- Email: [email protected]
Do bicycle safety helmets reduce severity of head injury in real crashes?
oool-4575/&7 s3.00+ .a0 0 1987 Pergamon Journals Ltd.
Accid. Anal. & Prev. Vol. 19, No. 3, pp. 183-190, 1987 Rimed in Great Britain.
DO BICYCLE SAFETY HELMETS REDUCE SEVERITY OF HEAD INJURY IN REAL CRASHES?* MARGARET M. DORSCH§, ALISTAIR J. WOODWARD? and RONALD L. SOMERS’ Road Accident Research Unit, University of Adelaide, G.P.O. Box 498, Adelaide, S.A. 5001. Australia (Received 30 September
1985; in revised form 5 May 1986)
Abstract-In the past, evaluation of helmet efficacy has been based on laboratory tests of limited relevance to real crashes. In the present study 894 South Australian bicycling enthusiasts returned mail questionnaires about their most recent bicycle crash and their helmet use at the time. 197 bicyclists reported a crash within the past five years in which they had struck their head or helmet. Helmet status at the time of the crash was reported as: no helmet used (n = 75), haimetstyle helmet (n =69), hard-shell with soft or no liner (n = 37), or hard-shell helmet with stiff liner (n = 16). Analysis of the crude, unadjusted data showed a statistically significant association between helmet use and reduced severity of head injury. The association persisted after adjustment for age and sex of rider, and severity of crash forces. Using an unpublished method developed by Somers, it was estimated that the risk of death from head injury was considerably reduced for helmeted relative to unhelmeted bicyclists, depending on helmet type.
INTRODUCTION
Helmets for bicyclists could do much to reduce deaths and injuries among crash-involved riders. While few people would doubt this assertion, there are currently no quantitative data demonstrating the efficacy of bicycle helmets in real crashes. Such information is desirable for devising appropriate and cost-effective intervention strategies. Data from several countries indicate that head injury is present in about 80-85 percent of fatally injured bicyclists [Tonge, O’Reilly and Davison, 1964; Cross and Fisher, 1977; Bennett and Downing, 1977; Gillies, 19801. However, those studies do not give a clear estimate of the proportion of bicycling deaths that are attributable to head injuries. A recent investigation from Dade County, Florida, reported that the head or neck was the most seriously injured body region in 86 percent of bicycling fatalities [Fife et al., 19831. Head injuries also contribute substantially to morbidity following bicycle crashes. Head injuries were reported for 68 percent of bicyclists involved in road crashes to which an ambulance was called in Adelaide, South Australia [McLean, Brewer and Sandow, 19791. However, not all of these head-injured bicyclists were subsequently admitted to hospital. Discharge statistics from all acute care hospitals in Western Australia show that about 48 percent of injured bicyclists admitted to these hospitals had suffered a head injury [Lugg, 19821. In a longitudinal survey of an entire Danish community, eight percent of bicycle crash victims seen at the hospital emergency room had sustained brain injury [Jorgensen et al., 19791. Other head injuries (i.e., skull, scalp or facial lesions) were recorded for 36 percent of the bicyclists. Overall, these figures highlight the need for protective headgear for bicyclists. Helmets for bicyclists can be evaluated by: 1. standardized laboratory testing for impact attenuation, and retention; 2. epidemiological studies of real crashes.
penetration
resistance
*A previous version of this paper was presented at the 28th Annual Conference of the American Association for Automotive Medicine in October 1984. OPresent Address: The Ageing Project, 229 Marion Road, Marleston, S.A. 5033, Australia. tPresent Address: Department of Community Medicine, University of Adelaide, G.P.O. Box 498, Adelaide, S.A., 5001, Australia. ‘Present Address: Epidemiology Branch, S.A. Health Commission, G.P.O. Box 1313, Adelaide, S.A., 5001, Australia. 183
184
M. M. DORSCHet al.
A review of the scientific and popular literature revealed that very little has been done in the latter area. Field assessment of helmets until now has been largely based upon anecdotal evidence, i.e., observations of crashed helmets coupled with information on head injuries sustained [Skid Lid, undated]. Without reference to a suitable control group of unhelmeted riders, conclusions derived from field studies may not be particularly useful. Likewise, it is of little help to compare the nature and frequency of head injuries between helmeted motor cyclists and unhelmeted bicyclists, as was done in a recent Australian study [McDermott and Klug, 1982].In such a study, a host of differences between precrash, within-crash and post-crash factors in the two groups could contribute to erroneous conclusions about the potential protective effects of bicycle helmets. Since laboratory test procedures rely heavily on performance criteria derived from experimental head injury research, their relevance to real world crashes has also been questioned [Aldman et al., 1978; Robbins, 19801. Given the limitations of current knowledge of the mechanisms of head injury, these tests might be viewed primarily as providing a means for comparing the performance of existing helmets under the circumstances of the test. However, their utility in terms of evaluating the head protection afforded by helmets is limited. Under present impact test methods, helmets are judged by their ability to meet specified criteria for linear acceleration, although rotational acceleration is thought to be an important mechanism of brain injury [Holbourn, 194.5; Strich, 1961; Ommaya and Gennarelli, 19741. No allowance is made in impact tests for the fact that bicyclists’ heads are connected to bodies and, furthermore, the characteristics of the headforms used bear little relationship to the human skull and brain [Robbins, 19801. It is not possible to quantify the degree of human protection provided by helmets in real crashes on the basis of laboratory performance. The present study represents an attempt to determine the effectiveness of bicycle safety helmets in real crashes, and to estimate the reduction in mortality risk associated with helmet use. METHODS
Sources of data
Data for the study were obtained by a mail questionnaire. Questionnaires were sent to current and non-current members of five of the seven registered bicycling clubs in South Australia. These included three track or road racing organizations, and two general interest bicycling groups. (A small club specifically for penny farthing enthusiasts was excluded, as was a BMX bicycle group. The latter failed to respond to requests for membership lists.) The questionnaire sought information about the bicyclists’ most recent crash. Respondents were also asked to give details of: their age and sex, the circumstances of the crash (including estimated speeds, location, objects struck, and relative directions of travel for all vehicles involved), helmet use, blows to the head/helmet, and any injuries sustained. Neither the questionnaire, nor the accompanying letter, indicated that the primary objective of the study was to determine the efficacy of bicycle helmets in diminishing trauma to the head. Responses were subsequently coded in a form suitable for computer analysis. Injuries were classified by body region and tissue type according to a previously described scheme [Somers, 19831 and each injury was then assigned an appropriate severity code from the Abbreviated Injury Scale (AIS). Skull and intracranial trauma were coded separately as recommended in the AIS manual [CIS, 19801. For the majority of victims reporting an intracranial injury, the AIS score was determined on the basis of duration of unconsciousness. Statistical methods
Only those bicyclists who reported having sustained a blow to the head or helmet during their most recent crash were included in the statistical analyses. Study subjects
Do helmets reduce severity of head injury
185
were then classified into four groups by helmet type. These groups consisted of bicyclists who wore: 1. no helmet; 2. a “hairnet” or racing-style helmet; 3. a “poor” hard-shell helmet, i.e., one having little or no inner-lining, or only soft liner material; 4. a “good” hard-shell helmet, i.e., a design incorporating a stiff liner of expanded polystyrene or rigid foam over most of the inner surface of the shell. Data on the nature and severity of head injuries sustained in the four groups were analysed at two levels. At the first, most basic, level of analysis simple frequency distributions of the raw data for each helmet group were compared. At the second level of analysis multiple linear regression [BMDP, 19811 was used to determine whether helmet use was associated with protection from head injury, independent of other factors affecting head injury. The concern here was that the designated helmet groups might vary in important ways (e.g., crash circumstances) and that such variations could account for observed differences in head injury. For this analysis head injuries were categorized into brain, skull, face/scalp soft tissue and facial bone lesions. The maximum AIS code recorded in each category was identified, and the two highest of these were defined respectively as “highest AIS code for head” and “second highest AIS code for head”. Each of these two head injury severity variables was then used respectively as the dependent variable in a multiple linear regression model. These two models proved useful in verifying conclusions reached in the first level analysis. The magnitude of the forces on the head in each crash could only be estimated in a relative sense on the basis of information provided by the respondent. These forces, referred to here as “crash violence”, were quantified with an ad hoc Likert-type scale [Maranell, 19741 which yielded a crash violence score for each victim. These scores were used in the second level analysis. The scale of crash violence incorporated information on the bicyclist’s speed at time of crash, the nature of the object (including road surface) impacted by the bicyclist’s head/helmet, whether a second vehicle was involved, and if so, the other vehicle’s speed. In addition another variable was used in this second level analysis to characterize the biomechanical forces of the crash. This variable was the severity of injuries sustained by the victim in body regions other than the head. Nonhead injury severity was quantified using the highest AIS code recorded for an injury outside the head region. RESULTS
Response rate and helmet use
A total of 1,321 questionnaires were mailed out to present and former members of bicycling clubs. Of these, 894 (67.7 percent) were completed and returned to us (Table 1). Among the 866 usable responses 197 (22.7 percent) were from bicyclists who had received an impact to the head or helmet in their most recent crash. There were 28
Table 1. Questionnaire
response and classification of respondents
Classification Crash with blow to head or helmet Crash with no blow to head or helmet No crashes to report Unusable questionnaire Questionnaire returned unopened Questionnaire not returned TOTAL
Number
% of Total
197 356 313 28 160 267
14.9 26.9 23.7 2.1 12.1 20.2
1321
100%
186
M. M. DORSCHet al.
replies which could not be used; these either contained information on more than one crash or clearly did not refer to the most recent crash. Of the 197 crash-involved riders who reported a head impact, 168 were male; the mean age for the entire group was 29 years (S.D. = 15). Seventy-five (38 percent) of these cyclists were not using a helmet. In 69 cases (35 percent) a hairnet helmet was worn. Hard-shell helmets of poor design were worn by 37 (19 percent), and the remaining 16 cyclists (8 percent) had been using a good hard-shell helmet at the time of the crash. No quantitative data on helmet wearing rates among the general South Australian bicycling population are available. However, personal observation suggests that the helmet usage rate reported here (62 percent) is relatively high. This reflects the preponderance of racing club bicyclists in our study sample. Effect of helmets on head injury
The results of the first level of statistical analysis are presented in Tables 2 and 3. Table 2 groups the 197 study subjects by the nature of the most severe head injury sustained and helmet status. The injuries, which range from no injury to severe concussion and skull fracture, have been ranked in severity by AIS code. The crash victims without helmets were more likely to suffer head injury than were bicyclists in any of the helmeted groups. Aggregation of the data into four severity groups, AIS 0, 1, 2 and 3+, and application of a standard X2 test confirms that the observed association of head injury severity and helmet status is statistically significant (X2 = 25.0, p < 0.005). In a related analysis each crash victim was assessed for presence or absence of lesions to four major anatomic divisions of the head: the brain, the skull, the facial bones and the soft tissue of face and scalp. Table 3 presents the findings by helmet status. Helmets appeared to be protective against brain injury (X2 = 9.4, p < 0.025) and external soft tissue damage (X2 = 12.5, p < 0.01). The Fisher Exact Probability Test [Siegel, 19561 was used to compare the presence of skull and then facial fractures in the unhelmeted group with that of the hairnet helmet group and the combined hard helmet groups. Helmets appeared to be protective against facial fractures (p < 0.01 for hairnets; p < 0.08 for combined hard helmets), but not skull fracture. It should be noted though that skull fracture was not well represented in the study sample as it was reported by only five crash victims. It is noteworthy that helmet use did not appear to increase the likelihood of neck injury in our sample of 197 crash-involved riders. In fact, neck injuries were relatively more frequent in the unhelmeted group than in any of the helmeted ones. At the second level of analysis seven variables were used in multiple linear regression models to predict head injury severity. The variables used were age, sex, crash violence, non-head injury severity and three dummy variables for helmet status, D 1 (for hairnet), 02 (for poor hard), 03 (for good hard helmet). Unhelmeted crash victims were coded zero on Dl, 02, and 03. As shown in Table 4 helmet status proved to be statistically significant in predicting head injury severity. (A few cases were excluded because of missing information on certain variables). Thus there is some evidence that helmets are protective against head injury, after controlling for group differences in age, sex and biomechanical forces. DISCUSSION
The results of this study are consistent, at each level of analysis, with the hypothesis that helmets (hairnet or hard shell designs) reduce the severity of head injury suffered by bicyclists in crashes. This finding is, to our knowledge, unique, since it is based on: 1. a field study, with real life crashes, including a direct comparison and unhelmeted bicyclists; 2. analysis which allows for the confounding effect of crash severity.
of helmeted
Data presented in Tables 2 and 3 suggest that helmets are effective, but these data do not allow for possible confounding factors in the association between helmet usage and
5
Total
2
Concussion > 24 hr Fracture base of skull/intracranial haematoma 75
1
1
Concussion l-24 hr
4
14 5 7
(100)
(4)
(1)
(9)
69
0
0
1
10 4 4
1
1
(27)
5
3
9
31
Hairnet
4
(40)
(19)
(%)
(100)
(0)
(1)
(6)
(22)
(26)
(45)
(%)
HELMET TYPE
12
1.5
14
No Helmet
Concussion 15-59 mm
I
Facial graze/ laceration < 5 cm/ fractured nose Scalp bruising laceration < 10 cm Concussion-no loss of consciousness Facial laceration > 5 cm/ scalp laceration > 10cm Concussion & loss of consciousness: cl5 min unknown duration
No head injury
Most Severe Head Injury
3
2
1
0
AIS Severity Code
37
0
1
1
1 4
5
(loo)
(3)
(3)
(11)
16
0
0 0
0
:
1
0
1
4
9
Good Hard
0
(16)
(13)
(54)
(“ro)
0
3
2
20
Poor Hard
Table 2. Dist~bution of bicycle crash victims by helmet type and nature of most severe head injury (N = 197)
(100)
(0)
(0)
(0)
(13)
(31)
(56)
(%)
188
M. M. DORSCHet
al.
Table 3. Proportion of bicycle crash victims in each helmet group with or without specified head injuries (N = 197)
Type of Head Injury
Brain Injury Skull Fracture Soft Tissue Injury To Face or Scalp Facial Fracture
Present/ Absent
No Helmet n = 75
Present Absent Total Present Absent Total Present Absent Total Present Absent Total
45 55 100 4 96 100 61 39 100 9 91 100
HELMET TYPE Poor Hairnet Hard n = 69 n = 37 33 67 100 3 97 100 36 64 100 0 100 100
Good Hard n = 16
32 68 100 0 100 100 32 68 100 3 97 100
6 94 100 0 100 100 44 56 100 0 100 100
head injury severity. Table 4 shows the factors which were considered, on theoretical grounds, to be possible confounders, and so were included in the multivariate analysis. Both the ad hoc crash violence score, and the variable representing severity of injuries to body regions other than the head, were found to contribute to the prediction of head injury severity. Nevertheless, when the influence of these factors (which together were taken as a measure of impact severity) and age were controlled analytically, the protective effect of helmets persisted. The present sample of bicyclists is almost certainly biased, i.e., deficient in cases of severe head injury compared with the actual mix of case severity which is incident in South Australia. The under-representation of severe cases in the study sample would not be so worrisome were these cases equally likely to be helmeted or unhelmeted. However, the present findings point to a definite association between helmet use and severity of head injury, which leads to the suspicion that severe head injury cases are more likely to be unhelmeted than helmeted. Cases of severe head injury were possibly missed from our sample through two mechanisms. First, persons killed or seriously disabled in a bicycle crash might be removed from the membership lists of the participating clubs. Second, such persons who did remain on the membership list, and who were consequently mailed a questionnaire, might not have been able or inclined to
Table 4. Results of the multiple linear regression analysis. Dependent variable was AIS code of most severe head injury (Sl) or second most severe head injury (S2). (N = 193) Independent Variable D3* 1 = good hard helmet 0 = no good hard helmet Dl* 1 = hairnet helmet 0 = no hairnet helmet Dii 1 = poor hard helmet 0 = no poor hard helmet Non-Head Injury** Crash Violence** Sex 1 = male 0 = female Age
Regression Coefficient Sl s2
Standard Error s2 Sl
Two-tail Significance s2 Sl
- ,832
- ,342
,303
,175
.0067
,052
- ,597
- ,221
,185
,107
.0015
,040
- ,405 .205 JO508
- ,179 ,166 - .000252
,229 ,107 .00296
,132 .0622 .00171
,078 ,059 ,088
.18 .0083 .88
- .336 .000395
.0176 - .00109
,227 .00564
,131 .00326
.14 .94
.89 .74
*These variables were used to code helmet use. Unhelmeted 03. **These variables were used to quantify crash severity.
cyclists were coded “0” for Dl, D2 and
Do helmets reduce severity of head injury
189
respond. Addition of even a few cases of severe head injury to the present unhelmeted group would increase the calculated advantage of helmet use. A third mechanism might have worked to decrease the representation of helmeted crash victims who sustained no head injury. The study was limited to persons who reported hitting their head or helmet during the crash. Helmeted bicyclists may have been less likely to notice a blow to the head than unhelmeted bicyclists. The unintentional exclusion of some uninjured helmeted bicyclists, like the de facto exclusion of some injured unhelmeted bicyclists, would work to decrease the apparent advantage of helmet use. Reporting bias is another source of systematic error which needs to be considered in the interpretation of our findings. In this study, the determination of helmet efficacy relied solely upon self-reported injury data. If the four helmet groups differed in their accuracy of reporting the severity of head injuries, then our estimates of the head injury reduction potential of helmets would be biased. Since we had no means of independently validating the injury descriptions for the great majority of our 197 subjects, we attempted to use a rather indirect method of doing this. On the assumption that helmet use itself does not influence injury severity to body regions other than the head, we performed a multivariate analysis identical to that described earlier except that this time “non-head injury severity” was used as the dependent variable. (The independent predictor variables then became: age, sex, crash violence, and the dummy helmet use variables.) We reasoned that any tendency towards inaccuracy of reporting non-head injuries among bicycle riders might also hold for head injuries. If helmet use was a predictor of non-head injury severity in our data set, then there would be grounds for suspecting a systematic bias in our estimates of the head protection afforded by helmets. The result of this multivariate analysis revealed that the helmet use variables were not in fact associated with non-head injury severity. Therefore, we have little reason to suspect systematic biases in the reporting of head injury severity between the sampled helmet groups. Our findings suggest that hard-shell helmets incorporating a complete or almost complete inner liner with good shock-absorption properties (e.g., expanded polystyrene or rigid foam) afford much better head protection in real crashes than do hairnet helmets, and hard-shells having inadequate or no liners. These results are in general agreement with previously published experimental data concerning the relative merits of various hard-shell, or hairnet-helmet designs [Gillies, 1980; Balderston, 1983; Jones and Mohan, 19841. We have used an unpublished method developed by Somers to roughly estimate the efficacy of bicycle safety helmets in saving lives. If a large group of bicyclists similar in age and sex to our 75 unhelmeted bicyclists was to be involved in crashes of similar severity, about 90% of the deaths due to head injury could be prevented with good hard helmets. Hairnet helmets could prevent 80 of the deaths and poor hard helmets could prevent 70% of the deaths. These estimates do not imply that safety helmets can be expected to prevent such large proportions of all bicycle deaths due to head injury; clearly some such deaths involve crashes of such violence that a helmet would not be effective. The estimates of helmet efficacy given here assume that each crash is “potentially survivable”, and further that helmets do not increase the chances of a bicyclist being involved in a crash, nor the chances that the bicyclist will hit his head given that a crash occurs. Further research is needed to confirm and refine our findings. Ideally, future study designs will achieve independent and objective assessment of both crash and injury severity, and will provide more detail on the morbidity arising from head injury amongst helmeted and unhelmeted riders. Acknowledgements-We are indebted to our colleagues, Michael Clark and Bruce Paix (NH&MRC Road Accident Research Unit), for writing the computer programs used in sorting membership lists and assisting with the “crash violence” score, respectively. Critical evaluations of our analyses were kindly provided by John Darroch (Flinders University, South Australia), Sander Greenland (UCLA) and John Mathieson (Newcastle Cycleways Movement). We also thank the committees and members of the participating clubs for their enthusiastic response to our survey.
M. M. DORSCH efal.
190
REFERENCES Aldman B., Lundell B., Thorngren L. and Turbell T., Helmet attenuation of the head response in oblique impacts to the ground. Proc. 3rd ht. IRCOBI Conf., Lyon: IRCOBI Secretariate, 1978. Balderston T., Bicycle helmets: Which one? Bicycling Magazine (US) XXIV, 124-147, 1983. Bennett M. and Downing C. S., Safety helmets for pedal cyclists-a pilot study amongst children. TRRL Suppl. Rep. 283, Transport and Road Research Laboratory, Crowthorne, Berkshire, U.K., 1977. BMDP, Biomedical Computer Programs, P-Series. (Edited by W. J. Dixon and M. B. Brown.) University of California Press, Berkeley, California, 1981. Committee on Injury Scaling (CIS), The Abbreviated Injury Scale 1980 Revision. American Association for Automotive Medicine, Morton Grove, IL 60053, U.S.A., 1980. Cross K. D. and Fisher G., A Study of Bicycle/ Motor Vehicle Accidents: Identification of Problem Types and Countermeasure Approaches. National Technical Information Service, Springfield, VA 22161, U.S.A., 1977. Fife D., Davis J., Tate L., Wells J. K., Mohan D. and Williams A., Fatal iniuries to bicvclists: the experience of Dade County, Florida. J. Trauma, 23, 745-755, 1983. Gillies N., Helmets for use by bicycle riders. Research Note 4/80. Traffic Accident Research Unit. Traffic Authority of N.S.W., Rosebery, New South Wales, Australia, 1980. Holbourn A. H. S., The mechanics of brain injuries. Br. Med. Bull., 3, 147-149, 1945. Jones I. S. and Mohan D.. Head Impact Tolerance: Correlation Between Dummv Imoacts and Actual Head Injuries. Insurance Institute for Highway Safety, Washington, D.C., 1984. ’ ’ Jorgensen K., Kruse T., Somers R. L. and Weeth R., Description of 3225 victims of road-traffic-accident trauma according to type of accident, severity of injury, and nature of lesions sustained. Proc. 4th Znt. IRCOBI Conf. Bron: IRCOBI Secretariate, 1979. Lugg M., Hospital morbidity statistics, special studies, pedal-cycle accidents, 1971-80. W. H. Health Statistics Series B, No. 1. Statistics Branch, Public Health Department, Perth, Western Australia, 1982. Maranell G. M., Scaling-A Sourcebook for Behavioural Scientists. Aldine, Chicago, 1974. McDermott F. T. and Klug G. L., Differences in head injuries of pedal cyclist and motorcyclist casualties in Victoria. Med. J. Aust., 2, 30-32, 1982. McLean A. J., Brewer N. D. and Sandow B. L., Adelaide In-Depth Accident Study, 1975-79, Part 3: Pedal Cycle Accidents. Road Accident Research Unit, University of Adelaide, Australia, 1979. Ommava A. K. and Gennarelli T. A., Cerebral concussion and traumatic unconsciousness: correlations of experimental and clinical observations on blunt head injuries. Brain, 97, 633-654, 1974. Robbins D. H., Impact Head Injury Data Base, Report No. UM-HSRI-80-21. Highway Safety Research Institute, University of Michigan, Ann Arbor, MI, U.S.A., 1980. Siegel S., Nonparametric Statistics for the Behavioral Sciences. McGraw-Hill, New York, 1956. Skid Lid Manufacturina Co. Popular Helmet Testing: Beware the Easv Answer. San Dieeo California. undated. Somers R. L., The probability-of death score: an\mprovement of the injury severityscore. Accid: Anal. & Prev., 15, 247-257,
1983.
Strich S. J., Shearing of nerve fibres as a cause of brain damage due to head injury. Lancet, ii, 443-448,1961. Tonge J. I., O’Reilly M. J. J. and Davison A., Fatal traffic accidents in Brisbane from 1935 to 1964. Med. J. Aust., 2, 811-820,
1964.
Accid. Anal. & Prev. Vol. 19, No. 3, pp. 183-190, 1987 Rimed in Great Britain.
DO BICYCLE SAFETY HELMETS REDUCE SEVERITY OF HEAD INJURY IN REAL CRASHES?* MARGARET M. DORSCH§, ALISTAIR J. WOODWARD? and RONALD L. SOMERS’ Road Accident Research Unit, University of Adelaide, G.P.O. Box 498, Adelaide, S.A. 5001. Australia (Received 30 September
1985; in revised form 5 May 1986)
Abstract-In the past, evaluation of helmet efficacy has been based on laboratory tests of limited relevance to real crashes. In the present study 894 South Australian bicycling enthusiasts returned mail questionnaires about their most recent bicycle crash and their helmet use at the time. 197 bicyclists reported a crash within the past five years in which they had struck their head or helmet. Helmet status at the time of the crash was reported as: no helmet used (n = 75), haimetstyle helmet (n =69), hard-shell with soft or no liner (n = 37), or hard-shell helmet with stiff liner (n = 16). Analysis of the crude, unadjusted data showed a statistically significant association between helmet use and reduced severity of head injury. The association persisted after adjustment for age and sex of rider, and severity of crash forces. Using an unpublished method developed by Somers, it was estimated that the risk of death from head injury was considerably reduced for helmeted relative to unhelmeted bicyclists, depending on helmet type.
INTRODUCTION
Helmets for bicyclists could do much to reduce deaths and injuries among crash-involved riders. While few people would doubt this assertion, there are currently no quantitative data demonstrating the efficacy of bicycle helmets in real crashes. Such information is desirable for devising appropriate and cost-effective intervention strategies. Data from several countries indicate that head injury is present in about 80-85 percent of fatally injured bicyclists [Tonge, O’Reilly and Davison, 1964; Cross and Fisher, 1977; Bennett and Downing, 1977; Gillies, 19801. However, those studies do not give a clear estimate of the proportion of bicycling deaths that are attributable to head injuries. A recent investigation from Dade County, Florida, reported that the head or neck was the most seriously injured body region in 86 percent of bicycling fatalities [Fife et al., 19831. Head injuries also contribute substantially to morbidity following bicycle crashes. Head injuries were reported for 68 percent of bicyclists involved in road crashes to which an ambulance was called in Adelaide, South Australia [McLean, Brewer and Sandow, 19791. However, not all of these head-injured bicyclists were subsequently admitted to hospital. Discharge statistics from all acute care hospitals in Western Australia show that about 48 percent of injured bicyclists admitted to these hospitals had suffered a head injury [Lugg, 19821. In a longitudinal survey of an entire Danish community, eight percent of bicycle crash victims seen at the hospital emergency room had sustained brain injury [Jorgensen et al., 19791. Other head injuries (i.e., skull, scalp or facial lesions) were recorded for 36 percent of the bicyclists. Overall, these figures highlight the need for protective headgear for bicyclists. Helmets for bicyclists can be evaluated by: 1. standardized laboratory testing for impact attenuation, and retention; 2. epidemiological studies of real crashes.
penetration
resistance
*A previous version of this paper was presented at the 28th Annual Conference of the American Association for Automotive Medicine in October 1984. OPresent Address: The Ageing Project, 229 Marion Road, Marleston, S.A. 5033, Australia. tPresent Address: Department of Community Medicine, University of Adelaide, G.P.O. Box 498, Adelaide, S.A., 5001, Australia. ‘Present Address: Epidemiology Branch, S.A. Health Commission, G.P.O. Box 1313, Adelaide, S.A., 5001, Australia. 183
184
M. M. DORSCHet al.
A review of the scientific and popular literature revealed that very little has been done in the latter area. Field assessment of helmets until now has been largely based upon anecdotal evidence, i.e., observations of crashed helmets coupled with information on head injuries sustained [Skid Lid, undated]. Without reference to a suitable control group of unhelmeted riders, conclusions derived from field studies may not be particularly useful. Likewise, it is of little help to compare the nature and frequency of head injuries between helmeted motor cyclists and unhelmeted bicyclists, as was done in a recent Australian study [McDermott and Klug, 1982].In such a study, a host of differences between precrash, within-crash and post-crash factors in the two groups could contribute to erroneous conclusions about the potential protective effects of bicycle helmets. Since laboratory test procedures rely heavily on performance criteria derived from experimental head injury research, their relevance to real world crashes has also been questioned [Aldman et al., 1978; Robbins, 19801. Given the limitations of current knowledge of the mechanisms of head injury, these tests might be viewed primarily as providing a means for comparing the performance of existing helmets under the circumstances of the test. However, their utility in terms of evaluating the head protection afforded by helmets is limited. Under present impact test methods, helmets are judged by their ability to meet specified criteria for linear acceleration, although rotational acceleration is thought to be an important mechanism of brain injury [Holbourn, 194.5; Strich, 1961; Ommaya and Gennarelli, 19741. No allowance is made in impact tests for the fact that bicyclists’ heads are connected to bodies and, furthermore, the characteristics of the headforms used bear little relationship to the human skull and brain [Robbins, 19801. It is not possible to quantify the degree of human protection provided by helmets in real crashes on the basis of laboratory performance. The present study represents an attempt to determine the effectiveness of bicycle safety helmets in real crashes, and to estimate the reduction in mortality risk associated with helmet use. METHODS
Sources of data
Data for the study were obtained by a mail questionnaire. Questionnaires were sent to current and non-current members of five of the seven registered bicycling clubs in South Australia. These included three track or road racing organizations, and two general interest bicycling groups. (A small club specifically for penny farthing enthusiasts was excluded, as was a BMX bicycle group. The latter failed to respond to requests for membership lists.) The questionnaire sought information about the bicyclists’ most recent crash. Respondents were also asked to give details of: their age and sex, the circumstances of the crash (including estimated speeds, location, objects struck, and relative directions of travel for all vehicles involved), helmet use, blows to the head/helmet, and any injuries sustained. Neither the questionnaire, nor the accompanying letter, indicated that the primary objective of the study was to determine the efficacy of bicycle helmets in diminishing trauma to the head. Responses were subsequently coded in a form suitable for computer analysis. Injuries were classified by body region and tissue type according to a previously described scheme [Somers, 19831 and each injury was then assigned an appropriate severity code from the Abbreviated Injury Scale (AIS). Skull and intracranial trauma were coded separately as recommended in the AIS manual [CIS, 19801. For the majority of victims reporting an intracranial injury, the AIS score was determined on the basis of duration of unconsciousness. Statistical methods
Only those bicyclists who reported having sustained a blow to the head or helmet during their most recent crash were included in the statistical analyses. Study subjects
Do helmets reduce severity of head injury
185
were then classified into four groups by helmet type. These groups consisted of bicyclists who wore: 1. no helmet; 2. a “hairnet” or racing-style helmet; 3. a “poor” hard-shell helmet, i.e., one having little or no inner-lining, or only soft liner material; 4. a “good” hard-shell helmet, i.e., a design incorporating a stiff liner of expanded polystyrene or rigid foam over most of the inner surface of the shell. Data on the nature and severity of head injuries sustained in the four groups were analysed at two levels. At the first, most basic, level of analysis simple frequency distributions of the raw data for each helmet group were compared. At the second level of analysis multiple linear regression [BMDP, 19811 was used to determine whether helmet use was associated with protection from head injury, independent of other factors affecting head injury. The concern here was that the designated helmet groups might vary in important ways (e.g., crash circumstances) and that such variations could account for observed differences in head injury. For this analysis head injuries were categorized into brain, skull, face/scalp soft tissue and facial bone lesions. The maximum AIS code recorded in each category was identified, and the two highest of these were defined respectively as “highest AIS code for head” and “second highest AIS code for head”. Each of these two head injury severity variables was then used respectively as the dependent variable in a multiple linear regression model. These two models proved useful in verifying conclusions reached in the first level analysis. The magnitude of the forces on the head in each crash could only be estimated in a relative sense on the basis of information provided by the respondent. These forces, referred to here as “crash violence”, were quantified with an ad hoc Likert-type scale [Maranell, 19741 which yielded a crash violence score for each victim. These scores were used in the second level analysis. The scale of crash violence incorporated information on the bicyclist’s speed at time of crash, the nature of the object (including road surface) impacted by the bicyclist’s head/helmet, whether a second vehicle was involved, and if so, the other vehicle’s speed. In addition another variable was used in this second level analysis to characterize the biomechanical forces of the crash. This variable was the severity of injuries sustained by the victim in body regions other than the head. Nonhead injury severity was quantified using the highest AIS code recorded for an injury outside the head region. RESULTS
Response rate and helmet use
A total of 1,321 questionnaires were mailed out to present and former members of bicycling clubs. Of these, 894 (67.7 percent) were completed and returned to us (Table 1). Among the 866 usable responses 197 (22.7 percent) were from bicyclists who had received an impact to the head or helmet in their most recent crash. There were 28
Table 1. Questionnaire
response and classification of respondents
Classification Crash with blow to head or helmet Crash with no blow to head or helmet No crashes to report Unusable questionnaire Questionnaire returned unopened Questionnaire not returned TOTAL
Number
% of Total
197 356 313 28 160 267
14.9 26.9 23.7 2.1 12.1 20.2
1321
100%
186
M. M. DORSCHet al.
replies which could not be used; these either contained information on more than one crash or clearly did not refer to the most recent crash. Of the 197 crash-involved riders who reported a head impact, 168 were male; the mean age for the entire group was 29 years (S.D. = 15). Seventy-five (38 percent) of these cyclists were not using a helmet. In 69 cases (35 percent) a hairnet helmet was worn. Hard-shell helmets of poor design were worn by 37 (19 percent), and the remaining 16 cyclists (8 percent) had been using a good hard-shell helmet at the time of the crash. No quantitative data on helmet wearing rates among the general South Australian bicycling population are available. However, personal observation suggests that the helmet usage rate reported here (62 percent) is relatively high. This reflects the preponderance of racing club bicyclists in our study sample. Effect of helmets on head injury
The results of the first level of statistical analysis are presented in Tables 2 and 3. Table 2 groups the 197 study subjects by the nature of the most severe head injury sustained and helmet status. The injuries, which range from no injury to severe concussion and skull fracture, have been ranked in severity by AIS code. The crash victims without helmets were more likely to suffer head injury than were bicyclists in any of the helmeted groups. Aggregation of the data into four severity groups, AIS 0, 1, 2 and 3+, and application of a standard X2 test confirms that the observed association of head injury severity and helmet status is statistically significant (X2 = 25.0, p < 0.005). In a related analysis each crash victim was assessed for presence or absence of lesions to four major anatomic divisions of the head: the brain, the skull, the facial bones and the soft tissue of face and scalp. Table 3 presents the findings by helmet status. Helmets appeared to be protective against brain injury (X2 = 9.4, p < 0.025) and external soft tissue damage (X2 = 12.5, p < 0.01). The Fisher Exact Probability Test [Siegel, 19561 was used to compare the presence of skull and then facial fractures in the unhelmeted group with that of the hairnet helmet group and the combined hard helmet groups. Helmets appeared to be protective against facial fractures (p < 0.01 for hairnets; p < 0.08 for combined hard helmets), but not skull fracture. It should be noted though that skull fracture was not well represented in the study sample as it was reported by only five crash victims. It is noteworthy that helmet use did not appear to increase the likelihood of neck injury in our sample of 197 crash-involved riders. In fact, neck injuries were relatively more frequent in the unhelmeted group than in any of the helmeted ones. At the second level of analysis seven variables were used in multiple linear regression models to predict head injury severity. The variables used were age, sex, crash violence, non-head injury severity and three dummy variables for helmet status, D 1 (for hairnet), 02 (for poor hard), 03 (for good hard helmet). Unhelmeted crash victims were coded zero on Dl, 02, and 03. As shown in Table 4 helmet status proved to be statistically significant in predicting head injury severity. (A few cases were excluded because of missing information on certain variables). Thus there is some evidence that helmets are protective against head injury, after controlling for group differences in age, sex and biomechanical forces. DISCUSSION
The results of this study are consistent, at each level of analysis, with the hypothesis that helmets (hairnet or hard shell designs) reduce the severity of head injury suffered by bicyclists in crashes. This finding is, to our knowledge, unique, since it is based on: 1. a field study, with real life crashes, including a direct comparison and unhelmeted bicyclists; 2. analysis which allows for the confounding effect of crash severity.
of helmeted
Data presented in Tables 2 and 3 suggest that helmets are effective, but these data do not allow for possible confounding factors in the association between helmet usage and
5
Total
2
Concussion > 24 hr Fracture base of skull/intracranial haematoma 75
1
1
Concussion l-24 hr
4
14 5 7
(100)
(4)
(1)
(9)
69
0
0
1
10 4 4
1
1
(27)
5
3
9
31
Hairnet
4
(40)
(19)
(%)
(100)
(0)
(1)
(6)
(22)
(26)
(45)
(%)
HELMET TYPE
12
1.5
14
No Helmet
Concussion 15-59 mm
I
Facial graze/ laceration < 5 cm/ fractured nose Scalp bruising laceration < 10 cm Concussion-no loss of consciousness Facial laceration > 5 cm/ scalp laceration > 10cm Concussion & loss of consciousness: cl5 min unknown duration
No head injury
Most Severe Head Injury
3
2
1
0
AIS Severity Code
37
0
1
1
1 4
5
(loo)
(3)
(3)
(11)
16
0
0 0
0
:
1
0
1
4
9
Good Hard
0
(16)
(13)
(54)
(“ro)
0
3
2
20
Poor Hard
Table 2. Dist~bution of bicycle crash victims by helmet type and nature of most severe head injury (N = 197)
(100)
(0)
(0)
(0)
(13)
(31)
(56)
(%)
188
M. M. DORSCHet
al.
Table 3. Proportion of bicycle crash victims in each helmet group with or without specified head injuries (N = 197)
Type of Head Injury
Brain Injury Skull Fracture Soft Tissue Injury To Face or Scalp Facial Fracture
Present/ Absent
No Helmet n = 75
Present Absent Total Present Absent Total Present Absent Total Present Absent Total
45 55 100 4 96 100 61 39 100 9 91 100
HELMET TYPE Poor Hairnet Hard n = 69 n = 37 33 67 100 3 97 100 36 64 100 0 100 100
Good Hard n = 16
32 68 100 0 100 100 32 68 100 3 97 100
6 94 100 0 100 100 44 56 100 0 100 100
head injury severity. Table 4 shows the factors which were considered, on theoretical grounds, to be possible confounders, and so were included in the multivariate analysis. Both the ad hoc crash violence score, and the variable representing severity of injuries to body regions other than the head, were found to contribute to the prediction of head injury severity. Nevertheless, when the influence of these factors (which together were taken as a measure of impact severity) and age were controlled analytically, the protective effect of helmets persisted. The present sample of bicyclists is almost certainly biased, i.e., deficient in cases of severe head injury compared with the actual mix of case severity which is incident in South Australia. The under-representation of severe cases in the study sample would not be so worrisome were these cases equally likely to be helmeted or unhelmeted. However, the present findings point to a definite association between helmet use and severity of head injury, which leads to the suspicion that severe head injury cases are more likely to be unhelmeted than helmeted. Cases of severe head injury were possibly missed from our sample through two mechanisms. First, persons killed or seriously disabled in a bicycle crash might be removed from the membership lists of the participating clubs. Second, such persons who did remain on the membership list, and who were consequently mailed a questionnaire, might not have been able or inclined to
Table 4. Results of the multiple linear regression analysis. Dependent variable was AIS code of most severe head injury (Sl) or second most severe head injury (S2). (N = 193) Independent Variable D3* 1 = good hard helmet 0 = no good hard helmet Dl* 1 = hairnet helmet 0 = no hairnet helmet Dii 1 = poor hard helmet 0 = no poor hard helmet Non-Head Injury** Crash Violence** Sex 1 = male 0 = female Age
Regression Coefficient Sl s2
Standard Error s2 Sl
Two-tail Significance s2 Sl
- ,832
- ,342
,303
,175
.0067
,052
- ,597
- ,221
,185
,107
.0015
,040
- ,405 .205 JO508
- ,179 ,166 - .000252
,229 ,107 .00296
,132 .0622 .00171
,078 ,059 ,088
.18 .0083 .88
- .336 .000395
.0176 - .00109
,227 .00564
,131 .00326
.14 .94
.89 .74
*These variables were used to code helmet use. Unhelmeted 03. **These variables were used to quantify crash severity.
cyclists were coded “0” for Dl, D2 and
Do helmets reduce severity of head injury
189
respond. Addition of even a few cases of severe head injury to the present unhelmeted group would increase the calculated advantage of helmet use. A third mechanism might have worked to decrease the representation of helmeted crash victims who sustained no head injury. The study was limited to persons who reported hitting their head or helmet during the crash. Helmeted bicyclists may have been less likely to notice a blow to the head than unhelmeted bicyclists. The unintentional exclusion of some uninjured helmeted bicyclists, like the de facto exclusion of some injured unhelmeted bicyclists, would work to decrease the apparent advantage of helmet use. Reporting bias is another source of systematic error which needs to be considered in the interpretation of our findings. In this study, the determination of helmet efficacy relied solely upon self-reported injury data. If the four helmet groups differed in their accuracy of reporting the severity of head injuries, then our estimates of the head injury reduction potential of helmets would be biased. Since we had no means of independently validating the injury descriptions for the great majority of our 197 subjects, we attempted to use a rather indirect method of doing this. On the assumption that helmet use itself does not influence injury severity to body regions other than the head, we performed a multivariate analysis identical to that described earlier except that this time “non-head injury severity” was used as the dependent variable. (The independent predictor variables then became: age, sex, crash violence, and the dummy helmet use variables.) We reasoned that any tendency towards inaccuracy of reporting non-head injuries among bicycle riders might also hold for head injuries. If helmet use was a predictor of non-head injury severity in our data set, then there would be grounds for suspecting a systematic bias in our estimates of the head protection afforded by helmets. The result of this multivariate analysis revealed that the helmet use variables were not in fact associated with non-head injury severity. Therefore, we have little reason to suspect systematic biases in the reporting of head injury severity between the sampled helmet groups. Our findings suggest that hard-shell helmets incorporating a complete or almost complete inner liner with good shock-absorption properties (e.g., expanded polystyrene or rigid foam) afford much better head protection in real crashes than do hairnet helmets, and hard-shells having inadequate or no liners. These results are in general agreement with previously published experimental data concerning the relative merits of various hard-shell, or hairnet-helmet designs [Gillies, 1980; Balderston, 1983; Jones and Mohan, 19841. We have used an unpublished method developed by Somers to roughly estimate the efficacy of bicycle safety helmets in saving lives. If a large group of bicyclists similar in age and sex to our 75 unhelmeted bicyclists was to be involved in crashes of similar severity, about 90% of the deaths due to head injury could be prevented with good hard helmets. Hairnet helmets could prevent 80 of the deaths and poor hard helmets could prevent 70% of the deaths. These estimates do not imply that safety helmets can be expected to prevent such large proportions of all bicycle deaths due to head injury; clearly some such deaths involve crashes of such violence that a helmet would not be effective. The estimates of helmet efficacy given here assume that each crash is “potentially survivable”, and further that helmets do not increase the chances of a bicyclist being involved in a crash, nor the chances that the bicyclist will hit his head given that a crash occurs. Further research is needed to confirm and refine our findings. Ideally, future study designs will achieve independent and objective assessment of both crash and injury severity, and will provide more detail on the morbidity arising from head injury amongst helmeted and unhelmeted riders. Acknowledgements-We are indebted to our colleagues, Michael Clark and Bruce Paix (NH&MRC Road Accident Research Unit), for writing the computer programs used in sorting membership lists and assisting with the “crash violence” score, respectively. Critical evaluations of our analyses were kindly provided by John Darroch (Flinders University, South Australia), Sander Greenland (UCLA) and John Mathieson (Newcastle Cycleways Movement). We also thank the committees and members of the participating clubs for their enthusiastic response to our survey.
M. M. DORSCH efal.
190
REFERENCES Aldman B., Lundell B., Thorngren L. and Turbell T., Helmet attenuation of the head response in oblique impacts to the ground. Proc. 3rd ht. IRCOBI Conf., Lyon: IRCOBI Secretariate, 1978. Balderston T., Bicycle helmets: Which one? Bicycling Magazine (US) XXIV, 124-147, 1983. Bennett M. and Downing C. S., Safety helmets for pedal cyclists-a pilot study amongst children. TRRL Suppl. Rep. 283, Transport and Road Research Laboratory, Crowthorne, Berkshire, U.K., 1977. BMDP, Biomedical Computer Programs, P-Series. (Edited by W. J. Dixon and M. B. Brown.) University of California Press, Berkeley, California, 1981. Committee on Injury Scaling (CIS), The Abbreviated Injury Scale 1980 Revision. American Association for Automotive Medicine, Morton Grove, IL 60053, U.S.A., 1980. Cross K. D. and Fisher G., A Study of Bicycle/ Motor Vehicle Accidents: Identification of Problem Types and Countermeasure Approaches. National Technical Information Service, Springfield, VA 22161, U.S.A., 1977. Fife D., Davis J., Tate L., Wells J. K., Mohan D. and Williams A., Fatal iniuries to bicvclists: the experience of Dade County, Florida. J. Trauma, 23, 745-755, 1983. Gillies N., Helmets for use by bicycle riders. Research Note 4/80. Traffic Accident Research Unit. Traffic Authority of N.S.W., Rosebery, New South Wales, Australia, 1980. Holbourn A. H. S., The mechanics of brain injuries. Br. Med. Bull., 3, 147-149, 1945. Jones I. S. and Mohan D.. Head Impact Tolerance: Correlation Between Dummv Imoacts and Actual Head Injuries. Insurance Institute for Highway Safety, Washington, D.C., 1984. ’ ’ Jorgensen K., Kruse T., Somers R. L. and Weeth R., Description of 3225 victims of road-traffic-accident trauma according to type of accident, severity of injury, and nature of lesions sustained. Proc. 4th Znt. IRCOBI Conf. Bron: IRCOBI Secretariate, 1979. Lugg M., Hospital morbidity statistics, special studies, pedal-cycle accidents, 1971-80. W. H. Health Statistics Series B, No. 1. Statistics Branch, Public Health Department, Perth, Western Australia, 1982. Maranell G. M., Scaling-A Sourcebook for Behavioural Scientists. Aldine, Chicago, 1974. McDermott F. T. and Klug G. L., Differences in head injuries of pedal cyclist and motorcyclist casualties in Victoria. Med. J. Aust., 2, 30-32, 1982. McLean A. J., Brewer N. D. and Sandow B. L., Adelaide In-Depth Accident Study, 1975-79, Part 3: Pedal Cycle Accidents. Road Accident Research Unit, University of Adelaide, Australia, 1979. Ommava A. K. and Gennarelli T. A., Cerebral concussion and traumatic unconsciousness: correlations of experimental and clinical observations on blunt head injuries. Brain, 97, 633-654, 1974. Robbins D. H., Impact Head Injury Data Base, Report No. UM-HSRI-80-21. Highway Safety Research Institute, University of Michigan, Ann Arbor, MI, U.S.A., 1980. Siegel S., Nonparametric Statistics for the Behavioral Sciences. McGraw-Hill, New York, 1956. Skid Lid Manufacturina Co. Popular Helmet Testing: Beware the Easv Answer. San Dieeo California. undated. Somers R. L., The probability-of death score: an\mprovement of the injury severityscore. Accid: Anal. & Prev., 15, 247-257,
1983.
Strich S. J., Shearing of nerve fibres as a cause of brain damage due to head injury. Lancet, ii, 443-448,1961. Tonge J. I., O’Reilly M. J. J. and Davison A., Fatal traffic accidents in Brisbane from 1935 to 1964. Med. J. Aust., 2, 811-820,
1964.