Study design
The study is a comparative, secondary analysis of a population-based cross-sectional health survey (2012) and final year health professions’ student surveys (2016 and 2017) undertaken in Switzerland.
Study population and data
The Swiss Health Survey (SHS) undertaken by the Swiss Federal Statistical Office (SFSO) is a nationwide survey on health status, health service utilization, and health-related behavior. The SHS employs telephone interviews and subsequent written questionnaires; it was first conducted in 1992 and is repeated every five years. For each survey year, a multistage probability sample is drawn of the permanent resident (including foreign nationals) population in Switzerland after stratification by the three predominant language/geographic regions (German, French, and Italian). Samples include individuals’ aged 15 years or older living in private households and excluding those living in institutions, i.e. hospitals, homes for the elderly, prisons, monasteries, and military barracks; only subjects conversant in either of the three languages are surveyed. Data were collected and administered by the SFSO under the regulation of the Federal Statistics Act (FSA) of 1992, which is a framework of law dedicated to federal data collection, data protection, and data security. Participants provide informed consent, which accommodates all future use of the data for research (FSA, 1992).
For our study, we obtained the most recent SHS data (2012). The respective net sample size comprised n = 21,597 respondents, representing 6,838,268 subjects in the general population. Data for HP students were derived from the National Survey of Final Year HP Students (National Graduate Survey of Health Professionals from Universities of Applied Sciences; Nat-ABBE). The Nat-ABBE is a nationwide census survey of final year HP students at six universities within the three major language regions (German, French, and Italian). The Nat-ABBE is part of a nationwide collaboration of Universities of Applied Sciences (https://www.cnhw.ch/en/) to develop a competence network to counter projected shortages in the health workforce. While the main focus of the Nat-ABBE encompasses education and professional development, it also comprises several questions on health status, and health-related behavior. The Nat-ABBE employs written online questionnaires; it was first conducted in 2016 and is repeated every year. We obtained Nat-ABBE data for the years 2016 and 2017 with a respective sample size of n = 1980. This sample was reduced to include only full-time students and students of large faculties, i.e. midwifery, nursing, nutritional sciences, occupational therapy, and physiotherapy (n = 1848). Excluded subjects (n = 132) comprised all students from Medical Radiology (n = 47) because this subject can only be studied in the French speaking part of Switzerland. Moreover, students of nursing and midwifery with a nursing diploma were also excluded (n = 85) because they worked already in the healthcare system, they studied part-time, and were much older than their fellow students.
For the comparison between the general Swiss population and HP students, data were pooled yielding an initial combined sample size of n = 23,445. Moreover, we extracted three demographically-stratified samples of female participants aged 21–30 years with secondary (n = 848), tertiary (n = 386), and secondary or tertiary education (n = 1234) from the SHS to match the corresponding female HP students (n = 1501) yielding a restricted pooled sample of n = 2349; n = 1887; and n = 2735. The comparison of HP students who were about to complete tertiary education with SHS respondents in the three respective samples served to assess potential education related differences in pain prevalence. More specifically, SHS respondents indicated their education level but we lacked information on whether they were completing a higher level of education at that time, i.e. respondents who completed secondary education may or may not have been studying at the tertiary level at the time the SHS was administered. Consequently, we used the three restricted samples to assess whether HP students’ pain prevalence were more similar to SHS respondents who completed tertiary education or to SHS respondents who completed secondary education or were more similar to a mixture of SHS respondents with either completed secondary or tertiary education.
Outcomes: Prevalence of LBP and NP
Self-reported LBP and NP, the target outcomes of this study, were derived from self-reported data. SHS participants were confronted with a list of health problems, including LBP and NP, and were asked to report for each health problem whether they had experienced it (Question: “I will read out different health problems. Please tell me for each of these health problems whether in the past 4 weeks you have had it”). Responses were captured using a three-point ordinal scale (no, a little bit, strong). Similarly, the Nat-ABBE asked respondents to report health problems (Question: “In the past year, have you had one or more of the following health problems?”); responses were captured using a four-point ordinal scale (no, rarely, occasionally, often). Unfortunately, the SHS wording of the question (“Bitte sagen Sie mir jedes Mal, ob Sie das in den letzten 4 Wochen überhaupt nicht, ein bisschen oder stark gehabt haben.”) is not very precise and its closeness to colloquial language makes it difficult to judge whether it referrers to the frequency or intensity of pain. Similarly, the three categories do not allow for a final judgment of what respondents had in mind when they answered the SHS question because the categories “a little bit” and “strong” may also refer to either frequency or intensity in many Swiss dialects. In contrast, the Nat-ABBE clearly refers to pain frequency. Despite this ambiguity and the difference in the number of response categories, we feel it is safe to assume that the common category “no” describes absence of pain in general, i.e. frequency and intensity of pain is zero while any other category describes the presence of pain in general. Therefore, we derived a subject-specific binary outcome for LBP and NP (yes/no), indicating the presence or absence of pain. Prevalence of LBP and NP was conceived as the proportion of respondents reporting pain. While the dichotomization of response categories for LBP and NP in our study was primarily driven by the need to make outcomes more comparable across the two surveys, it is also not unusual for prevalence studies of LBP and NP to use dichotomous outcomes [30]. The systematic review of Hoy et al. [31] show that the majority of studies (661 of 893) do not specify the minimum episode duration necessary for inclusion, while one day was the most used when reported. As such, prevalence of LBP and NP used in our study is comparable to other studies. However, our self-reported prevalence of LBP and NP may differ from prevalence estimates in which minimum episode duration was specified [31].
Collapsing of categorical variables, whilst valid, involves loss of information and may lead to reduction in efficiency in the statistical analysis under consideration [32]. Moreover, changing outcome categories can affect the effect estimates as well as the inferences drawn from the data [33, 34]. In order to address the latter issue and justify the dichotomization of our dependent variables, we assessed the association between age, gender, education, and LBP and NP respectively [35]. We compared the results obtained for the dichotomous variables using a logistic model with the results obtained for the original ordered categorical variables using a cumulative odds model (results not shown). The results yielded by the logistic model were confirmed by the alternative cumulative odds model, which incorporated the original three (SHS) and four (Nat-ABBE) ordered categories of LBP and NP, i.e. we found similarity of results regarding the size and statistical significance of effects. Furthermore, collapsing the original dependent variables into two instead of three categories was motivated by the need to derive clearly distinguishable categories from two different scales and simultaneously provide measures of LBP and NP, which achieve the best possible comparability with other studies, i.e. roughly 74% of the studies considered by Hoy et al. [31] reported prevalence of LBP based on dichotomous outcome measures (presence/absence of LBP).
Prevalence differs substantially according to prevalence period, i.e. point, four-week, yearly or lifetime. In students, yearly prevalence of LBP and NP was assessed (via the Nat-ABBE), while the national population was assessed (via the SHS) based on four-week prevalence. We derived frequency weights based on a well-cited systematic review of LBP, reporting and comparing 145 four-week and 271 yearly prevalence of LBP [31], to estimate comparable four-week prevalence in students. The systematic review found that on average, yearly prevalence was 1.25 times higher than four-week prevalence. Frequency weights were calculated as following:
$$ {\omega}_1={1.25}^{-1} $$
(1)
$$ {\omega}_2=\frac{N-\left({\omega}_1\bullet {n}_1\right)}{n_2} $$
(2)
$$ N=\sum {\omega}_i $$
(3)
Where:
N is to total number of students in the sample
n1 is the number of students with the pain condition
n2 is the number of students without the pain condition
ω1 is the frequency weight for students with the pain condition
ω2 is the frequency weight for students without the pain condition
Frequentist and Bayesian statistical approaches
Binary regression models were employed to estimate crude and adjusted prevalence. Adjustment comprised age, gender, and education. We used a conventional frequentist statistical approach for the comparison between HP students and the general Swiss population. However, a Bayesian statistical approach was employed for the comparison among HP students because of its flexibility to derive many different models, i.e. dyadic comparisons among the five health professions, from the posterior distribution. The posterior distribution was determined using Markov-Chain-Monte-Carlo (MCMC) sampling. In order to assess convergence, we initially used 4 chains with 4000 iterations and monitored the corresponding trace plots. For the final estimates, a single chain with 20,000 iterations was used. The first 2000 iterations were discarded (burn-in phase). Non-informative priors, i.e. N(0, 5), were used for all parameters in the binary models.
Statistical analysis
We used R Version 3.4.3 (R Foundation for Statistical Computing, Vienna, Austria), the package ‘survey’ [36, 37], and Stan [38] for statistical analyses. We reported estimated LBP and NP prevalence with corresponding 95% confidence intervals (95% CI) or 95% highest posterior density intervals (95% HPDI), respectively. Differences between the national population and students were assessed using design-based F-tests, which take into account the complex survey structure of the SHS [39]. Statistical significance was established at p < 0.05 [40]. Differences among students were assessed using predicted posterior mean differences in pain prevalence between student groups with corresponding 95% HPDI.