This cross-sectional descriptive epidemiological study was approved by our institutional review board. We utilized cases of shoulder dislocation in the Consumer Product Safety Commission's (CPSC) National Electronic Injury Surveillance System (NEISS) database. A full description of the sample, design, and utilization of this complex probability sample of all injuries presenting to emergency departments in the United States has been published on the CPSC electronic web page7-9. The NEISS sample included 100 hospitals, which were originally designated by stratified, randomized sampling of all United States hospitals with emergency departments. Stratification was based on both geographic location and emergency department volume data. Data were gathered on all injuries presenting to the 100-hospital probability sample, and each injury was assigned a weighted estimate. In order to create sample weights, participating hospitals were classified into five strata: one representing children's hospital emergency departments and four representing hospital emergency departments of various sizes. The CPSC conducted yearly sampling frames of all active United States emergency departments, which included information regarding total emergency department visits. With use of these data, adjustments were made to the sampling frame to ensure that hospitals conformed to required specifications, and the sampling frame was utilized to ratio-adjust the statistical sample weights to account for changes in strata emergency department visits. Variables included in the standard NEISS case record were treatment date, age, sex, race, diagnostic category, body part injured, patient disposition, location where the injury was sustained, and two descriptive narrative fields. Data were input on a daily basis, and any missing information was obtained by telephone interviews with patients within the first week after the injury whenever possible. The NEISS has served as a reliable and reproducible source for a wide range of epidemiological subjects, ranging from intentional self-poisonings to injuries resulting from rugby and from martial arts in children10-15.
In the current study, the NEISS database was queried in one-year intervals for all injuries between January 1, 2002, and December 31, 2006, classified as dislocation-type injuries of the shoulder region. This query yielded 10,701 records. Query results were pooled and analyzed for any redundancy with use of the unique case identifier (CPSC case number). Narrative descriptions of each case were individually analyzed to further characterize cases with regard to variables not included in the standard NEISS matrix. These additional variables included characteristics such as laterality, mechanism of injury, and activity at the time of the injury. We excluded from the final database 1761 cases with descriptive diagnoses listing other conditions. These included 1694 acromioclavicular joint injuries; twenty-eight proximal humeral fracture-dislocations; eighteen shoulder sprains, rotator cuff tears, or contusions; sixteen sternoclavicular joint injuries; and five fractures. Records in which shoulder dislocation injuries were described were retained in the corrected sample for further analysis. Following these refinements, the retained database included 8940 shoulder dislocation injuries in the NEISS sample population from 2002 through 2006.
The initial analysis focused on overall demographics of patients with shoulder dislocation. The NEISS database query resulted in sample records from the 100 hospitals included in the model. The numbers of sample records are denoted as n values in this report. With use of the complex NEISS statistical model, SAS statistical software version 9.1.3 (SAS Institute, Cary, North Carolina), and specific SAS programming code provided by the CPSC for the NEISS model, sample-record instability events (n) were converted to probability model estimates of all United States cases (N) with 95% confidence intervals. Gross sample data (denoted as n) are provided in this report for reference purposes. Consistent with model design, proportions were calculated on the basis of weighted United States estimates (N). Weighted proportions were therefore utilized to analyze the proportional demographic data of the complete NEISS sample population as well as those of variable-delimited subgroups (i.e., age, sex, and race) with respect to dislocation patterns (i.e., laterality, direction, and recurrent injury), location at which the injury event occurred, patient disposition, mechanism of injury, and participation in sports or recreational activity at the time of injury. Statistical analyses including the chi-square and Wald chi-square tests were performed to identify significant differences between subgroups. Additionally, United States Census Bureau population estimates were utilized to calculate at-risk person-years for both the United States population and the variable-delimited subgroups (age, race, and sex) throughout the defined time period, allowing the calculation of incidence rates per 100,000 person-years and incidence rate ratios with 95% confidence intervals. As a result of the off-cycle nature of United States population estimates (index date, July 1) relative to the NEISS sample (data range, January 1 through December 31 of a given year), at-risk person-years were calculated by including the full population estimates for each year from 2003 to 2006 and half the population estimate for 2002 and 2007. In all statistical analyses, a p value of <0.05 was considered significant.
Source of Funding
This study received no external funding support.
A total of 8940 shoulder dislocations were recorded in the NEISS sample population during the five-year period from 2002 through 2006. This represented an estimated 349,486 dislocations in the United States population (a rate of 69,897 per year). These dislocations represent an estimated incidence rate of 23.9 (95% confidence interval, 20.8 to 27.0) per 100,000 person-years.
Dislocation Characteristics
The direction was recorded for only 418 (4.7%) of the dislocations. Laterality was known for half of the sample (n = 4606, N = 187,292), and the injuries occurred with similar frequencies in right shoulders (51.4%, n = 2347, N = 96,319) and left shoulders (48.4%, n = 2250, N = 90,577). Simultaneous bilateral dislocation was identified in nine cases (N = 396). The status of a dislocation with regard to whether it was a recurrence was not required to be reported for the injuries in the NEISS record and therefore recurrence was probably underreported. Described recurrent dislocations accounted for 2.1% of all cases (n = 183, N = 7352).
Injury Event Characteristics
The place where the injury was sustained was documented in 78.1% of the records (n = 6984, N = 278,303), and homes (47.7%, n = 3371, N = 132,670) and places of sports or recreation (34.5%, n = 2350, N = 96,024) were the most common sites. A disproportionate percentage of the injuries in males occurred at a place of sports or recreation (86.7%, n = 2040, N = 83,265), and a disproportionate percentage of those in females occurred at home (42.5%, n = 1435, N = 56,877). There was a sufficient narrative description of the mechanism of injury in 6881 records (77.0%, N = 269,290). Falls (58.8%, n = 4047, N = 158,461) and direct blows (8.9%, n = 626, N = 23,963) were the most common mechanisms. Sports or recreation-related injuries accounted for 89.8% (n = 1855, N = 71,985) of the 2059 dislocations (N = 80,195) with an unknown mechanism. Overall, sports or recreation-related injuries accounted for nearly half of all dislocations (48.3%, n = 4303, N = 168,730). Males had a significantly higher proportion of sports or recreation-related dislocations (p < 0.001) than females, and the younger age groups had a significantly higher proportion than the older age groups (p < 0.001). Football and basketball each accounted for more than twice as many sports or recreation-related injuries as any other sports or recreational activity (17.3%, n = 814, N = 29,238, and 16.7%, n = 780, N = 28,167, respectively), and the two sports combined accounted for 34% of all sports or recreation-related injuries.
Sex
The majority of the dislocations occurred in males (71.8%, n = 6400, N = 250,861 compared with 28.2%, n = 2537, N = 98,472 in females). The overall male incidence rate was 34.9 (95% confidence interval, 30.1 to 39.7), while the female incidence rate was 13.3 (95% confidence interval, 11.6 to 15.0). The incidence rate ratio for males was 2.64 (95% confidence interval, 2.4 to 2.9), with female sex as the referent category (Table I).
Age
The mean age at presentation was 35.4 years, with a mode of seventeen years. It was found that 19.4% (n = 1835, N = 67,630) of all dislocations occurred in individuals between the ages of fifteen and nineteen years, and this age group had the largest proportion of dislocations of any five-year age group (Fig. 1). When segregated by decade, the group between the ages of twenty and twenty-nine years had the largest proportion of instability injuries (27.5%, n = 2405, N = 96,001). Therefore, 46.8% (n = 4240, N = 163,631) of the dislocations occurred in individuals between fifteen and twenty-nine years of age. When incidence rates for all individuals were calculated according to age, the results revealed a bimodal distribution, peaking in the third decade of life (twenty to twenty-nine years old) and the ninth decade of life (eighty to eighty-nine years old) and corresponding to the peak male incidence (79.2 [95% confidence interval, 67.4 to 90.9]) and the peak female incidence (38.8 [95% confidence interval, 30.8 to 46.7]), respectively (Fig. 2). Male incidence rates were significantly greater (p < 0.05) in those between the ages of zero and thirty-nine years, while trends of higher female incidence rates in the sixty to ninety-nine-year age group never reached significance. The overall incidence rates were lowest in the zero-to-nine and fifty to fifty-nine-year age groups (0.92 [95% confidence interval, 0.56 to 1.3] and 12.9 [95% confidence interval, 10.5 to 15.3], respectively). There was a peak age-related incidence rate ratio of 3.70 (95% confidence interval, 3.15 to 4.25) in individuals between twenty and twenty-nine years old, with the age group of fifty to fifty-nine years old as the referent category (Table I).
Race
Race was characterized in 6566 records (N = 253,429), which showed that 73.9% of the patients were white (n = 4551, N = 187,189), 17.5% were black (n = 1401, N = 44,294), 5.4% were Hispanic (n = 379, N = 13,657), 2.0% were Asian (n = 133, N = 4,952), 0.8% were Native American (n = 34, N = 2085), 0.06% were Indian (n = 6, N = 142), and 0.4% were other not specified (n = 62, N = 1110). Incidence according to race was then examined in the following subgroups: white, black, and other (consisting of Hispanic, Asian, Native American, and Indian individuals). Cases with undefined race (n = 2374) were excluded from this analysis; therefore, the incidence rates according to the defined racial subgroups likely underestimate actual values. Statistical analysis failed to identify any significant differences in incidence between black and white subgroups (incidence rate ratio = 1.25 [95% confidence interval, 0.66 to 1.85], with white as the referent). The subgroup "other" was at significantly less risk than whites (incidence rate ratio = 0.42 [95% confidence interval, 0.22 to 0.61], with white as the referent).
The purpose of this study was to determine the incidence rate of shoulder dislocation presenting to emergency departments in the United States as well as to elucidate the demographic risk factors for this injury. To our knowledge, this study represents the largest population sample of shoulder dislocations in the literature. The incidence of shoulder dislocation in the United States population was 23.9 (95% confidence interval, 20.8 to 27.0) per 100,000 person-years, which matches the incidence rate of traumatic dislocations of 23.9 described by Nordqvist and Petersson in Sweden5 but is higher than the incidence rate of 17 reported by Krøner et al. in Denmark4 and is more than double the United States incidence rate of 11.2 previously reported by Simonet et al.6 (Table II). While Nordqvist and Petersson did not differentiate between dislocations and fracture-dislocations, we found only twenty-eight fracture-dislocations (through analysis of the narrative descriptions), and these were excluded from our study. The substantial difference in the incidence rate between the current study and the study by Simonet et al.6 warrants discussion. We postulate a number of explanations for this discrepancy. The prior study excluded posterior, idiopathic, and voluntary dislocations as well as recurrent dislocations in patients in whom the first dislocation occurred prior to 1970 (the first year of the study period). In addition, the makeup of Olmsted County, Minnesota, in the 1970s was largely rural and may not have been an accurate model for the United States population as a whole. Lastly, and perhaps most importantly, demographics and lifestyles in the United States have certainly changed since the 1970s and now may differ even more substantially from those of the largely rural Olmsted County population.
The significance of an incidence rate of 23.9 per 100,000 person-years is debatable. Although it is more than double the previously reported rate for the United States population, it still lags behind rates of other common musculoskeletal injuries presenting to emergency departments. Unpublished data from the NEISS database showed rates of distal radial fractures of sixty-two per 100,000 person-years and rates of ankle sprains of 215 per 100,000 person-years. While shoulder dislocation has a lower incidence than these other injuries, it has been shown to have a high incidence in military16 and athletic populations17. We believe that our research has confirmed that there are similar trends of risk in the United States population as a whole, with young male subpopulations being at higher risk and with the majority of their injuries occurring during sports activities. We have therefore confirmed the utility of targeting this group in future studies, so that effective preventive strategies can be developed and tested.
The current study identified a significantly higher incidence rate and incidence rate ratio for male sex relative to female sex (Fig. 2, Table I). While the finding of a higher incidence in males was reported by Simonet et al.6 and Nordqvist and Petersson5, the magnitude of this difference was greatest in the current study (Table II). Krøner et al.4 also demonstrated a higher incidence rate in males, although the rate was not significantly higher. The current study documents a sex-related incidence rate ratio, with males found to be more than 2.5 times more likely to sustain a glenohumeral dislocation. While this has been shown in selected populations such as the United States military16, to our knowledge this is the first general population study to demonstrate this finding. That we were unable to definitively differentiate between traumatic and atraumatic injuries may also help to explain the differences noted between the sexes.
Age-related differences were also demonstrated in our population. The overall distribution of dislocations based on raw numbers confirms that the vast majority of shoulder dislocations occur in young people, a finding that parallels clinical experience. The mean age of the patients with a dislocation in our study was similar to that in the study by Simonet et al.6 and lower than that in the studies from the European countries. When converted to an incidence rate, dislocations demonstrated a bimodal distribution, with peaks in young adulthood and in the elderly, which corresponded to the peak male and female incidence rates (Fig. 2). While the bimodal distribution may seem unexpected, similar findings were reported by Simonet et al.6 and Krøner et al.4 (Table II). A majority of the dislocations in young males were related to sports activities and recreational activities at sporting grounds, while the largest portion of injuries in elderly women were sustained during falls at home. The latter situation likely reflects injury events similar to those of patients sustaining proximal humeral fractures (which were excluded from our study). Of note, the peak male and female incidence rates in this study were both higher than any previously reported in the general population4-6. Analysis of age-delimited incidence rate ratios demonstrated a peak value of 3.70 in patients between the ages of twenty and twenty-nine years, illustrating the high risk of glenohumeral dislocations in the third decade of life. The correlative incidence rate for this age group in the current study was 47.8 (95% confidence interval, 41.0 to 54.5). It is clear that young, active males are an ideal population in which to study traumatic shoulder dislocations and should be the focus of future prevention strategies. This finding also helps in the understanding of the endemic rates recently reported in military and athletic populations. These include incidence rates of 169 per 100,000 person-years in the general United States military population16 and 435 per 100,000 person-years in military academy cadets18.
A previous large cohort study of the United States military population showed that both white and "other"-race service members had higher rates of injury than black service members16. In the current study, we did not find significant differences between white and black races, but we did find that the "other" group had significantly fewer dislocations than whites. However, the lack of a clear racial identity in the "other" group makes it difficult to derive conclusions from these data. It should also be noted that race data were recorded for only 73% of our cases.
The current study had numerous strengths relative to the other studies discussed. While the studies by Krøner et al.4 and Nordqvist and Petersson5 have contributed substantially to the understanding of glenohumeral dislocations, they were based on urban European populations, which in many ways differ from the United States population, and their conclusions may not be generalizable to the United States population for that reason. While Simonet et al.6 reported the first in-depth study of the epidemiology of shoulder dislocations, variables such as occupation and levels of recreational activities have likely changed since the evaluation of that cohort from the 1970s. The NEISS sample used in the current study is based on a probability sample of the entire United States population within a modern sampling frame (2002 through 2006) and therefore represents the nation's current population. The utility of this method depends on the validity of the model. The NEISS database has been used as a tool for estimating United States injury data in multiple studies10-15. In addition, the model draws from hospitals from five different tier categories, providing data that are representative of the cumulative experience of numerous levels of health care.
While the model provides a number of advantages, it also has inherent weaknesses. The final data are only as accurate as the initial raw sample. Sampling bias is expected and accounted for in the design of the model. However, as the NEISS model includes only patients presenting to emergency departments, this study excluded patients who presented only to other facets of the health-care system (e.g., primary-care offices) and those who failed to seek formal medical care. A second issue was the basic NEISS coding for "shoulder" and "dislocation," which potentially allowed inclusion of related diagnoses (such as acromioclavicular and sternoclavicular dislocations) in the initial sample data. An attempt was made to minimize this effect by performing a line-by-line analysis and excluding erroneously coded injuries. The third, and possibly greatest, limitation of our data is the lack of determination of whether a dislocation was a recurrent injury. The presence of recurrences within our data set may have erroneously increased the incidence rate. Finally, databases are limited by their initial formatting. While standard coding protocols provided nearly complete data with regard to age, sex, location where the injury was sustained, and patient disposition, the NEISS system for coding of race was less precise. Narrative sections could be utilized to characterize other nonstandard variables such as laterality, recurrence, mechanism of injury, and participation in sports or recreational activities, but the resulting data for variables not explicitly included in the standard NEISS record were inherently prone to reporting bias, selection bias (both by treating physicians and coders), and categorization bias based on our interpretations. Results in these categories should therefore be interpreted with these limitations in mind.
The purpose of this study was to determine the epidemiology of traumatic shoulder dislocations. Given this goal, the use of an emergency department database was appropriate. However, the use of this database did not allow us to definitively exclude atraumatic shoulder dislocations, as this determination and mechanistic data were not part of the required data entry. While the nature of our database suggests that our data would consist primarily of traumatic injuries, this must be considered a limitation of our study.
In the current study, we determined the incidence rate of shoulder dislocations presenting to emergency departments within the United States to be 23.9 per 100,000 person-years. Additionally, we identified an age between twenty and twenty-nine years and male sex as significant demographic risk factors. Nearly half of all injuries occurred during sports or recreational activities. Future evaluation of prevention strategies and treatment options focused on patients with these risk factors are likely to have the largest impact on the burden that glenohumeral dislocations and their resultant morbidity present to the United States health-care system.