Table 1 Predefined potential effect modifiers of interest

  We expected a larger treatment effect in type I compared to type II.

Type I Modic changes are more strongly associated with low back pain [11]. In animal models, injecting C.acne into intervertebral discs induced type I Modic changes [12, 13].

  We expected a larger treatment effect in patients with previous disc surgery compared to those without.

Low-grade discitis might be a complication of disc surgery due to introduction of bacteria into the disc during the surgical procedure. The randomised trial from 2013 with a high number of patients with previous surgery found effect of antibiotic treatment [4], while a case series with few patients with previous surgery was negative [14].

  We expected a larger treatment effect in patients with a positive Springing test (patient reported pain with pressure applied to lumbar transverse processes) compared to those with a negative Springing test.

Spinal tenderness may indicate regular spondylodiscitis [15]. Springing test is found to be borderline significant discriminator between patients with and without Modic changes [16].

  We expected a larger treatment effect in those with higher CRP. The predefined cut-off values for CRP were changed from 3 mg/L and 10 mg/L to 5 mg/L due to too few patients in the predefined categories.

CRP in serum is associated with bacterial infection and inflammation.

  We expected a larger treatment effect in those with disturbed sleep than in those without. Disturbed sleep was defined as a ≥ 2 score on the Oswestry Disability Index item 7 (sleep scale), i.e. less than 6 h sleep to no sleep because of pain.

Night-time pain may indicate infectious spondylodiscitis [15].

  We expected a larger treatment effect in those patients with constant low back pain compared to those with fluctuating low back pain.

Constant pain may indicate regular spondylodiscitis [15].

  We expected a larger treatment effect in those with short duration of symptoms compared to those with longer duration of symptoms. The predefined categorization (< 1 year, 1–2 years and ≥ 2 years) was dichotomized into < 2 years and ≥ 2 years due to too few patients with symptoms < 1 year.

Recent disc herniation could have increased perfusion in the disc as part of disc repair, thereby increasing absorption of amoxicillin.

  We expected a larger treatment effect in patients < 40 years of age compared to those ≥40 years of age.

C.acne could be more prevalent in discs of young patients [17].

  We expected a larger treatment effect in men compared to women.

C.acne could be more prevalent in discs in men than in women [17].