Stratifying workers on sick leave due to musculoskeletal pain: translation, cross-cultural adaptation and construct validity of the Norwegian Keele STarT MSK tool

Translation of the Keele STarT MSK tool

The translation of the Keele STarT MSK tool into Norwegian went through seven phases according to established guidelines for linguistic validation and cross-cultural adaptation [12].

Phase 1: Contact was made with the developers of the Keele STarT MSK tool to receive consent to collaborate and their approval of the present project.

Phase 2: The original version of the tool was independently translated from English into Norwegian by two native speakers fluent in English. One translator was a physiotherapist and the other had no medical background. Both translators provided a written report with their comments on linguistic challenges.

Phase 3: A synthesis of these two translations was performed, resulting in a version of the forward-translation. The challenges from each report were identified, discussed and resolved.

Phase 4: Two translators blinded to the original version of the Keele STarT MSK tool then independently translated the forward-translation back into English. Both translators were bilingual native English speakers with a medical background.

Phase 5: An expert committee consisting of two methodologists, five health professionals, and the four translators compared the backward translation with each other and with the original tool. This fifth phase resulted in a written report of each step in the process and a pre-final version of the Norwegian Keele STarT MSK tool.

Phase 6: The pre-final version was then tested among 42 primary care patients with musculoskeletal pain who were randomly selected from one university-affiliated rehabilitation clinic and one private physiotherapy clinic in Oslo, Norway. Fourteen (33.3%) of these patients were on sick leave due to their condition. Each person completed the pre-final version and had the opportunity to write down comments, answer questions, or both regarding their understanding of the instructions, questions, and response options, as well as the instrument’s wording. A few patients (7%) wanted more response options to items 2–10, rather than yes/no answers. The ambiguity with the word “physical activity” in item no. 9 was reported, as this word is not widely used in certain parts of the population. Thus, a parenthetical with examples of activities was added behind the word to clarify the concept for the responder: “physically active (e.g., walking, lifting, exercising, bending)”. Following this process, the expert committee discussed the findings and proposed a final version.

Phase 7: The final Norwegian version of the Keele STarT MSK tool was submitted with a written report to the developer of the tool (Appendix S1).

Study design and participants

We conducted a prospective cohort study using the Norwegian Labour and Welfare Service (NAV) web-based registry. This is a secure Internet site where NAV communicates with individuals receiving any form of sickness absence or disability benefits in Norway. Workers sick-listed (>4 weeks) due to musculoskeletal pain were invited to read the project information and to accept participation via digital consent. Thereafter, individuals who consented to participate completed an electronic baseline questionnaire that included the Norwegian version of the Keele STarT MSK tool and a set of self-reported measurements to assess validity. The data used in the present study were collected from November 2018 to February 2019.

Eligible participants were sick-listed workers aged 18–67 years in Norway. All participants had to be on sick leave for a minimum of 4 weeks with a diagnosis within the musculoskeletal (L) chapter of the International Classification of Primary Care, Second edition (ICPC-2) [13]. Exclusion criteria were insufficient Norwegian or English language skills to participate in the study.

The Keele STarT MSK tool

The Keele STarT MSK tool was developed to predict outcomes across patients with back, neck, upper limb, lower limb or multisite pain [8, 9]. The conceptual framework of the tool is founded on a formative model, as all of the items are indicators representing different prognostic factors for prolonged disabling pain in people with musculoskeletal disorders (Figure 1). The tool consists of 10 items; the first item is an 11-point numeric rating scale (NRS) to measure pain intensity over the past 2 weeks (0=“no pain” and 10=“pain as bad as it could be”). Responses of 0–4 are counted as 0 points, 5–6 as 1 point, 7–8 as 2 points, and 9–10 as 3 points. The next nine items concern different functional and psychosocial factors known to be predictive of persistent disability (e.g., bothersomeness, pain management, fear-avoidance, multisite pain). These items have a dichotomous response option: yes (1 point) or no (0 points). The Keele STarT MSK tool is scored by summing the 10 item scores, with a possible score of 0–12. Based on the established cut-off points, a total score of ≤4 points indicates low risk, a total score between 5 and 8 points medium risk, and ≥9 points high risk of chronicity [9].

Figure 1:

Graphical presentation of the Keele STarT MSK tool and its constructs.

Figure was adapted from van den Broek et al. [10].

Comparator instruments

A set of patient-reported measures were included as comparator instruments in the validity assessment. They measure constructs representing core domains of musculoskeletal pain (such as pain intensity, physical function, anxiety, depression, sleep and work disability) [14], [15], [16].

The Örebro Musculoskeletal Pain Screening Questionnaire Short Form (ÖMPSQ-SF) measures the psychosocial risk factors of musculoskeletal patients in five domains: self-perceived function, pain experience, fear-avoidance beliefs, distress, and return to work (RTW) expectancy. The ÖMPSQ-SF is based on a formative model and consists of 10 items. Each item is scored on a scale from 0 to 10, resulting in a total score between 0 and 100 with a higher score indicating higher risk. Based on previous studies [17, 18], a cut-off score of ≥50 (out of 100) has been found reliably classify participants into the high-risk group. The ÖMPSQ-SF has been shown to have similar discriminative ability as the original 25-item ÖMPSQ [17]. The Norwegian version of the original ÖMPSQ has good test-retest reliability and moderate predictive validity [19].

The Musculoskeletal Health Questionnaire (MSK-HQ) consists of 14 items assessing musculoskeletal health status in people with musculoskeletal pain. It is based on a formative model, covering aspects that have been shown to be highly relevant to patients across a range of musculoskeletal pain disorders and settings (e.g., pain, fatigue, physical function, sleep, self-efficacy, and psychological well-being) [20]. The total score ranges from 0 to 56, with a higher score indicating better musculoskeletal health status. The MSK-HQ has demonstrated good reliability and validity in subjects with a range of musculoskeletal disorders [20], including patients with inflammatory arthritis [21]. The Norwegian version of the MSK-HQ used in this study has shown good test-retest reliability and construct validity [22].

Quality of life was measured using the EuroQol 5 Dimensions (EQ-5D-5L) which consists of five items scored from 0 (worst imaginable health) to 5 (best imaginable health). The scores are transformed into an index value ranging from −0.59 to 1, where 1 represents perfect health [23].

Participants were also asked to indicate their average pain intensity over the past week using an 11-point NRS, ranging from 0 (“no pain”) to 10 (“worst pain imaginable”).

To ensure full completion and no missing data for the key questionnaires, a requirement to answer all questions on the Keele STarT MSK tool, ÖMPQ-SF, and MSK-HQ was implemented in the electronic baseline questionnaire.

Main study parameters

We conducted the construct validity assessment by following the recommendations of the Consensus-based Standards for the selection of health Measurement INstruments (COSMIN) initiative [24]. Construct validity is defined as the degree to which the scores of an instrument are consistent with hypotheses based on the assumption that the instrument validly measures the construct that is to be measured [25]. As recommended by the COSMIN initiative [26], we assessed construct validity using a hypotheses-based approach. A priori hypotheses were defined by providing evidence of convergent, discriminant, and known-group validity.

Convergent and discriminant validity

Eleven predefined hypotheses about the expected relationships between the Keele STarT MSK tool and comparator measures were used to assess related (convergent) or unrelated (discriminant, also called divergent) constructs. As the Keele STarT MSK tool is based on a formative model, we compared the scores of separate items to comparator measures. The hypotheses of convergent validity were formulated based on the moderate to high correlation with a similar prognostic tool (ÖMPSQ-SF) and with single items from other formative models (e.g., ÖMPSQ-SF and MSK-HQ) measuring similar constructs, whereas discriminant validity was determined by low correlation with measures of unrelated constructs [25].

A priori we expected a high (r≥0.6) positive correlation between the Keele STarT MSK total score and ÖMPSQ-SF. This hypothesis was based on the rationale that both tools are prognostic risk models based on a formative model in which all of the items form a corresponding construct that covers persisting pain disability [9, 17]. Next, we expected a high (r≥0.6) positive correlation between item 1 (pain) and pain intensity on the NRS, and item 10 (pain duration) and the item on pain duration from the ÖMPSQ-SF. We expected a moderate (r≥0.3–0.6) to high (r≥0.6) negative correlation between Keele STarT MSK item 2 (pain self-efficacy) and the single item covering symptom management from the MSK-HQ, and item 3 (bothersomeness) with the single item on bothersomeness of the MSK-HQ. We expected a moderate (r≥0.3–0.6) to high (r≥0.6) positive correlation between Keele STarT MSK item 4 (disability) and the mobility item from the EQ-5D-5L, item 6 (long-term expectancy) and the single item covering expectation of pain persistence from the ÖMPSQ-SF, item 8 (depression) and the anxiety/depression item from the EQ-5D-5L, and item 9 (fear of movement) and the single item covering fear-avoidance from the ÖMPSQ-SF. The rationale for these hypotheses was based on a recent validation of the Dutch STarT MSK tool [10]. As no comparator measures of similar constructs were available for items 5 and 7, discriminant validity was applied. Therefore, we expected a low (r<0.3) negative correlation between Keele STarT MSK item 5 (comorbid pain) and the single item from the ÖMPSQ-SF covering return to work expectancy, and item 7 (clinically relevant comorbidity) and the single item from the EQ-5D-5L concerning usual activities.

Known-group validity

To assess known-group validity (also called discriminative validity), we defined four a priori hypotheses containing a statement about the magnitude of the differences between the risk subgroups (low, medium, and high) of the Keele STarT MSK tool. For each increase in the Keele STarT MSK tool risk subgroup, we expected to find a higher sum score on the NRS and ÖMPSQ-SF, and a lower score on the EQ-5D-5L and MSK-HQ. The magnitude of the expected difference was set to ≥10% of the total score of each comparator instrument. The rationale for these hypotheses was based on the assumption that the Keele STarT MSK tool should be able to differentiate between participants who score differently on instruments measuring constructs related to prolonged disabling pain in people with musculoskeletal disorders.

Floor and ceiling effects

Floor and ceiling effects were considered to be present if >15% of the respondents achieved the minimum or maximum possible score [27].

Sample size

For construct validation, a minimum sample size of 50 participants is recommended, but a larger sample size (>100) is preferred. For known-group validation, a minimum of 50 patients per subgroup is recommended. Similar validation studies on the STarT Back tool reported that approximately 10% (range, 7.5–15.0%) of the total sample was stratified into the risk group with the fewest participants [5, 28], [29], [30]. Considering these figures, we aimed for a sample size of at least 500 participants in the current study.

Statistical analysis

Descriptive statistics were used to present sociodemographic and participant characteristics. The risk profiles distribution and their characteristics were reported. Missing data on the Keele STarT MSK tool and the comparator instruments were explored. To investigate the extent to which the sample is representative with respect to age, gender, occupational group, annual salary, geographic location, and diagnosis, descriptive statistics were used to compare the responders to non-responders (workers on sick leave >4 weeks due to musculoskeletal pain in the same timespan as in the current study) using data from the NAV registry. A maximum 5.0% deviation was agreed to indicate a representative sample.

To assess the convergent and discriminant validity, the Pearson’s correlation coefficient (r) was calculated. Values <0.3, between 0.3 and 0.6, and >0.6 were considered to indicate low, moderate, and high correlation, respectively [31]. As recommended by the COSMIN group [25], we assessed known-group validity by exploring expected differences between characteristics across risk groups via a comparison of mean scores. To present the differences between risk groups, we created box plots including minimum and maximum values, mean and median values, and the interquartile range. Construct validity was considered satisfactory if ≥75% of the hypotheses were confirmed [32]. All statistical analyses were carried out in STATA version 16.1 (StataCorp. 2015. College Station, Texas, USA).

Participant characteristics

A total of 720 individuals answered the online questionnaire, 160 (22.2%) of whom were excluded from the analysis because they did not have a musculoskeletal disorder according to the ICPC-2. We excluded another 15 (2.1%) individuals because they were not on sick leave at baseline, and 5 (0.7%) who had not been on sick leave the last 4 weeks prior to the baseline assessment. Thus, a total of 549 participants met the inclusion criteria and were included in the study. No data were missing at the item level for the majority of the measurements and sociodemographic data, except for the EQ-5D-5L (missing items of n=549, 1.5%), cohabitating (0.5%), and education level (0.2%).

The characteristics of the study population are presented in Table 1. The mean age was 48.6 ± 10.7 years (range 18–67 years), and 56.3% of the participants were female. The main medical causes of sick leave as specified on the medical certificates using ICPC-2 were: 121 (22.0%) with an upper limb condition, 107 (19.5%) with a low back condition, 90 (16.4%) with other musculoskeletal conditions, 54 (9.8%) with a joint/inflammatory condition, 51 (9.3%) with injuries/trauma, and 36 (6.6%) with a neck condition. The distribution of the musculoskeletal diagnoses is presented in TableS1.

Using the Keele STarT MSK tool, >80% of the participants reported that they had been bothered by their pain in the last 2 weeks (item 3) and that they were worried that their musculoskeletal pain would be long-lasting (item 6). Regarding risk profiles, 74 participants (13.5%) were categorised as low risk, 314 (57.2%) as medium risk and 161 (29.3%) as high risk. Age, gender, cohabitation, and sick leave days the last year prior to baseline were comparable across risk groups (Table 1).

Representativeness of the sample

In general, the demographic and socioeconomic characteristics of the participants resembled non-participants (n=168,137), which broadly confirmed that the sample was representative (Table S1). Gender was equally distributed in participants and non-participants (56.3% vs. 54% women, respectively), but participants were slightly older in our sample (mean age, 48.6 years vs. 44.0 years). Differences in income categories were <2.4%. Regarding occupational groups, differences were <2.5%, except for health and social services, which was somewhat underrepresented in our sample (20.2% vs. 27.6%). Differences in geographic location across all the Norwegian counties were <4.8%. Regarding diagnoses, our sample included a higher percentage of persons with an upper limb diagnosis (22.0% vs. 16.4%), whereas injuries/trauma diagnosis (9.3% vs. 15.6%) were somewhat underrepresented.

Construct validity

Known-group validity

Participant characteristics in the low, medium and high risk subgroups are given in Table 1. The box plot graphs in Figure 2 present the distribution of scores on the comparator measures for each risk subgroup of the Keele STarT MSK tool. For each increase in the Keele STarT MSK risk group, we found a corresponding increase (≥10% of the total score) in pain intensity on the NRS and psychosocial risk factors for work disability (ÖMPSQ-SF), and a decrease in musculoskeletal health status (MSK-HQ). The Keele STarT MSK tool was not able to distinguish between participants in the low and medium risk groups based on scores from the EQ-5D-5L. Therefore, 3 of 4 (75%) hypotheses were confirmed, which indicates satisfactory known-group validity.

Figure 2:

Box plot showing the distribution of comparator measures according to the Keele STarT MSK tool risk profiles.

(A) Pain intensity (0–10, higher score indicates more pain). (B) ÖMPSQ-SF (0–100, higher score indicates higher risk). (C) MSK-HQ (0–56, higher score indicates better musculoskeletal health status). (D) EQ-5D-5L (−0.59–1, higher score indicates better quality of life). The box provides the lower and upper quartiles, the central line is the median, the orange reference line is the mean, the whiskers are the 5% and 95% values, and the score frequency is noted by blue dots.

Floor and ceiling effects

No floor and ceiling effects were present in the Keele STarT MSK tool, as the lowest (0 points) and highest (12 points) scores were found in only 2 (0.4%) and 6 (1.1%) participants, respectively.