Responsiveness and minimal important change for the ProFitMap-neck questionnaire and the Neck Disability Index in women with neck–shoulder pain

Neck pain is highly prevalent with a reported 1-year prevalence estimated to be 30 to 50 % in the general population [1]. Neck pain also contributes to activity limitations in 11 to 14 % of workers [2]. In the largest group of neck pain patients, the underlying cause of the pain is uncertain [3, 4]; hence, the designation is non-specific neck pain. The alleviation of symptoms and restoration of functional limitations are particularly important for neck pain sufferers without a clear pathophysiology. To evaluate and establish effective treatment and rehabilitation strategies, access to reliable and valid patient-reported outcome measures, i.e., standardized questionnaires measuring specific constructs of interest, is a necessity. There are a number of questionnaires available to measure pain and disability in people with neck pain. However, weaknesses in measurement properties of several questionnaires were recently recognised, and important methodological aspects to improve were, for example, content validity regarding the relevance and comprehensiveness of items and the use of better statistical methods in responsiveness studies [5, 6]. Also, Wiitavaara and co-workers [7] found a low correspondence between neck–shoulder pain questionnaires and the symptoms experienced by the sufferers, implying a questionable content validity of the questionnaires. One potential explanation for this may be that the neck pain sufferers’ experiences are seldom taken into account in the developmental process of the neck–shoulder pain questionnaires [7], even though it is recommended in the literature [6, 8‐11].

The Profile Fitness Mapping neck questionnaire (ProFitMap-neck) is a questionnaire developed in collaboration with neck pain patients, designed to assess symptoms and functional limitations in people with neck pain [12]. It consists of a functional limitation scale and a symptom scale of which the latter is subdivided in separate indices for the intensity and frequency of symptoms. The two scales can also be combined in a compound total score. The content of ProFitMap-neck symptom scale had the best correspondence with experienced symptoms among subjects with chronic neck pain, compared with 9 other neck-specific questionnaires [7]. The function scale of ProFitMap-neck has not been compared in the same way, but items of this scale have shown associations with sensorimotor function tests in different groups of people with neck pain [13‐16]. The overall validity and reliability of the questionnaire has been tested on patients with chronic whiplash-associated disorders, as well as chronic non-traumatic non-specific neck pain [12]. However, the validation study of Björklund and co-workers [12] had a cross-sectional design that assessed validity of single scores. To evaluate the ability of an instrument to detect change over time in the construct to be measured, a measurement property referred to as responsiveness [17], longitudinal study designs are necessary.

An issue related to responsiveness concerns the interpretation of a change score, i.e., the change of a score from baseline to a follow-up. It is important to know if a change score of an instrument reflects a change in the patient’s status that he/she would consider important. The cut-off score with the best discriminative ability between patients that have improved and not improved is often referred to as the minimal important change (MIC) of the instrument, defined as the smallest measured change score that patients perceive to be important [17, 18]. The knowledge of a questionnaire’s responsiveness and MIC is crucial for its use in the evaluation of treatment and rehabilitation. In clinical practice, it can be used to judge whether a patient has reached a change of importance, and in research, the measurement properties are useful for the analysis and interpretation of study results. The primary aim of the present study was to determine the responsiveness and MIC of the ProFitMap-neck and the Neck Disability Index (NDI) [19] in women with chronic non-specific neck–shoulder pain. A secondary aim was to compare the responsiveness between ProFitMap-neck and NDI. We chose to compare with NDI since it is the most frequently used and evaluated neck-specific questionnaire [5, 20, 21].

Data for the current study were derived from two randomized controlled trials (ISRCTN trial registration numbers ISRCTN92199001 [22]) and ISRCTN49348025 [23]. Both trials had an observer-blinded three-arm parallel group design with baseline measures and follow-ups 1 week, 6 months and 12 months after an 11-week intervention. For the purpose of the current study, only the measurements at baseline and 1 week after intervention were used. Both trials were approved by the Ethical Review Board in Uppsala, Sweden, and informed consent was obtained from all individual participants included in the study. The two trials with their adherent samples will from here on be called trial I—sample I [22] and trial II—sample II [23], respectively.

The purpose of trial I was to evaluate the effects of neck coordination exercise, compared to either strength training for the neck and shoulder regions or massage treatment, in 108 women with non-specific neck–shoulder pain [22]. The inclusion criteria for the study were women, age 25–65 years, with more than 3 months of non-specific neck pain with the neck region indicated as the dominant pain area on a pain drawing [24] and disability with limitations in performing everyday activities involving the neck, shoulders and arms according to DASH [25]. Excluded were those that had trauma-related neck pain, diagnosis of a psychiatric, rheumatic, neurological, inflammatory, endocrine or connective tissue disease, fibromyalgia, cancer, stroke, cardiac infarction or diabetes type I, surgery or fracture to the back, neck, or shoulder in the last 3 years, shoulder luxation in the last year or reported strenuous exercise >3 times/week during the last 6 months. All interventions comprised of 22 individually supervised treatment sessions. The neck coordination exercise was performed with a training device that participants wore on their head [26]. The exercise task was to control, through visual feedback via mirrors, the movement of a metal ball placed on the device with the aim to improve the fine movement control of the cervical spine. The strength training intervention consisted of isometric and dynamic exercises for the neck- and shoulder muscles, inspired by the training programme of Ylinen and co-workers [27]. The massage treatment consisted of classical massage for the back, neck and shoulders.

In trial II, the purpose was to evaluate individualized treatment compared to non-individualized treatment or treatment as usual (participants received no treatment from the study and no restriction to what they were allowed to do) in 120 women with non-specific neck–shoulder pain [23]. The inclusion and exclusion criteria were the same as in trial I with the following exceptions: The age span in trial II was 20–65 years, pain duration was minimum 6 weeks, and participants were required to have between mild and severe disability according to NDI [19] (participants did not answer DASH in trial II) and impaired capacity to work due to neck problems [28]. Also, in trial II, strenuous exercise was not an exclusion criteria, but concurrent low back pain was. Participants of the two intervention groups received treatments two to three times per week for a period of 11 weeks. The individualized treatment was tailored to the individuals’ functional limitations and symptoms, as decided from a decision model comprising the five categories cervical mobility, neck–shoulder strength and motor control, eye–head–neck control, trapezius myalgia and cervicogenic headache. The non-individualized treatment included the same available treatment components but applied quasi-randomly [23].

In both trial I and II, the participants answered a comprehensive set of questionnaires at each test occasion. This set included ProFitMap-neck [12] NDI [29] and a global rating of change scale (GRCS, only administered after intervention). In the present study, the GRCS is used as a comparator instrument and external anchor of change in relation to ProFitMap-neck and NDI.

The number of participants that completed the intervention was 89 in trial I and 104 in trial II. Four participants were excluded from the analysis because they rated <3 on GRCS (one participant from sample I and three from sample II). Of the remaining 88 participants in sample I, 47 rated an improvement in health after the intervention (i.e., 6 or 7 on the GRCS), and 41 were categorized as no or little change (i.e., rated 3, 4, or 5 on the GRCS). Of the remaining 101 participants in sample II, 54 rated an improvement and 47 did not do so. The characteristics and baseline measurements of the samples are shown in Table 1. The maximum possible score was reached at follow-up for five and six participants for the ProFitMap-neck function index and NDI, respectively. No participant reached the maximum possible score in any of the indices at baseline. Table 2 presents the change scores for each category in the two samples, including the proportion of missing items in the questionnaires.

The AUC with 95 % confidence interval for the two samples is shown in Table 3. Overall, the ProFitMap-neck performed similarly to NDI, and the AUCs tended to be larger for sample II compared to sample I but the confidence intervals showed substantial overlap. Among the ProFitMap-neck indices, the function index had slightly lower AUC than the symptom indices.

In Table 4, the MIC and its corresponding sensitivity and specificity are shown for all indices in both samples. NDI had the lowest MIC in both samples. For sample I, this NDI-MIC value had the lowest sensitivity and specificity, but in sample II its sensitivity was higher. The highest combination of sensitivity and specificity was observed for the ProFitMap-neck symptom-intensity index in sample II. The highest MIC in both samples was obtained for the ProFitMap-neck symptom-frequency index. Overall, the MIC tended to be lower in sample II for all indices.

In the present study, we aimed to investigate the ProFitMap-neck performance by assessing its responsiveness, and compare that to NDI, in two samples of women with non-specific neck–shoulder pain. The results suggest that both measures possess similar ability to detect change in self-rated perceived health with AUC exceeding 0.7 which is a cut-off value that has been used to delineate adequate responsiveness [40‐43]. While this was the first examination of responsiveness for ProFitMap-neck, several previous studies exist on this measurement property for NDI [30, 34, 36, 40‐42, 44‐47]. Most of these show results in concordance with the present study, except for two studies that found lower AUC for NDI (0.57 [36] and 0.59 [44]). In a review of measurement properties of eight neck-specific pain and disability questionnaires, where NDI but not ProFitMap-neck was included, it was concluded that NDI was one of two questionnaires that had better than limited evidence of responsiveness [5].

Correlation analyses between change scores and GRCSs showed significant associations for both ProFitMap-neck indices and NDI, which indicates that the GRCSs were valid anchors for our study [37, 48]. In contrast to the more general GRCS used in trial I, the GRCS in trial II explicitly expressed neck–shoulder problems and may therefore have better construct validity as an external anchor [32, 49]. This could have affected our results; however, correlations were only slightly higher in trial II, and earlier findings of similar reliability for questions on general perceived recovery compared to perceived change in neck pain [50] indicate that both types of questions could be used. Global rating of change scales of general perceived recovery seem to be the most common external anchors (see e.g. [30, 36, 40, 41, 46, 47]).

Minimal important change of normalized values in the two samples examined ranged between 6.6 and 13.6 % for the ProFitMap-neck indices and was 6.3 and 5.2 % for the NDI. The symptom-frequency index had the highest MIC in both samples. This may reflect the often existing temporal variation of symptoms in neck pain individuals [7, 51]. The symptom-frequency index had also the highest measurement error in the previous validation study of ProFitMap-neck [12]. However, pain frequency may still be important to measure in chronic pain clinical trials since temporal aspects of pain have shown to be a valid dimension discerned from pain intensity, therefore recommended as an outcome [31]. The MICs obtained for NDI are rather low compared with previous studies in chronic neck pain, showing a range of 5–19 % [30, 34, 36, 40‐42, 44, 47, 52].One explanation for this may be the low mean NDI baseline scores of 28 and 23 NDI% in sample I and II, respectively. Association between NDI baseline scores and MIC was recently demonstrated, showing larger MIC for those above (i.e., with higher disability) compared to those below (i.e., with lower disability) median baseline score [42, 44, 52]. The same effect of baseline values on MIC in neck pain patients was also shown for pain intensity numerical rating scale [53], but not for Neck Pain Disability Scale [42]. In the comparison of MIC values of NDI and the ProFitMap-neck indices, the latter were slightly higher. However, the combination of sensitivity and specificity for the MICs was higher in all ProFitMap-neck indices in sample I and in the majority of the ProFitMap-neck indices in sample II. The comparison of the MIC of ProFitMap-neck with MIC of other neck-specific questionnaires beside NDI is hampered by the low number of studies and differing methodology to determine MIC. For comparable studies, Neck Pain and Disability Scale [41, 42] and Neck Bournemouth Questionnaire [54] had MIC of similar magnitude as ProFitMap-neck, whereas MIC reported for the Core Outcome Measure Index summary score was higher (20 and 27 %) [55, 56].

Methods to determine MIC can be sorted into anchor-based or distribution-based approaches. Distribution-based methods are conceptually different in being based on statistical characteristics of the sample distribution. These methods rather deal with minimal detectable change than any indication of the importance for the patient of the observed change, which is the ground for anchor-based methods [48, 57, 58]. In the current study, we used anchor-based methods for determining responsiveness and MIC, thereby considering patient perception as a key factor for the MIC [59] in accordance with its conceptual definition [17].

However, the reliance of anchor-based methods poses several challenges. The first concerns the validity of the external anchor. In line with many other studies [30, 34, 36, 40‐42, 44‐47, 53, 60], we used GRCS as the external anchor to discern improved versus no or little change. This method has been criticized, one reason being recall bias [32]. The COSMIN (Consensus-based Standards for the selection of health status Measurement Instruments) checklist points out that GRCS should not be regarded as a gold standard, and suggests that no gold standard exists for patient-reported outcomes except for longer versions of the same outcome as the one under test [17]. However, the same checklist recommends using a GRCS of the same construct as the instrument under study as a useful comparator with high face validity, and evidence supports the use of GRCS with 7–11 response alternatives [32]. Also, in a review on methodological quality of neck questionnaire studies, GRCS was deemed appropriate and the best criterion available [6]. A second challenge of anchor-based methods, brought up by de Vet and co-workers [57], is that they do not include any aspect of measurement precision, thereby leaving out information whether the MIC lies within measurement error, i.e., is smaller than minimal detectable change, of the tested scale or not. The MIC of the ProFitMap-neck indices established in the present study was smaller than the smallest detectable change earlier determined from test–retest of 45 subjects with non-specific neck pain [12]. The same situation applies to our result on the MIC for NDI, i.e., they were smaller than minimal detectable change observed in most other studies. As a matter of fact, MIC was always smaller than minimal detectable change in NDI (see compilation, Table 1 in [52]), meaning that MIC may be confounded with measurement error [58]. Thus, using minimal detectable change instead of MIC as cut-off in NDI and ProFitMap-neck increases the certainty of that measurement error will be exceeded and should therefore be the choice when a high rate false positive (low specificity) should be avoided. The MIC, expressed as the optimal point on the ROC curve for high sensitivity and specificity equally weighted, may be used as an alternative cut-off in situations where a low rate of false negative (high sensitivity) is equally important. Finally, the use of anchor-based methods to determine responsiveness is not suitable if the proportion of improved versus not improved are severely skewed with only few individuals in one category [61]. This was, however, not the case in either sample (Table 2).

Limitations of the study include the long time period of 12 weeks between measurements which may increase recall bias for the GRCS questions. Another aspect to consider is the generalizability of the results to other women with subacute and chronic non-specific neck pain. The recruitment procedure in both trials was partly done by advertising [22, 23], and samples should therefore be considered as convenience samples which constituted of women with relatively mild pain and disability. This may reduce the generalizability of results. Also, findings cannot be generalized to men with neck pain. A further limitation is that the interventions given could potentially have influenced the MIC differently, but separate analyses of each intervention group were not possible due to small group sample sizes. Finally, the small differences between trial I and II in respect of the inclusion criteria and wordings of the external anchors, and the differences in characteristics and baseline measurements, made us unwilling to pool the data into one sample. This could be seen as a drawback due to reduced sample size, but the number of participants in each sample was most likely adequate for our purpose [62]. With that in mind, the separate samples used could be regarded a strength of the study since confirmation of responsiveness across samples is recommended [37].

This study extends the knowledge of measurement properties of the ProFitMap-neck questionnaire by endorsing its validity for change scores in two groups of women with non-specific neck–shoulder pain. In both groups, adequate ability to discriminate between improved and not or little changed participants was demonstrated and values of important change presented. The responsiveness of the ProFitMap-neck was similar to that of NDI which, in turn, was similar to earlier findings corroborating NDI and ProFitMap-neck as responsive measures. Continuing future validation of the ProFitMap-neck is warranted and should include other neck pain conditions as well as men.