In this longitudinal cohort study, the 3-year incidence of new MRI findings was quite low (range 2-8%), and consequently the majority of participants who developed incident symptoms did not concurrently develop new MRI findings. When examining associations between new MRI findings and specific spine-related outcomes, only three large magnitude effects were found for our primary MRI predictors of interest which corresponded with our expectations: the association of incident annular fissures with chronic LBP, and the association of disc extrusions and nerve root impingement with radicular symptoms. Annular fissures were not specific for incident chronic LBP, in that 3 subjects (50.0%) with new annular fissures did not develop chronic LBP. In contrast, all participants with disc extrusions and nerve root impingement developed symptoms, and these findings are therefore quite specific for the production of radicular symptoms. No other incident MRI findings showed large effects on symptoms.
There are scarce prior reports examining longitudinal associations between incident lumbar MRI findings and symptoms, and direct comparisons with our findings are complicated by the different symptom outcomes used in earlier work. Borenstein et al. reported on 31 individuals who received longitudinal lumbar MRI assessments separated by 7 years, where incident symptoms were assessed using a single questionnaire recalling symptoms over the 7 years prior . Having any incident MRI finding conferred a relative risk of 3.5 for incident LBP, but no individual MRI findings were strongly linked with incident symptoms. Elfering et al. reported on 41 individuals who received longitudinal lumbar MRI assessments separated by 5 years . Individuals with incident disc degeneration (disc dessication and/or height narrowing), and individuals with incident vertebral endplate changes showed absent to weak associations with LBP by various definitions, and no associations were seen between annular fissures and LBP. Jarvik et al. reported previously on the same LAIDBACK participants described in the current article, who were followed for 3 years and assessed for the association between incident MRI findings and incident symptoms, using a composite outcome of any LBP and/or radicular symptoms. They did not detect statistically significant associations (p < 0.05), but noted that all subjects with new extrusions, nerve root impingement, and central canal stenosis developed new symptoms. The current work differs from the original LAIDBACK analysis in that we applied more specific symptom definitions that separated LBP from radicular symptoms, corresponding to the distinctions made in actual clinical practice. The importance of distinguishing LBP, radicular symptoms, and composite outcomes combining these two symptoms is more than just semantic: a recent systematic review of associations between lumbar MRI findings and chronic LBP excluded the LAIDBACK cohort from consideration due to the use of this composite symptom outcome . Our findings, and that of prior longitudinal studies, demonstrate that the annual incidence of new MRI findings is extremely low, rendering all published studies severely underpowered to detect statistically significant associations with incident symptom outcomes. Our study showed no large magnitude longitudinal associations between disc degeneration (height narrowing, dessication, bulging) or endplate changes and chronic LBP, which is generally consistent with prior longitudinal studies. We had not expected disc degeneration to be associated with LBP due to the modest effects seen in prior cross-sectional studies . However, we had expected to see a large magnitude association between endplate changes and LBP, based on cross-sectional studies reporting positive associations with effects ranging from OR 2.0 to 6.1 [18, 29]; our 95% confidence intervals do not rule out associations of this magnitude. Our finding of a large magnitude longitudinal association between annular fissures and LBP is consistent with some cross-sectional studies showing positive associations with LBP (OR 2.5-4.6 [5, 18, 30]), but inconsistent with one cross-sectional study suggesting a protective effect on LBP , and the longitudinal study by Elfering, which detected no association .
Earlier studies by Borenstein et. al and Jarvik et. al showed conflicting results for the association between incident disc extrusions and incident symptoms; our secondary analysis of the same cohort reported on by Jarvik et. al suggests that disc extrusions are associated with radicular symptoms, but not with LBP. Similarly, we found nerve root impingement to have a large magnitude association with radicular symptoms, but not with LBP. These findings are consistent with the common clinical reasoning that extrusions and nerve impingement primarily produce radicular symptoms due to nerve involvement, and in this setting LBP may or may not be present. These results suggest that when a new MRI demonstrates a disk extrusion or nerve root impingement that was not previously seen, in a patient with new radicular symptoms, the new MRI finding is a probable cause of radicular symptoms. However, the overwhelming majority of individuals with new radicular symptoms will not have relevant new MRI findings.
A strength of our study is the clear distinction made between the symptom outcomes of LBP and radicular symptoms. We conducted separate analyses for specific MRI findings with specific symptom outcomes, based on prespecified relationships that were thought to have greater conceptual importance. Our analyses examining how MRI finding effects varied with the symptom outcome confirmed several instances where the expected association with one symptom outcome was not seen with the other symptom outcome (i.e. the association of annular fissures with LBP but not radicular symptoms, and the association of disc extrusions/nerve impingement with radicular symptoms but not LBP). This highlights the importance of the case definitions used in back pain research. Variation in case definition, and the tendency to combine symptoms into composite outcomes in epidemiologic imaging studies, may obscure real and important biological relationships. Although we made efforts to optimize case definition, even these definitions had important limitations. For instance, our definition of radicular symptoms resulted in a 3-year prevalence of 57% for incident pain, weakness, numbness, or tingling, which is substantially higher than would be expected for sciatica. Our definition of radicular symptoms likely included not only true cases of sciatica, but other apparent ‘cases’ where limb symptoms were explained by factors unrelated to a lumbosacral radicular syndrome (i.e. limb pain due to hip/knee osteoarthritis, or occasional limb paresthesias due to transient nerve impingement during leg crossing). An interview and examination by a trained clinician during the episode of radicular symptoms may have improved case definition accuracy. Another potential source of misclassification in longitudinal studies of imaging associations with symptoms is recall bias. The potential for misclassification due to inaccurate recall is high for conditions like acute LBP and sciatica, where the natural history may involve transient episodes of severe pain that resolve quickly, but may also recur or otherwise vary over time. Our study, in contrast to earlier studies [27, 28], sampled participants at regular intervals every 4 months for 3 years, and was likely to have minimized such recall inaccuracy. Nevertheless, sampling at 4-month intervals has shortcomings as well: a period of LBP lasting only 1 month could have bridged two 4-month sampling periods and met criteria for LBP using our case definition, while not exceeding the usual 3-month or 6-month threshold used for defining persistent LBP.
Certain other limitations of our study warrant further mention. First, the low 3-year incidence of MRI findings limited the statistical stability of our point estimates, and future longitudinal MRI studies will require a longer duration of follow-up. The low incidence of new MRI findings also limited our ability to simultaneously adjust for multiple incident MRI findings in the same multivariable models; for a multifactorial condition such as LBP, this might obscure important relationships. Second, our analysis made multiple statistical comparisons, raising the concern that identified associations may represent type I error and not reflect a true biologic effect. This is of particular concern for the association of annular fissures with LBP, where prior association studies have had mixed results [5, 18, 30, 31]. Although we attempted to limit multiple comparisons by prioritizing seven ‘primary predictors’ where associations were expected, and relegating other factors to secondary status, longitudinal studies are needed to replicate our findings. Third, our study ascertained MRI findings at only two timepoints. Since specific MRI findings are known to improve over time, particularly with disc extrusions (which often resorb [32, 33]), the lack of an interim imaging timepoint may have created misclassification by failing to detect some MRI findings which developed at the precise time of symptom onset, but had improved by the 3-year follow-up MRI. Alternatively, new MRI findings on the 3-year follow-up MRI may have developed after new symptoms had occurred. An ideal design for future longitudinal MRI studies would include pre- and post- MRI assessments separated by ≥ 6 years, with the addition of an interim MRI assessment at the time of incident symptoms, and/or at an interim control period. Fourth, although interobserver variability between the two neuroradiologist readers were calculated using the baseline MRIs, interobserver variability for changes in MRI findings (i.e. incident findings) was not assessed. Therefore the reliability estimates for MRI interpretation that we estimated may not apply to scans that demonstrated incident findings. Fifth, the use of dichotomies to define the abnormality itself could introduce inaccuracy, since there is a dearth of prior studies examining what thresholds for each of these MRI findings best correlate with symptoms, and what level of reliability pertains to each threshold. Lastly, the current study includes the same participants that were described in an earlier report . This analysis, however, was distinct from that performed previously in that different (and more specific) symptom outcomes were used, with a focus entirely on associations between longitudinal changes in MRI findings and incident symptoms. As described above, the symptom outcome definitions for chronic LBP and radicular symptoms were specified prior to conducting the analyses described here.
Although our findings did not detect multiple large magnitude and statistically significant associations between incident MRI findings and incident LBP or sciatica, they should not be taken to suggest that research studies of imaging for LBP are not a vital area for future research. Imaging research into the structural correlates of LBP is an essential step towards delineating that component of the LBP experience that is explained by structural or anatomic factors, understanding that these biological correlates of pain are only one part of a complex process most accurately viewed within the context of a biopsychosocial framework . Currently there is no evidence that imaging improves LBP outcomes in standard clinical practice; however, without ongoing back pain imaging research, there will be no opportunity for imaging methods and approaches to improve to the point where they can refine clinical decision-making. Instead, we emphasize the need for future back pain imaging research, while taking care to distinguish between what is of value from a research perspective, and what is ready for translation into actual clinical practice. Given the low specificity of most MRI findings in both prevalence and incidence studies, we suspect that the time of translation into clinical practice is still far off in the future.