Summary of findings
Our review included 61 studies published between 1991 and 2020 that reported prevalence and/or incidence rates. In survey-based studies, we found prevalence rates ranging from 10.8–55.2% for a reference period of 12-months or more, 11.7–42.4% for a reference period of 2–6 weeks, 2.0–34.2% for a reference period of 7-days, and 20.9–26% for point-prevalence. The prevalence estimates when primary care health records were used for data ascertainment ranged from 1.0 to 4.8%. The annual incidence rates varied from 7.7–62 per 1000 person years. Women were more likely to report shoulder pain than men, and studies conducted in higher income countries generally produced higher prevalence estimates than those from lower income countries.
It can be expected that the reference period has a considerable effect on the reported prevalence rates. A longer reference period could lead to higher prevalence rates. However, because the prevalence ranges are so large even within a certain reference period there was a considerable overlap in prevalence results between reference periods. Therefore, based on the results of this review it cannot be concluded that the reference period has an effect on the prevalence rate. A possible explanation of the large ranges found between studies with the same reference period could be the differences in used case definition. However, our findings suggest that even between studies using the same case definition and the same reference period (studies performed in the COPCORD program) the differences in prevalence rates are still substantial. For these studies the main reason for differences in reported prevalence rates appear to be due to one or more differences in study population.
Using Mexico as an example, two studies [12, 13] were conducted as part of the COPCORD project, using the same sampling methods, data ascertainment method, and reference period, yet produced very different prevalence estimates (Cardiel et al.: 5.28%; Rodriguez Amado et al.: 15.2%). The significant difference between these studies is the study population, with Cardiel et al. recruiting subjects from an urban community in Mexico City, and Rodriguez Amado et al. recruiting from a mixed urban and rural community in Nuevo Leon in North-eastern Mexico. Similar differences in prevalence were observed in studies conducted in Iran [14,15,16].
Several factors which were outside the scope of this review, such as age, occupation, presence of co-morbid health conditions, and even health literacy levels and access to healthcare may influence the likelihood of people reporting musculoskeletal pain. In particular, the relationship between age, occupation and musculoskeletal pain is complex, and we were not able to explore this in our review due to limitations in the available data. The prevalence of shoulder pathology increases linearly with increasing age, however the prevalence of pain appears to decrease after the age of 65 . This could be explained by the impact of physical activity and occupation on shoulder pain, a relationship which has been demonstrated elsewhere . The global population is getting older and many countries are increasing their retirement age despite evidence that healthy working life expectancy is not increasing at the same rate . It is likely therefore that the burden of shoulder pain problems in working adults will increase over time. Future epidemiological studies will need to consider these factors.
Strengths of our review
The main strength of our review is the expansion of included studies over the previous review. Eighty per cent of the included studies were conducted after the most recent review of shoulder pain prevalence in 2004, and we obtained significantly more estimates from South America, North America, Asia, and Australasia. This represents a significant expansion of available data compared to the most recent review .
We also conducted an extensive search of several databases and used a team of three reviewers to verify study selection and data extraction.
Limitations of our review
The first potential limitation of our review is that, as with all systematic reviews, our search protocol may not have captured all available studies. We created a thorough search protocol and applied it in four major publication databases, however we did have to keep the terms of the search sufficiently narrow to prevent an overwhelming number of results that would have been impractical to work with.
We also decided to exclude non-English language studies. This is often reported as a weakness of studies as it introduces a potential source of bias. However, in our case only 3 studies were excluded based on publication language alone, and our search was sufficiently broad to capture data from many non-English speaking countries. Furthermore, there is evidence that exclusion of non-English language papers from systematic reviews may not have a significant impact on the overall conclusions .
The final limitation of our review was that we were not able to conduct a meta-analysis. This was due to the heterogeneity among the included studies which we have discussed above. Instead, we opted for a narrative synthesis of the results.
Limitations of the included studies
There were 45 studies with an overall low risk of bias and 16 studies with an overall moderate risk of bias. None of the included studies had a high overall risk of bias. The most frequent limitation across the included studies was the use of a target population that was not representative of the national population. This was particularly problematic in countries where the socioeconomic conditions varied considerably across the country, such as between urban centres and rural areas.
There was also a lack of standardised case definitions and data ascertainment methods. Many studies utilised the COPCORD questionnaire or the Nordic musculoskeletal questionnaire but the application of these was not always consistent across studies.
Several studies did not obtain an adequate response rate leaving them at more risk of selection bias, and in over half of studies there were issues with data presentation, mainly not reporting numerators and denominators with the prevalence or incidence estimate.
Implications for practice and research
Overall, this review shows that there are substantial differences in reported prevalence and incidence rates of shoulder pain. These differences could be large and explained by variations in research methods used in these studies. We found differences in definitions of shoulder pain, data ascertainment, and use of prevalence and/or incidence period which undeniably led to divergence in results. These differences underline the importance of a standardized research methodology to compare epidemiologic study results and allow for clear interpretation.
Interestingly, studies performed under the COPCORD program with a standardized research methodology showed considerable differences in reported prevalence even within a country. This variance even with standardized methodology supports the evidence that the study population has a considerable influence on the estimated prevalence of shoulder pain.
Healthcare workers and policy makers should be aware of the variations in research methods used and the population examined when interpreting prevalence or incidence rates reported by a single study and take caution when generalizing these results to other populations.