Effectiveness of bicycle helmets and injury prevention: a systematic review of meta-analyses

The results section is divided into three main parts. First, the findings from the bicycle crash data studies are presented, followed by the findings from the laboratory and simulation studies, and third the results of the injury severity and cycling studies are discussed. The first section where the bicycle crash data studies are investigated is further divided into the following subsections: bicycle crash data studies: publication bias, bicycle crash data studies: cyclist behavior, bicycle crash data studies—final remarks.

Bicycle crash data studies: findings on helmet effectiveness

Findings on helmet effectiveness concerning different body regions (such as neck or face) that were found by the three meta-studies with their respective 95% CI are presented in Table 2. Interestingly, as a disclaimer, Olivier and Creighton19 pointed out that the investigation of exactly one injury type is not as straightforward, because most injuries come paired with others. They emphasize that police report data might document a more severe Traumatic Brain Injury (TBI) while ignoring a minor facial injury. If multiple types have been documented, the more severe one generally determines the overall injury severity category.

Studies included in the meta study of Attewell et al.16 vary in size from 21 to 3390 cases and include a variety of injury types. All age groups were represented with children being overrepresented. Head injuries were found to be reduced significantly with helmet usage by 60 percent, brain injuries by 58 percent and facial injuries were reduced by 4 percent. The impact on neck injuries was shown to be insignificant. Fatal injuries were shown to decrease significantly by a prominent 73 percent if the cyclist was wearing a helmet. Only the subgroup of children resulted in a higher injury rate, which might be due to hospital admission as an inclusion criterion. A broader and more recent study that was based on the crash data suggested that for children helmet wearing decreases the risk of severe injury23. Attewell et al.16 concluded that wearing a helmet reduces the overall risk of an injury, even at conservative upper confidence intervals. Only seven of the total sixty-three articles that Attewell et al.16 included in their research did not endorse helmets.

Olivier and Creighton19 included a larger number of studies, but excluded self-report and data published in abstracts only. They found a significant reduction in all their injury groups, especially severe injuries. Except for the impacts on neck injury, which yielded near to a null effect. Olivier and Creighton19 stressed out that the magnitude of injury reduction turned out to be higher in serious injuries compared to groups with any injury severity. Serious head (69 percent) and fatal head (65 percent) injuries saw a clear reduction in severity.

Høye20 included 55 records in the analysis (including abstract exclusive data) and studied a slightly different injury group. The research included a summary group of any injury type and any severity, but instead of a neck they considered cervical spine injury. Again, all but one group showed a significant decrease in OR, a decreased risk of injury while wearing a cycling helmet, mostly for a fatal head injury. Compared to the results obtained by Olivier and Creighton19, the resulting ORs are similar. ORs of fatal head injury and unspecified severity are even slightly smaller but insignificant. The serious head OR is 9 percentage points higher and the face injury group is 10 percentage points higher with both being statistically significant. Høye20 expected varying types of facial accidents to be affected in varying magnitudes but the only group with an OR above 1 was the cervical spine group. As expected, wearing a helmet during a crash does not significantly decrease such injury, but neither amplifies it.

When examining the safety effects of a mandatory bicycle helmet legislation with case-control and before-after study designs, Høye21 found that such legislation led to a reduction in injuries. Again, especially serious injuries were decreased. There were two interesting findings; first, the legislation only applying to children does only not significantly reduce serious injury among them, but also among the adults and second, the effects are even greater for all age groups when the mandatory helmet legislation is issued without age differentiation.

Bicycle crash data studies: publication bias

All discussed meta-studies addressed to varying degrees publication bias (PB) as a possible limitation. Publication bias describes that possibly studies with statistically significant results are more likely to be published. A visual cue are funnel plots, which plot standard error against its estimate, which should be symmetrical around the average and narrow at the top. Publication bias can be one reason for skewness of the funnel, which has been used as an indicator (for details and example please refer to Attewell et al.16). Also, Elvik17,18 inferred publication bias from this ten years later and addressed it using the trim-and-fill (T&F) method that argues to make the funnel plots symmetric before taking the final estimate. Olivier and Creighton19 stressed that applying the trim-and-fill technique may lead to underestimating of ORs if publication bias does not exist. Additionally, Olivier and Creighton19 used formal criteria such as rank correlation test to determine publication bias. No strong evidence of publication test was found (τ = − 0.08, P = 0.25), which made the need of the trim-and-fill obsolete in the first place. Høye20,21 used funnel plots and corrections by applying the trim-and-fill method where the funnel plots were asymmetric, which reduced the effect of publication bias. It is important to point out that the symmetrization of the funnel plot does not lead to the elimination of publication bias, since such skewness may have other causes.

Bicycle crash data studies: cyclist behavior

Cyclist behavior and particularly shift in risk taking in response to wearing a helmet was found to be an important point of consideration in the reviewed studies. To study this phenomenon, Olivier and Creighton19 made an adjustment for effects that might be induced by a theoretical risk compensation, and found the adjusted odds to be nearly identical, suggesting that such effect does not exist. The decrease of fatalities or serious injuries (KSI) in cyclist groups likely outweighs any effect of behavioral adaptation that wearing a helmet might have. Nevertheless, it remains unclear if behavioral effects are positive or negative on injuries20.

When comparing studies in areas with and without a mandatory bicycle helmet legislation, a tendency towards greater protective effects on head injuries when use is mandatory are generally shown21. After an introduction of a mandatory helmet legislation, a decrease in cyclists might occur, but such effect did not necessarily last for a long time. Surveys showed that other factors are much more important in making cycling a valid or invalid option for the general population. Høye21 argued that a possible self-selection effect would decrease the average crash risk in general, due to a variety of characteristics cyclists who decide to wear a helmet have and often exhibit a safer cycling behavior in general. Studies on behavior adaptation found no clear causality between helmet use and risk-taking behavior21.

Bicycle crash data studies: final remarks

While all the reviewed studies agree that helmet usage protects against head injuries, most studies also emphasize that helmets are most effective in preventing serious and fatal injuries. Also, the effects of bicycle helmets are larger in single bicycle crashes20. Attewell et al.16 concluded that their result of increased neck injury should continue to be monitored and might be related to the helmet type. Other studies that explored this phenomenon did not reproduce this finding. Wearing a helmet while cycling has been found to have a clear benefit on injury reduction, for all ages, independent of severity, and in bike crashes that may or may not involve a motorized vehicle.

Lab and simulation studies

To test the effectiveness of bicycle helmets without risking the well-being and health of cyclists, some studies investigated their protective capability in the laboratories. Studies employing simulation allow for standardization of test conditions and do need the real-world environment to collect the data. Each helmet that is approved on the market must comply with local laws and therefore needs to be subjected to testing in consistent conditions. In Europe, the regulations are guarded by En 1078:2012 (En1078 2012), in the USA by CPSC 16 CFR 1203-08 (16 CFR Part 1203 1998), Japan JIS T 8134:2007, Australia and New Zealand AS/AnZ 2063: 2008, and China applies their GB 24429–2009. All the outlined standards only use the peak linear acceleration (PLA) at the head’s center of gravity24. The standards fail to account for oblique impacts, where rotational dynamics exist. The rotational dynamics are known to be the most common scenarios in the real cycling conditions25. The literature search yielded three papers on the complexity of this topic. The first identified study by Bland et al.26 meant to identify the helmets’ differences in terms of their protective capabilities between the standard lab conditions and the real-world scenarios. The impacts were tested with a standard drop rig of four helmet models and compared the standard specified conditions to the ones commonly found in the real-world environment. To compute further injury criteria Deck et al.27 added a rotational acceleration sensor on a head model that represented a more realistic rotational inertia. Using the acceleration data, the study simulated a brain finite element (FE) model to obtain an indicator based on tissue level brain injury criteria, which can predict a moderate diffuse axonal injury (mDAI) such as moderate neurological injuries or short coma. Finally, Wang et al.24 aimed to mirror accurately different scenarios using a full-scale multibody (mB) simulation. Digital models of helmet’s effectiveness were created and validated through drop tests before being used in full-scale simulations for a cyclist. Impact scenarios with a cyclist’s head impacting a curb and cyclist skidding were simulated with and without the helmet models for comparison. Wang et al.24 made use of further methods of computational biomechanics—nine helmet models were first modeled by laser scanning and the material properties were determined experimentally. By comparing brain injury severities when wearing a helmet versus not wearing a helmet, the helmet head-protection effectiveness was scored. The cyclist model in Wang et al.24 was found to be the most complete and featured a head-neck complex of the Total Human model for Safety THUmS finite element model, coupled to the full-scale multibody model pedestrian model (again 50th percentile adult man), sitting on a validated model of a typical road cycle. First, all three studies came to the results that vary significantly depending on the helmet model, nevertheless, some interesting results were found.

Compared with a cyclist not wearing a helmet, the risk of skull fracture decreased across all helmets if cyclist was wearing a helmet24. On average, there was an 80 percent reduction in skull fractures for curb-impact and a 65 percent reduction for skidding. When comparing protective capabilities between the standard and real-world scenarios, Bland et al.26 showed that the risk of severe brain injury is negligible (0.2 percent to 2 percent) at lower velocity. Four of the ten models (including all non-road-style helmets) were found to have an unacceptably high risk of 88 percent to 97 percent for the temporal impact. In line with these results, Deck et al.27 obtained the most critical head injury criterion (HIC) values for the lateral and as well as occipital impacts. With the brain injury criterion (BrIC), they identified oblique impacts causing rotation around the axis represented by the neck to be the most severe. While the skidding impact shows only minor differences, a substantial difference was found between the global kinematics for the curb-impact scenario with and without a helmet. In parity with the consistently reduced global parameters (PLA and HIC based), Wang et al.24 examined both and concluded that brain deformation criteria also point to helmets to be protective in every scenario studied, even if the level of protection varied. The maximum principal strain (mPS) was greatly reduced, and cumulative strain damage measure (CSDm) varied by the helmet type.

Protective effectiveness of the bicycle helmets varied from model to model at standard impact velocity, yet this is not apparent to the consumers. The variation in injury risk varies more at the temporal location. Bland et al.26 considered a larger radius of curvature, larger contact area, and associated higher stiffness to be the factors contributing to the unfavorable energy absorption resulting in a higher risk of injury. Frontal impacts at the rim below the standardized lines are common, but some helmets fail to provide protection at these points. Optimization only at certain standardized zones might be disadvantageous at other impact points. Bland et al.26 concluded the protective effectiveness of cycling helmets to differ between real-world and standard conditions. The researchers point out that the current standards are important and testable areas should be expanded, not replaced. Although rotational acceleration has been known to be relevant in cyclist injuries, it is still missing in standardized testing today. Using full body simulation, Wang et al.24 confirmed that rotational acceleration is indeed increased when wearing a helmet. A standard incorporating such criteria could reduce the introduced effect by changing the helmet design. Deck et al.27, found no correlation between any of the global kinematic parameters and the simulated model-based brain tissue injury. Even if the effects may vary, helmets’ overall protective effect against injury is confirmed by Wang et al.24 in terms of the head injury criterion (HIC) and skull fracture for both curb-impact and skidding impact scenarios.

The results suggest that the shape and size of the head itself also play a key role in the protective effects of bicycle helmets. All three studies used a fifty-percentile male head and body forms. There is no reason to assume that helmets standardized for a specific head shape will be safest for individuals with other anthropometric characteristics. To be equitable, the future standards should not be exclusive to studying average men but should include much broader and more diverse population.

STAR protocol developed by the Virginia Tech Helmet Lab that incorporates the oblique impacts28, suggests that the voluntary indication of the protective capabilities provided to consumers should become more common.

Simulation and dummy tests are a powerful tool in assessing the effectiveness of bicycle helmets, but only to some degree, are capable to mirror the reality. Both approaches are qualitatively different. While dummy tests take place physically, simulations approximate laws of nature by computation. Surely, both cannot entirely account for the human nature, but as wisely used tools, they can provide more insights into the protective effectiveness of helmets.

Injury severity and cycling

To paint a more comprehensive picture and complement the above-mentioned analysis, it is important to examine the findings considering the injury severities among the cyclists. Behnood and Mannering29 investigated the severity of crashes between bicycles and motorized vehicles, while Myhrmann et al.30 used statistical models to estimate the severity of single-bike crashes by accounting for class-specific heterogeneity.

Research has shown that there are several key factors that can influence the severity of injuries of cyclists. These factors primarily include the characteristics of vehicles and drivers involved, as well as the environmental conditions such as the quality of the infrastructure or weather31. Past studies that have partitioned datasets under explanatory variables do not always account for unobserved heterogeneity in the data. Nevertheless, a wide range of models that are capable to account for unobserved heterogeneity have been widely used to handle big datasets with many variables32. Using advanced statistical methods, Behnood and Mannering29 analyzed the Los Angeles police-reported bicycle-vehicle crash data.

Over a seven-year period, 5,653 crashes were recorded, and one of the three injury categories noted: no visible injury, minor injury, severe injury (including death). For each crash the police data captured variables about the bicyclist’s and driver’s characteristics, their movement preceding the crash, environmental and location-related factors, and other variables such as helmet use. Myhrmann et al.30, on the other hand, took hospital’s emergency department data from Aarhus (the second most populous municipality in Denmark) (years recorded 2010–2015, N = 4250 injured bicyclists). Their dataset included information about road victim’s age, gender, helmet use, injury severity, as well as road type, surface condition, time and location of the crash. The authors combined the above-stated dataset with the road maintenance data, and their final dataset included 1,720 single-bicycle crashes with one of the injury severity categories: no injury, slight injury, severe injury.

Since it is not possible to collect all the factors that affect injury severity, it is essential to account for the unobserved heterogeneity in the data. This idea originates from the fact that all the explanatory variables do not account for the full extent of heterogeneity in the conditional mean (and possibly variance) across the dependent variable29.

Behnood and Mannering29 analyzed the Los Angeles data, allowing for crash-specific unobserved heterogeneity and estimated random parameters multinomial logit model with heterogeneity in the means and variances of random parameters on bicyclist’s injury severity. The model is estimated by simulated maximum likelihood using 1000 Halton draws.

Using the data from the emergency department in Demnark, Myhrmann et al.30 choose a latent class ordered probit model, where the injury severity of a cyclist was estimated using the multinomial logit regression (MNL). Model estimation was made using maximum likelihood estimation (MLE). Both studies computed the marginal effects to formulate their findings.

Furthermore, both studies found that there are several factors that contribute to an increased likelihood of serious injury for both—bicycle-vehicle crashes and single-bicycle crashes. Behnood and Mannering29 found for police-reported bicycle-vehicle crashes that driver’s race and gender, alcohol use by any party, older age, using the wrong side of the road, speeding driver, not wearing a helmet were found to increase the likelihood of a serious injury. Single-bicycle crashes were more likely to become severe crashes on shared road sections compared to bike lanes as well as on poorly maintained bike lanes or roads with little traffic and after dark. Regarding the helmet usage, not wearing a helmet in the Los Angeles study, was found to decrease the likelihood of a non-visible injury as well as was increase the likelihood of minor and major injuries29. The Aarhus hospital’s emergency department data showed no change for severe injury when wearing a helmet but a significant increase in the probability of not being injured and an inverse decrease in the probability of being slightly injured30.

Various factors affect the severity of bicycle crashes. Wearing a bicycle helmet was shown, in both studies, to significantly decrease the likelihood of slight/minor injuries and increase the likelihood of no (visible) injury.

link

Leave a Reply

Your email address will not be published. Required fields are marked *