Ethical approval to conduct the study was obtained from the Health Research Ethics Committee (HREC) of the University of the Witwatersrand (Ethical Clearance Number: M170717), and permission to translate and validate the STarT back screening tool into isiZulu was obtained from the original authors at the Keele University in North Staffordshire, UK.

Minor linguistic differences emerged during the forward and back translation phases, which were addressed in consultation with the expert committee. The word ‘sometime’ in questions 1 and 2 of the English tool was translated to mean ‘other time’ in isiZulu, with no word meaning exactly the same in isiZulu. The words ‘more slowly’ in question 4 were translated to mean ‘long time’ in isiZulu, as again no isiZulu word was found for ‘slowly’. In question 5, the words ‘physically active’ were initially translated to mean ‘mostly exercise’ with a bit of other activity, but the expert committee agreed that it relates to being physically active during activities of daily living (ADLs) and not specific exercises. The translation was thus changed to mean ‘general activity’, which better aligns with the original text. The translators agreed that the term ‘worrying thoughts’ in question 6 could be translated to mean worry or stressful. It was decided that engikhatazayo which means ‘bothering me’ was the closest translation, as worry and stressful are understood as the same concept in isiZulu. The forward and back translators found the word ‘terrible’ in question 7 difficult to translate as no specific word exists in isiZulu. The expert committee agreed to ngizwasengathi, which means ‘I feel like’ and better covers the original meaning of the word. Difficulty was found in translating ‘in general’ and ‘enjoy’ in question 8, as no direct translated words exist in isiZulu. The expert committee revised the full sentence to convey the meaning of both these terms – Angizange ngizithokozele izinto engivame ukuzijabulisa ngazo njengenjwayelo, which translates as ‘I did not enjoy the things I usually enjoy’. The word ‘bothersome’ in question 9 is not a word often used in Africa, and does not exist in isiZulu, so the expert committee agreed that the most common translation kukuhluphe kangakanani will be used, meaning ‘to bother or trouble’.

The above changes were made, and the preliminary final translation was resubmitted to the expert committee for editing. No further changes were required, and this preliminary final translation was used in the validation process.

A summary of the demographic characteristics of the sample population is given in Table 1. A total of 37 patients was recruited to participate in our study, with seven patients not returning for the second round of testing. Two questionnaires were declared invalid because of missing data. Overall, 76% of the sample population consisted of women with 40% aged between 50 and 59 years. Educationally, 53% were high school educated, 60% were unemployed, and 50% had no previous history of back pain. Finally, 60% of the sample population was allocated to the high-risk group after tests 1 and 2.

The content validity was measured in two components, I-CVI and S-CVI. The item content validity for relevance was excellent (means I-CVI = 1.00) and was satisfactory for clarity, simplicity and ambiguity (means I-CVI = 0.94). The S-CVI was calculated by adding the I-CVI for each category (relevance, clarity, simplicity and ambiguity) and dividing by the total number of categories to reach the average. The S-CVI for the final translated isiZulu STarT back screening tool was 0.955, agreeing with Polit and Beck’s (2006) recommendation of an S-CVI being considered acceptable with a value of 0.80 or higher.

The test–retest reliability was determined by calculating Spearman’s correlation coefficient (ρ) between test 1 and test 2, resulting in ρ = 0.73, which demonstrates acceptable test–retest reliability (Glen 2018) (Table 2).

The internal consistency was evaluated by calculating Cronbach’s alpha coefficients (item by item) for questions 1–9. An alpha index of more than 0.70 is considered to be acceptable internal consistency (Bolarinwa 2015).

Cronbach’s alpha was calculated for the total score (questions 1–9) as 0.68, which indicates less than acceptable internal consistency, and for the psychosocial subscale (questions 5, 6, 7, 8 and 9) as 0.68, which also showed less than acceptable internal consistency.

Figure 1 demonstrates usability of the developed questionnaire: 33% found the translated tool very easy to understand, whilst 3% found it very difficult to understand.

Figure 2 demonstrates the ease of completion of the developed questionnaire: 40% found the translated tool very easy to complete, whilst 10% found it moderately difficult to complete.

Figure 3 demonstrates the time taken to complete the developed questionnaire: 20% took 20 min to complete the translated isiZulu tool, 13% completed it in 15 min, 44% took 10 min and 23% completed it in 5 min.

Table 3 demonstrates suggestions from the participants: 27% of the participants requested assistance with the explanation of the questions.

The translation procedure applied in this study followed the guidelines set out by Beaton et al. (2000). The English STarT back screening tool and the final isiZulu STarT back screening tool are provided in Figures 4 and 5, respectively.

Finding equivalent words without altering the meaning of the original sentence was the major problem experienced during the translation phase and was resolved through discussions within the expert committee.

Firstly, the closest possible translation for a specific word was used; secondly, the closest possible translated meaning for a phrase was used; and thirdly, sentence structure was adjusted to convey the closest possible meaning to the English version. Some words had no equivalent in isiZulu and were translated to the closest possible meaning, such as ‘sometimes’ in questions 1 and 2, ‘worrying’ in question 6 (similar difficulty was faced in the French study; Bruyere et al. 2012), ‘terrible’ in question 7 (similar difficulty was faced in the Finnish study; Piironen et al. 2016), ‘enjoy’ in question 8 and ‘bothersome’ in question 9 (similar difficulty was faced in the Finnish, Piironen et al. 2016, and the Danish, Morso et al. 2011 studies). Certain phrases such as ‘physically active’ in question 5 posed difficulties for both forward and back translators, as well as ‘in general’ in question 8. Similar difficulties were experienced in the German (Aebischer et al. 2015), Swedish (Betten et al. 2015), Italian (Maggiani & Abenavoli 2019) and Finnish (Piironen et al. 2016) translation studies. Ultimately the expert committees in all these studies decided to keep the wording as close to the original English version as possible so as not to lose the meaning entirely. All translation challenges and changes to the isiZulu version were addressed and approved by the expert committee before testing the validity, reliability and usability of the translated tool.

South Africa has one of the most diverse populations on the African continent, with 11 official recognised languages. Almost one-quarter (23%) of the population speaks isiZulu as their home language, and 58% speak it as their second language. English is the dominant language in major cities across South Africa but is only spoken as a second language by black South Africans in most other and rural areas of the country, including Witbank/eMalahleni, Mpumalanga, where up to 24% speak isiZulu (The 11 languages of South Africa 2019).

The majority of the sample population consisted of women (77%) aged between 50 and 59 years, which is similar to the literature showing a higher prevalence of low back pain onset in women aged between 30 and 50 years (Nicholas et al. 2011; Yoshimoto et al. 2017), and which does not differ in the developed world versus the developing world (Louw et al. 2007). A number of other translation studies identified employment status of their participants (Hill et al. 2008; Karstens et al. 2015; Piironen et al. 2016; Robinson et al. 2017), highlighting that the majority of patients were employed or received disability benefits, whereas most participants (60%) in our study were unemployed and did not receive employment or disability benefits.

This is an expected difference, as our study is one of the only studies carried out in developing Africa, where resources are limited, income inequality exists and the majority of the population does not have access to disability benefits or unemployment funds (United Nations Development Program (UNDP) 2017).

Most of the sample population was high school educated (53%), which was similar to the Swedish (64%) (Betten et al. 2015), Portuguese (32%) (Raimundo et al. 2017) and Brazilian–Portuguese (60%) (Pilz et al. 2014) studies. None of the other translation studies, including the original English study, examined the educational levels of their sample populations, perhaps suggesting that educational levels would not influence their translation process. Educational levels may not be a concern in developed countries; however, in the developing world where disparity in education exists, level of education/literacy could impact on completing the questionnaire (United Kingdom [UK] Essays 2016). The legacy of apartheid presents a significant challenge for education and health systems in South Africa. Black South Africans people were on the receiving end of education inequality, resulting in lower educational levels, a lack of educational resources and a shortage of qualified teachers (UK Essays 2016). This has led to a lack of basic information regarding health and wellness, particularly amongst those above the age of 50 years (UK Essays 2016). This is demonstrated by the majority of this study’s participants (77% in the 50–59-year age category), and the age prevalence of NSLBP (30–60 years) (Maher, Underwood & Buchbinder 2017). Lower levels of education demonstrated in our study suggest that the participants do not have access to basic healthcare information as a result of education inequalities and healthcare challenges.

Most participants were stratified into the high-risk group (60%), which may suggest that the patients are not actively seeking treatment for their back pain, were not getting the correct treatment when their back pain began, or were unaware of the back care treatment options available. The possibility also exists that the staff at local primary care clinics in Mpumalanga was not educated on the correct treatment pathways for these patients, and thus did not know that these patients should be referred for physiotherapy or for interdisciplinary management.

Measuring a questionnaire’s content validity measures the questionnaire’s quality (Bolarinwa 2015), which is a necessary part of developing an appropriate tool. Acceptable content validity shows that the content is relevant and appropriate, particularly when adapting a specific tool culturally, as seen in our study. The Dutch translation study (Bier et al. 2017) and the original English tool (Hill et al. 2008) tested content validity based on floor and ceiling effects, contrary to our study. For floor and ceiling effects to be present, some items need to be missing in the lower and upper ends of the scale, which results in limited content validity (Terwee et al. 2007). Floor and ceiling effects were not tested in our study, as the recommended sample size (Terwee et al. 2007) to test these effects is at least 50 patients. This study comprised only 30 patients, and the sample size calculation in our study was primarily based on a similar translation study performed by Morris et al. (2012).

Other translation studies (Morris et al. 2012) tested face and content validity simultaneously, using a group of patients to report the ease of understanding and ease of completing the translation. Our study did not test face validity, as it is an even more subjective test than content validity (Bolarinwa 2015), but used a panel of experts to test content validity, and the sample population to report on usability of the isiZulu tool. Criterion-related (predictive and concurrent) validity and construct validity were tested in other translation studies (Mphahlele, Mitchell & Kamerman 2008, Sharafi et al. 2017; Tsang Chi Chung et al. 2017), but were not tested in our study because the STarT back screening tool is a screening tool and does not predict future change. Furthermore, there is a lack of single reference isiZulu translated standard tools that screens for similar constructs.

The test–retest reliability of the preliminary isiZulu STarT back screening tool showed acceptable reliability (ICC of 0.73; 95% confidence interval [CI]: 0.38–0.82) for the total score, which is similar to that of the original version for the total score (ICC of 0.73; 95% CI: 0.57–0.84) (Hill et al. 2008). The Brazilian–Portuguese (Pilz et al. 2014), Portuguese (Raimundo et al. 2017), Finnish (Piironen et al. 2016) and German (Karstens et al. 2015) studies also demonstrated acceptable reliability (ICC ranging from 0.70 to 0.78).

Internal consistency in our study showed less than acceptable results for both the total score (0.68) and the psychosocial subscale (0.68). This was less than the original English version (0.74 total score and 0.74 psychosocial subscale) (Hill et al. 2008) as well as the Brazilian–Portuguese (Pilz et al. 2000), Portuguese (Raimundo et al. 2017), Finnish (Piironen et al. 2016), French (Bruyere et al. 2012), Persian (Abedi et al. 2015) and Japanese (Matsudaira et al. 2016) versions. Difference in the results of our study could be because of the way the constructs of the original English version were translated, with many words and phrases not existing in the isiZulu language. It may also be because of the level of education of our study sample not understanding the translated concepts clearly, particularly the negative wording of some of the questions (Bolarinwa 2015). Less than acceptable internal consistency resulted in less than acceptable reliability of the tool, which suggests the need to retest the translated tool on a different sample population.

Usability, tested in 30 participants, was found to be acceptable, easy to understand and easy to complete, whereas all the other translated versions, including the original English version (Hill et al. 2008), used a smaller sample size (n = 12–30), and also found good acceptability, ease of understanding and ease of completion (Abedi et al. 2015; Aebischer et al. 2015; Betten et al. 2015; Bier et al. 2017; Bruyere et al. 2012; Gusi et al. 2011; Hill et al. 2008; Karstens et al. 2015; Luan et al. 2014; Maggiani & Abenavoli 2019; Morso et al. 2011, Piironen et al. 2016; Pilz et al. 2000; Raimundo et al. 2017). The participants in our study took longer to complete the isiZulu questionnaire than the other aforementioned translation studies, which may be because of educational levels noted in the sample population, or because of this being the first time this sample population has been exposed to any form of health questionnaire.

The strengths of our study included the use of internationally recognised guidelines, the use of first-language isiZulu speakers throughout the process, which strengthened the linguistic accuracy of the translation, as well as the use of a wide inclusion criterion for the validation phase, and patient recruitment from a variety of geographical outpatient clinics which contributed to the acceptable external validity and reliability of our study. The limitations of our study included its inability to measure criterion or construct validity because of the lack of a single reference standard isiZulu translated tool against which the isiZulu STarT back screening tool could be tested. A smaller sample size than recommended by Terwee et al. (2007) could also have limited the validation results. The isiZulu tool also did not account for the treatment effect that NSLBP patients may have received between the first and second test–retest reliability testing, which could affect the outcome of the test–retest reliability test. However, the original STarT back screening tool was not designed to gather this type of information. The isiZulu tool was also only tested in the Mpumalanga province of South Africa, where many but not the majority of isiZulu-speaking people reside.

It is necessary to conduct another study to evaluate the psychometric properties in a larger, representative sample. Translation and validation of the Orebro Musculoskeletal Pain Screening Questionnaire (OMPSQ) into isiZulu, which then could be used as a reference standard to test the isiZulu STarT back screening tool could assist in fully validating the tool. It is also necessary to test the clinical effectiveness and cost-effectiveness of the isiZulu translation as well as the original English version in a South African context by comparing this stratified approach with a non-stratified best practice approach. This would determine whether stratified care leads to a significant decrease in healthcare costs and faster recovery periods specifically in South Africa.

isiZulu is the most widely spoken language in South Africa, so the widespread use of the isiZulu STarT back screening tool could have a significant effect on the way NSLBP is treated. Therefore, making the isiZulu tool available for use by clinicians in local clinics and private practices would improve clinical decision-making and treatment outcomes for isiZulu-speaking patients with NSLBP, and possibly improve physiotherapy and inter-professional referrals as needed. Training in ‘psychologically informed physiotherapy’ is also needed for treatment of the high-risk group.