sessions were individual, real life, 30-minutes face-to-face sessions. The purpose of the CTET was to change inappropriate beliefs and perceptions in combination with PNE by gradually confronting the patient with movements and activities that were feared and/or avoided using a time-contingent approach. The exercises gradually progressed towards more complex and physically, cognitively, and psychosocially demanding situations. The same exercises were also implemented in an individual home exercise program. Across all sessions, patient's cognitions and perceptions about their problem and about exercises were addressed. In addition to CTET the experimental group also received individual CBT-I (including general sleep education, sleep restriction therapy, stimulus control, sleep hygiene instructions, and cognitive therapy) within the same sessions. More details of both intervention groups can be found elsewhere.25 Both interventions were delivered by trained physical therapist with a master's degree and took place at the University Hospitals of Ghent and Brussels. Physical therapists were trained by experts in the field of chronic pain rehabilitation (JN, AM). Only the experimental therapists received CBT-I-training by a psychologist/behavioral somnologist with expertise on this matter (OM).

Participants were randomized by an independent researcher (AM) uninvolved in the treatment or assessment. Randomization lists were made available separately for both treatment centers and stratified for sex and dominant pain problems.

The participants, assessor, and statistician were blinded to the maximal extent for the study hypothesis and randomization. Therapists were not blinded (i.e., the experimental therapists knew they had to include CBT-I). However, to avoid therapist bias and contamination of the therapy arms, therapists involved in the experimental intervention were not involved in the control intervention, and vice versa.

Online questionnaires (in Dutch) were used to assess secondary self-reported outcomes, socio-demographic, and medical data. All outcome measures were assessed at baseline and immediately after treatment. Accelerometry was used to measure physical activity levels.

Statistical analyses were performed with SPSS 26.0 (IBM, Armonk, NY, USA). Descriptive statistics were retrieved for all demographic characteristics and variables of interest at baseline. The normality assumptions were checked using histograms, Q-Q plots, and Kolmogorov-Smirnov tests. Correlations among all independent variables and variance inflation factors (VIF) for all variables included in the regression analysis were calculated to check for multicollinearity indicated by an absolute correlation coefficient ≥0.8 or a VIF ≥5.57 Outliers were determined as a difference of 2.2 times the interquartile range.58 Because previous studies in the general populations show there are differences in physical activity based on sex, age, and educational level, these variables were considered as control variables.59-63

The multivariate regression model explained 33.3% of the total variance in MVPA at baseline (Table 3). Physical and mental fatigue were both negatively associated with MVPA (b = -0.9; 95%CI: -1.59, -0.15, and b = -1.0; 95%CI: -1.67, -0.43). Body mass index (BMI) was also negatively related to the amount of MVPA (b = -0.5; 95%CI: -0.93, -0.16). Lower subjective sleep quality was related to greater MVPA (b = 0.7; 95%CI: 0.22, 1.17). Fewer limitations in functioning as a result of emotional problems was associated with greater MVPA (b = 0.1; 95%CI: 0.03, 0.10).

The final model included the variable “pain duration” and the interaction “education × time” (Table 4). A main effect for sex was found (estimate [SE]: -4.5 [2.0], p=.03). No other significant main and/or interaction effects improved the prediction model. Three post-treatment data points and two pre-post differences in MVPA were identified as outliers. No collection errors from the actigraph in the participants with data outliers were identified. Consequently, all outliers were considered a realistic part of the data set and maintained in all analyses.

To our knowledge, this is the first study to investigate predictors of (baseline) MVPA and determinants for a change in MVPA after a treatment in people with nCSP and comorbid insomnia. The data were derived from a randomized trial using valid and reliable research questionnaires. However, this study has some limitations. First, only short-term changes after the treatment were investigated. Second, the findings might not be generalizable to other chronic pain populations. However, nCSP is one of the most common chronic pain condition and similarities were found in other populations. Third, the included variables are limited to the measured outcomes in the original trial. Possibly several influencing factors were not included (e.g., pain catastrophizing, social factors, and environmental factors). Fourth, there might be a potential lack of power given that the sample size calculation for the linear regression model was based on four predictors with a medium effect size. Possibly some weaker predictors could not be identified. Nevertheless, we were able to identify five predictors. No determinants for a change in MVPA after treatment could be identified. No a priori power analysis was conducted for the corresponding linear mixed model analysis because of the explorative nature of the analysis (i.e., included variables and interactions in the final model where unknown at study initiation). Fifth, there is lack of information about the validity of wrist-worn accelerometers to assess physical activity (in free-living conditions) in people with nCSP with comorbid insomnia. One study in older adults investigated the validity of wrist-worn actigraphy and found that different physical activity intensity levels can be identified accurately.38 However, they propose to use different cut-off points for identifying intensity.38 Similarly, validation studies identifying optimal cut-off values in specific populations with chronic conditions are warranted. Last, while the English equivalents of the questionnaires used in this study are validated, the psychometric properties of some of the Dutch versions are not yet investigated. Still, the psychometric properties of the Dutch version of the SF-36 and CSI are well-established.46,92 As most questionnaires are developed for and validated in English-speaking populations, the questionnaires should be cross-culturally adapted and validated before being used in clinical studies.

Nevertheless, this study can serve as a basis for future studies investigating whether targeting one of the associated variables with MVPA as part of a treatment has an impact on the physical behavior and short- and long-term disability. Furthermore, current findings can be useful to identify people early who are more likely to be inactive and who might experience or develop negative consequences related to inactivity or overactivity. Early identification makes it possible to address undesirable physical behavior early.