This protocol follows the Preferred Reporting Items for Systematic Reviews and Meta-Analysis Protocols (PRISMA-P) guidelines [
40,
41], provided as Additional file
1. The review was registered with the International Prospective Register of Systematic Reviews on 31 October 2019 (PROSPERO ID: CRD42019145257).
Search strategy and study selection
We will search the following electronic databases from inception to current:
-
MEDLINE (Ovid) (1946 to current)
-
EMBASE (Ovid) (1980 to current)
-
CINAHL (EBSCO) (1982 to current)
-
Cochrane Central Register of Controlled Trials (CENTRAL) in the Cochrane Library, current issue
-
-
EU Clinical Trials Register (eudract.ema.europa.eu)
-
WHO International Clinical Trial Registry Platform (apps.who.int/trialsearch/Default.aspx)
Our search strategies incorporate the recommended strategies from the Cochrane Back and Neck Group (4) to identify randomised trials of low back pain and terms for the interventions of interest [
36]. The search strategy for MEDLINE is listed in Additional file
3. We will also search previous systematic reviews and the reference lists of included studies to identify any additional trials. Records identified through all searches will be downloaded and managed in a custom relational database.
We will conduct record screening in Covidence systematic review software [
57]. We will conduct two stages of screening: (i) title and abstract and (ii) full text. Two reviewers will independently screen studies for eligibility at each stage. Disagreements will be resolved through discussion, with arbitration from a third author (JHM) if required. We will contact a study’s corresponding author up to three times to obtain additional information to determine eligibility, and if no reply is received, we will exclude the study from this iteration of the review. Studies in languages other than English will be translated. We will summarise the literature search using an adapted PRISMA flow diagram [
58].
Record management
We will manage the included records in the relational database. We will conduct record linkage to establish unique studies for data extraction, which may consist of multiple records. We will search for the protocols and trial registrations of all included trials. We will use the following hierarchy to prioritise records for data extraction: (i) primary report (typically the journal article reporting the results of the primary analysis of the trial), (ii) secondary report (secondary analysis of the trial), (iii) conference abstract (a report of a secondary analysis), (iv) trial registration, (v) other secondary records, and (vi) other conference abstracts.
Two reviewers will independently extract and enter data from included trials into standardised spreadsheets. Review authors will not extract data from any trial in which they have had any involvement. Data will be taken from previous reviews conducted by the authors when possible. Disagreements between reviewers will be resolved through discussion, with arbitration from a third author (JHM) if required. We will not extract data from interventions that do not meet the eligibility criteria for this review.
We will extract data on:
Trial characteristics: country, setting, and number of trial sites; sample size; and study duration.
Participants: diagnosis, duration of LBP, age, male/female ratio, arm-level pain intensity at baseline (as mean (standard deviation [SD])), experience or naivety with the trial intervention, and co-morbidities, including alternate sites of pain.
Interventions: medicine(s) tested, control; duration of intervention; dosage regimen; routes of administration; and usage of rescue medication.
Outcomes: type and dimensions of the scale/measure used to assess pain or function and the time from randomisation at which the end of treatment data were obtained in individual trials. We will extract the definition of ‘adverse effect’ and ‘serious adverse effect’ used in each study. We will extract data on study results including participant allocation to each intervention group; compliance to the intervention (including the definition of compliance); the number of participants who discontinued due to an adverse event; the event rate and descriptions of all reported adverse effects; and pain intensity and function at the completion of treatment.
If studies report more than one measure for pain, we will prioritise extraction in the following order: 100 mm VAS, 10 cm VAS, 11-point NRS, rating scale for pain intensity from a composite measure of pain (e.g. McGill Pain Questionnaire), and ordinal scale. We will preferentially extract the outcome score and measure of variance at the end of treatment (or closest time point) for each group, followed by the change from baseline and measure of variance. If data are not available for each trial arm, we will extract the between-group statistics at the end of treatment.
If studies report more than one measure for function, we will prioritise extraction in the following order: ODI, RMDQ, rating scale for functional ability from a composite measure, and ordinal scale. We will preferentially extract the outcome score and measure of variance at the end of treatment (or closest time point) for each group, followed by the change from baseline and measure of variance. If data are not available for each trial arm, we will extract the between-group statistics at the end of treatment.
Missing data
We will contact a trial’s corresponding author up to three times via email to request missing data, which will be considered unobtainable if no reply is received within 6 weeks. If data for outcomes of pain and function are not presented in an appropriate form for meta-analysis (such as median and range instead of SDs, standard errors,
t-statistics, or
p values), we will attempt to impute these using established methods [
54,
59]. We will conduct sensitivity analyses for pain at the end of treatment and safety if we impute missing data for either of these outcomes.
Risk of bias
We will appraise each trial’s risk of bias using the Cochrane ‘Risk of bias’ tool, version 5.1 [
54] and recommendations by Furlan et al. [
4]. Two reviewers will independently appraise trial-level risk of bias for 13 items across the domains of selection, performance, attrition, detection, reporting, and other sources of bias. If an item is typically rated at outcome level, which may differ between our two primary outcomes (pain intensity and safety), we will use the more conservative rating (e.g. using high risk over unclear risk). Review authors will not appraise risk of bias for any trial in which they have had any involvement (e.g. trial investigator). Risk of bias assessments will be taken from previous reviews of analgesic medicines conducted by our author team, where the same approach was used.
We will determine an overall risk of bias for each trial by adapting the process from Furukawa et al. [
47]: low overall risk is determined when three or fewer items are rated ‘unclear’ risk and no domains are rated ‘high’; moderate overall risk is determined if a single item is rated as ‘high’ risk of bias, or no item is rated as ‘high’ risk but four or more are rated as ‘unclear’; and high overall risk otherwise.