Estimates from 2003 place the number of leisure skiers in the United States at 13 million, with about 57 million visits to ski areas annually.1 Of the 600,000 ski and snowboard-related injuries each year, an estimated 15% to 20% are head injuries.2–5 Traumatic brain injury is the major cause of mortality in snow sports.5
Helmets have been reported to reduce the risk of head injury among bicyclists6,7 and motorcyclists.8,9 We conducted a case-control study to measure the association between helmet use and the occurrence of head injury among skiers and snowboarders who fell in the western United States. Our study was conducted concurrently with Canadian10 and Norwegian11 studies examining helmet use in relation to head injuries among skiers and snowboarders. Although both of those studies reported a protective association between helmet-wearing and head injury among skiers and snowboarders who were in falls or collisions, the adjusted odds ratios (ORs) (0.72 [95% confidence interval (CI) = 0.55–0.92] and 0.45 [0.34–0.59], respectively) differed substantially, and our study provides additional estimates. We estimated the association between helmet use and injuries to the head, face or neck.
METHODS
Study Design
We used a case-control design in which the source population was persons involved in a fall or collision. Cases were persons reported by the ski patrol to have an injury to the head, face, or neck; most skiers with head, face, or neck injuries of any importance are seen by the ski patrol. We selected controls from among all skiers and boarders who fell, to estimate the prevalence of helmet use and other characteristics. To do this, we again used ski patrol reports to identify persons who fell and had injuries only below the neck. This group approximates the helmet-use prevalence of all persons who fell, provided that 2 assumptions are true: (1) helmeted persons are no more or less likely to see the ski patrol, aside from any effect helmets had on the risk of injury in a fall or collision, and (2) helmets do not cause or prevent injuries to body regions below the neck, given that a fall occurred.12,13 Our first assumption was tested in a sensitivity analysis (see below). We further assume that the proportion of skiers who were seen by the ski patrol was small relative to the source population of all skiers who fell; thus, odds ratios from logistic regression should approximate risk ratios for the associations of interest.
Data Source
Approval was received from the Institutional Review Board of the University of Washington before conducting this study. At ski resorts, clients usually seek care from the ski patrol for injuries. We obtained data from all ski patrol reports at 3 ski resorts in 2 western states. These included a large destination resort and 2 smaller resorts with larger proportions of local skiers. All 3 areas allowed snowboarding. The 2 smaller resorts provided paper ski patrol reports, which we electronically scanned or photocopied on-site after masking identifying information. Relevant information was subsequently key-entered. At the third resort, information from reports had already been key-entered, and we used these electronic data without identifiers.
We included only reports of injuries occurring as a result of a fall or collision while skiing (any type of ski) or snowboarding. We excluded injuries from other events or illnesses. We were able to read text and checkbox portions of report forms for 2 resorts to determine whether reports met inclusion criteria. At one of these resorts, where the ski patrol also aided local motor vehicle crash victims, 3502 (73%) met inclusion criteria, and at the other, 4579 (95%) met inclusion criteria. At the third resort, electronic data were screened for injuries and events meeting inclusion criteria; 13,818 of 14,500 reports (95%) met inclusion criteria.
Case and Control Selection
Cases were persons who suffered any injuries above the shoulders in a collision or fall while skiing or snowboarding. Three case groups were defined: head injury (including the scalp or skull above the hairline, the ears, and injuries to the brain), facial injury (between the hairline and lower jaw, including eyes, mouth, teeth, and jaws), and neck injury. Individuals could have injuries to more than one anatomic area and thus could belong to more than one case group. Some cases also sustained injuries to other body parts below the neck.
Head injury was indicated by the presence of any of the following: a “yes” reported for “head,” “loss of consciousness,” or “concussion” checkbox variables, or the presence of any comments or variables indicating a probable head injury on the report form. Brain injury was defined as any indication of abnormal mental functioning. Examples of written comments indicating a brain injury include “coma,” “lost consciousness,” “blacked out,” loss of memory for the event, or an abnormal Glasgow Coma Score. In a few reports, the presence of a head injury was unclear; these were reviewed and discussed by the study team, and determined by consensus.
Controls were persons who were seen by the ski patrol for any injury below the neck (but not the neck or above) that occurred in a fall or collision while skiing or snowboarding.
Exposure Variable
The exposure of primary interest was helmet use, as indicated by the “helmet worn” item on the ski patrol report, or if indicated anywhere in the text of reports from the 2 resorts for which we had access to paper reports.
Potential Confounding Variables
We examined several variables for their potential influence on the association between helmet use and injury. Variables were treated as categorical indicator terms unless otherwise noted: age (quadratic splines with knots at 12, 24, and 39 years and with tails constrained to be linear14); sex; self-reported skiing ability (beginner, intermediate, expert); ski season (linear term for the 5 seasons from 2000–2001 through 2004–2005); ski resort; equipment type (snowboard versus other); equipment ownership (owned versus rented or borrowed); crash site trail rating (beginner, intermediate, expert, and terrain park); person in ski or snowboard school; a high fall (terrain park elevated jump, cliff, or other indication of likely greater height), medium fall (mogul or small jump), or fall on level surface; collision with a person; collision with an object; any collision; fell while getting on or off a lift; wind (calm, breeze, windy); and clear visibility.
Missing Data
Some information was missing for most variables (Table 1);the largest proportion of missing data was for trail difficulty rating (14%).
;)
TABLE 1: Characteristics of Cases and Controls Among Injured Skiers and Snowboarders Involved in Falls or Collisions
We first analyzed records with complete data to identify the most important confounding variables. Using these important confounding variables, we multiply imputed 20 sets of data that were identical in regard to known information, but could differ on imputed values for missing information. The imputation models also inlcuded case status (terms for all 3 case groups) and age, helmet use, self-reported ability, equipment type, ski resort, and interaction terms—between age and helmet, age and equipment type, and ability and equipment type. We used the method of chained equtions (regression switching), sampling imputed values from the posterior predictive distribution of the missing data.15–18
Analysis
We used odds ratios from logistic regression to estimate risk ratios for injury among helmeted persons compared with unhelmeted persons. To select from the 18 potential confounding variables, we entered each alone in the logistic model and estimated the percent change in the size of the odds ratio for helmet use: 100 × (initial OR − adjusted OR)/initial OR.19,20 We added the confounder that produced the greatest absolute change and repeated this process until no additional variable changed the OR by more than 1%. Using this method, we selected 6 variables as important confounders: high fall, equipment ownership, any collision, sex, season, and being in ski school. We checked whether transformations of age might exert more confounding influence than quadratic splines; the spline terms were sufficient.21
We then carried out the multiple imputation process. Using the imputed data, we repeated the confounder selection process and obtained the same results regarding confounder selection.
Odds ratios from each of the 20 imputed data sets were averaged on the log scale and we estimated confidence intervals that accounted for both the within-imputation and between-imputation variances.22 All ORs are based on the multiply-imputed data, except where specified. Odds ratios for the association of helmet use with head injury were adjusted for the 6 selected confounders, as well as age (spline terms), self-reported ability, equipment type, ski resort, and for interactions between age (classified as 0–12, 13–24, and 25 or more years) and equipment type, and interactions between ability and equipment type. Variations in ORs by sex, age, ski resort, ability, and equipment type were examined with interaction terms and homogeneity was tested with the likelihood ratio test.23 For the outcomes of facial and neck injury, we used the cases that had those injuries and the same control group as used for the head injury analysis to estimate ORs that were adjusted for the same confounders, except for the interaction terms (as they had negligible confounding influence).
Sensitivity Analysis
Our goal was to estimate the association of helmet use with head injury among all skiers or snowboarders in falls or collisions. However, rather than sampling all persons with falls or collisions as controls, we used persons involved in these events with injuries below the neck who were treated by the ski patrol, under the assumption that being seen by the ski patrol for these injuries would be unrelated to helmet use. To test our assumption, we re-estimated the OR for head injury using the same cases, but selecting as controls persons who were thought by the ski patrol to have fractures below the neck, under the assumption that most controls with fractures would have little choice regarding the use of ski patrol assistance.12,13
Assessment of Helmet Use and Head Injury Reporting
Study staff members went on-site to observe 54 skiers/snowboarders in the ski patrol huts, without interacting with the ski patrol or injured persons. There was complete agreement between the study staff assessment of head injury and head injury as recorded by the ski patrol; 8 head injuries were indicated by both assessments, 45 individuals were noted as not having a head injury by both assessments, and in one instance, both assessments indicated that the presence of a head injury was unclear (100% agreement, κ statistic = 1.00). There was one disagreement between study staff and recorded helmet-use information; 22 skiers used helmets and 31 skiers had no helmets as indicated by both assessments, and the ski patrol, but not the observer, indicated a helmet was worn for one skier (98% agreement, κ statistic = .96).
RESULTS
Description of Cases and Controls
There were 2537 cases with a head injury, 1122 with a face injury, and 565 with a neck injury. There was some overlap among groups: of the 3701 subjects who were cases, 2050 had head injuries only, 837 had face injuries only, 314 had neck injuries only, 249 had head and facial injuries without neck injury, 215 had head and neck injuries without face injury, 13 had face and neck injuries without head injury, and 23 had all 3 injury types.
Helmet use was similar for head-injured cases (21%) and controls (22%) (Table 1). Compared with controls, cases with head injury were more often teenagers and young adults (13–24 years), more often male, reported greater ability, were more likely to use snowboards, were more likely to own their equipment, and were more likely to have had a high fall or collision. Helmet use was more common among younger persons, males, experts, snowboarders, persons who owned their own equipment, and those involved in a high fall (Table 2).
TABLE 2: Selected Subject Characteristics by Helmet Use
Odds Ratio Estimates for Head Injury
The crude OR for head injury among helmeted compared with unhelmeted persons who fell or collided was 0.96 (95% CI = 0.87–1.07). The most influential confounding variables, based on our change-in-OR criteria, were a high fall, equipment ownership, any collision, and male sex; as these variables were added sequentially to the regression model, the adjusted OR became 0.91, 0.88, 0.86, and finally 0.84. Each of these variables was associated both with a greater risk of head injury (Table 1), and a greater prevalence of helmet use (Table 2). The fully adjusted odds ratio was 0.85 (0.76–0.95) (Table 3).The adjusted odds ratio based on the known, nonimputed data, was 0.82 (0.72–0.94). Recalculation of the adjusted OR using data from the 5721 controls with fractures and all head-injured cases did not change these results: OR = 0.85 (0.75–0.96).
TABLE 3: Adjusted Odds Ratios for Head Injury Comparing Skiers Wearing Helmets With Skiers not Wearing Helmets, by Selected Characteristics
Variation in Odds Ratios for Head Injury
The adjusted OR for the association of helmet use with head injury showed little variation by ski resort: 0.85 at resort 1, 0.78 at resort 2, and 0.88 at resort 3 (P = 0.61 for a test of homogeneity) (Table 3). The adjusted OR for this association were similar for alpine skiers (0.84) and snowboarders (0.85) (P = 0.83 for a test of homogeneity). There was some variation in the size of the ORs by reported ability: 0.69 for beginners, 0.86 for intermediates, and 0.92 for experts (P = 0.15 for a test of homogeneity). There was also some variation in the ORs for head injury among males (0.80) and females (0.98, P = 0.09 for a test of homogeneity). The OR for head injury for helmeted compared with unhelmeted persons varied with age: 0.60 for age 1 to 12 years, 0.80 for age 13 to 24 years, and 1.13 for age 25 years and older (P < 0.001 for a test of homogeneity).
Odds Ratio Estimates for Facial Injury and for Neck Injury
The adjusted OR estimate for facial injury among helmeted subjects compared with those who were not helmeted was 0.93 (CI = 0.79–1.09). The adjusted OR for neck injury among helmeted subjects compared with persons not helmeted was 0.91 (0.72–1.14).
DISCUSSION
We estimated that skiers or snowboarders who crashed or fell had a 15% (95% CI = 5–24) reduction in the risk of any head injury if they were wearing a helmet, compared with similar skiers or snowboarders in similar events who were not wearing a helmet. An apparent protective effect of ski helmets has also been reported in 2 other studies,10,11 although the magnitude of the estimated effects varied.
The adjusted OR of 0.85 differed from the crude OR of 0.96. This difference was chiefly due to 4 variables: having a high fall, owning one's own equipment, having any type of collision, and being male. Cases were more likely to have these characteristics than controls. To compare helmeted persons with unhelmeted persons who were otherwise similar and in similar falls, we adjusted for these variables to remove bias due to confounding.
Our goal was to estimate the consequence of wearing helmets in the event of a fall or collision. We thought of our hypothetical source population as all skiers or snowboarders who fell or crashed.12,13 To sample controls from all subjects who fell, surely a large number, we used as controls skiers who fell and were seen by ski patrol, under the assumptions that (1) helmeted controls with injuries below the neck would use ski patrol as often as persons not wearing helmets and (2) in a fall, helmets would not influence the risk of injury to body regions below the neck. If these assumptions are not true, our results may be biased due to selection of controls who do not represent all eligible skiers who fell (Fig. 1). 24
FIGURE 1.:
Directed acyclic graph showing our study design in which all skiers fell or collided and selection of controls was conditional on their being seen by the ski patrol for injuries below the neck. If helmet-wearers with injuries were more or less likely to see the ski patrol or if helmet use increased or decreased the risk of injuries below the neck, given that a fall occurred, biased selection of controls could affect our estimates of any causal association between helmet use and head injury.
Limitations
Whether our design can be used for inference about the effects of helmet-wearing among all skiers depends on assumptions that we cannot test with our data. If causal relationships are correctly illustrated in Figure 2, then bias could arise if aggressive skiing style is not measured and controlled for in a study of all persons skiing. In our study, limited to those who fell, a fall may be thought of as a proxy for skiing style, and our design should remove this potential source of confounding. If causal relationships are as illustrated in Figure 3, our design might not produce valid estimates for all skiers, as our sampling scheme does not account for the possibility that helmet-wearing might either increase or decrease risky skiing behavior.
FIGURE 2.:
Directed acyclic graph showing relationships among helmet use, crashes, and injuries, under the assumption that helmet use does not directly cause or prevent crashes, but aggressive skiers may be more or less likely to use helmets than other skiers. In this scenario, helmet use may be related to head injury in a fall (path: helmet use → head injury) and through a backdoor pathway (helmet use → aggressive skiing → skier crash → head injury), and therefore aggressive skiing style may confound the association of helmet use with head injury. Limiting the study design to subjects who crashed would remove confounding of helmet effects by aggressive skiing style.
FIGURE 3.:
Directed acyclic graph showing the relationships among helmet use, crashes, and injuries, under the assumption that helmet use may cause skiers to be more or less aggressive than other skiers. In this scenario, helmet use might increase or decrease the risk of a head injury by influencing the risk of falling (path: helmet use → aggressive skiing → skier crash → head injury), and could also be related to the risk of head injury when a fall occurs (path: helmet use → head injury). A study set in the population of all persons who ski could estimate the overall influence of helmet use. Limiting the study to subjects who crashed could estimate only the influence of helmets on head injury risk in a crash.
Due to limitations in the ski patrol data and variation in how often the ski resorts used ambulance transfer or referral to a hospital, we could not estimate ORs for severe head injuries based on these variables. Missing data was a limitation; some data were missing for 12% of the records used in our final regression model for the head injury outcome, and 10% for the other 2 outcomes. We used multiple imputation methods, which require the less stringent assumption that data were missing at random, conditional on the values of variables used for imputation.25 However, bias could still be present if the pattern of missing values were strongly related to unmeasured characteristics of subjects or their crashes.
Our reliance on information reported in ski patrol reports is another limitation. Event descriptions were sometimes unclear, and we were unable to validate the presence of a head injury or helmet use. However, agreement was excellent between our own observations and ski patrol-reporting of these variables when compared for a separate series of injured skiers. In a Canadian study, ski patrol information on helmet use was compared with self-report based on follow-up phone interviews or mail questionnaires, with good agreement (κ = 0.88, 95% CI = 0.87–0.90).26 Finally, delayed head injuries diagnosed after skiers left the resort or other injuries not reported to the ski patrol were not included in our study.
Comparison With Previous Studies
Hagel et al10 conducted a case-control study at 19 ski areas in eastern Canada. As in our study, ski patrol reports were used to identify cases and controls, but the investigators also obtained information about several risk factors by mail or telephone interview. That study estimated that among persons who fell or collided, the OR for head injury among helmeted persons, compared with those not helmeted, was 0.71 (95% CI = 0.55–0.92). This estimate did not differ greatly from our estimate of 0.85 (P = 0.21 for a test that the 2 estimates came from similar populations).20
Sulheim et al11 used data from 8 ski resorts in Norway. Cases were identified by the ski patrol, but controls were selected from skiers and snowboarders waiting in lift lines. Comparing helmeted with unhelmeted persons, the OR for head injury was 0.40 (95% CI = 0.30–0.55). This OR, which was intended to apply to all people who ski, not just those in a fall or collision, differed from both our estimate and the estimate in the study by Hagel et al10 (P ≤ .005 for a test of similarity for either comparison). The choice of control group in the Norwegian study did not explain this difference, as the investigators reported that when they selected controls who fell from among injured persons seen by the ski patrol the OR estimate was 0.45 (95% CI = 0.34–0.59). It seems doubtful that uncontrolled confounding alone could explain the differences among the 3 studies: confounders used for adjustment overlapped considerably across the 3 studies and the confounder with the greatest influence in the Norwegian study (age) was used for adjustment in all the studies.
We found evidence that the OR for head injury given helmet use varied with age; helmets appeared to offer more protection for the youngest subjects and little or no protection for older persons. We are unsure about the validity of this finding as: (1) it is not apparent to us why this variation might arise and (2) the studies from Canada and Norway found no clear evidence that the ORs varied by age.10,11 This variation by age was present in the data from each of the 3 ski resorts, giving some credence to the finding.
Our study and the previous 2 studies10,11 all reported that the odds ratio for neck injury was less than 1 for helmet wearers compared with those who did not wear helmets. There is little statistical evidence of heterogeneity across the studies (P = 0.43 for a test that the estimates came from the same population) and a summary OR estimate (using inverse variance weights20) across all 3 studies was 0.85 (95% CI = 0.69–1.05) for neck injury among helmet wearers compared with similar persons who were not wearing helmets when they crashed.
If the results from our study and the others10,11 are causal, they suggest that skiers and snowboarders involved in falls or collisions may reduce their risk of a head injury if they wear helmets. Given the variation in the available estimates, however, the size of this possible protective effect is uncertain.
There is no evidence from our study, or from prior case control studies, that helmets increase the risk of neck injury. This is despite some concern that their use may actually increase the occurrence of spinal or neck injuries.27 To our knowledge, our study is the first to examine facial injuries in relation to ski helmet use; although we found evidence of a modest protective effect, the confidence intervals were wide.
ACKNOWLEDGMENTS
We thank Marni Levy and Jo Russell for their assistance with data collection, and the ski resorts for their help and cooperation.
REFERENCES
1. Lynch J. Snow Sports Industry Intelligence Report: State of the Industry. McLean, VA: Snow Sports Industries America, 2003.
2. Levy AS, Smith RH. Neurologic injuries in skiers and snowboarders.
Semin Neurol. 2000;20:233–245.
3. Chow TK, Corbett SW, Farstad DJ. Spectrum of injuries from snowboarding.
J Trauma. 1996;41:321–325.
4. Levy AS, Hawkes AP, Hemminger LM, et al. An analysis of head injuries among skiers and snowboarders.
J Trauma. 2002;53:695–704.
5. Hagel BE, Meeuwisse WH, Mohtadi NG, et al. Skiing and snowboarding injuries in the children and adolescents of Southern Alberta.
Clin J Sport Med. 1999;9:9–17.
6. Thompson RS, Rivara FP, Thompson DC. A case-control study of the effectiveness of bicycle safety helmets.
N Engl J Med. 1989;320:1361–1367.
7. Thompson DC, Rivara FP, Thompson RS. Effectiveness of bicycle safety helmets in preventing head injuries. A case-control study.
JAMA. 1996;276:1968–1973.
8. Evans L, Frick MC. Helmet effectiveness in preventing motorcycle driver and passenger fatalities.
Accid Anal Prev. 1988;20:447–458.
9. Norvell DC, Cummings P. Association of helmet use with death in motorcycle crashes: a matched-pair cohort study.
Am J Epidemiol. 2002;156:483–487.
10. Hagel BE, Pless IB, Goulet C, et al. Effectiveness of helmets in skiers and snowboarders: case-control and case crossover study.
BMJ. 2005;330:281.
11. Sulheim S, Holme I, Ekeland A, et al. Helmet use and risk of head injuries in alpine skiers and snowboarders.
JAMA. 2006;295:919–924.
12. Cummings P, Koepsell TD, Weiss NS. Studying injuries with case-control methods in the emergency department.
Ann Emerg Med. 1998;31:99–105.
13. Cummings P, Koepsell TD, Roberts I. Case-control studies in injury research. In: Rivara FP, Cummings P, Koepsell TD, Grossman DC, Maier RV, eds.
Injury Control: A Guide to Research and Program Evaluation. New York, NY: Cambridge University Press, 2001:139–156.
14. Greenland S. Dose-response and trend analysis in epidemiology: alternatives to categorical analysis.
Epidemiology. 1995;6:356–365.
15. van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis.
Stat Med. 1999;18:681–694.
16. Royston P. Multiple imputation of missing values.
Stata J. 2004;4:227–241.
17. Royston P. Multiple imputation of missing values: update.
Stata J. 2005;5:188–201.
18. Royston P. Multiple imputation of missing values: update of ice.
Stata J. 2005;5:527–536.
19. Mickey RM, Greenland S. The impact of confounder selection criteria on effect estimation.
Am J Epidemiol. 1989;129:125–137.
20. Rothman KJ, Greenland S.
Modern Epidemiology. Philadelphia: Lippincott-Raven, 1998;238:, 255–259, 275, 660.
21. Royston P, Altman DG. Regression using fractional polynomials of continuous covariates: parsimonious parametric modelling (with discussion).
Applied Stat. 1994;43:429–467.
22. Little RJA, Rubin DB.
Statistical Analysis With Missing Data. Hoboken, NJ: Wiley, 2002;86–87.
23. Hosmer DW, Lemeshow S.
Applied Logistic Regression. New York: Wiley, 2000;15–16.
24. Hernán MA, Hernández-Diaz S, Robins JM. A structural approach to selection bias.
Epidemiology. 2004;15:615–625.
25. Raghunathan TE. What do we do with missing data? Some options for analysis of incomplete data.
Annu Rev Public Health. 2004;25:99–117.
26. Hagel BE, Pless IB, Goulet C, et al. Quality of information on risk factors reported by ski patrols.
Injury Prevention. 2004;10:275–279.
27. Hunter RE. Skiing injuries.
Am J Sports Med. 1999;27:381–389.
