Figure 1 illustrates the development process of the questionnaire.
Initial content of the questionnaire
The Multidimensional Pain Questionnaire in Dance (MPQDA) is based on the survey by Lampe et al. [14, 15] concerning pain in semi- and non-professional dance, which was then modified and adapted for professional dancers. The theoretical basis is Birbaumer and Schmidt's pain model [9]. In addition, the time course was considered with regard to acute and chronic pain [9, 16, 17]. The questionnaire includes the following blocks:
a) prevalence and localizations
b) subjective sensation of pain
c) temporal course of pain
d) pain assessment
e) pain behavior
Block a) contained the three-month pain prevalence collected via a no/yes selection during or within 24 hours after working time (training, rehearsals, and performances). Only participants who reported pain answered the subsequent pain questions. The period of three months was based on the survey period of the validated Pain Sensation Scale (SES) according to Geissner [18]. Pain localizations in the previous three months were inquired by multiple selection from six categories of the head and trunk region, six categories of the upper extremity (right and left, respectively) and eight categories of the lower extremity (right and left, respectively). The selection options were based on existing assessments such as the Nordic Musculoskeletal Questionnaire (NMQ) [19] or the Self-Estimated Functional Inability because of Pain (SEFIP) [12]. In the subsequent question, the most affected region was to be selected from these categories of localizations, to which all further pain questions referred. On a four-point ordinal scale from ‘not’ (= 0) to ‘very’ (= 3), accompanying symptoms (tension, redness, swelling, warming, restrictions of mobility and resilience) were asked.
Block b) contained the average perceived pain intensity on the 11-point Numeric Rating Scale (NRS) from ‘no pain’ (= 0) to ‘worst conceivable pain’ (= 10) [20]. Furthermore, pain quality was assessed by rating twelve sensory and eight affective adjectives on a four-point ordinal scale ranging from ‘does not apply’ (= 0) to ‘applies exactly’ (= 3). The scale was based on existing pain adjective lists: McGill Pain Questionnaire (MPQ) by Melzack [21], the Pain Sensation Scale (SES) by Geissner [18] and the Pain Description List (SBL) by Korb and Pfingsten from the German Pain Questionnaire (DSF, version 2015.2) [22, 23]. The adjectives were chosen from content considerations which were based on the dancers' descriptions of pain in Thomas and Tarr's study [3].
In block c), the duration of pain was surveyed using five categories ranging from a maximum of one week to longer than six months. The time periods were oriented to the classification of acute, subacute and chronic pain [9, 17]. The pain frequency in the selected period could be chosen as a subcategory from ‘only once’ (= 0) to ‘permanently’ (= 3). Furthermore, the manner of pain occurrence was asked, i.e., if it occurred as a sudden traumatic event, as a creeping process, as an occurrence of pain within 24 hours after working time or ‘in another way’ (with the option of an open entry). The frequency of pain occurrence under the mechanical stimuli of weight-bearing and movement, or at rest, was asked on a four-point ordinal scale from ‘never’ (= 0) to ‘permanently’ (= 3).
In block d), an 11-point bipolar rating scale was used to record the pain assessment with regard to ‘good/positive’ pain as rather harmless and not disturbing (values from + 1 to + 5) and ‘bad/negative’ pain as alarming and disturbing (values from − 1 to -5). There was the opportunity to choose the middle of the scale (= 0) if the pain could not be classified as ‘good’ or ‘bad’.
Block e) asked whether work was done in training, rehearsals or performances despite experiencing pain (subjective ability to work); response options were ‘No’ (= 2), ‘Yes, with limitations’ (= 1) and ‘Yes, without limitations’ (= 0). If work was carried out despite pain, motives for this behavior were subsequently asked. Fourteen items formulated on the basis of the known motives for dancing with pain or injury from the literature [1, 4, 24] were rated on a four-point ordinal scale from ‘does not apply’ (= 0) to ‘applies exactly’ (= 3).
Sociodemographic and health-related information were recorded as co-variables; these included gender, age, height, weight, injuries, diseases and smoking. Information on the participants' professional practice included the years of professional work as a dancer, the employment relationship (freelance/salaried), the predominant dance style (classical/neoclassical, contemporary/dance theater, musical/revue, other) and information on the average amount of training, rehearsals and performances in hours per week.
Online construction and pretest of the questionnaire
The questionnaire was provided via the survey server SoSci Survey [25] in German and in English. The authors translated the questionnaire from German into English and the translation was subsequently checked by a native speaker.
In order to examine qualitatively the practicability and comprehensibility of the German and English versions, an online pretest was conducted [26]. The pretest was conducted from 22/05/2019 to 07/06/2019 by five dancers with different linguistic backgrounds. Two questionnaires were completed in the German version and three in the English version. In the pretest, comments were permitted in addition to the regular response. Furthermore, additional questions (e.g., on the meaningfulness of the questions and ordering) were asked, based on recommendations available on the construction of written surveys [27–29]. Minor adjustments were then made as a result of the pretest (e.g., the word ‘currently’ was added to the question about injuries).
Study design and population
In a pilot study, the questionnaire was field-tested and optimized using cross-sectional data. A sample size of at least n = 50 was sought, a size which is recommended for validation studies [30]. The target group included professional dancers in Germany above the age of 18 with sufficient German or English language skills. Excluded individuals were persons younger than 18 years as well as dancers from the semi-, non- and pre-professional areas. A cover letter was included which informed the participants about the study. The participants were only included if they actively agreed to the consent question on the first page. An ethical vote was obtained via the Ethics Committee of the Department of Medicine of the Goethe University Frankfurt am Main (No. 25/19).
Data collection
The sampling was conducted by a non-probabilistic method. Access to the questionnaire was via the electronic distribution of the questionnaire's link at dancers' workplaces (e-mails to theaters), dance associations (Tanzmedizin Deutschland e. V. (ta.med), Bundesdeutsche Ballett- und Tanztheaterdirektor*innen-Konferenz (BBTK), Stiftung TANZ, Dancersconnect, Dachverband Tanz) and social media (Facebook, tanznetz.de). The survey period was from 27/02/2020 to 07/04/2020.
Data processing
A total of n = 82 participants completely answered the questionnaire. The raw data were checked according to the inclusion and exclusion criteria, the logical sequence of the response and the plausibility. Five participants had experienced no pain in the last 3 months; these cases were eliminated from further analysis as they had consequently not answered the main questionnaire containing the in-depth questions on pain. Four cases were eliminated in which professional work as a dancer appeared questionable (n = 3 with a total weekly work time of < 10 hours and n = 1 being a tournament sports trainer). Another case with > 50% missing values was also eliminated. Thus, a total of n = 72 cases were included in the analysis (n = 36 each in the English and German language versions).
Data analysis
Data were analyzed using Stata/IC 14.2. Based on descriptive statistics, for the dichotomous variables, items were dropped or combined if fewer than 5 participants in the TS had selected the respective item; for the four-point ordinal scales, items were dropped if more than 70% in the TS had selected the ‘not’ (= 0) or ‘does not apply’ (= 0) category of the respective item. Descriptive statistics were then calculated for the total sample (TS), English version (EV) and German version (GV). Differences between the language versions were calculated using the Chi2-test, Fisher's exact test, the Mann-Whitney-U-test or the t-test for independent samples. The significance level was α = 0.05. For non-normally distributed metric variables, the median (\(\tilde{x}\)) and interquartile range (IQR) were specified and non-parametric test procedures were used.
Scales with numerous items (dichotomous: pain localizations according to the subscales ‘head and torso’, ‘upper extremity’ and ‘lower extremity’; four-point ordinal scales: accompanying symptoms, sensory and affective pain quality, pain occurrence under mechanical stimuli and psychosocial motives for working with pain) were checked in several steps by statistical parameters. Items were dropped or combined if necessary. Firstly, for an examination of the dimensionality or homogeneity of the items of a scale, inter-item correlations were computed by means of tetrachoric correlation for the dichotomous variables and polychoric correlation for the four-point ordinal scales. Items were dropped if they did not correlate with any of the other items of the scale (rpol or rtet < 0.2). In addition, item removal was considered when the correlation was very high (rpol or rtet > 0.9) [30]. Secondly, in the course of convergent and divergent validity, Pearson correlations between the items of a scale and the rest sum score of the scale (without including the respective item to be tested in the sum score) was calculated [31]. The threshold value for adequate convergent validity was set at r = 0.3; this value is known from the literature as the threshold value in the course of item-total correlation [30]. If the correlation was not high enough, item removal was considered. Divergent validity is defined by the proportion of items that have a greater correlation coefficient with the sum score of their own dimension than with the score of another scale [31]. Thirdly, as a reliability measure, internal consistency was calculated using Cronbach's alpha (> 0.9 excellent, > 0.8 good, > 0.7 acceptable, > 0.6 questionable, > 0.5 poor and < 0.5 unacceptable) [30, 32]. On the basis of the different calculation procedures, which were carried out for the TS as well as for the EV and GV, and the content-related considerations, the authors decided whether a sum score formation of the respective scales would be appropriate.
The sum scores or the individual items were included in the further analysis. Since most of the pain dimensions did not show significant differences between the EV and the GV, the TS was used for further analysis.
Receiver-Operating-Characteristics (ROC) analysis was used to examine the extent to which the pain dimensions of the MPQDA can classify the subjective ability to work in training. For this purpose, the variable of subjective work ability in training was dichotomized into no work ability (‘No’) or limited work ability (‘Yes, with limitations’) (= 1) and full work ability (‘Yes, without limitations’) (= 0). Nominal variables of pain dimensions were each coded as dichotomous dummy variables. Furthermore, the items of the psychosocial motives for working with pain were recoded into ‘does not apply’ (= 3) to ‘applies exactly’ (= 0), since a full ability to work (= 0) is more likely with the stronger manifestation of the motives. Subsequently, correlations (point-biserial correlation, Somers' D, Cramer's V) of the pain dimensions, as well as the sociodemographic, health and occupational data, with the subjective work ability in training (dichotomous) were calculated. Next, in the course of the ROC analysis of the pain dimensions, the Area under the Curve (AUC), as a measure of discriminability [30, 33] of the single pain dimensions, was determined with a 95% confidence interval. Discriminability was assumed if the AUC was > 0.5 [33]. An AUC of ≥ 0.7 was considered acceptable [30]. Moreover, for metric and ordinal variables, it was judged whether the ROC curve was a proper or improper curve. An improper curve was judged as being present when the curve crossed the 0.5 change diagonal [33].