Changes in pain sensitivity and spinal stiffness in relation to responder status following spinal manipulative therapy in chronic low Back pain: a secondary explorative analysis of a randomized trial

Clinical guidelines recommend spinal manipulative therapy (SMT) for chronic low back pain (LBP) [1], given its effect on pain and disability is comparable to that of other recommended conservative therapies [2]. Toward optimizing SMT for chronic LBP, we recently published a randomized clinical trial (RCT) [3]. In which, we experimentally categorized each segment as having either high stiffness or high mechanical pain sensitivity. In this trial, the participants received four SMT sessions directed at either the segment with the highest stiffness or highest pain sensitivity in a randomized manner. While significant within-group changes emerged in patient-reported LBP, the randomization did not yield any between-group differences. We also re-measured the experimental outcomes (stiffness and mechanical pain sensitivity) at each time point in the course of treatment. While we did not find differences in lumbar stiffness, mechanical pain sensitivity decreased significantly in the group that received SMT at segments characterized by high mechanical pain sensitivity (pain segment group), compared to the group that received treatment at segments characterized by high stiffness (stiff segment group). Interestingly, this increase did not imply clinical pain reduction. On this basis, we hypothesized that a segment-specific neurophysiological reflex most likely mediated the observed effect on mechanical pain sensitivity as opposed to a curative effect on a mechanical dysfunction [3].

Our prior paper’s analysis did not investigate the potential impact of being a clinical responder or non-responder within the allocated subgroups. For instance, preliminary research suggests decreases in lumbar stiffness following SMT, but only when this corresponded with improvements in disability [4, 5]. A similar pattern may emerge in this cohort when applying a responder threshold.

Notably, the finding that a segmental effect on mechanical pain sensitivity is not a proxy for clinical improvement appears inconsistent with previous studies that have found a hypoalgesic effect after overall clinical improvement [6, 7]. However, interpreting our novel finding of a modulating segmental impact of SMT requires a thorough responder analysis. Possible within-group differences in mechanical pain sensitivity could have affected our prior results. Besides, our previous study only included mechanical deep pressure pain sensitivity. Including other types of stimuli associated with chronic LBP [8], like peripheral heat pain sensitivity, may be helpful in an explorative and secondary analysis, as would be the impact of segmental versus regional measurements.

In this explorative analysis, we used patient-reported disability and low back pain intensity to determine responder status and examine its relationship to various experimental outcome measures (changes in lumbar stiffness, mechanical pain sensitivity, and heat pain sensitivity).

A pre-planned secondary explorative analysis of data used for a primary analysis in an RCT (registered at Clinical.​Trial.​gov identifier: NCT04086667) [3]. One-hundred-and-thirty-two participants enrolled in the study. All participants received lumbar spinal manipulative therapy directed randomly at either the segment of highest stiffness or the segment of highest mechanical pain sensitivity. Seven participants did not complete the study. Two others were lost to follow-up, leaving 123 participants with complete data sets.

We included participants from a regional Spine Center in Denmark. The inclusion criteria were 18 to 60-year-old patients with LBP of benign or degenerative origin for more than three months. No prior spinal surgery and no indication for current spinal surgery. No history of SMT in the preceding four weeks. Body mass index had to be below 35. We limited the daily opioid intake to 40 mg of morphine at the time of inclusion.

Exclusion criteria were: failure to complete 75% of the allocated SMT interventions, receiving manual therapy of the lower back in another, non-project setting, or change pain medication during the study period. We excluded no patients after initiating the project based on these criteria.

All participants gave oral and written informed consent for the study approved by the regional research ethics board (S-20160201).

The primary study protocol is described in full detail in the prior study [3]. A short overview is provided here.

We extracted demographic data from the SpineData questionnaire, a clinical registry in use at The Spine Center of Southern Denmark [9].

The baseline lab session consisted of i) completing patient-reported clinical outcomes, ii) identification and marking of each spinous process for each lumbar segment using ultrasonography (Sonosite Titan Linear, L38 probe) [10], iii) measures of lumbar stiffness and pain sensitivity (mechanical and heat), iv) segment randomization (high stiffness or high mechanical pain sensitivity) and v) initial SMT application.

Three additional SMT applications were provided over the following 14 days.

A final follow-up lab session took place approximately 14 days after the post-SMT session. Again, we repeated items i - iii. This concluded the study.

We conducted the responder analyses with two variables using three different responder thresholds. Disability, assessed by the Oswestry Disability Index (ODI) (version 2.1), was used to dichotomize responder status using three different responder thresholds: i) more than or equal to a 50% improvement to better be able to compare with similar literature [4, 5, 12], ii) more than or equal to a 30% improvement as consensus recommended [13], and iii) more than or equal to a 0% improvement indicating the absolute dichotomization between worsening and improving. The ODI has been translated to Danish and validated as a reliable instrument to measure LBP changes [14].

Similarly, we used patient-reported low back pain to set responder status at the same three different responder thresholds: i) a 50% improvement, ii) a 30% improvement, and iii) 0% improvement. The pain intensity score used a mean score (0–10) from the self-reported LBP numerical rating scale (NRS) [15], which consists of three scales: Current LBP, Worst LBP in the previous 14 days, and Average LBP in the last 14 days.

We only performed each dichotomization at the final follow-up time point.

The randomization process is described in full detail in the primary study [3]. In short, segmental stiffness was determined using the raw force-displacement data (mm) from the VT’s heaviest individual load (typically 6 kg), and segmental mechanical pain sensitivity was determined as the mean value of the three PPT measures. To resolve situations where the segments with the highest pain sensitivity and stiffness were identical or adjacent, a ratio ranging between − 1 and + 1 was calculated for each segment based on these two variables in combination. A ratio approximating − 1 would indicate high stiffness and low pain sensitivity, and correspondingly a ratio approaching + 1 would indicate high pain sensitivity and low stiffness. SMT was directed at the segment with the highest (or lowest) ratio index.

Frequencies (count) are presented for responders/non-responders as defined by each threshold, and the cumulative proportion of responders is presented graphically for both ODI and NRS.

We performed a three-way mixed model analysis with subject as a random intercept using an unstructured variance-covariance structure. The interacting fixed effects were segment-allocation (target site), time, and responder status to determine within-group changes and between-group differences. For a concise description of the mixed model approach, see Bates et al. [22]. Responder status consisted of six different predictor variables (minimum or equal to 50, 30, and 0% change in both ODI and NRS). The mixed model assumptions were upheld for the models. They evaluated: i) normal distribution of the residuals error using Q-Q plots, and ii) the homogeneity of variance by visually inspecting the residuals versus the predicted values.

We had to omit approximately 11% of the data points due to inaccuracies when identifying the segments at different time points. See the primary study for further detail [3].

The mixed models are presented as mean changes within-group from baseline to post-SMT and follow-up along with 95% confidence intervals. Where significant within-group changes are present in any outcomes, a table is present to further describe between-group mean differences (responder versus non-responder).

Table 1 present the responder/non-responder distribution of the 123 participants who completed the intervention.

Figure 1 illustrates the cumulative proportion of responders for ODI and NRS at follow-up. The vertical dashed lines represent thresholds for responder/non-responder at 50% (blue), 30% (green), and 0% (red). The line extends negatively beyond 0, indicating that 28 (23%) of the cohort experienced worsening of ODI at follow-up, and 29 (24%) demonstrated worsening of NRS. It also shows that the two groups respond at an equal rate. Furthermore, while an equal number of participants has increased (worsening) ODI and NRS, the variance is much greater in NRS changes, approximating 150% versus 50% for ODI.

We present the results of the responder analysis in Tables 2 and 3. Table 2 lists changes in objective outcome measures (GS, PPT, and HPT) immediately post-SMT (4th session) and at follow-up for the two randomization groups (SMT at the stiffest segment and most pain-sensitive segment) and subgrouped by ODI responder status defined as 0, 30, and 50% improvement thresholds. Table 3 presents similar data for NRS.

When categorizing segments into i) target, ii) adjacent, and iii) other, no significant within-group differences were observed, and no discernible patterns emerged (see Figure f.seg). The same pattern is present for all outcomes (GS, PPT, HPT) for each segment and each responder threshold in ODI and NRS. For simplicity, we illustrate 30% ODI improvement in Fig. 3. The remaining figures are similar and can be found in Additional file 2.

The underlying causes of low back pain are often obscure, and the connection to spinal stiffness is unclear. In degenerative joint and disc disease, spinal stiffness is likely affected, albeit pain is not always present. In other words, segmental spinal stiffness may be relevant to disability but potentially irrelevant to pain. Conversely, pain will tend to affect disability, irrespective of spinal stiffness.

Therefore, SMT may change pain perception through neurophysiological reflex mechanisms, thus changing both clinical pain and overall disability without actually improving segmental spinal stiffness. However, the difference in deep pressure pain sensitivity, which accompanies clinical improvement, may be unrelated to the SMT. Furthermore, SMT’s reflex-mediated segmental effect on deep pressure pain sensitivity, which does not translate into a clinical improvement, is probably of limited relevance.

We observed a difference in the number of clinical responders when using disability compared to pain intensity. We consider the most likely reason for this difference to be that a reduction in pain intensity and improvement in disability does not necessarily correlate entirely [40]. Furthermore, the responsiveness of the two scales varies. The numerical rating scale is restricted to one domain with scores between 0 and 10, whereas ODI spans a range of 0 to 50 and records ten different domains. The smaller resolution of the 11-item NRS could limit the responsiveness, thereby translating into a smaller number of responders. However, this is speculative, and the authors are not aware of any data to support this.

Participants in the present study were LBP patients referred for assessment in a hospital setting. One of the formal criteria for such referral in Denmark is the insufficient effect of conservative management in the primary care setting. In other words, we must assume that LBP patients who respond favorably to SMT are underrepresented in this cohort compared to LBP patients in general [2]. The fact that around 23% of the participants experienced worsened disability after the intervention supports this assumption. This may further negatively impact the likelihood of positive clinical outcomes with SMT and affect the relationships between disability, pain, and experimental outcomes. This is speculative, however.

Another potential limitation is that we choose to dichotomize responder status at the final follow-up time point. We may miss some rapid responders directly following the fourth treatment to provide greater detail of the changes. However, the mean improvement at post-SMT and follow-up appears to be of equal size [3].

As stated in our previous paper [3], this was not a placebo-controlled study, and thus, the clinical improvement observed could be due to something other than the SMT. Therefore, the present findings do not speak to SMT’s clinical efficacy but rather to the underlying mechanism of any such effect. Furthermore, these results are limited to the cohort in question, as different possible outcomes could be observed in a primary care setting. Finally, the pressure algometry was applied by a double-headed probe. It is unknown whether this affects how the PPT scores change following SMT. Hence, caution should be taken when comparing our data to similar literature.

There are also strengths to consider; this was a relatively large cohort of chronic LBP patients seen in the secondary care sector. We measured the experimental outcomes both immediately following SMT and at 14-days follow-up and were able to correlate this with clinical improvements.