Central Sensitisation and functioning in patients with chronic low back pain: protocol for a cross-sectional and cohort study

Introduction A relevant subsample of patients with chronic low back pain (CLBP) have manifested augmented central pain processing, central sensitisation (CS). Patients with CLBP have limited functioning and participation. Theoretically, physical functioning in patients with CLBP can plausibly be linked to CS; however, evidence to explain such association is scarce. Moreover, there is no gold standard for CS diagnosis. The objectives of the study are: (1) to analyse the association between instruments assessing reference symptoms and signs attributed to CS; (2) to analyse whether reference symptoms and signs attributed to CS are associated with functioning measurement outcomes; and (3) to analyse whether changes (between baseline and discharge) in reference symptoms and signs attributed to CS are related to changes in each of the functioning measurement outcomes. Methods and analysis A cross-sectional and longitudinal observational study is performed with measurements taken at baseline and discharge of an interdisciplinary rehabilitation programme. A sample size of 110 adult patients with CLBP has been calculated for the study. CS measurements are: Central Sensitisation Inventory, quantitative sensory testing and heart rate variability. Functioning measurements are: lifting capacity, maximal aerobic capacity, accelerometry and reported functioning. Statistical analyses to be performed are: (1) correlation between CS measurements, (2) multiple regression between functioning (dependent variable) and CS measurements (independent variable), and (3) multiple regression between changes in scores of functioning (dependent variable) and CS measurements (independent variable), and corrected for sex and age. Ethics and dissemination The study obtained the clearance to its implementation from the Medical Research Ethics Committee of the University Medical Center Groningen in July 2017. The results will be disseminated through scientific publications in peer-reviewed journals, presentations at relevant conferences, and reports to stakeholders. Trial registration number NTR7167/NL6980.

osteoporosis, and spinal fractures. Do the authors wish to focus this study on all types of low back pain or rather non-specific low back pain with the exclusion of pathological low back pain? It is important to consider this question as low back pain is a heterogeneous condition and it is well accepted that low back pain due to pathological causes s very different to non-specific low back pain. Moreover, what about patients that have had surgery or plan to have low back pain surgery? The inclusion and exclusion criteria need to be reconsidered. There is a significant amount of information provided on each of the instruments to be used, including their specificity and sensitivity. However, limited information is presented that justifies the choice of the instruments relative to other instruments that are available. I would suggest further information is provided to justify the choice of instruments. It is planned that some of the data will be collected 6 weeks prior to baseline and other data will be collected at baseline. Together, these data will be considered "baseline" data. However, there are issues associated with this as participants' low back pain may significantly change during the 6 weeks period prior to baseline. Why not collect all the baseline data at one time period? Statistical analyses: There are a few aspects of the statistical analyses that require attention. Firstly, it is proposed that multiple correlations are to be performed. However, there is no mention of the possible issues with this and how they might be overcome. Secondly, given there is limited studies in this field, the authors propose that their sample size is based on the number of variables to be included in the analyses. However, this method has several limitations and in addition, a search identifies available studies. Strength/limitations section (Pg 3). It is unclear what the following means: "Statistical analyses are performed with proxies of CS and functioning measurements based on their relevance and contribution; the decision may not be beyond debate. Introduction 2. On page 5 the authors provide two possible reasons that an association between CS and physical function may exist. However, it is unclear what the authors are trying to say, in part because this is wordy and the grammar is poor. This section is overall quite cumbersome to decipher. Response: We agree with the reviewer in the use of pairwise deletion in the correlation analyses it is more appropriate. We have adapted it in the manuscript accordingly.
Changes to the manuscript: page 16 line 16.

What will you do if the data is no MCAR?
Response: Please see response to Item 13.
12. How will you handle non-normal data?
Response: Please see response to Item 13.
13. Overall, your decisions here should be backed by references from statistics literature.

Response: The decision of how should be handled the MAR and NMAR values and non-normal
data will be decided depending on the specific characteristics of the database. We aim to keep any adaptation to a minimum while maintaining the statistical power according to the specific characteristics of the data available and to fully report the approach applied.

MAR and NMAR values are likely to undergo imputation[van Ginkel et al. 2018]; and non-normally distributed data will be analyzed with a non-parametric method when available, such as Spearman's rank correlation for the multiple correlations. If no non-parametric method meeting the study's objectives is available, may follow transformation by categorization and/or logtransformation.
We have added the description of non-normally distributed data (median and interquartile range) to the presentation of the characteristics of the sample, we also have made a suggestion on how to handle no MCAR missing data, and added the type of correlation coefficients that could be expected from the analyses. 14. How will you decide what to keep in the models? How will you account for type I error given the multitude of analyses? How will you address any interactions?

Reference
Response: We agree with the reviewer that there are a multitude of correlation analyses to be done. The idea of performing Bonferroni adjustments has been considered for the multiple correlation analyses, but:

1) Due to the large amount of correlations in the multiple correlation analyses, a Bonferroni adjustment would lead to an increase in Type II error when trying to avoid Type I error. 2) Bonferroni would only adjust for the significance. Establishing results and conclusions only on
significance can be flawed [Ziliak and McCloskey 2008]; moreover, we are interested in the size of effect of the associations. All things considered we have decided to have a stricter significance level for multiple correlation analyses, i.e. 0.01; and to provide 95% Confidence Intervals of the correlation coefficients in the results as they can be more insightful of the values of the population. We believe that with these measures we will be able to extract meaningful information.
For the interactions, the variables that have shown to be confounders in the multiple regression models will be explored for possible interaction/moderation effect, provided the variables are theoretically and statistically sound.
Aims 16. The protocol presents 3 aims. However, it is not clear how Aim 2 and 3 are different. Both are looking at the association between central sensitisation and physical functioning. The first is examining whether there is an association and the second is looking at the direction of the associationwhich will be answered by meeting the first aim. I would suggest combining these aims in to one.  1-2, 7-9 and 28-29. 17. The protocol uses the word "trial" in the manuscript. I think this is confusing as this study is not a clinical trial rather an observational study. I would suggest using the word "study" instead. 19. There is a significant amount of information provided on each of the instruments to be used, including their specificity and sensitivity. However, limited information is presented that justifies the choice of the instruments relative to other instruments that are available. I would suggest further information is provided to justify the choice of instruments.

Changes to the manuscript: page 4 lines 22-25 and page 5 lines 1-3.
20. It is planned that some of the data will be collected 6 weeks prior to baseline and other data will be collected at baseline. Together, these data will be considered "baseline" data. However, there are issues associated with this as participants' low back pain may significantly change during the 6 weeks period prior to baseline. Why not collect all the baseline data at one time period?

GENERAL COMMENTS
Thanks to the authors for considering the feedback and make changes to manuscript based on the suggestions provided. While these have improved the quality and content of the manuscript, there are issues that need further comment.
I would suggest that the authors include their changes to the manuscript within the response document in future to assist the review process.
1. Background. The authors state that they performed the following: We have emphasized the scarcity in CS and functioning literature in the Introduction section, and supported the symptoms and signs of CS in the Introduction section with the paper of Smart et al. 2012. Changes to the manuscript: page 5 lines 9-10 and lines 22-25.
The authors have essentially stated that the literature is scarce. However, both reviewers have referred to information in the literature on this topic and suggested that a summary of the current literature be included. I would suggest that this still needs to be addressed. Moreover, while changes were made in paragraph 4 of the introduction, there are no references provided to support the new information added.
2. Justification of choice of instruments. It is still not clear why these particular instruments have been chosen compared to other options. There needs to be a clear discussion about what other instruments are available and why instrument A was chosen over instrument B.
3. Inclusion and exclusion criteria: This is no mention of whether participants that have had or are planning surgery are included.
4. Statistical issues: The statistical issues raised in the manuscript need to be also mentioned in the limitations section of the discussion.

VERSION 2 -AUTHOR RESPONSE
Author's response to Reviewer #1, dr. Samantha Meints, comments: I commend the authors in their addressing reviewer concerns. However, consistent with reviewer #2, the rule of thumb is not sufficient for the sample size estimate. As noted by reviewer #2 as well, you should identify similar studies to determine estimated effect sizes and use this to complete your sample size calculation. As you noted, there are differences between what has already been published in this area and what you are proposing. However, these differences can be noted in the manuscript to suggest that the sample sizes used are just a guide. Response: Thank you for the proposal. We have looked at previous literature published in this topic with patients with CLBP and with a similar design. Although, there are no studies performing a multiple regression model to base on to calculate samples sizes for Objectives 2 and 3. We have based on the publication of Huysmans et al.2018 to define medium effects for the multiple regression analyses and adapted the methods and discussion accordingly. Author's response to Reviewer #2, Assoc. Prof. Donna Urquhart, comments: Thanks to the authors for considering the feedback and make changes to manuscript based on the suggestions provided. While these have improved the quality and content of the manuscript, there are issues that need further comment. I would suggest that the authors include their changes to the manuscript within the response document in future to assist the review process.

GENERAL COMMENTS
Per reviewer #2's comment, the authors still did not summarize the past literature regarding the relationship between CS and disability. Instead, they added a couple references to show there is a scarcity of work. This does not sufficiently address the reviewer concerns.
Otherwise, the authors did a nice job addressing reviewer concerns.

VERSION 3 -AUTHOR RESPONSE
Reviewer #1, dr. Samantha Meints, comments: Per reviewer #2's comment, the authors still did not summarize the past literature regarding the relationship between CS and disability. Instead, they added a couple references to show there is a scarcity of work. This does not sufficiently address the reviewer concerns.
Otherwise, the authors did a nice job addressing reviewer concerns.  , Huysmans et al. 2018, Tanaka et al. 2019, Cohen et al. 2000, Gockel et al. 2008, Hübscher et al. 2013, Georgopoulos et al. 2019  • Citations needed for expected medium correlations between CS measurements. Also, because there are multiple correlation analyses, there is a higher risk for Type I error. How will this be addressed?

VERSION 4 -AUTHOR RESPONSE
Reviewer #1, dr. Samantha Meints, comments: 1. Augmented pain processing does not always mean central sensitization. Indeed, it could be peripheral sensitization. This should be addressed Response: Thank you for the remark, we have now specified that it is augmented central pain processing: "A relevant subsample of patients with Chronic Low Back Pain (CLBP) have manifested augmented central pain processing, Central Sensitisation (CS)." Changes to the manuscript: page 2 line 3.
In this situation and all things considered to have a stricter significance level for multiple correlation analyses, i.e. 0.01; and to provide 95% Confidence Intervals of the correlation coefficients in the results because they can be more insightful of the values of the population. We believe that with these measures we will be able to extract meaningful information. Reference: Ziliak S, McCloskey DN. The cult of statistical significance: How the standard error costs us jobs, justice, and lives. University of Michigan Press; 2008." Additional changes to the protocol: 1. During our revision in the original DFNS protocol and reference values by Rolke et al.(2006), we have come to realize that it would not be possible to compare our measurements to theirs ̶ DFNS measurements include the face, hands and feet, whereas ours include the upper back, low back, and upper legs. As a result z-scores for our study based on their reference values will not be possible. We have decided to present an overall mean for each QST test. We believe that with this strategy we'll be able to provide informative results about the general somatosensory profile of patients with CLBP. 2. The QST assesses the somatosensory system with a battery of different sensory tests. It might be possible that the different QST tests relate differently to the functioning domains. The pre-selection of a single test representative of the whole battery of QST tests that could relate to all functioning domains, might be too limited and much information may be missed. To overcome this, the selection of the QST test to enter the multiple regression models for RQ2 and RQ3, will be based on the highest correlation coefficient per functioning dependent variable. Changes to the manuscript: Page 17 line 2 and page 21 line 11.