Table 2.

Analyses—pre-planned and post hoc

	Analysed population
Pre-planned analyses
ANCOVAa—dependent variable: RMDQ score at 1 year; independent variables: baseline RMDQ score, the potential effect modifier, treatment group, their interaction term (effect modifiera treatment), age, former disc herniation surgery	PP and ITT
ANCOVAa—as above with ODI score replacing RMDQ score	PP and ITT
ANCOVAa—as above with LBP intensity score replacing RMDQ score	PP and ITT
ANCOVAa—to assess the independency of any effect modification for STIR composite and/or MC composite; dependent variable: RMDQ score at 1 year; independent variables: baseline RMDQ score, STIR composite, MC composite, age, treatment, STIR composite*treatment, MC composite*treatment, age*treatment, former disc herniation surgery	PP
Post hoc analyses—to evaluate STIR3 results
Comparison of treatment groups—baseline factors, concomitant treatment; no statistical testing	STIR3 group
Responder analyses—number needed to treat to achieve > 30%, > 50%, and > 75% reduced RMDQ score from baseline to 1 year (for those with complete data)	STIR3 group with data
Distribution of responders—number and proportions of patients with > 30%, > 50%, and > 75% reduced RMDQ score at 1 year (those with complete data) by treatment group and STIR group	All with data
Linear mixed-effects models—to assess treatment effect over time; using Akaike’s information criterion to decide which covariance matrix to apply; dependent variable: RMDQ (5 time points), ODI (3 time points), or LBP intensity (17 time points); independent variables: time, treatment, time*treatment, age, prior disc herniation surgery	PP STIR3 group
Scatterplot of change in RMDQ score—to visualise change in RMDQ score from baseline to 1 year in each treatment group for STIR1, STIR2, and STIR3 patients (those with complete data)	PP with data
Bangs blinding index—for each treatment group based on their response at 1 year to ‘Which study medicine do you think you received?’ (antibiotics/placebo/unsure); range: − 1 (all report incorrect treatment) to 1 (all report the correct treatment); 0 = random reporting of treatment	STIR3 group

PP, per protocol; ITT, intention to treat; RMDQ, Roland-Morris Disability Questionnaire; ODI, Oswestry Disability Index; LBP, low back pain; STIR, short tau inversion recovery; MC, Modic change

^aMissing values of RMDQ, ODI, or LBP intensity were substituted with imputed values from the multiple imputations performed in the trial. This multiple imputation model used 50 imputations, predictive mean matching, and the following predictors: age, leg pain, comorbidity, fear avoidance, emotional distress, physical workload, former surgery for disc herniation, study centre, MC type group, and treatment group