To perform a systematic review of the global prevalence of low back pain, and to examine the influence that case definition, prevalence period, and other variables have on prevalence.
To perform a systematic review of the global prevalence of low back pain, and to examine the influence that case definition, prevalence period, and other variables have on prevalence.
We conduced a new systematic review of the global prevalence of low back pain that included general population studies published between 1980 and 2009. A total of 165 studies from 54 countries were identified. Of these, 64% had been published since the last comparable review.
Low back pain was shown to be a major problem throughout the world, with the highest prevalence among female individuals and those aged 40–80 years. After adjusting for methodologic variation, the mean ± SEM point prevalence was estimated to be 11.9 ± 2.0%, and the 1-month prevalence was estimated to be 23.2 ± 2.9%.
As the population ages, the global number of individuals with low back pain is likely to increase substantially over the coming decades. Investigators are encouraged to adopt recent recommendations for a standard definition of low back pain and to consult a recently developed tool for assessing the risk of bias of prevalence studies.
Low back pain is one of the most common health problems and creates a substantial personal, community, and financial burden globally (1–4). As part of estimating the global burden of low back pain, with low back pain defined as “activity-limiting low back pain (+/− pain referred into 1 or both lower limbs) that lasts for at least 1 day” (5), country-specific prevalence data were required.
The most recent global review of the prevalence of low back pain in the adult general population was published in 2000 and showed point prevalence of 12–33% and 1-year prevalence of 22–65% (6). Since then, 2 additional global reviews have been conducted, one of which focused on the elderly (2) and the other on adolescents (7). A key finding from these reviews was the extent of methodologic variation between studies, especially regarding the case definition and prevalence period used, and the nature and extent of measures taken to minimize bias (2, 6–10).
Although these previous reviews made a major contribution to our understanding of low back pain, a large number of prevalence studies have been published subsequently. The specific aim of the current study was to perform an up-to-date systematic review of the global prevalence of low back pain for informing the Global Burden of Disease (GBD) study, and in doing so, to examine the influence that case definition, prevalence period, and other variables have on prevalence.
All population-based studies published from 1980 to 2009 in which the prevalence of low back pain was reported were considered for inclusion. Studies were excluded if they clearly were not representative of the general population (e.g., clinic patients, pregnant women, miners), were limited to a subset of individuals with low back pain (e.g., those with spondylolisthesis), had a sample size of <150, or were reviews.
The Ovid Medline, EMBase, and CINAHL electronic databases were searched. There were no age, sex, or language restrictions. The terms “back pain,” “lumbar pain,” “back ache,” “backache,” and “lumbago” were used individually and combined with each of the following: “prevalence,” “incidence,” “cross-sectional,” and “epidemiology.” Reference lists of included studies were inspected to identify additional relevant studies. One reviewer (DH) assessed the titles and abstracts of all retrieved references to identify studies that appeared to fulfill the inclusion criteria, and all potentially eligible articles were retrieved in full text.
The relevant study information was extracted (by DH) into a Microsoft Excel database (13). If a study presented age- and/or sex-specific estimates, the total counts were not extracted. If data were stratified by age and sex separately, the total and sex-specific data were not extracted. Age/sex bands with sample sizes of <50 were merged with one or more adjacent age/sex bands in the study. If a study presented both raw and standardized data, standardized data were ignored; if a study presented only standardized data, these were extracted. Case definitions were partitioned as follows: anatomic area, minimum episode duration, and whether or not cases had to have activity limitation.
Variables extracted included the following: region, country, year of publication, citation, study type, data ascertainment, sample size, case definition (overall), case definition (anatomic), case definition (minimum episode duration), case definition (activity limitation), coverage, urbanicity, each item from the risk-of-bias tool, year start of data collection, year end of data collection, prevalence period, age, sex, denominator (number of cases at risk), numerator (number of cases with low back pain), prevalence, standard error (SE), design effect, and whether data were standardized. Double entry of data took place for a randomly selected sample of the studies (10% [n = 16]) and demonstrated a high level of accuracy (99.4%). The 8 inaccuracies related to text fields (e.g., incorrect spelling) did not influence the numerical data.
One reviewer (DH) assessed the risk of bias for each included study, using a tool that was developed for this purpose and was shown to be reliable (14). The tool includes 10 items that assess measurement bias, selection bias, and bias related to the analysis (all rated as either high or low risk) and an overall assessment of risk of bias rated as either low, moderate, or high risk (see Appendix 1, which is available at the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131).
A second reviewer (RB) assessed the risk of bias on a sample of 16 studies (10%) to ensure that the criteria were applied consistently and that consensus could be reached, as recommended by the Cochrane Handbook for Systematic Reviews of Interventions (15). Overall agreement between the reviewers was 78% with a kappa value of 0.69 (95% confidence interval 0.55, 0.79), indicating moderate agreement (kappa values from 0.41 to 0.60 indicate moderate agreement, and values from 0.61 to 0.80 indicate substantial agreement). Following discussion, agreement was reached for all differences. In the majority of instances, the initial assessment by DH was confirmed by the consensus.
Uncertainty for each estimate was recorded as a standard error. If an estimate was not reported, it was calculated from the reported confidence interval or sample size, as described in Appendix 2 (available at the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). The database was screened for outliers, inconsistencies, and unexpected and missing values. Scatterplots were used to inspect the outliers. Outliers were subjectively defined as prevalence estimates that appeared to be substantially outside the plausible range. Outliers were excluded from the analysis if the study risk of bias was moderate or high, and if data were available from another study in the same country with an equal or lower risk of bias.
Heterogeneity between estimates was assessed using the I2 statistic (16). A value of zero indicates true homogeneity, while values of 25%, 50%, and 75% indicate low, moderate, and high heterogeneity, respectively (17).
Statistical analysis was performed using Stata version 10.1 (18). The influence that individual and summary risk-of-bias items, case definition, prevalence period, sex, age, year of data collection, urbanicity, and economy have on prevalence was assessed using pairwise correlations for continuous variables, t-tests for independent samples for binary variables, and one-way analysis of variance for variables with multiple categories to detect differences between groups. In a multivariate regression analysis, data were log transformed to achieve normality, using the following formula: log(prevalence + 0.2). The value 0.2 was chosen, because it provided the best approximation to normality. The outcome variable was “overall prevalence of low back pain” and was unrestricted by prevalence period. The standard error was calculated using the formula (19): SE(log[prevalence + 0.2]) = (SE[prevalence])/(prevalence + 0.2). Linear, quadratic, and cubic associations of prevalence with age were assessed by including the midpoint of the age group, centered age squared, and centered age cubed in the multivariate model.
The centered age squared and centered age cubed were calculated by taking the average of all midpoints of the age groups and subtracting this value from each midpoint of the age group to derive a centered age value, which was then squared and cubed, respectively. Midpoints of the age groups were categorized as follows: 0–9, 10–19, …, and 90–99. Prevalence trends over time were assessed using a pairwise correlation t-test for independent samples to compare data collected before 1998 with data collected during or after 1998. The influence of economic status on prevalence was assessed by grouping countries according to the World Bank income group classification system (20) and performing a t-test for independent samples. A pairwise comparison was undertaken to compare the Human Development Index with prevalence (21).
The quality of the overall evidence from the systematic review was summarized using the GRADE (Grading of Recommendations Assessment, Development and Evaluation) system (15, 22), which has the following options: 1) high quality—further research is very unlikely to change our confidence in the estimate, 2) moderate quality—further research is likely to have an important impact on our confidence in the estimate and may change the estimate, 3) low quality—further research is very likely to have an important impact on our confidence in the estimate and is likely to change the estimate, and 4) very low quality—any estimate is very uncertain.
Three prespecified sensitivity analyses were undertaken to assess their impact on the estimates, as follows: 1) removal of high risk-of-bias estimates, 2) removal of standardized estimates, and 3) removal of estimates with a prevalence period of >1 year. In addition, to attempt to control for some of the methodologic heterogeneity, a prediction of the overall mean prevalence estimate was made by including the following variables in a multivariate regression model: sex, midpoint of age group, centered age squared, centered age cubed, prevalence period, anatomic case definition, minimum episode duration, activity limitation, coverage, urbanicity, and the 10 individual risk-of-bias items. The resulting estimates were for the national-level mean point and 1-month prevalence of activity-limiting low back pain lasting for more than 1 day on the “posterior aspect of the body from the lower margin of the twelfth ribs to the lower gluteal folds” (5).
The electronic database search yielded 8,727 studies (Figure 1). Irrelevant titles (n = 8,211) were excluded, leaving 516 eligible titles. Of these, 139 abstracts met the inclusion criteria. An additional 20 eligible studies were identified from inspection of the reference lists of included studies. Nine full-text articles could not be located, leaving 150 studies that met the inclusion criteria. Two German studies (one with a high risk of bias and one with a moderate risk of bias) (23, 24) were excluded, because their prevalence estimates were considered to be outliers (point prevalence ranged from 76% to 92% in elderly Germans). Two other German studies (one with a moderate risk of bias and one with a low risk of bias) had estimates (point prevalence ranging from 20% to 50%) that were more in keeping with those of most other studies (25, 26). Of the remaining studies, one contained data from studies in 17 countries (27), and the other contained data from studies in 2 countries (23). Thus, a total 165 studies provided 966 age- or sex-specific prevalence estimates for 54 countries.
An overview of the included studies is provided in Appendix 3 (available at the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). All of the included studies had cross-sectional designs and ascertained data through an interview or self-completed questionnaire. The majority of studies included both sexes, a broad age range in the adult population, and both urban and rural populations. Of the 966 estimates, 161 were standardized by age, sex, and/or some other factor. The mean year of publication was 1999 (median 2000 [range 1982–2009]), and 64% of the studies had been published since the last comparable systematic review (6). Overall, 36 studies (22%) were rated as having a low risk of bias (353 estimates), 82 (50%) were rated as having a moderate risk of bias (434 estimates), and 47 (28%) were rated as having a high risk of bias (179 estimates). High risk-of-bias ratings were most common for item 1 (national representativeness/target population), item 4 (nonresponse bias), item 6 (case definition), and item 7 (study instrument) (see Appendix 4 and Appendix 5 [which shows the risk of bias ratings for all included studies], available at the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131).
The mean overall prevalence of low back pain, which was defined as all prevalence regardless of prevalence period, was 31.0%. The mean point prevalence was 18.3%, and 1-year prevalence was 38.0% (Table 1). For prevalence plotted by age, I2 = 99.6%, indicating high heterogeneity, and additional stratification by prevalence period and varying case definitions had only a minor impact on reducing the I2 value (to 99.1%). Multivariate regression showed that several study-level variables had a significant influence on prevalence (Table 2).
|Prevalence||No. of estimates||Quantile||Mean ± SD%|
|Point||243||6.3||10.3||15.0||24.2||35.5||18.3 ± 11.7|
|1 month||145||14.8||21.3||32.1||38.0||49.0||30.8 ± 12.7|
|1 year||271||14.3||21.0||37.4||53.0||64.8||38.0 ± 19.4|
|Lifetime||133||6.2||15.1||42.0||60.4||66.4||38.9 ± 24.3|
|Back||268||9.9||15.8||26.6||36.4||53.6||28.5 ± 16.4|
|Low back||302||7.2||12.8||26.1||43.1||56.0||29.1 ± 18.8|
|R12 to lower GFs*||254||11.0||17.4||35.2||52.0||63.7||35.5 ± 19.7|
|Minimum episode duration|
|Not specified||661||8.7||15.0||31.5||48.8||62.5||33.2 ± 20.3|
|1 day||146||14.1||22.1||34.0||44.0||56.4||33.8 ± 15.8|
|3 months/“chronic”||86||8.7||12.8||19.2||24.3||33.6||20.1 ± 9.8|
|With or without activity limitation||912||9.1||15.8||29.1||45.5||58.2||31.8 ± 19.0|
|Activity-limiting only||54||5.0||8.1||12.2||18.8||30.8||17.0 ± 15.4|
|Covariate||Regression coefficient||95% CI||P|
|Centered age squared||<0.01||<0.01, <0.01||<0.001|
|Centered age cubed||<0.01||<0.01, <0.01||0.147|
|1 month||0.30||0.24, 0.36||<0.001|
|1 year||0.45||0.40, 0.49||<0.001|
|Anatomic case definition|
|Low back||0.06||0.01, 0.11||0.029|
|R12 to lower GFs†||0.56||0.45, 0.67||<0.001|
|Minimum episode duration|
|1 day||−0.10||−0.15, −0.04||<0.001|
|3 months/“chronic”||−0.28||−0.37, −0.19||<0.001|
|With or without activity limitation||0|
|Activity-limiting only||−0.20||−0.29, −0.10||<0.001|
|Risk of bias item 1, target population|
|High risk||−0.07||−0.14, 0.01||0.071|
|Risk of bias item 2, sampling frame|
|High risk||0.03||−0.05, 0.12||0.457|
|Risk of bias item 3, random selection|
|High risk||0.11||−0.01, 0.23||0.066|
|Risk of bias item 4, nonresponse bias|
|High risk||−0.04||−0.08, 0.00||0.079|
|Risk of bias item 5, was a proxy used?|
|High risk||−0.51||−0.62, −0.40||<0.001|
|Risk of bias item 6, case definition|
|High risk||0.44||0.35, 0.53||<0.001|
|Risk of bias item 7, study instrument|
|High risk||0.12||0.07, 0.16||<0.001|
|Risk of bias item 8, data collection mode|
|High risk||0.47||0.34, 0.60||<0.001|
|Risk of bias item 9, prevalence period|
|High risk||−0.07||−0.15, 0.00||0.050|
|Risk of bias item 10, numerator/denominator|
|High risk||0.01||−0.18, 0.20||0.913|
Five of the 10 individual risk-of-bias items were shown to significantly influence prevalence (Table 2). A high risk of bias for 3 items (case definition [item 6], whether the study instrument had been tested for reliability and validity [item 7], and comparability of mode of data collection [item 8]) was associated with a higher prevalence, while high risk of bias for 2 items (whether the data were collected directly from subjects as opposed to a proxy [item 5] and prevalence period [item 9]) was associated with a lower prevalence.
Most estimates (n = 661) did not specify the minimum episode duration required for a case to be counted (Table 1). Of those that did specify the minimum duration, the most common durations were 1 day (n = 146), 3 months (n = 38), and 1 week (n = 34). In addition, 48 estimates specified that disease “chronicity” was required for inclusion. For estimates in which the minimum episode duration was not specified, the mean prevalence was significantly higher than that for estimates for which durations were specified as 1 day (T = −3.73, P < 0.001) and 3 months/“chronic” (T = −6.61, P < 0.001).
For anatomic locations, the “low back” was the most common category (n = 302), followed by the “back” (n = 268), and the “posterior aspect of the body from the lower margin of the twelfth ribs to the lower gluteal folds” (n = 254) (Table 1). Prevalence differed significantly across anatomic definitions (F = 20.91, P < 0.001); the definitions of “low back” (T = 2.05, P = 0.041) and “posterior aspect of the body from the lower margin of the twelfth ribs to the lower gluteal folds” (T = 10.52, P < 0.001) were associated with significantly higher mean prevalence compared with the “back.”
Only a small proportion of the estimates were restricted to activity-limiting cases (n = 54) (Table 1). The mean prevalence of activity-limiting low back pain was approximately half that of low back pain with or without activity limitation (T = 5.63, P < 0.001). Activity limitation continued to be significantly related to prevalence in the multivariate regression analysis (T = −4.02, P < 0.001).
The most common prevalence periods were point (n = 243), 1 month (n = 145), 1 year (n = 271), and lifetime (n = 133) (Table 1). Prevalence differed significantly according to prevalence period (F = 29.15, P < 0.001). The mean point prevalence (18.3%) was significantly lower than the 1-month prevalence (30.8%) (T = −9.8, P < 0.001), and the 1-month prevalence was significantly lower than the 1-year prevalence (38.0%) (T = −4.0, P = 0.001) (Figure 2). There was no significant difference between the 1-year prevalence and the lifetime prevalence (38.9%). Regression analysis demonstrated that 1-month, 1-year, and lifetime prevalences were all significantly higher than the point prevalence (Table 2).
The median overall prevalence of low back pain was higher among females than among males across all age groups (Figure 3). The overall mean prevalence of low back pain was significantly higher among females compared with males (Table 3) (T = 4.1, P < 0.001), and this difference continued to be evident in the regression analysis (T = 6.04, P < 0.001). Both the mean point prevalence and the mean 1-month prevalence were significantly higher among females compared with males (T = 2.31 [P = 0.022] and T = 2.26 [P = 0.025], respectively), but there was no significant difference between the sexes for 1-year prevalence and lifetime prevalence.
|Prevalence||No. of estimates||Quantile||Mean ± SD%|
|Female||344||12.5||21.0||33.7||48.1||64.0||35.3 ± 18.8|
|Male||323||7.8||15.0||25.9||40.0||56.5||29.4 ± 18.5|
|Rural||62||1.8||13.1||31.1||45.2||63.0||31.9 ± 21.8|
|Urban||270||7.2||14.2||25.3||44.3||62.1||30.7 ± 20.4|
|Low income||13||0.5||0.8||18.2||21.7||25.9||16.7 ± 15.7|
|Middle income||216||5.2||10.6||21.4||38.6||52.0||25.4 ± 18.3|
|High income||737||10.3||16.9||30.3||46.6||60.9||32.9 ± 19.0|
A cubic representation of the age curve provided the best fit to prevalence. Both the mean prevalence and the median prevalence were high during adolescence, declined among those ages 20–29 years, progressively increased until peaking somewhere between 40 and 69 years (this peak occurred earlier for men than women), and then progressively declined (Figure 3). However, the difference in mean prevalence between adolescents and individuals ages 20–29 years was not significant, whereas there were significant differences between those ages 20–29 years and those ages 40–69 years (T = −3.18, P = 0.002) and between individuals ages 40–69 years and those ages 80–99 years (T = 3.14, P = 0.002).
The regression analysis showed that the quadratic association with age was more significant than the cubic and linear associations. That is, the association characterized by increasing prevalence until middle age followed by a decline during older age was more significant than those characterized by gradually increasing prevalence across all ages (T = 4.36, P < 0.001) and increasing prevalence in adolescence, followed by a decline in the 20s, an increase during middle age, and a decline during the oldest ages (T = 1.45, P = 0.147). This remained so when the analysis was limited to point prevalence estimates.
The prevalence of low back pain increased very slightly over the past 3 decades (r = 0.096, P = 0.003). There was no significant difference in the mean prevalence between urban and rural areas (Table 3). The mean prevalence in countries with high-income economies was higher than estimates from countries with middle-income (T = 5.09, P < 0.001) and low-income economies (T = 3.03, P = 0.003). There was no significant difference in mean prevalence between middle-income and low-income economies. There was a strong positive correlation between a country's Human Development Index and overall mean prevalence (r = 0.088, P < 0.001), and this continued to be significant when the analysis was limited to point prevalence estimates (r = 0.122, P = 0.023).
The quality of the overall evidence from this review was moderate; that is, further research is likely to have an important impact on our confidence in the estimate and may change the estimate.
Excluding high risk-of-bias estimates from the analysis resulted in a significant increase in the overall mean prevalence, from 31.0% to 32.7% (T = 2.64, P = 0.008) and a nonsignificant increase in point prevalence, from 18.3% to 18.7%. Excluding standardized data resulted in nonsignificant increases in overall prevalence (to 31.1%) and point prevalence (to 18.7%). If estimates associated with prevalence periods longer than 1 year (5 years and lifetime) were excluded, the overall prevalence decreased significantly to 29.7% (T = −2.02, P = 0.044).
When the regression analysis results were used to adjust the overall mean ± SD prevalence (31.0 ± 0.6%) to reflect our GBD 2010 study case definition for low back pain (activity-limiting low back pain lasting more than 1 day on the “posterior aspect of the body from the lower margin of the twelfth ribs to the lower gluteal folds”), point prevalence was reduced to 11.9 ± 2.0%, and 1-month prevalence was reduced to 23.2 ± 2.9%. These values were also lower than the unadjusted mean ± SD estimates for point prevalence (18.3 ± 0.8%) and 1-month prevalence (30.8 ± 1.1%). P values less than 0.05 were considered significant.
Our updated systematic review of the global prevalence of low back pain showed that low back pain is a major problem throughout the world and is most prevalent among females and persons ages 40–80 years. After adjusting for methodologic variation, the mean ± SD point prevalence of activity-limiting low back pain lasting more than 1 day was estimated to be 11.9 ± 2.0%, and the 1-month prevalence was estimated to be 23.2 ± 2.9%. Due to significant methodologic heterogeneity between the included studies, single summary measures, such as mean prevalence, should be interpreted with caution.
This systematic review of the global prevalence of low back pain is the first to assess the risk of bias in the included studies and is the first study in which a sensitivity analysis was performed to assess the impact of including estimates with a high risk of bias (11, 12, 28). The sensitivity analysis showed that the overall mean prevalence would have been significantly higher if estimates with a high risk of bias had been excluded. In addition, 5 of the 10 individual items on the risk of bias tool had a significant influence on prevalence. These findings provide empirical data about the direction of the bias and its potential effect.
We observed a substantial increase in the number of studies of low back pain prevalence since the last comparable review (6). Similar to other reviews, we observed considerable methodologic variation between studies, which particularly related to the prevalence period and case definition (2, 6, 9, 29). A standardized definition of low back pain will assist future reviews, enable greater comparisons between countries, and ultimately lead to a far-improved understanding of low back pain.
Dionne et al (30) recommended using the following questions in prevalence studies of low back pain: 1) In the past 4 weeks, have you had pain in your low back? and 2) If yes, was this pain bad enough to limit your usual activities or change your daily routine for more than one day? Those investigators emphasized the importance of describing the specific anatomic area and, when possible, using a diagram of the body with the low back area shaded. The area they recommend for the low back is “the posterior aspect of the body from the lower margin of the twelfth ribs to the lower gluteal folds” (30). Given that low back pain is quite common, point prevalence estimates are also useful to capture and are easily interpreted by policy-makers.
In addition, a detailed description of the study population aids the validity of comparisons between populations. Factors of interest include age, sex, history of low back pain, occupation, job satisfaction, educational status, stress, anxiety, depression, social support in the workplace, body mass index, and family history of low back pain (31).
Consistent with other research, we observed a higher mean and median prevalence of low back pain among females compared with males (9, 32). Possible explanations for this difference include 1) pain related to osteoporosis (33), menstruation (34–36), or pregnancy (37–39), 2) individual or societal influences resulting in sex differences in the likelihood of reporting somatic symptoms (32, 40, 41), and 3) the divergent growth patterns between the sexes during adolescence, which may influence pain in this period (7).
We observed that the prevalence of low back pain was high during adolescence, which concurs with a previous review showing that the prevalence of low back pain increases throughout adolescence, and this peak often appears earlier in girls than in boys, possibly as a result of an earlier onset of puberty (7). In our review, the prevalence of low back pain was highest during middle age, which represents some of the most productive years of a person's working life. This results in a major economic impact for many individuals, families, businesses, and governments (42–44).
A curvilinear distribution of the prevalence of low back pain over age was also reported in a review by Dionne et al (2). Those investigators demonstrated that this was apparent for all low back pain; however, when they restricted their analysis to more severe forms of low back pain, they observed that the prevalence kept increasing in the older age groups. Consistent with these findings, there is some evidence that older individuals have a greater threshold for lower levels of pain but a reduced tolerance to more severe pain (45).
Dionne et al (2) suggested that many factors could explain the decrease in the prevalence of less severe low back pain that occurs with aging, including cognitive impairment, depression, decreased pain perception, and increased tolerance to pain. In addition, surveys often exclude persons living in institutions such as nursing homes (9), and these individuals may have a higher prevalence of low back pain compared with older persons living in the community.
Despite an increase in the amount of data since earlier reviews (6, 10), there continues to be a paucity of information on low back pain in countries with low-income and middle-income economies. Our data are consistent with a previous review showing that low back pain was less prevalent in countries with low-income and middle-income economies compared with countries with high-income economies (10). The lower prevalence of low back pain in developing countries has been speculated to be attributable to higher levels of exercise, shorter height, higher pain thresholds, and less access to industrial insurance compared with countries with high-income economies (10).
Methodologic issues are also likely to explain some of this difference, including survey planning methods and differing case definitions and sample population age and sex structures. Related to this, researchers from countries with low-income and middle-income economies may, in some cases, experience greater barriers in trying to publish studies. For example, the majority of peer-reviewed journals accept submissions only in English. Moreover, difficulties in constructing accurate sampling frames and accessing remote regions and villages can greatly add to the challenge of publishing academically rigorous studies.
The mean lifetime prevalence of low back pain (38.9%) was much lower than expected and was particularly influenced by low rates from studies conducted in China (46–48), Nepal (49), Cuba (50), and Pakistan (51). The low prevalence of low back pain observed in these countries with low-income and middle-income economies may have several influences, some of which were discussed earlier. In addition, chronic low back pain may make up a larger proportion of all low back pain in these countries, making the ratio of lifetime prevalence to other prevalence periods lower in these countries compared with countries with high-income economies. Although no data support this, a study in Tibet showed a relatively low ratio of 1-year–to–point prevalence (42%:34%), suggesting that a high proportion cases of low back pain are chronic in nature (52). The relatively low lifetime prevalence observed in these studies may also be attributable to selection, measurement, and recall bias.
Similar to most systematic reviews, our study is likely to be subject to publication bias that may have inflated the prevalence estimates of low back pain (53). We attempted to limit the potential for publication bias by conducting an extensive search for potentially relevant studies and placing a specific focus on capturing information from countries with low-income or middle-income economies. In addition, we carefully examined the risk of bias for each included estimate.
Based on the results of this systematic review, low back pain continues to be a very common problem globally. With aging populations, the absolute number of people with low back pain is likely to increase substantially over the coming decades. Further research is needed to identify risk factors and culturally appropriate interventions to prevent and treat low back pain. Researchers are encouraged to adopt recent recommendations on defining low back pain in epidemiologic studies to assist future reviews, enable comparisons between countries, and improve our understanding of low back pain. Furthermore, the tool for assessing the potential risk of bias of included estimates could be used to improve the design of future epidemiologic studies.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Hoy had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Hoy, Bain, Williams, March, Brooks, Blyth, Woolf, Vos, Buchbinder.
Acquisition of data. Hoy, March, Blyth.
Analysis and interpretation of data. Hoy, Bain, Williams, March, Brooks, Blyth, Woolf, Vos, Buchbinder.
We would like to thank the following individuals who were kind enough to provide us with data upon request: Professor Fereydoun Davatchi, Dr. Arash Tehrani, Dr. Rowsan Ara, and Professor Atiqul Haq. In addition, we are thankful to Dr. Emma Smith for her work on the GBD 2010 study, Dr. Karla Meursing for translating a number of the articles, Dr. Rungthip Puntumetakul, Melinda Protani, and Dr. Rumna De for their involvement in testing of the risk-of-bias tool, and Karen Carter and Dr. Linda Cobiac for their useful insights.