Background

Low back pain (LBP) is a leading global cause of disability [1] and a large majority of patients have nonspecific back pain without a clear pathoanatomical diagnosis [2]. It has been proposed that patients with Modic Changes (MCs) may represent a subgroup of nonspecific chronic LBP that could benefit from a specific management [3,4,5,6]. MCs are magnetic resonance imaging (MRI) signal changes in the vertebral bone marrow extending from the vertebral endplate. An association between LBP and MCs has been supported in systematic reviews [7, 8], but the association is inconsistent [9].

MCs have been proposed to be associated with a specific clinical profile [4, 10, 11]. In a cross-sectional study of a population based sample of 40-years old Danes, the clinical profile differed between people with MCs and disc degeneration compared to those with disc degeneration alone [4]. Of the 23 variables tested, the groups differed in duration and severity of LBP, previous disc herniation, heavy physical workload, heavy smoking, reduced physical activity at work and in leisure time, sick leave, pain on movement, inability to activate lumbar multifidus muscles and lumbar pain tolerance. A study of Chinese individuals found a similar picture, but only for patients with MCs in the lower lumbar spine [12].

MCs are defined into three types on MRI [13] that are considered to represent different stages of the same histopathological process [14]. Type 1 (edema type) is hypo-intense on T1- and hyper-intense on T2-weighted MRI, type 2 (fatty type) is hyper-intense on T1- and iso- or hyper-intense on T2, and type 3 (sclerotic type) is hypo-intense on T1- and T2. These definitions apply to non-fat saturated T2 images only. Individual MCs can change type over time [15]. Histopathological, inflammatory, biochemical and genetic studies support the differentiation between MC types. There is inflammation of the bone marrow, possibly more in type 1 than type 2 MCs [16, 17]. Ohtori et al. found more TNF-immunoreactive cells in endplates adjacent to type 1 compared to type 2 MCs [18] and Rannou et al. found higher CRP values in type 1 compared to both type 2 or no MCs [11].

Clinical features of the different MC types are poorly studied. Subjects with type 1 MCs have reported more intense pain than patients with type 2 MCs, and conversion from type 1 to type 2 has been associated with decreased pain [6, 11, 19,20,21,22,23]. Knowledge about clinical features is important, forming the primary basis for further management. For instance, patients with type 1 MCs may respond differently to treatment compared to patients with type 2 MCs [24,25,26]. Knowledge about potential important clinical differences between MC types can potentially reduce unnecessary treatment and imaging. Differences in clinical profiles between type 1 and type 2 MCs would also suggest that these two imaging phenotypes deserve to be viewed as separate entities. Hence, it is important to look at clinical differences other than just pain intensity. Finally, elucidating a clinical profile for each MC type could point to its etiology.

Biopsies of MCs have indicated that inflammatory changes, which could affect clinical features, may be present in both type 1 and type 2 MCs [16]. Findings on fat suppression MRI series have indicated that edema may also be present in type 2 (fatty type) MCs [27]. Thus, it is relevant to assess clinical features in relation to inflammatory edema across both MC types. Short tau inversion recovery (STIR) series are sensitive to edema and can show signal increase that reflect symptomatic or asymptomatic edema-like bone marrow lesions.

The primary aim of the present study was to compare clinical characteristics between patients with type 1 versus type 2 MCs. The secondary aim was to explore associations between clinical characteristics and MC related STIR signals.

Methods

The present study is a cross-sectional study based on baseline data from a randomized controlled trial comparing amoxicillin to placebo in patients with chronic LBP and MCs (the Antibiotics In Modic changes (AIM)-study) [28]. The AIM-study was approved by the Regional Committees for Medical Research Ethics in south east Norway (REK Sør-Øst), was registered at ClinicalTrials.gov by December 2014 under the identifier: NCT02323412, and monitored by the Clinical Trial Unit, Oslo University Hospital. The trial was performed and reported in accordance with the Helsinki declaration. All patients gave written, informed consent to participate in the trial. Funding was granted by a governmental organisation (Helse Sør-Øst and Helse Vest), which did not have any role in planning, performing or reporting of the trial. A patient representative was a member of the Scientific Board of the study, which effected all the major decision from planning and design of the study, to the dissemination of the study results. The patient representative assessed the burden of the time and efforts required to participate in the trial.

Eligibility criteria and study population

Participants with chronic LBP from all health regions in Norway were recruited at six participating hospitals’ outpatient clinics between June 2015 and September 2017.

Adults 18 to 65 years of age who presented with LBP of more than 6 months duration and type 1 and/or type 2 MCs were eligible for participation in the trial. Patients had to have a pain intensity score of ≥5 on a 0–10 Numerical Rating Scale (NRS) (mean of three NRS scores; current LBP, the worst LBP within the last 2 weeks, and usual/mean LBP within the last 2 weeks). MCs had to be confirmed on a study-specific MRI, have height ≥ 10% of vertebral height and diameter > 5 mm, and be present at a level with previous lumbar disc herniation verified on MRI within the preceding 2 years.

Patients were excluded if they had any specific diagnosis that could explain the low back symptoms (e.g. tumor, fracture, spondyloarthritis, infection, spinal stenosis), former low back surgery (L1 – S1) other than for disc herniation (e.g. fusion, decompression, disc prosthesis), or former surgery for disc herniation < 12 months before inclusion. Further exclusion criteria were use of opioids except codeine/tramadol, and patient unlikely to complete the AIM-study. See the trial protocol for a complete list of eligibility criteria (available at ClinicalTrials.gov).

MRI

The study-specific MRI used in this investigation was performed at median 22 (interquartile range 15–29) days before the baseline measurements, and included standard T1- and T2 weighted fast spin echo images and sagittal STIR images. The same MRI protocol and 1.5 T scanner type (Magnetom Avanto with B19 software, Siemens) were used at all study sites. An integrated spine array coil was applied, but no surface coils. Imaging parameters for the sagittal images used in the present study were as follows: T1: repetition time 575 ms, echo time 11 ms, matrix 384 × 269, echo train length 5; T2: repetition time 3700 ms, echo time 87 ms, matrix 384 × 269, echo train length 17; STIR: repetition time 5530 ms, echo time 70 ms, inversion time 160 ms, matrix 320 × 224, echo train length 20. Field of view was 300 mm × 300 mm and slice thickness / spacing was 4.0 mm / 0.4 mm for all three sequences.

All sagittal slices were used to grade T1/T2- and STIR findings. MC types were defined by T1/T2 characteristics alone; type 1 as clearly hypo-intense on T1 and hyper-intense on T2, type 2 as hyper-intense on T1 and iso- or hyper-intense on T2. Patients were allocated to the type 1 MC group if their current study MRI showed MCs of primary (most extensive) or secondary type 1 at a level (superior or inferior endplate) with prior disc herniation verified on MRI within the last 2 years. Patients were allocated to the type 2 MC group if their study MRI showed MCs of primary or secondary type 2 – but not primary or secondary type 1 – at a level with MRI-verified disc herniation within the last 2 years. Patients with both type 1 MCs and type 2 MCs (at previously herniated disc levels) were hence allocated to the type 1 MC group. Thus, we were able to compare patients with type 1 MCs (and possibly type 2) versus type 2 MCs only (and not type 1).

As this was required for inclusion, all MCs used for MC group classification had height ≥ 10% of vertebral height and diameter > 5 mm on T1−/T2-weighted fast spin echo images. No criteria were predefined for which minimum size STIR signal changes should have to be reported.

We assessed MC related STIR signal increase (compared to normal vertebral body marrow) at 12 endplates (Th12-S1) using the following variables defined and selected prior to analysis:

  1. I.

    Volume of STIR signal increase

Each of the 12 endplates was given a STIR volume score based on the volume of STIR signal increase in percent out of the total vertebral body volume (0 = no STIR signal, 1 = < 10%, 2 = < 25%, 3 = 25–50%, 4= > 50%). The STIR signal volume was visually estimated by taking into account the affected area on all images. A total sum score (possible values 0–48) for the 12 endplates was then calculated by summing up the score values for each individual endplate.

  1. II.

    Maximum STIR signal intensity

The maximum intensity of MC related STIR signal increase at any endplate, recalculated and reported as a percentage on a STIR signal intensity scale ranging from normal vertebral body intensity (0%) to cerebrospinal fluid intensity (100%) (possible values 0–100). If maximum STIR signal intensity was reported for more than one endplate, the highest value was used.

  1. III.

    Number of endplates with STIR signal increase

The number of endplates with MC related STIR signal increase (possible values 0–12).

Two radiologists independently classified patients into the type 1 or the type 2 MC group (kappa = 0.65, good inter-observer agreement [29]), and solved all disagreements on MC type group by discussion. Both radiologists independently evaluated the presence of MC related STIR signal increase (kappa ≥0.83, very good agreement), its volume, and its intensity relative to normal bone marrow and cerebrospinal fluid. If they disagreed on presence or volume of STIR signal increase, a third radiologist evaluated the STIR images, and the majority rating was used. For intensity measurements, we used the mean of two radiologists’ values. All radiologists had more than 10 years’ experience in musculoskeletal MRI.

Clinical information and outcomes

All clinical information from history and examination was collected and reported by trial care givers (medical doctors or physiotherapists), who had available patients’ MC type group, but not their STIR findings. The patient-reported outcome measurements included the Roland and Morris Disability Questionnaire (RMDQ), LBP and leg pain intensity, Oswestry Disability Index (ODI) and the health-related quality of life (the EQ-5D). Background characteristics included age, gender, body mass index (BMI), ethnicity, marital status, educational level, work status, physical work load, leisure time activity, smoking habits, subjective health complaints (SHC) [30], emotional distress (Hopkins Symptom Checklist–25 HSCL-25) [31], fear-avoidance beliefs (FABQ work/physical activity) [32], LBP history/duration (including former treatment), comorbidities and pain medications. A detailed description of all outcomes collected in the trial is found in the trial protocol (available at ClinicalTrials.gov).

Table 1 describes the clinical characteristics of interest, pre-specified before data were available, with a rationale behind why they were chosen.

Table 1 Clinical characteristics of interest with rationale
Full size table

Statistical analysis

Differences in baseline characteristics between patients with type 1 versus type 2 MCs were analyzed by student’s t-test for normally distributed variables, by Mann-Whitney U test for non-normally distributed variables and by Chi-squared test for categorical data.

The diagnostic accuracy of each clinical variable to distinguish between type 1 and type 2 MCs was analyzed by calculating the area under the receiver operating characteristic curve (AUC) with its 95% CI (the term ‘diagnostic accuracy’ does not imply that MC type 1 s a diagnosis). The AUC represents the probability that a randomly chosen individual with type 1 MCs is (correctly) rated or ranked with greater suspicion than a randomly chosen individual with type 2 MCs, and can be interpreted as the clinical characteristic’s ability to discriminate between the two MC types [43]. Statistical analyses were performed with type 1 MCs defined as abnormal index test. An AUC > 0.5 was interpreted as an ability of the clinical characteristic to favor those with type 1 MCs, while AUC < 0.5 was interpreted as an ability of the clinical characteristic to favor those with type 2 MCs. For dichotomized variables we calculate sensitivity, specificity, positive likelihood ratio (PLR) and negative likelihood ratio (NLR). PLR and NLR were used instead of positive predictive value and negative predictive value, as the latter two require a valid estimate of prevalence of MC type 1 in relation to MC type 2 [44].

We analyzed 118 cases (type 1 MCs) and 62 controls (type 2 MCs), which meant we were able to detect (α = 0.05, β = 0.1) AUC larger than 0.628 (using http://www.biosoft.hacettepe.edu.tr/easyROC/). AUC values < 0.6 have been regarded as uninformative, and values 0.6 to 0.7 as indicating poor discrimination [34]. We thus regarded our sample size to be sufficiently large.

We used Pearson r, or Spearman rho, to analyze the associations between continuous clinical variables and STIR findings. We regarded correlation coefficient values < 0.10 as negligible correlation, values 0.10–0.39 as weak, values 0.40–0.69 as moderate, values 0.70–0.89 as strong and values > 0.90 as very strong correlation [45]. Categorical clinical variables were dichotomized by using Liu’s method for estimating the cutoff point on ROC curves for MC types. Dichotomous variables were analyzed by a t-test for associations with volume and intensity of STIR signal increase, and by Mann-Whitney U-test for associations with number of endplates with STIR signal increase.

We did not correct for multiple testing, as it was not the individual clinical characteristic’s association with the MC type that was of interest, but rather an exploration of various clinical characteristics together. Occasional false positive associations were therefore of less concern. Also, we did have an a priori justification for checking each clinical characteristic.

All analyses were performed using software package Stata version 15.

Results

We included 180 patients to the trial and in this study, 118 in the type 1 and 62 in the type 2 MC group. Reasons for exclusion into the trial are described in Fig. 1. Background characteristics are described in Table 2. Patients with type 1 MCs had somewhat lower BMI (mean difference: -1.3, 95% CI − 2.5 to 0.0) and were less likely to take opioids than patients with type 2 MCs (24% versus 44%, p = 0.006). There were no other differences in the background characteristics between patients with type 1 and type 2 MCs. No clinical characteristic of interest had more than three (1.7%) missing observations.

Fig. 1
figure 1

Flowchart. LBP Low back pain

Full size image
Table 2 Background characteristics
Full size table

Clinical characteristics and MC types

The distribution of each clinical characteristic by type 1 and type 2 MCs is summarized in Table S1 and Figure S1 in the Supplementary Appendix.

We found no statistically significant difference in any clinical characteristic between the two MC types (Table 3). The clinical characteristic variables showed poor ability to distinguish between type 1 and type 2 MCs with AUCs ranging from 0.42 to 0.55. Estimates of the diagnostic accuracy of all the clinical characteristics for MC types are summarized in Table 3.

Table 3 Diagnostic accuracy of various clinical characteristics to separate type 1 from type 2 Modic changes
Full size table

Clinical characteristics and STIR findings

There were only negligible or weak correlations between the clinical characteristics and the STIR variables (Tables S3, S4, S5 in the Supplementary Appendix). Total volume of MC related STIR signal increase was weakly correlated to age (r 0.18) and low BMI (r − 0.14) (Tables S3a) and was mean 1.3 points lower on the 0–48 point scale for patients with versus patients without a positive Springing test (Table S3b). Maximum STIR signal intensity was weakly correlated to the number of days with back pain last 4 weeks (r 0.19) and duration of back pain (r − 0.13) (Table S4a). The number of endplates with STIR signal increase was weakly correlated to age (rho 0.17) (Table S5a) and was smaller in patients who had versus patients who had not been disc operated (p 0.040, median 2 in both groups) (Table S5b).

Discussion

This cross-sectional study found no differences in predefined clinical characteristics between patients with type 1 and patients with type 2 MCs. All correlations between MC related STIR signals and clinical characteristics were weak or negligible. We did observe some statistically significant associations and small differences, but due to the large number of tests performed there were no more than what we could expect by chance. These results suggest that one cannot distinguish patients with type 1 MCs from patients with type 2 MCs only based on clinical symptoms and signs. Our findings are in accordance with a study on Dutch military personnel which reported no differences in various clinical tests between type 1 and 2 MCs [48] and a study which did not report any difference in duration of symptoms between type 1 and type 2 MCs in hospitalized patients [11].

The slightly higher BMI in the type 2 versus the type 1 MC group may be due to multiple testing and is unlikely to be clinically relevant. A previous study reported that type 2 MCs were associated with fat mass and suggested a metabolic mechanism behind the fatty marrow in type 2 MCs [49]. However, type 2 MCs were not related to weight in a study of LBP patients [50].

Our observation that volume of and number of endplates with MC related STIR signal tended to increase with higher age might suggest an underlying degenerative process. Indeed, MCs are linked to disc degeneration in both cross-sectional and longitudinal studies [4, 51, 52]. However, our observed correlations were weak and do not exclude other non-degenerative mechanisms behind MCs.

Our finding that patients with a positive Springing test had slightly less volume of STIR signal increase was opposite of what we expected and makes little biological sense [4]. It is difficult to explain and may be due to chance.

As we did not find sufficient candidates of clinical characteristics, and to avoid further problems with multiple testing, we did not perform a multivariate regression analysis.

To our knowledge, no previous study has investigated the association between clinical characteristics and MC related STIR signal increases in patients with nonspecific chronic LBP. It has been shown that vertebral bone marrow edema on gadolinium-enhanced MRI is related to LBP and lumbar tender points in elderly patients with degenerative scoliosis [53] and similarly, bone marrow edema on MRI related to pain in knees and ankles [54, 55].

The main limitation of our study is that it was based on a sample that was not representative of LBP patients in general, but included patients with particularly strong symptoms and already verified MCs at the level of a previous disc herniation. The fact that we were able to enroll many more patients with type 1 MCs than type 2 MCs supports the possibility of selection bias. In addition, restricting analysis to a subgroup with a limited range of pain intensity is likely to lower correlation coefficients for variables related to pain intensity [56]. Further, as we did not include patients without LBP, we cannot conclude about an association between presence of LBP and type of MCs or degree of MC related STIR signal increase.

Another limitation is that the type 1 MC group also included patients with both type 1 and type 2 MCs. This might have obscured any differences in clinical characteristics between a more pure type 1 MC group and a pure type 2 MC group. Further, we based the MC type group on MCs found at a level with a previous disc herniation, since such MCs were target for the treatment tested in our trial, and some patients had other types of MCs at other levels. However, as many patients have a mix of MC types, comparing clinical features between pure MC type groups would be less clinically relevant.

The assessors that performed the physical examinations had a heterogeneous clinical experience that could reduce the reliability of the examinations, and thus increase the risk of overlooking their true relations to other variables [57]. The reliability was acceptable for the Springing test [58], but poor or modest for many physical tests, in previous studies [59]. Accordingly, more clinical experience of the clinicians may not improve reliability [59].

Strengths of our study include strictly standardized MRI technique and MRI evaluation by multiple experienced radiologists, which can improve the reliability of MRI reports [60]. Further, data collection was systematic and prospective, and the data were almost complete. Despite the study had limitations, we would expect it to have revealed at least some relevant associations if strong true associations actually existed in the chronic LBP population.

Conclusion

Our study of selected patients with chronic LBP showed no differences in clinical features between patients with type 1 MCs and patients with type 2 MCs. There were only weak associations between clinical patient characteristics and MC related STIR signal increase.