In a secondary care setting, differences between neck pain subgroups classified using the Quebec task force classification system were typically small – a longitudinal study

This study was a longitudinal observational study of baseline, 3, 6 and 12-month measurements extracted from systematically collected, routine clinical data.

These data were collected at the Medical Department of the Spine Centre of Southern Denmark, which is an outpatient secondary care facility that performs multidisciplinary assessments of patients with back pain referred from either general practitioners or chiropractors in primary care. The main function of the Department is to provide comprehensive patient evaluations via patient history, clinical examination, MRI and other special tests, together with management plans for use by their referring primary care clinician. Short courses of treatment are sometimes offered to clarify treatment response. Data were collected from the Spine Centre’s electronic registry, SpineData. Each patient completed a comprehensive self-reported baseline questionnaire on a touch screen prior to his/her first consultation. The clinician received a summary of the patient’s self-reported data before the clinical examination. Following the clinical examination, the clinician entered the results of a core set of examination procedures into the registry. Patients were also invited to complete two internet-based or postal follow-up questionnaires. Before 1 January 2012, the first follow-up questionnaire occurred 3 months after the date of the initial consultation. After that date, the first follow-up was collected at 6 months, as this was believed to be a more suitable outcome time point in the clinical course of people with chronic pain. The second follow-up questionnaire has always occurred at 12 months.

All patients aged 18 years or more referred with neck pain as their main complaint and seen at the Spine Centre between 1 January 2011 and 31 December 2012 were selected for the analysis. This main complaint was self-selected by the patient via a body chart on a touch screen. Patients were excluded if data on pain intensity, or pain location (based on a pain drawing completed as part of the electronic questionnaire), or from the neurological clinical examination were missing.

Data collection in the SpineData database and its analysis in scientific projects has been approved by the Scientific Ethics Committee of the Region of Southern Denmark (Project ID S-200112000-29). Under Danish law, the secondary analysis of such de-identified data does not require separate ethics approval (The Act on Processing of Personal Data, December 2012, Section 5.2; Act on Research Ethics Review of Health Research Projects, October 2013, Section 14.2).

Patients were classified into four subgroups adapted from the Quebec Task Force Classification System [ 15].

Baseline characteristics were presented as proportions with 95 % confidence intervals (CI), means with standard deviations (SD) or medians with interquartile ranges (IQR). Subgroup differences and outcome responder/non-responder differences were tested using chi-squared, ANOVA or Kruskal-Wallis’ test depending on the type and distribution of the variable.

The predictive value of the Quebec Task Force subgroups was then compared to the prediction of a more detailed model that consisted of the separate factors of the Quebec Task Force classification: arm pain intensity, dominating arm pain, arm pain location and neurological signs. This comparison was based on linear regression with 12 months of pain and activity limitation outcomes as the dependent variables. To simplify this comparison, it used a single-time outcome rather than a longitudinal model. Variables with associations displaying p-values <0.2 were retained in the model. Predictive capacity was quantified in terms of adjusted R-squared and root mean squared errors (RMSE).

No data imputations were made since only patients with complete baseline variables required for classification into the Quebec Task Force subgroups were included and longitudinal linear mixed models cope with missing outcome values. All analyses were performed using STATA/IC 13.1.

A total of 2,446 patients with neck pain, with or without arm pain, and above 18 years of age were registered within the inclusion period of the study. Of those, 2,330 fulfilled the inclusion criteria, 478 patients were excluded due to missing or conflicting data that were required for the classification process, and therefore 1,852 could be classified into the pre-defined subgroups (64 % females, mean age 49 years) (Fig.  1). As the time period for the initial follow-up changed from 3 to 6 months during the inclusion period, there were only 930 potential responders to the 3-month follow-up and 1,400 to the 6-month follow-up.

Follow ups after 3, 6 and 12 months were completed by 45 %, 32 % and 40 % of those invited to participate at each time point. The proportion of non-responders did not differ across subgroups. There were some statistically significant differences between responders and non-responders. Non-responders were younger (by 3 years at the first follow-up and 5 years at the second), had slightly more activity limitation at baseline (1 and 2 points higher Neck Disability Index scores in non-responders at the first and second follow up respectively), and had a larger proportion of females (62 %) than males (56 %) who did not respond at the 12-month follow-up. Baseline neck pain intensity and arm pain intensity did not differ between responders and non-responders.

Patient self-reported baseline characteristics are summarised in Table  1 and statistically significant differences across subgroups were observed for many factors. Generally, those with Local NP were the least severely affected and those with NP + NRI had the most severe profile. These subgroups did not differ significantly on workload factors or sick leave. The differences observed in duration indicated that patients with NP + NRI were referred to the outpatient hospital department earlier than other patients, but nonetheless, a large proportion of all subgroups had long-lasting pain, as less than 25 % of patients in all subgroups had an episode duration of 3 months or less and most had a duration of 6 months or more.

A small improvement in pain was observed over time in all subgroups (Fig.  2). There was an overall statistically significant interaction between subgroups and time, but effect sizes were small (Table  2) and below reported thresholds for clinical importance [ 25]. The local NP subgroup improved slightly less after 3 months as compared with all other groups, but remained with the lowest level of pain. After 6 and 12 months, those with NP + pain above the elbow had improved the least (0.6 points improvement after 12 months) and patients with NP + NRI had experienced the largest improvement in pain intensity (1.3 points after 12 months) (Table  2).

Similar results were obtained for activity limitation (Fig.  3). Improvements were small, and although the overall interaction between subgroups and time was statistically significant, the differences between the trajectories of the subgroups were minor (Table  2) and below reported thresholds for clinical importance [ 25]. The local NP group had the least activity limitation at all the time points, whereas the other groups did not differ significantly from each other at any follow up (Fig.  3).

The proportion of the total cohort reporting improvement (dichotomised global perceived effect) after 3, 6 and 12 months, were 24 % (19 to 30 %), 33 % (28 to 38 %) and 35 % (31 to 39 %) respectively. There was no statistically significant association between subgroups and improvement (Fig.  4.). The estimated odds ratio, as compared with local NP, was 1.1 (95%CI 0.6 – 2.1) for NP + pain above the elbow; 1.2 (0.7 – 2.2), for NP + pain below the elbow and 1.7 (1.0 – 2.8) for NP+ NRI. As there was no significant interaction between time and subgroups, these odds ratios apply to all three time points.

The model including only the predefined Quebec Task Force Classification System subgroups as the predictor variable explained 1.0 % of the variance in the 12-month pain intensity outcome (RMSE 2.41) and 1.4 % in activity limitation (RMSE 10.15). In the more detailed predictive model, arm pain intensity, arm pain location, dominating arm pain and having more than one neurological sign were significantly associated with 12-month pain and explained 13 % of the variance in the pain intensity outcome (RMSE 2.26). Arm pain intensity and dominating arm pain were the only significant factors in the more detailed predictive model with activity limitation as the outcome, and this model explained 11 % of the variance (RMSE 9.66).

The strengths of this study are that the large sample size increased the precision of the estimates, and as these data were collected as part of routine clinical procedures, the results may have generalizability to similar secondary care settings. However, that the spectrum of neck pain patients within sample was predominantly those with more severe and chronic neck pain, might weaken the generalizability of the results. The routine data collection may also weaken the internal validity of the study, as these data may have been of varying quality and more subject to bias than those collected in more rigorous research situations. An additional potential weakness of the study is some uncertainty in the definition of the nerve root involvement, as the reliability and validity of the clinicians’ neurological examinations are unknown and may have resulted in some misclassification of patients. In addition, the Upper Limb Tension Test, recommended to rule out cervical radiculopathy by Rubinstein et al. [ 9], was not part of the standard neurological examination used in the collection of these data. Also, clinicians were not blinded to their patient’s questionnaire when they completed the clinician questionnaire, although this does pragmatically mirror usual practice. We also did not record or control for the type or amount of treatment received over the follow-up period and this could have affected patient outcomes.

It should be recognised that some of the aims of clinical subgrouping are to better target treatment and to stratify care pathways, but information on diagnosis or treatment could be taken into account within the subgrouping or outcome measurement in this study. Hypothetically, if it were possible to systematically and effectively target treatment to the Quebec Task Force classifications, the predictive ability of the subgroups in this study could have been different due to effective treatment differentially improving patient outcomes.

At baseline, those with Local NP were generally the least severely affected and those with NP + NRI the most affected. This reflects the findings of another cross-sectional study that found combined neck and arm pain was associated with a lower overall health status compared with either local neck pain or radicular symptoms alone [ 33]. It is also similar to the findings of a low back pain study that described cross-sectional differences in patients subgrouped using the Quebec Task Force Classification System [ 34].

On the outcomes of pain and activity limitation, very little improvement was observed over time in all subgroups with small and mostly non-significant differences between groups, although the Local NP subgroup had slightly less severe pain at all time points. Similarly, for the outcome of global perceived effect, there were no significant differences between the subgroups at any time point, although the NP + NRI group did have slightly higher odds for improvement that was borderline significant. The above findings are in contrast to a recent study of low back pain in the same setting, where the presence of neurological signs and leg pain had prognostic implications for pain and activity limitation, but not for global perceived effect [ 18] . It could be that the presence of neurological signs actually is less important for prognosis in patients with neck pain, or an alternative explanation might be that in this particular sample of neck pain patients, the neurological examination or recording of its results were less precise than in the low back pain study.

The results in this study indicate that simple subgrouping based on the Quebec Task Force Classification System explained very little about the prognosis of individual patients in this chronic neck pain cohort. This suggests that if prognostic models for this setting are to be created that have precision at an individual patient level, they are likely to require more and different factors. It also may be that precision in prognostic estimates at an individual patient level requires predictor variables to remain in their original metrics, as ‘dumbing them down’ into a simple four-category classification loses too much information, no matter how useful the classification might be for other purposes.

Hayden et al. [ 35] suggested that prognostic research should occur within a stepwise framework where identifying and testing prognostic factors is an important initial step towards subsequently understanding prognostic pathways. So, an additional contribution from our results is the indication that individual variables such as pain location, arm pain intensity and number of neurological signs should be considered as candidate variables for prognostic models of neck pain. Although the individual variables that make up the Quebec Task Force Classification System had greater prognostic strength than the Quebec Task Force subgroups, it still did not have sufficient predictive accuracy for us to consider it clinically useful in this setting. However, the Quebec Task Force subgrouping may still usefully reinforce diagnostic categories. Again, a caveat is that this Quebec Task Force Classification may have more prognostic value in less severe and less persistent neck pain populations, especially where treatment is subgroup-targeted. A recent study found that baseline characteristics in patients with nerve-related neck and arm pain predict the likely response to neural tissue management [ 36], and identifying and matching prognostic factors with relevant treatment pathways has been found to be effective in a large trial of patients with low back pain [ 37].

The identification of clinically important and validated subgroups of neck pain patients remains a challenge. There is a lack of consistent evidence for prognostic factors, other than baseline pain intensity and baseline disability [ 38]. To our best knowledge the only externally validated rule for prediction of neck pain outcomes is a rule suggested by Shellingerhout and colleagues which performed well in an external validation sample of chronic neck pain patients [ 39]. While that model in that rule did include radiating arm pain, it also included other factors that are not contained in the Quebec Task Force Classification. Our finding that the simple Quebec Task Force Classification was not useful for prediction of outcome in a very severely affected group of neck pain patients, suggests a need to also test this more comprehensive Shellingerhout model in populations similar to ours.