The prevalence of low back pain (LBP) has been reported in the literature for different populations. Methodological differences among studies and lack of methodological rigor have made it difficult to draw conclusions from these studies. A systematic review was done for adult community prevalence studies of LBP published from 1981 to 1998. The technique of capture-recapture was performed to estimate the completeness of the search strategy used. Established guidelines and a methodological scoring system were used to critically appraise the studies. Thirteen studies were deemed methodologically acceptable. Differences in the duration of LBP used in the studies appeared to affect the prevalence rates reported and explain much of the variation seen. It was estimated that the point prevalence rate in North America is 5.6%. Further studies using superior methods are needed, however, before this estimate can be used with confidence to make health care policies and decisions relating to physical therapy.
One definition of low back pain (LBP) is any back pain between the ribs and the top of the leg, from any cause.1 Low back pain is an important public health problem in all industrialized nations.2,3 Although most people appear to recover quickly from an episode of LBP, disability resulting from back pain is more common than any other cause of activity limitation in adults aged less than 45 years and second only to arthritis in people aged 45 to 65 years.1
Prevalence is the number of people in a defined population who have a specified disease or condition at a point in time.4 Prevalence is usually measured by surveying a particular population containing individuals with and without the condition of interest.5 Thus, prevalence equals number of people with a health problem at a point in time divided by the total defined population alive at this point in time. Prevalence rates are usually reported as percentages. Prevalence is the result of many factors: (1) the periodic number of new cases, (2) the immigration and emigration of people with disease, and (3) the duration of illness.4 Because LBP is a condition that often resolves completely, individuals can be counted as having prevalent cases if pain reoccurs. This is different from chronic conditions that do not resolve (eg, rheumatoid arthritis).
Point prevalence is measured at a single point in time (ie, the number of people reporting LBP on the day of a survey).6 Period prevalence is measured over a specified time period, usually 1 year (ie, those people who report having had LBP in the past 12 months). In contrast, incidence refers to the number of new cases occurring during a period of time among a group initially free of the disorder.4 The cumulative incidence or lifetime incidence is the total number of people who have or have had the condition during their lifetime. Deyo and Tsui-Wu7 also refer to this as the cumulative lifetime prevalence. Thus, it appears that there is confusion in the literature with respect to the use of the terms “lifetime incidence” and “lifetime prevalence.” Figure 1 illustrates 10 cases of LBP in a hypothetical population of 20 individuals from 1993 to 1998. Calculations of period prevalence, point prevalence, annual incidence, and cumulative incidence are provided as examples to clarify definitions.
Numerous authors have estimated the prevalence of LBP in various populations. Unfortunately, methodological differences among studies and lack of methodological rigor make it difficult to draw accurate conclusions. Methodological issues relevant to studies that estimate LBP prevalence have been identified.8,9 Table 1 lists these issues. In summary, it is important to have (1) a random sample that is representative of the target population, (2) a measure of outcome that yields valid and reliable results, (3) a definition of LBP, and (4) a report of response rates and a comparison of respondents and nonrespondents so that the generalizability of the sample can be determined. Recently, formalized guidelines for the critical appraisal of prevalence literature have been developed by Loney et al.10 Table 2 describes the criteria needed to critically assess articles determining prevalence. The reader is referred to Loney et al10 for further elaboration and clarification of the criteria. Articles that have sound methodological rigor will provide more precise estimates of prevalence, with minimal bias.
Accurate determination of the prevalence of LBP is important to physical therapists and other health care professionals for a number of reasons. First, once the true prevalence rate is known, the societal impact of LBP in terms of cost and disability can be assessed accurately. Second, knowledge of the prevalence rate would aid in the organization and prioritization of information for health service planning and development. Currently, numerous practitioners are diagnosing and providing varying types of care for patients with LBP.7 Some interventions being used have not been shown to be effective.11 Third, with accurate prevalence estimates, increased funding for research aimed at improving the diagnosis and treatment of this condition may be possible. Finally, physical therapy educators might think of expanding course curricula to provide more study time for the most prevalent conditions seen by the physical therapy profession.12 If LBP is more prevalent than other conditions seen by physical therapists, then the management of people with back pain may need to be re-evaluated.
The purposes of this review are (1) to apply the afore-mentioned methodological guidelines to the literature on community prevalence of LBP and assign an overall methodological score, (2) to compare studies determined to be methodologically acceptable in an attempt to draw conclusions about the prevalence of LBP in the world population, (3) to estimate of the point prevalence of LBP in North America, and (4) to make suggestions for improving the methodological quality of these types of studies.
Some authors13 have suggested that several overlapping search strategies should be used to ensure that as many as possible of the available articles are included in a literature review. A search of MEDLINE, Cumulative Index to Nursing and Allied Health (CINAHL), and Science Citation CD-ROM systems from 1981 to 1998 was conducted for this review. The Medical Subject Headings (MeSH) headings “low back pain” (focused to epidemiology) and “prevalence” were used. The search was limited to studies printed in English and to the study of adults (ages 19–64 years). Only community-based prevalence studies were used for this review. Studies of groups in the community (eg, health care workers, industrial workers, military) were excluded.
The search strategy for this review was as follows:
Step 1. The reference lists of articles identified by MEDLINE (1981–present) were searched for other relevant publications. We refer to this process as a hand search. MEDLINE citations and hand-searched articles were grouped together and called the MEDLINE search strategy.
Step 2. CINAHL CD-ROM (1981–present) was searched. Selected articles were hand searched for relevant publications. Relevant citations and hand-searched articles were grouped together and called the CINAHL search strategy.
Step 3. Science Citation CD-ROM (1981–present) was searched for appropriate articles.
Estimating the Completeness of the Search
Capture-recapture methods are derived from the technique used in ecological studies in which animals are captured, marked, released, and recaptured for population estimates.14 Recently, these methods have been applied in epidemiology to estimate the degree of overlap between 2 or more search strategies. Simple formulas are used to obtain an estimate of the total size of the population, also known as the horizon,14 or, in our search of the literature, the total number of articles available. This procedure enables a researcher to estimate the number of publications that were not identified by the search strategy in order to evaluate the completeness of a systematic literature search.
The first step of our process was to define the search strategy and the steps for obtaining articles for the review. We did this to ensure the reproducibility of the search, which will allow other researchers to judge the validity of our strategy. Spoor et al14 used articles in a journal found by MEDLINE versus an independent hand search of the same journal. The search strategy for our review is defined above. In this case, the hand-search component was combined with the associated CD-ROM search because articles obtained by hand searching were found from references obtained by the CD-ROM search. Next, we constructed a contingency table to show the number of articles identified by each search strategy and the amount of overlap between strategies. Only 2 search strategies (MEDLINE and CINAHL) were appropriate, because no articles were found in Science Citation. Thus, a 2 × 2 table was constructed (Tab. 3). The contingency table shows the extent of overlap in the number of articles found and not found by the MEDLINE and CINAHL search strategies. In this table, the cell of “not found” by both search strategies (X) is unknown, and it is this cell that will ultimately be estimated.
It is beyond the scope of this article to provide the derivation of the formulas used in the capture-recapture calculation, and the reader is referred to Bishop et al.15 Using the formulas provided, the total size of the population of articles (N) can be estimated.15 Once N is known, X can be calculated using simple algebra. Finally, 95% confidence intervals can be obtained for the point estimate. All calculations are provided in Table 3.
Critical Appraisal and Scoring
The primary author appraised each article using a worksheet that summarizes the criteria outlined by Loney et al10 (see Appendix for a sample of the worksheet used). A scoring system was developed to rate the quality of the studies reviewed (Tab. 4). We agreed that each item would be weighted equally because we did not believe that one item was more important than the other items. Each item was given a score of 10 points, with the maximum score being 90 points. The weighting of items equally has support in the literature. Streiner and Norman16 explained that unequal weighting of items in a scale consisting of relatively homogeneous items contributes relatively little, except added complexity. Dixon et al17 used equal weighting in their rating scales for the critical appraisal of articles on clinical agreement, diagnosis, causation, therapy, prognosis, and overview.
A sample size of 300 or more has been suggested as adequate for population surveys examining dementia.18 The prevalence of LBP, however, is likely higher than that of dementia. According to Kachigan,19 the sample size required to estimate a proportion with a specified degree of precision can be determined using simple formulas. Proportions nearer 50% require larger sample sizes than smaller proportions.19 Using a conservative sample size estimation for proportions (prevalence=50%), with an error in estimate of less than 3% at the 95% confidence level, the calculated sample size is 1,067.19 Thus, we considered a sample size of 1,000 to be adequate for the purposes of this review. A response rate in population surveys of two thirds to three quarters has been suggested to be generalizable to the population samples.20 Thus, a response rate of 70% was chosen as acceptable. A priori, a total methodological score of 70 points was deemed acceptable. This cutoff was chosen because we believed that these studies would be methodologically sound to enable generalization and to provide a basis for discussion and conclusions.
Specification of a Comparison Measure
Point prevalence was chosen as the primary epidemiologic comparison measure (ie, subjects' report of LBP at the time of the survey). Although other measures of prevalence are provided in many studies (ie, period and lifetime prevalence), point prevalence has the advantage of not being based on recollection. “Memory decay” is a phenomenon associated with increasingly forgetting events (eg, past episodes of LBP) with the passage of time.9 The longer time has passed, the more likely a person is to forget the event. This would tend to cause an underestimate of the reporting of LBP event, thus lowering the prevalence rate. Recollection bias was supported in a study by Carey et al.21 In contrast, the phenomenon of “forward telescoping” (tendency to recollect events as occurring more recently then they actually did) tends to have the opposite effect on the prevalence rate.9 Forward telescoping tends to increase the reporting of LBP in a specified time period, thus overestimating the prevalence rate. Carey et al21 stated that these recall biases may offset each other. The use of point prevalence, however, is supposed to eliminate such assumptions and thus was used as the main comparison in our review of the literature, although other prevalence rates will be provided and compared.
Eighteen studies were reviewed from the years 1981 to 1998. Twelve of these studies were conducted in Europe, 5 studies were conducted in North America, and 1 study was conducted in China. All studies dealt with the prevalence of LBP in adults. The studies yielded different types of prevalence estimates (ie, point, period, lifetime) and included different subgroups of the community. The researchers used different durations of LBP in their estimates. Generally, the definition for LBP was consistent. Some authors, however, did not define LBP. Many researchers used a body diagram for clarification of the definition of LBP.
Completeness of Search Strategy
Using the capture-mark-recapture technique, the estimate of the total population size of the available articles was 18 (95% confidence interval=16,20), rounded to the nearest whole number (Tab. 3). The number of articles missed by this search strategy was estimated to be 0. Thus, the overall search strategy can be considered complete.
Table 5 presents the results of the critical appraisal of articles reviewed and the overall methodological quality scores calculated.7,22–38 Three studies22,26,29 were given scores of 70 points or more and were considered of high quality. Studies given scores of 45 points or more were considered to be of moderate quality and were selected for further comparison to determine trends in methodology and prevalence rates. Van Tulder et al39 used a cutoff of one half of the total methodological score in their critical review of the effectiveness of conservative treatment for LBP. Thirteen studies were given a score of 45 points or more, and we deemed these studies to be methodologically acceptable. The most common methodological problems identified in this group of studies were the following: (1) failure to provide validity and reliability data on the survey questions that participants were asked, (2) lack of precise estimates (95% confidence intervals) of the prevalence rates provided, and (3) no comparison of study participants and nonparticipants to determine the generalizability of the sample obtained.
Figure 2 shows the point prevalence of LBP from studies that were identified as methodologically acceptable (ie, scored ≥45 points) and provided point prevalence estimates. The 95% confidence intervals are shown for the studies by Cassidy et al26 and Hillman et al.29 Confidence intervals were not provided by other authors and thus are not shown. Prevalence estimates are further grouped by duration of LBP for comparison. The duration of LBP was not provided in the study by Skovron et al35 and is marked as unknown. Prevalence estimates varied from study to study. Generally, studies in which LBP was examined on the survey day have higher prevalence rates than studies dealing with the seriousness of LBP or LBP lasting greater than 2 weeks. The point prevalence rates in this group of studies varied-from 4.4% to 33.0% (X=19.2%, SD=9.6%). Figure 3 shows the 1-year prevalence rates summarized by study. Again, a range of rates is presented (3.9%–63%). In these methodologically acceptable studies, the mean was 32.37% (SD=23.6%). The trend of higher rates for shorter duration of LBP examined is seen in both period and point prevalence estimates.
Point prevalence estimates by age, when available, from the studies deemed methodologically acceptable are presented in Figure 4. The results suggest that younger subjects tend to have lower overall prevalence rates compared with older subjects up to the age of 60 years. Beyond the age of 60 years, it appears that prevalence rates decline.7,28
A large variation in community prevalence rates was observed among methodologically acceptable studies. The duration of LBP applied to define a prevalent case varied among studies, and this feature appeared to explain much of the variation in the reported prevalence rates. For example, if a person had an episode of LBP lasting only several days, this would not count as a prevalent case of LBP in a study investigating the prevalence of LBP lasting at least 2 weeks. There is also evidence that qualifying statements about duration may improve the accuracy of reporting.40 Long-term memory is related to the duration of a painful experience and the frequency with which it is reoccurring.40 In our opinion, therefore, definitions of the duration and severity of LBP need to be standardized for adequate comparisons among studies.
A second reason for the range of prevalence rates that we observed is the varying methodological quality among studies, as indicated by the scores obtained. The cutoff of 45 points or more as methodologically acceptable seems generous to us. The point prevalence rates in the 3 studies22,26,29 that were given very high quality scores (≥70) were 13.7%, 28.7%, and 19%, respectively. One-year prevalence rates were similar in the studies of Biering-Sorensen22 (44.9%) and Hillman et al29 (39%). Cassidy et al26 reported a 6-month prevalence rate of 68.8%. In these studies, similar definitions were used for duration of LBP. We feel more confident in making conclusions about LBP prevalence from these studies because of their methodological strength.
The disparity in LBP prevalence rates among studies of similar methodological rigor, although smaller in methodologically superior studies, leads us to consider factors other than methodological issues to help explain differences. Age appears to be one factor influencing the prevalence rates in the studies reviewed. The methodologically acceptable studies that provided subgroup estimates by age demonstrated varying LBP point prevalence with age. The younger patients (aged 20–35 years) had lower prevalence rates in all studies reviewed. Prevalence estimates then increased in the middle years (ages 40–60 years), with the exception of the group surveyed by Biering-Sorensen.22 After 60 years of age, the point prevalence rates tended to decline again. Deyo and Tsui-Wu7 stated that there may be a cohort effect, such that older people simply experience less back pain than younger people do. Selective mortality and poorer recall also may explain the fall in prevalence rates among people over 60 years of age.7 The overall prevalence rates observed in studies7,33 that included younger subjects were lower than those of studies that examined only subjects in their middle years.
A major methodological challenge in determining the prevalence of LBP in the general population is that there is no standardized tool for diagnosis. Radiographs, computed tomography scans, and magnetic resonance imaging have aided with diagnosis, but they do not always indicate the cause. Thus, we believe there is a need to rely on other indexes of disease such as patient reports of LBP.35 Reports of pain, whether by phone or by interview, can lead to inaccurate results. Biering-Sorensen and Hilden40 reported that subjects responding to a mailed questionnaire consistently answered questions of ever having LBP 84% of the time on repeated occasions. Walsh and Coggon41 also examined the repeatability of a self-administered questionnaire and found agreement of 89% among reports of LBP. The associated error is likely due to recall bias. Carey et al21 suggested that false reporting was minimal (3%). Thus, until a standardized diagnostic tool is developed, substantial measurement error in prevalence rates is likely and rates will vary among studies using different durations of pain.
Other methodological challenges include using survey questions that have been shown to have sound measurement properties. Many authors of the studies we reviewed did not provide reliability and validity data for the questions used. Leboeuf-Yde et al32 tested the LBP questions in their study and provided references of reliability and validity. Cassidy et al26 used a questionnaire with what we consider acceptable reliability and validity. Choosing the method of questionnaire administration also will influence outcome.40 Telephone interviews and self-administered questionnaires are more feasible than personal interviews, but they may not be the most accurate.16 Finally, obtaining high response rates in population surveys can be a challenge. We contend that it is imperative that comparisons of respondents and nonrespondents be provided so that the generalizability of the study sample can be determined.
Future research on the community prevalence of LBP is needed before an accurate assessment of the societal impact of LBP on society with respect to disability and cost can be determined. Improved methodological quality and homogeneity among researchers are needed in all areas identified above. Generalizations to the community can be made with confidence from the 3 studies that we deemed to be methodologically superior. The point prevalence rate was estimated to be 13.7% in Glostrup, Denmark22; 28.7% in Saskatchewan, Canada26; and 19% in Bradford, United Kingdom.29 The prevalence of LBP in North America was estimated by 3 studies reviewed.7,26,33 The mean point prevalence rate estimated by Deyo and Tsui-Wu7 and Lee et al33 was 5.6%. This estimate is considerably lower than the estimate by Cassidy et al26 (28.7%), likely because of the definition of duration of LBP used in each study. Cassidy et al26 investigated LBP on the day of the survey, whereas Deyo and Tsui-Wu7 and Lee et al33 examined the seriousness of LBP and LBP of greater than 2 weeks' duration, respectively. Using the estimate of 5.6%, the population of North American adults is roughly 178 million.42,43 Thus, roughly 10 million people are experiencing LBP on any given day. Many of these individuals will need medical care. Thus, it appears justified to contend that further research on the models of care and the effectiveness of treatments for LBP is needed and that more accurate prevalence estimates would aid in that research.
Concept and research design, data collection, and project management were provided by Loney and Stratford; writing and data analysis, by Loney; institutional liaisons and manuscript review, by Stratford.
- Received June 15, 1998.
- Accepted January 13, 1999.
- Physical Therapy