Determinants of pain occurrence in dance teachers

Study design and sample

The present study is a retrospective cohort study. The data were collected anonymously by means of online questionnaires. The following inclusion criteria were defined for participation:

• – full-time or part-time dance teaching activity in the semi- or non-professional area,

• – age: at least 18 years old.

Content and development of the questionnaire

The pain aspects of the questionnaire were based on the Birbaumer and Schmidt scheme of pain perception and behavior [9]. The pain scheme includes aspects of pain perception, including the sensory and affective component, pain assessment (cognitive component) and pain behavior [9]. The operationalization of the pain aspects was carried out by modifying existing pain measuring instruments, taking into consideration dance-specific characteristics described in dance-specific literature [11], [12], [13], [18]. The questionnaire contained closed and partially open questions. The following blocks of questions were considered in the subsequent data analysis:

• – (a) pain prevalence,

• – (b) pain location,

• – (c) pain assessment.

Block (b) contains information on pain locations in total, including the most severely affected body region, radiating pain (yes/no) and concomitant symptoms on an ordinal 4-point Likert scale from “not” (=0) to “very” (=3) (e.g., “Restricted in mobility” and “Less resilient”). For this research, only the pain locations in total are reported, as the other questions of this block contain information on a specific pain region which is not relevant to the research question of this paper. For this study a validated self-assessment of functional disability due to pain (SEFIP) with 14 body regions to assess pain was used [21]. In order to differentiate specific regions of complaints in dance, the authors listed the locations in more detail and included a right-left distinction for the extremities (Table 1).

Block (c) contained classifications on an ordinal 4-point Likert scale from “does not apply” (=1) to “applies exactly” (=4) on the influence of dancing in pain (caused, improved, aggravated) and on statements as to whether the pain is assessed more as a natural consequence of dancing (performance pain) or as injury pain based on the study by Anderson and Hanrahan [13]. A statistically validated assessment to assess the constructs of “good” and “bad” pain in dance is currently not available.

In addition, information on dance pedagogical activity was requested: employment relationship (part-time or full-time employed, or freelance), dance style, years of dance and teaching experience, lessons per week, as well as preparations for competitions and performances during the 3-month period (yes/no). Furthermore, typical sample parameters (gender, age, height, weight, injuries (yes/no), diseases (yes/no)) were collected. Injuries and diseases could be named in an open text field.

The questionnaire was created online via the survey server SoSci Survey. The practicability and quality of the questionnaire were checked in advance by representatives of the target population in a pre-test. In the pretest, the pages of the questionnaire could be commented in a comment field. Additional questions were asked to determine the reasonableness of the questions and their order, the reasonableness of the processing effort and possible answer tendencies.

Data collection

The data collection took place over a period of 3 months between March and May, 2017. A non-probabilistic method was used in the sampling, which means that a non-randomly, convenience sample was generated [22]. The acquisition of participants was carried out through various associations with broad networks and numerous members in the dance community in Germany (Tanzmedizin Deutschland e. V. (tamed), German Professional Association for Dance Education e.V. (DBfT), Deutscher Tanzsportverband (DTV), Royal Academy of Dance (RAD), Dachverband Tanz), via e-mails to individual dance and ballet schools and via social networks on the Internet.

Data analysis

The statistical evaluation was carried out with Stata/IC 14.2. The descriptive statistics of the study sample and the variables collected were based on frequency, position and dispersion measures. Median (X∼) and interquartile distance (I₅₀) were given for non-normal distributed metric variables (age, weight, body mass index, year of dance teaching, hours per week). A binary logistic regression model was used to determine the most influential determinants of binary scaled pain occurrence within the 3-month period. The most influential determinants were determined from the sample parameters and the data on dance pedagogical activity. To reduce the number of possible determinants, variables with missing values of more than 20% were excluded as independent variables (e.g., “preparation for competitions”). In the next step, differences between DT with pain and DT without pain were statistically tested (Chi² test, Fisher exact test, Mann-Whitney-U test, t-Test in independent samples). Since there were no significant differences (p<0.05), the authors chose a higher limit of the p-value for the purpose of pre-selecting determinants for the regression: Only variables with a p-value <0.5 were included in further analysis. For the identified metric variables, pairwise Pearson correlations were performed. For correlations r≥0.5 [23], one of the two variables was eliminated. In binary logistic regression analysis, categorial variables were dummy-coded. In order to determine a meaningful regression model with as few variables as possible, a complete model with all preselected independent variables was tested against a reduced model omitting the variable with the highest p-value. The model was selected using pseudo R² (McFadden’s), Akaike’s information criteria (AIC) and Bayesian information criterion (BIC) [24], [, 25]. The larger the pseudo R², the better the quality of the model fitting results [25], whereas, the lower the AIC and BIC, the better the model matching is achieved [24]. The evidence for the superiority of a model is divided into weak (0–2), positive (2–6), strong (6–10) and very strong (>10) [24] according to the absolute difference between the BIC and AIC. For the final binary logistic regression model, the statistical parameters were determined. In order to quantify the influence of the determinants, both the odds ratios and the predicted probabilities (average marginal effects (AME)) with 95% confidence intervals were determined [25].

Study sample

A total of n=166 DT (female: n=151; male: n=15) participated in the study. The participants were a median 45.0 (I₅₀=18.0) years old. The average height was 167.8 (sd=6.3) cm and the median weight was 60.0 (I₅₀=10.5) kg. The BMI (kg/m²) was X∼=21.1 (I₅₀=3.2). According to the BMI, n=10 (6.0%) were underweight (BMI<18.5), n=134 (80.7%) were normal weight (BMI 18.5–24.9), n=15 (9.0%) were pre-obese (BMI 25.0–29.9) and n=3 (1.8%) participants had obesity (BMI>30). 122 (73.5%) of the DT were full-time employed (including n=106 self-employed or freelance and n=15 employed). 31 (18.7%) worked part-time as DT (including n=23 self-employed or freelance and n=8 employed). Most of the DT were active in artistic dance directions including ballet, modern/jazz dance, contemporary dance, hip hop. Only n=3 (1.8%) were dance sport teachers. On average, the DT had danced for 32.5 (sd=12.4) years. The median teaching experience was 18.0 (I₅₀=17.0) years. In the median, the DT taught 15.0 (I₅₀=12.0) hours per week. In the 3-month period, n=108 (65.1%) of the DT had prepared performances and n=19 (11.5%) had entered competitions.

The majority of DT reported neither to have suffered from diseases (n=116, 69.9%) nor from injuries (n=92, 55.4%). Of the n=49 DT with diseases, musculoskeletal findings, such as osteoarthritis (n=9), hip dysplasia (n=3) or scoliosis (n=2), as well as internal diseases, such as hypothyroidism (n=4), hypertension (n=3), diabetes (n=2), asthma (n=2) or rheumatism (n=2), were frequently mentioned. Occasionally, there were also psychological (depression, n=2), neurological (multiple sclerosis, stroke, n=2) or oncological (breast cancer, n=2) findings. Among the n=73 injured DT, ligament and tendon lesions (n=15), meniscus damage (n=14), disc protrusions (n=10) and muscle injuries (n=6) were common.

Pain prevalence

75.3% (n=125) of the DT reported pain while teaching in the past 12-month and 78.9% (n=131) reporting pain within 24 h thereafter. 70.5% (n=117) had been in pain while teaching in the past 3- month, with 78.3% (n=130) being in pain within 24 h thereafter. In total, 86.1% (n=143) of the DT reported pain within the 3-month period.

Pain locations

The total pain locations of the 3-month period of n=143 DT are shown in Table 1. The three most common pain locations included the area of the lower back, the neck or cervical spine and the knees.

Pain assessment

Table 2 shows the pain ratings of n=143 DT. Most DT (n=91, 63.6%) believed that dance contributed to their pain to some extent. A comparable number of DT stated that their pain was worsened by dancing to some extent (n=83, 58.0%). In each case 9.8% (n=14) were in complete agreement that dance was the exact cause of their pain and that pain was worsened by dancing. In contrast, only 41.3% (n=59) believed that dance offered a relief to their pain.

14% (n=20) were convinced that their pain had been induced by an injury and for 31.5% (n=45) this is in part true. In the belief of pain as a possible warning signal of an injury, a relatively even distribution of responses at the extremes and in the partial agreements is shown. 75.5% (n=108) agreed with the statement in parts or exactly. Only about half of the DT (n=65, 45.5%) assessed the pain as partially harmless.

Determinants of pain occurrence

Possible determinants of the occurrence of pain were age, BMI, the presence of diseases, employment (self-employed or freelance/employed) and the number of years of dance teaching experience (see Table 3). The years of teaching experience were not taken into account in the further analysis as there was a very strong correlation with age (r=0.82, p<0.001). There was no correlation between age and BMI (r=−0.01, p=0.895).

Only for 146 of the DT were the data sets of the possible determinants complete. When comparing the full model with the reduced model, the reduced model with the determinants age, BMI and diseases (pseudo R²=0.05, AIC: 111.4, BIC: 123.4, n=146) proved to be more suitable than the full model with all variables (pseudo R²=0.05, AIC: 113.1, BIC: 128.0, n=146). The evidence for this was “weak” according to the AIC (absolute difference=1.7) and “positive” according to the BIC (absolute difference=4.6). Using pseudo R² there no differences were observed.

Since the selected 3 determinants (diseases, age, BMI) were completely available in the total sample (n=160), this sample was used for further calculations. According to the calculations, the probability that the DT had been in pain during the 3-month period was significantly influenced by the variables included in the model (LR-Chi²=7.8, p<0.05, pseudo R²=0.06, n=160). The estimates of odds ratios (OR), their 95% confidence interval (CI) and changes in OR as a function of changes in determinants are presented in Table 4. The independent variables were not statistically significant (p>0.05) individually when the other variables were kept within their mean. It was found that the higher the characteristic values of the determinants were, the higher were the odds of pain occurrence during the 3-month period. The strongest increase in odds was observed in the presence of a disease. Here the odds increased by a factor of 2.8 (CI: 0.78–10.15) or by 180.8% in cases where a disease was present in comparison to the DT where no disease was present. However, the lower limits of the 95% CI of OR of the identified variables are below 1, which means that an increase in odds of pain occurrence is not clearly determined by the variables.

Figure 1 shows the odds ratios (OR) and the predicative average marginal effect (AME) with the 95% CI of the three determinants. The AME of the probability of pain occurrence in the presence of a disease was 0.098 (CI: −0.001–0.197); i.e., in the case of disease, the probability of pain occurrence increased, on average, by 9.8 percentage points compared to the non-existence of disease. The AME of age was 0.004 (CI: −0.0008–0.008) and the AME of BMI was 0.016 (CI: −0.008–0.0399).

Figure 1:

(a) Odds ratios (OR) and (b) average marginal effects (AME) with 95% CIs of pain occurrence in dance teachers for the determinants disease, age and BMI during the three-month period.

The predictive marginal effects with 95% CI in the presence of disease, age and BMI are shown in Figure 2. If no disease was present, the probability of pain was 83.5% (CI: 76.9–90.2). This increased to 93.3% (CI: 85.99–100.6) if a disease was present. At the lowest age of 19 years, the probability was 75.6% (CI: 58.9–92.4) and at the age of 70 years it was 94.0% (CI: 86.8–101.2). At the lowest BMI of 16 kg/m², the probability of pain was 75.2% (CI: 54.8–95.5) and at the highest BMI of 35 kg/m², it was 97.7% (CI: 91.1–104.3).

Figure 2:

Predictive marginal effects with 95% CIs of the probability of pain occurrence in the three-month period for dance teachers with the determinants (a) disease, (b) age, and (c) BMI.

Limitations and recommendations for future research

The representativeness of the sample may be limited by the non-probabilistic sampling method used [22]. An indication that the sample of this study may be representative of DT in Germany provided by the study by Wanke et al. [1] on the health status of n=165 DT in Germany in 2012. The randomized sample in Wanke et al. [1] shows parallels in terms of gender distribution (female: n=154, male: n=11), weight (X‾=59.8 kg), height (X‾=167.1 cm), age (X‾=46.1) and years of dance teaching activity (X‾=17.0) with the sample in this study.

The pain prevalences of this study may be overestimated in so far that the self-selective participation in the study primarily motivated DT with pain to participate. Therefore, there is a potential selection bias and a limited generalizability of results. The number of participants without pain in the 3-month period was very low, therefore, relatively little information on sample parameters and dance pedagogy among DT without pain was included in the analysis. Furthermore, a recall bias is conceivable, as the questionnaire asks for pain during the last 12 or 3-months.

When interpreting the results, it should be taken into account that with a significance level of α=0.05 no significant differences between DT with pain and DT without pain are shown (Table 3). For the pre-selection of the determinants of the regression model, a probability of error to reject the null hypothesis of less than 50% was chosen (p<0.5) in order to identify a higher number of potential determinants for the model.

The diseases and injuries in this study were self-reported so there were no uniform definitions. To what extent these were medically diagnosed or, rather, self-diagnosed diseases and injuries remains unclear. In addition, the differentiation between disease and injury does not always seem clear since, in two cases, for example, “arthrosis” was mentioned among the injuries.

Future studies should aim at random sampling to improve the representativeness of the sample and thus the generalizability of results. In order to draw conclusions as to the extent to which professional practice of dance influences the occurrence of pain, it is recommended to compare the study group with a control group without dance history. In view of dance history in DT, the years of dance experience, which were surveyed globally in this study, should be differentiated according to previous non-professional, pre-professional and/or professional dance episodes. The validity of the determinants identified in this study should be further investigated. In addition, the information on dance teaching should include questions on the age and dancing level of the students, as well as on the percentage of the type of dance teaching (e.g., sitting, indicating, full movement demonstration). Furthermore, it should be taken into account that DT can also be active as dancers or have been active as such. This information should also be collected.