Costs of Injuries Resulting from Motorcycle Crashes:
A Literature Review
Overview

Table of Contents / NHTSA home / Next

1.Executive Summary
Analysts at the Pacific Institute for Research and Evaluation, working under a NHTSA contract, identified 25 motorcycle safety studies for review. Most were published in the 1990s, and most addressed the subject of the costs of injuries resulting from motorcycle crashes. Most of the studies reviewed employed data from a state, locality, or medical institution in the United States. Seven of the studies linked data from multiple sources -- e.g., police crash reports and hospital records -- with varying degrees of success, in order to compile detail on individual crashes and their victims. A few other studies drew data from multiple sources in computing aggregate estimates of crash costs, but most of the studies worked from a single medical dataset. Most of the studies focused on either the benefits of wearing motorcycle safety helmets or the impact of state laws requiring the wearing of such helmets.

The studies that examined the impact of safety helmets or helmet laws consistently found that helmet use reduced the fatality rate, the probability and severity of head injuries, the cost of medical treatment, the length of hospital stay, the necessity for special medical treatments, and the probability of long-term disability. This work reinforces similar conclusions from earlier studies.

A number of the reviewed studies examined the question of who pays for medical costs. Only slightly more than half of motorcycle crash victims have private health insurance coverage. For patients without private insurance, a majority of medical costs are paid by the government. Some crash patients are covered directly through Medicaid or another government program. Others, who are listed by the hospital as “self-pay” status, might eventually become indigent and qualify for Medicaid when their costs reach a certain level.

A few studies examined the frequency of alcohol use by motorcycle crash victims. They found high rates of alcohol use and intoxication, particularly among unhelmeted crash victims.

While the literature has widely explored acute medical costs, research is sparse in the areas of long-term medical and work-loss costs. For victims of serious head injury, acute hospital care might be only the first stage of a long and costly treatment program. For many crash victims, lost wages from missed work days will outweigh medical costs. And for victims who are permanently disabled, their earnings might be reduced for the rest of their lives. More research is needed on these subjects to provide a more comprehensive picture of the full cost of motorcycle crash injuries.

2. Introduction
Good public policy decisions require accurate facts. But the challenges faced by the lay public when trying to read, understand, and use academic literature can often result in misinformation. Rightly or wrongly, findings about motorcycle safety, the value of helmet use, and who pays for helmet non-use can call forth controversies rivaling the findings on the causative link between gun proliferation and firearm deaths, between smoking and cancer, or between heavy drinking and injury.

The National Highway Traffic Safety Administration (NHTSA) commissioned this independent study, whose primary purpose is to summarize and critically review the literature on motorcycle injury costs published in the 1990s. The review supplements an earlier review of helmet effectiveness studies (US General Accounting Office 1990 summarized and reviewed in Appendix). It describes what we know, the gaps in what we know, and what we need to learn. It emphasizes studies of prevention approaches, including helmets, protective gear, and rider training. It also focuses on who pays the costs of motorcycle injury. (1)

Decision-makers,safety advocates, and motorcyclists need an accessible, scientifically credible review of the literature. This review addresses that need by identifying the best studies and summarizing their results. It also offers a handy, understandable summary of flawed studies and documents their major problems.

Background
Almost 4 million motorcycles are registered in the United States. Overall, they accounted for 11.3 billion person-miles of travel in 1998, averaging 2,640 miles per motorcycle (Federal Highway Administration. 1998 Highway Statistics, Washington DC: FHWA, 1999).

Motorcycles are by far the most fuel-efficient class of highway vehicle, at 50 miles per gallon (FHWA 1999). Because they are capable of high speeds but offer minimal occupant protection, they also are the most hazardous highway vehicles: they have the highest crash costs per person-mile (Miller et al. 1999). Helmets are the best-evaluated way to reduce motorcycle deaths and injuries. They are 29-35 percent effective at preventing motorcycling deaths and substantially more effective against deaths from brain injury. They also significantly reduce nonfatal brain injury (NHTSA 1989; GAO 1990; CODES Report to Congress). Annually, more than 2,500 motorcyclists die in traffic crashes. The number of motorcyclists dying on the highway fell to a historic low in 1997, and the number of motorcyclists injured in crashes has fallen by 40 percent from the 1990 level. However, since 1997, motorcyclist fatalities have increased over more than 40 percent and data indicate that motorcycle crash-related injuries are also increasing. Many motorcyclists believe rider training has been instrumental in reducing motorcycle deaths. This belief has been tested only modestly in the literature. Even less examined is the impact of protective clothing on motorcyclist injury. And there simply is no literature evaluating the training of auto drivers about driving safely around motorcycles.

Perhaps because many motorcyclists prize their independence, government attempts to reduce motorcycling deaths, injuries, and costs have met active resistance. Proven rider protection measures, notably helmets, are not required in many states. The federal government has twice enacted and then repealed laws designed to promote state helmet laws.

In the debates over helmets and motorcycle safety, both rider groups and safety advocates are increasingly framing their arguments in terms of statistics from published studies. The complexity of the methods required to analyze data covering limited subsets of motorcycle crashes, however, often makes the results of the studies difficult to interpret. Moreover, sometimes the studies themselves are flawed by data limitations or inadequate methods. In this context, the present survey is intended to aid in evaluating extant studies and arguments based upon them.

The next section summarizes our general approach. Section 4 provides an overview of the topics, data, and methods of the studies we reviewed, highlighting recurring problems and their solutions. Section 5 summarizes and analyzes the findings of the reviewed studies. And Section 6 suggests gaps in the current state of knowledge.

3. Literature Review Methods
Identifying and selecting articles for review
We began the process by considering a number of articles and sources recommended by our team's senior economists and by NHTSA staff. To supplement these, we conducted searches on Medline and Transportation Research Information Service (TRIS) for articles from the medical and transportation fields, respectively, that addressed motorcycle injury costs. (Appendix D shows the search terms that were used.) We also contacted researchers in other countries for suggestions of articles we might have missed in the United States. All together, we located references to nearly 200 publications from the motorcycle literature of the 1990s.

Only a few of these articles met the criteria for inclusion in this study. We dropped most articles that did not directly address the costs of motorcycle injuries, excepting a few articles that made unique contributions to the literature. We also eliminated studies that did not include human subjects, articles that did not present or review original research (e.g., documents that merely expressed the author’s opinion without presenting new facts or data), and studies that were not in the English language. We summarily eliminated more than one-third of the articles based on a cursory look at the title and abstract. We eventually narrowed the list to fewer than 80 articles that looked like they might meet our criteria, and we set about obtaining these articles. We found that most of these articles did not directly address the costs of motorcycle injuries, and we slowly further narrowed the list to the 25 studies on which this literature survey is based (see Table 1). (References to reviewed studies can be found with the reviews in Appendixes A and B, while non-reviewed studies are listed in Appendix C.)

Writing the reviews
Our reviews are all arranged in a common format. Each review begins with an article summary -- an expanded abstract edited for accuracy, clarity, and completeness. The summary describes the reviewed study’s objectives, study population, data, research methods, results, and conclusions. Results related to costs were highlighted and, when necessary, supplemented with tables. The remainder of the review critiques this information, pointing out methodological strengths and weaknesses. In a few instances, we included multiple articles in a single review, where there was overlap in authorship, data, and methods.

Every review was a joint product of two reviewers. It was drafted by a primary reviewer and then critiqued and supplemented by a second reviewer. Both reviewers had to sign off on a review before it was considered final. For articles written by PIRE staff, a non-PIRE reviewer drafted the review and retained final editorial control in order to ensure objectivity. In reviewing publications, the reviewers were guided by checklists, which served as reminders of the topics of interest and the methodological issues requiring scrutiny (see Appendix E). While completed reviews were sent to NHTSA, NHTSA staff did not offer or provide comments on the reviews in order to maintain the integrity of the independent review. NHTSA staff did offer editorial comments on the final report, but these comments did not change the context of the literature review.

4. Overview of Data and Methods Employed by Motorcycle Safety Researchers
Motorcycle safety issues
The most-studied topic in motorcycle safety is the efficacy of helmets in preventing and mitigating head injuries. Most of the publications evaluated for this project studied either helmet use or state helmet laws. The link between head injury and riding without a helmet is by now well established. Little research has been published, however, on helmet design. Other sorts of safety equipment, notably protective clothing for riders, are less studied. Other factors in motorcycle safety include alcohol intoxication and the roles of rider training and experience. Both of these issues have received some attention in the literature, but much less than helmets.

A further dimension in motorcycle safety analysis of interest for public policy is the cost of motorcycle crashes, especially insofar as this cost is borne by the public. It is easy enough to determine the costs of various safety equipment and programs, but it is more complicated to determine the costs that are saved when injuries are prevented or mitigated by successful safety measures. The major types of costs resulting from injury are 1) the cost of medical treatment, 2) the value of lost work, and 3) decreased quality of life. Most of the publications reviewed in this study made some effort to look at short-term medical costs. But few looked at long-term medical costs or work loss, and only two considered quality of life. Accurate medical cost estimation is complicated by institutional factors that differ between states -- and sometimes between hospitals or payers. The other costs are difficult to ascertain because their computation requires information from a lengthy follow-up period after the crash, particularly in cases where the victim is permanently disabled.

In addition to questions concerning the level of costs, policy discussions often involve the question of who bears the costs -- the rider, other drivers, private insurers, the government, or another party. If an institution is shown to bear a large share of the costs, then it will have a proportionate incentive to reduce those costs. A number of the publications surveyed in this study examined the question of who paid a crash victim's medical costs.

Classifying studies of motorcycle crash costs
The studies reviewed for this project can be categorized by 1) their purpose or focus and 2) by their data and methods. Of the 25 published studies of motorcycle crash costs that we reviewed, nearly half (12) focused on evaluating the benefits of helmet use. Another four addressed the effectiveness of state laws mandating universal helmet use by motorcycle riders. The remaining nine articles addressed a number of different topics -- methods, costs, epidemiology, training, and other safety measures.

A majority of the articles (15) employed state or local data from crash reports, hospitals, or mortality files. Eight of these studies linked data from two or more sources at the level of the individual rider in order to construct a more comprehensive picture of each crash episode than would be possible from a single data source. Such linkage is always a difficult process, and it was more successful in some instances than in others. Three of the state/local studies examined data from years before and after passage of a comprehensive helmet law in order to evaluate the impact of the law (Max et al., McSwain & Belles, Muelleman et al.). Five studies employed detailed records from a single hospital, while two studies were based on U.S. national samples. Three studies were based on data from other countries -- Germany (Hell & Lob), New Zealand (Begg, Langley, & Reeder), and Taiwan (Tsauo et al.).

Table 2.
Classifications of Reviewed Articles

* One of the 12 helmet-use studies employed two different data sources in its analysis – one from a single hospital, and one from the state/local level.

Data problems
Ideally, a study of motorcycle crashes would begin with a large representative sample -- or a complete census -- of all motorcycle crashes in the geographic jurisdiction and time frame being studied. Five of our 25 studies (Karlson & Quade, McSwain & Belles, Shankar et al., Wang et al., and NHTSA) attempt to do this by starting with all police-reported crashes, whether these crashes resulted in injury or not. Even this method will miss some minor single-vehicle incidents that did not result in injury or property damage, and which were therefore not reported, but it results in a much more representative survey of crashes than alternative methods. The rest of our remaining articles looked only at crash victims injured severely enough to be treated in specific settings -- emergency department, trauma center, or hospital -- or who died as a result of the crash. A crash-involved rider who was saved from injury by a helmet would not be captured by studies of this sort. This selection bias prevents these studies from gauging the benefits of helmet use in a comprehensive way.

Identifying motorcycle crash victims in medical datasets is usually accomplished by use of external-cause-of-injury codes (E-codes), which “record events, circumstances, or conditions that caused or contributed to the occurrence of an injury” (ICD-9-CM coding manual). E-codes allow for the identification of crash victims as motorcycle drivers or passengers, among other categories. However, in some cases where the hospital data coder cannot determine whether the motorcyclist was an operator or a passenger, the victim’s role might be coded as “unspecified,” making it impossible to identify him as a motorcyclist from the E-code. Begg et al. found that in New Zealand 1.5 percent of hospitalized motorcycle crash victims and 5 percent of motorcycle fatalities were coded as “unspecified,” which means that the usual method of identifying motorcyclists would have missed them. Moreover, no dataset has complete E-coding. Practically, a 95 percent rate of injury E-coding is the best that any dataset achieves, (2) and some do not even reach 90 percent. Datasets selected by E-codes alone will usually miss some cases.

The other major data problem besides getting a representative sample of motorcycle crash victims is obtaining data on all the variables of interest, including:

• Whether the rider was wearing a helmet or employing other safety equipment.

• Whether the involved operators were under the influence of alcohol or other drugs.

• The conditions of the crash -- e.g., vehicles involved, velocities, crash configuration, road surface, weather, visibility, time of day.

• Injury diagnoses, medical treatment, costs (including long-term), and sources of payment, for both crash survivors and fatalities.

• Patient outcomes -- e.g., how long the patient was out of work or disabled, and whether the patient fully recovered.

• Each rider’s riding patterns -- e.g., annual mileage, purpose for riding.

• Each rider’s age, experience, training, and insurance coverage.

In reality, researchers in motorcycle safety never have such comprehensive data. Some of the variables listed above (e.g., riding patterns) are rarely available, short of detailed follow-up surveys with individual crash victims. Other variables might be available but of low quality, with questionable accuracy or a lot of missing values (e.g., helmet use).

A particular problem for this project is that actual medical costs are usually not available. Hospital charges are often used as a proxy, but they are far from ideal for this purpose because 1) some important cost elements, including emergency transport, physicians’ fees, and medicines, are not included in hospital bills; and 2) hospitals over-charge, in the knowledge that Medicaid and most private insurers will pay only a fraction of the full charge. Although these two elements represent biases in opposite directions, the accepted wisdom is that hospital charges over-state medical costs, on average. (For further discussion of medical costs, see Appendix F: Glossary.)

Even if all of the desired variables could be found, they would not all be available from a single data source. Typically, helmet use, crash details, and alcohol or drug use are recorded, if at all, by the investigating police department or by the state agency responsible for highway safety. Medical and demographic information, meanwhile, are recorded by the hospital or other medical institution where patients are treated. And some information about riders might be available only from the riders themselves, through follow-up surveys. Therefore, it is necessary to have personal identifiers on each dataset, which allow records from the various data sources to be linked at the patient or incident level. Confidentiality concerns, however, often make it difficult to obtain identification information sufficient to permit linkage. Even when such concerns can be overcome, personal identifiers can be difficult to match from one dataset to another because of differences in coding conventions and information availability.

Methodological problems
Working with problem data requires well as caution in framing questions and drawing conclusions. A number of pitfalls await the analyst who does not allow for the sample’s nonrepresentativeness, missing data, and other problems.

A common finding in analysis of data on medical treatment of motorcycle crash victims is that, while helmeted victims are always found to have lower rates of head injury than nonhelmeted victims, they typically have higher rates of limb injury and roughly equal short-term medical costs. A hasty researcher might conclude that wearing a helmet does not reduce the victims’ expected medical costs. But this is actually a phenomenon of the biased sample resulting from selecting only those crash victims who are treated for injury in a trauma center or hospital. Such data omit crash victims whose head injuries were prevented or mitigated by wearing a helmet, unless they sustained injuries in some other part of the body. The sample is thus not representative of the entire population of crash victims. Every patient in the dataset must have been injured severely enough to require treatment -- if not in the head, then somewhere else. Thus, helmet-wearing patients in the dataset are likely to have serious injuries in some body location not protected by a helmet. While limb injuries do not carry the potentially serious long-term consequences of a brain injury, they can still be quite expensive to treat. Thus, a helmeted crash victim admitted to a hospital will probably show high medical expenses. This does not mean the helmet did not reduce the patient’s medical costs.

A handful of articles (Kelly et al., Murdock & Waxman, Offner et al., Rutledge et al.) acknowledged the selection bias and addressed it creatively, stratifying the data by the severity of a patient’s non-head injuries. While the adjustment did not completely compensate for the bias, it was sufficient to allow a clearer picture of the data and stronger conclusions.

Most studies report only the percentage of admissions covered by different payment sources, rather than the percentage of costs paid by different sources or the percentage of payments received from different sources. A small number of catastrophic cases involving brain or spinal cord damage can account for a highly disproportionate share of costs. Since the helmetless riders who sustain a majority of expensive head injuries are also more likely to be uninsured, the public pays for a higher percentage of dollars than claims (Kelly et al., Nelson et al.). Also, since Medicaid pays for catastrophic cases disproportionately often (as self-pay patients become indigent), it also may pay a much higher percentage of dollars than claims.

Some studies analyze injury survivors exclusively, or wisely separate them from fatalities. Others, however, commingle hospitalized survivors with victims who die after admission -- or occasionally even ones pronounced dead at the scene or in the emergency department. Despite the obvious severity of fatal injuries, they tend to entail low medical costs, since medical treatment is discontinued at death.

When an important data element is missing from some cases, many studies simply drop those cases without considering whether this might bias the results. Some studies do not even reveal how many cases were dropped, making it impossible to evaluate the size of the data exclusion problem. In some motorcycle studies, the rider’s helmet use status is frequently missing from the data. The better articles will include a “helmet-use unknown” column next to the helmeted and unhelmeted columns in tables and discuss the make-up of this group. In studies using linked datasets, large numbers of recorded crashes might be dropped if they cannot be successfully linked with medical or Department of Motor Vehicles (DMV) data. This could introduce bias into the results if a record’s linkage probability is correlated with key target variables. For example, if the rider’s license number is used to link to DMV records, all riders under age 16 and riders from other states would fail to link.

A final type of methodological shortcoming is the failure to adequately adjust for differences between years in multi-year data. Costs should be adjusted to a common year’s dollars -- a simple adjustment that is often overlooked in the literature reviewed. Similarly, annual counts of crashes and injuries should be adjusted for changes in exposure from one year to the next. The ideal exposure measure is vehicle miles traveled (VMT), but this is seldom available. Second-best exposure measures are the number of licensed motorcycles or, in states with special licenses for motorcyclists, the number of motorcycle driver’s licenses. With these latter measures, however, the analyst should also consider how the riding population is changing. For example, a decline in the number of riders could coincide with a much smaller decline in miles ridden if marginal motorcyclists drop out while the true motorcycling devotees continue riding.

5. Overview of Results
Acute medical costs
Of all the costs that can result from injury -- acute medical, long-term medical, short-term work loss, long-term disability, and lost quality of life -- only acute medical costs have received extensive attention in the motorcycle crash literature. Medical costs are of particular interest to policymakers, since they are often borne by someone other than the motorcyclist. Moreover, a proxy for acute medical costs -- the total hospital charge -- is readily available from most state hospital discharge datasets.

Comparison of dollar figures reported from different articles is problematic for a number of reasons. First, most articles do not specify what year’s dollars their costs are reported in. One might assume they are reporting in current dollars of the years their data come from, but this does not lead to a straightforward solution to the problem for studies that employ multi-year datasets. Even if all dollar figures could be converted to a common year’s dollars, they still would not be directly comparable. Costs vary between states, not only according to differences in the general cost of living, but also according to institutional differences in their medical care and cost regimes. In addition, different articles define their cost measures differently. Some include physician charges, along with hospital charges, but most do not. A few attempt to measure actual costs, but most settle for hospital charges, which can be substantially higher than actual costs. (In California, for instance, Max et al. found that charges were 2.3 to 2.6 times as high as costs.) We will, therefore, report dollar figures as they appear in the respective articles and sidestep the comparability problem by focusing on ratios computed from each study’s costs to analyze the impacts of helmet use and head injury.

Table 3 summarizes the hospital charges reported by various articles by treatment level and helmet status. Despite the range of years represented (1984-1992), the charges within each treatment level were of similar magnitude. As one might expect, charges were lower for fatalities, since treatment ceases at death. Similarly, average charges are relatively low for patients treated in the Emergency Department (ED), as this category included many cases less severe than those treated in trauma centers or admitted to the hospital. It should be noted that the ED category includes patients who were subsequently admitted to the hospital, as well as those who were released. Trauma centers are hospitals that specialized in injury care, and they disproportionately treat the most severely injured patients.

Every study that examined the question found that nonhelmeted patients incurred higher average hospital charges than helmeted patients. The extent of this difference, however, varied widely, from less than 10 percent in two studies to about 200 percent in two others. Three articles reported a difference of about 30 percent, with four articles reporting a higher figure and four lower. Therefore, 30 percent can be taken to represent both the median and the mode of the distribution of ratios for the twelve studies.

Table 3.
Average Hospital Charges per Case
by Helmet Status and Place of Treatment

*Includes physician charges.
**Admitted to trauma center or died.

The other question examined by a number of articles was how the costs of a head injury compare with the costs of other injuries. Each of the four articles that examined this question found that head-injured patients incurred greater hospital charges, on average, than those without a head injury. The charge differences reported ranged from 79 percent to 178 percent, as shown in Table 4. Shankar et al. also examined this question, but their results are presented by helmet use and lower limb injury, as well as head injury, making them harder to summarize. For crash victims without a helmet, they find that hospital charges are higher with a head injury. But for victims who were wearing a helmet, charges are higher without a head injury, perhaps because head injury severity was moderate.

Table 4.
Average Hospital Charges per Case by Head Injury Status

1 Includes physician charges.
2 Estimated costs, rather than charges.
3 Brain injury.
4 Severe head injury.

Max et al. estimated that aggregate hospital costs of motorcycle-related head injuries in California fell from $36.6 million in 1991 to $15.9 million in 1992, after the state’s universal helmet law went into effect. During this time, the share of medical costs accounted for by head injuries fell from 46 percent to 31 percent.

Begg et al. take up a different question. They compare the costs of traffic collisions, traffic noncollisions, and nontraffic crashes for motorcyclists in New Zealand. This is an important distinction for motorcycles, unlike most other vehicles. Because of a motorcycle’s relative instability, injuries can frequently occur without collision with another vehicle or object. And motorcycles are often ridden off-road. Begg et al. find that traffic collisions incur much higher average medical costs (NZ$6,942) than traffic noncollisions (NZ$3,342) or nontraffic crashes (NZ$2,617).

Nonmonetary measures of medical outcomes
A number of studies reported on various outcomes of motorcycle-related injuries that, while not expressed in monetary terms, would have an impact on the cost of medical treatment or other social costs.

A few studies provided fatality rates. Stutts et al. reported that 7.1 percent of their trauma registry sample died, compared with just 2.7 percent of the broader crash file. Braddock et al. found a population-based fatality rate of 1.2 per 100,000. They also found that the fatality rate was 3.4 times higher for nonhelmeted riders. Four other articles also looked at the impact of helmets on fatality rates. The fatality rate was higher for nonhelmeted patients in Rowland et al.’s hospital-admitted sample (4.7 percent vs. 2.9 percent), Kelly et al.’s ED sample (7.35 percent vs. 1.72 percent), and McSwain & Belles’s Louisiana crash sample (6.2 percent vs. 1.6 percent), while in Offner et al.’s trauma center sample, nonhelmeted patients had a lower fatality rate (7.7 percent vs. 9.1 percent), though this last difference was not statistically significant. Miller et al. estimated that 100 percent helmet use would have reduced the number of motorcycling fatalities in 1991 from 2,760 to 2,270. NHTSA found that wearing a helmet reduces the probability of fatality by 35 percent.

Given the long recovery times and potential long-term nature of motorcycling injuries, an important crash outcome for survivors is disability. Bried et al. found that 12 of their 71 hospital-admitted patients did not return to baseline functioning, while the other 59 fully recovered in an average of 23 weeks. Murdock & Waxman found that nonhelmeted patients treated in a trauma center were more likely to be disabled than helmeted patients (9 percent vs. 5 percent). Disability results not only in increased medical costs, but in reduced lifetime earning potential and reduced quality of life.

Two articles consider the effect of helmet use on injury severity scores (ISS). Kelly et al. found that the average ISS is higher for nonhelmeted ED patients (11.9 vs. 7.0). Rowland et al. reported that 4.6 percent of nonhelmeted hospital patients had ISS of 16 or greater, compared with 2.3 percent of helmeted hospital patients.

Several articles looked at the relationship between head injury and helmet status. As shown in Table 5, wearing a helmet was consistently found to be associated with a lower probability of head injury. The figures in the table cannot be compared in a straightforward way. Besides their origins in different treatment settings, which would be associated with different injury severity levels, they are not all measuring exactly the same thing. The final column provides detail on what is being measured, usually involving a minimum severity level. In general, the ratio of nonhelmet to helmet head injury percentages is higher for head injuries of higher severity levels. This implies that wearing a helmet is associated not only with fewer head injuries, but with less severe head injuries, as well.

Table 5.
Shares of Motorcycle Crash Victims Who Suffer
Head Injury by Helmet Status and Place of Treatment

Other articles considered head injuries from other angles. NHTSA found that wearing a helmet reduces the probability of brain injury by 67 percent, while Weiss’s model predicts that helmets lead to a 42 percent increase in the number of riders with no head injury. Max et al. found that the hospitalization rate for head injury in California dropped from 230 per 100,000 registered motorcycles in 1991 to 129 in 1992, after the state’s universal helmet law went into effect. Hell & Lob found that, of all crashes with AIS of 2 or greater, 43 percent involved head injury. Braddock et al. found that 22 percent of hospital-admitted injuries involve the head, neck, or spine. Bried et al. reported that head injury was associated with an increased need for ICU and ventilator treatment.

In a similar vein, a couple of articles looked at the relationship between helmets and spine injuries. Some opponents of motorcycle helmet laws, citing a badly flawed study by Goldstein (reviewed in Appendix B), have suggested that wearing a helmet increases the risk of spinal injuries, particularly in the neck region. But Orsay et al. found that, among trauma center patients, the percentage with spinal injury was twice as high for patients who did not wear a helmet as for those who did (8 percent vs. 4 percent). Max et al. found that the hospitalization rate for spinal injury in California dropped from 15 per 100,000 registered motorcycles in 1991 to 10 in 1992, after the state’s universal helmet law went into effect.

Some of the articles that focused on hospital-admitted samples looked at the length of stay. For hospital-admitted patients, length of stay is a key determinant of medical cost. Bried et al. found the average length of stay to be 13 days. Bray et al. reported an average length of stay of 21.2 days, but their sample consisted exclusively of open fracture patients, whose injuries are of high severity. Offner et al. reported that the average length of stay was 15.5 days for nonhelmeted patients and 10.8 days for helmeted patients. Similarly, Rowland et al. found the average lengths of stay to be 12.6 days without a helmet and 9.9 days with a helmet, while McSwain & Belles’s Louisiana sample showed 11.8 days for nonhelmeted and 5.8 days for helmeted patients.

Tsauo et al., in their Taiwan study, reported a mean length of stay of 14.3 days for patients with a helmet. For patients without a helmet, they reported average lengths of stay depending on the Glascow Outcome Score, which measures head injury severity. Stays ranged from 17.2 days for a patient who fully recovered to 85.6 days for patients who ended up in a vegetative state. Those who died averaged 7.8 days.

A few articles considered the impact of helmets on the probability of various sorts of medical treatment. Kelly et al. found that ambulance transport to the ED is required for 63.4 percent of nonhelmeted patients, but only 46.4 percent of helmeted patients. Murdock & Waxman found, in their trauma center sample, that 19 percent of unhelmeted patients required a ventilator, compared with 8 percent of helmeted patients. Offner et al. reported that, in their trauma center sample, 36 percent of nonhelmeted patients required intubation, compared with 22 percent of helmeted patients. Rowland et al. found that nonhelmeted patients required readmission more often (5.2 percent vs. 2.3 percent) and were also transferred to another hospital for further treatment more often (13.1 percent vs. 8.1 percent).

Two other articles considered the probability of special medical treatments for motorcycle crash victims without relating them to helmet use. Bried et al. found that 66 percent of their hospital-admitted sample required surgery, and 36 percent required a second surgical procedure; 52 percent were held in the ICU for an average of 3.3 days; 30 percent required a ventilator, for an average of 6 days; 38 percent required blood transfusions. They also found that the average hospital-admitted patient sustained 2.4 fractures. Rutledge & Stutts found that 6.8 percent of patients admitted to North Carolina trauma centers were discharged to a nursing home or rehabilitation center.

Work loss and other costs
Apart from acute medical costs, relatively little attention has been devoted to the costs of motorcycle injury in the literature. Yet Miller et al. estimate that total medical costs (not just acute) account for only about 6 percent of the total costs of motorcycle injuries. Work loss represents 29 percent of the total cost, while pain, suffering, and reduced quality of life represent 63 percent.

The only detailed breakdown of total medical charges and costs appears in the article by Max et al. Most other authors considered only acute hospital charges -- i.e., emergency department and initial hospitalization charges. Max et al. found that these charges represent only about 70 percent of the total medical charges for a motorcycle injury. Other charges are for emergency medical services, readmissions to the hospital, professional fees, ambulatory care services, rehabilitation, and nursing home care. Costs are less than half the size of charges -- Max et al. find cost/charge ratios ranging from 0.42 to 0.46 in their three-year study period in California. The initial hospitalization and ED treatment account for only 67-68 percent of the total medical cost.

Max et al. also considered the productivity losses resulting from California motorcycle fatalities. About 80 percent of the motorcyclists who died were under 40 years of age, so the years of potential life lost were quite large. In 1991, 512 fatalities resulted in an estimated 24,435 years of potential life lost, or 3,824 years per 100,000 registered motorcycles. At a discount rate of 3 percent, the productivity losses came to $603 million. In 1992, after enactment of a universal helmet law, 327 fatalities resulted in 15,108 lost years (2,591 per 100,000 registered motorcycles), valued at $380 million. The helmet law coincided with a 37 percent drop in productivity losses due to motorcycle fatalities.

The fatalities examined by Max et al. do not exhaust the productivity losses resulting from motorcycle crashes. Survivors of serious injuries will lose work while they are being treated at a hospital or rehabilitation facility, and some will continue to convalesce at home for some time after discharge. The worst injuries can result in long-term disability, which might impair the crash victim’s earning capacity for life. Even without these nonfatal work loss costs, Max et al. estimate the cost of productivity losses at roughly six times the cost of medical treatment. Miller et al., who attempt a more comprehensive national estimate of productivity losses due to fatal and nonfatal injury, put work loss costs at nearly five times medical costs.

Only two studies considered the cost of pain, suffering, and lost quality of life. Miller et al. estimated the total lost quality of life from US motorcycle crashes in 1993 at $11.5 billion -- about 80 percent larger than medical costs and productivity losses combined. Thus, Miller et al. estimate that 63 percent of the total costs of motorcycle crashes come from lost quality of life. Wang et al. estimate “economic” costs (medical, work loss, and property damage) at $6.5 billion per year in 1989-93 and “comprehensive” costs at $22.6 billion. The difference between Wang et al.’s two measures, $16.1 billion, is the estimated cost of intangible losses, such as pain and suffering. Thus, Wang et al. estimate intangible costs as 71 percent of the total costs.

The scant attention paid to non-medical costs in the motorcycle crash literature is not in proportion to their importance to society. Fatalities, in particular, incur low average medical costs, since medical treatment ceases at death. The high cost of motorcycle fatalities, therefore, is not captured by medical costs. Society’s loss from the deaths of motorcyclists is better measured by the lifetime of work that they will not be able to perform.

Having estimated comprehensive costs of motorcycle crashes, both Miller et al. and Wang et al. proceed to report costs in several different ways in order to provide a clearer picture of the social costs of motorcycle crashes. Table 6 shows some of their more interesting cost calculations. (The tables accompanying the respective reviews of the two articles in the appendix contain many more such figures.) In every instance, costs are higher for motorcycles than for all vehicles (a category dominated by cars, but also including trucks and vans, as well as motorcycles). The difference is relatively small for costs per vehicle lifetime because motorcycles do not last as long as most other vehicles. The most revealing figures are in the first row: The expected costs of crashes come to about $2 per mile for motorcycles -- over 10 times the cost per mile for other vehicles.

Table 6.
Selected Measures of Comprehensive
Costs from Two Articles (1997 dollars)

NA = Not Available
* = Cor/UAN

Who bears the burden of the medical costs of motorcycle crash?
Most of the medical costs resulting from motorcycle crashes are borne not by the victims, but by private insurers and the government. Therefore, both insurers and governments, as well as the general public who buys insurance policies and pays taxes, have an interest in ascertaining not only the size of these costs, but also how their impact is distributed among payers.

A typical state hospital discharge dataset gives a breakdown of a dozen or so different categories of expected payers, normally including Medicare, Medicaid, other government, worker’s compensation, Blue Cross/Blue Shield, commercial insurance, HMO, PPO, self-pay, charity, and other. But some states break out other payers (e.g., other federal programs or state-specific programs) while grouping the various non-government payers together. The consensus solution in the motorcycle literature is to group all payers into three categories:

• Public or government, always including Medicare and Medicaid and sometimes including other government payers, such as the Indian Health Service.

• Private or commercial insurance, which usually includes worker’s compensation, as well as Blue Cross and various forms of managed care.

• Uninsured or self-pay.

Some detail is lost in these groupings, but they make for easier comparisons between different states and time periods.

It should be noted that the standard practice is for hospitals to record the expected payer at the time of admission or discharge. The actual ultimate payer might be different. A 1986 study published by the New York State Department of Health found that the ultimate payer matched the expected payer in more than 90 percent of cases where the expected payer was Medicare, Medicaid, or commercial insurance, but the match rate was lower for other expected payers - notably just 67 percent for self-pay. (“The Accuracy of the Expected Primary Source of Reimbursement as Reported by New York State Hospitals,” Gerald I. Kaufman, Albany, NY: 1986.)

Stutts et al. compared the expected payers of victims of motorcycle injuries to those of other road trauma patients. They found that motorcyclists were more likely to be uninsured (42.7 percent vs. 35.5 percent) but less likely to rely on Medicare or Medicaid (7.9 percent vs. 13.9 percent). This should not be surprising, given that motorcyclists tend to be younger than the driving population at large, and therefore rarely eligible for Medicare, but old enough to drive, and therefore rarely eligible for Medicaid coverage of children. Motorcyclists were found to be about as likely to be privately insured as other road trauma patients (49.4 percent vs. 50.9 percent).

Table 7 shows the cost distributions reported in the articles we analyzed. The differences in distributions are to be expected, given the wide variations in institutional arrangements for health coverage between states. The share of patients whose expenses were to be paid by government programs ranged from 7 percent to 38 percent. The range for commercial insurance was 44-63 percent. And patients were recorded as self-payers in 14-43 percent of cases.

Table 7.
Expected Payer Distributions for
Hospital-Admitted Motorcycle Crash Patients

* These three studies provide the distribution of charges by payer. All other figures in this table show the distribution of patients by payer. (Note that Shankar & al. and Max & al. calculate the payer distribution both ways.)
** Excludes patients with unknown helmet use status. Numbers do not sum to 100% because of rounding.

The key distinction is between commercial insurance and all else. Many uninsured patients are never able to pay their bills, and the costs are ultimately divided between the government and the hospital. Only patients with adequate private health insurance coverage represent no medical cost burden to the government. This means that in some states, the government is probably paying more than half the medical costs of injured motorcyclists.

A number of studies found that nonhelmeted patients were more likely to be uninsured. Kelly et al. found that 54.5 percent of nonhelmeted patients were uninsured, compared to 44.2 percent of helmeted patients. Smaller differences in the same direction were found by Nelson et al. (33 percent vs. 26 percent), Offner et al. (7 percent vs. 5 percent), Orsay et al. (27.8 percent vs. 21.1 percent), and Shankar (30.6 percent vs. 29.3 percent). Rutledge & Stutts, on the other hand, found no difference (40 percent vs. 40 percent). The finding of correlation between lack of insurance and riding without a helmet probably reflects the fact that younger riders, who are more likely to ride helmetless (Orsay et al.), are also more likely to be uninsured. The implication is that, insofar as nonhelmeted riders incur greater medical costs, the burden of these expensive cases will fall disproportionately on the public.

Two studies provided full payer distributions by helmet status. As shown in Table 8, both studies found that helmeted patients were more likely to be privately insured, while nonhelmeted patients were more less likely to be covered by the government privately insured.

Table 8.
Expected Payer Distributions for Hospital-Admitted
Motorcycle Crash Patients by Helmet
Use Status (Helmeted / Nonhelmeted)

Alcohol intoxication and motorcycle crashes
A number of studies reported on alcohol intoxication of motorcycle crash victims. Bried et al. found that 24 percent of patients were legally intoxicated, while Bray et al. found that 55 percent of those tested were intoxicated at admission. Most found that helmetless riders were more likely to have been drinking. Nelson et al. reported the most extreme finding, based on their sample of fatalities: 51 percent of unhelmeted victims were legally intoxicated, compared with just 18 percent of helmeted victims. Less extreme differences were reported by Offner et al. (39 percent vs. 27 percent) and Murdock & Waxman (32 percent vs. 21 percent). Murdock & Waxman also reported percentages of patients with positive, but not necessarily illegal, BAC levels: 47 percent of unhelmeted patients and 29 percent of helmeted patients. Shankar reported that helmet usage in his sample was 35 percent overall, but only 16 percent among those with a history of drug or alcohol conviction. Finally, Stutts et al. found police-reported alcohol involvement in 14.4 percent of patients in the crash file, compared with 24.3 percent of those in the trauma registry, which suggests that alcohol use might be associated with more serious injuries.

Caution must be exercised in interpreting all of these results on alcohol use. Except where state law or hospital policy requires that all patients be screened for alcohol use, one might reasonably surmise that the patients who are tested at the hospital are those who are suspected of alcohol use, while those not suspected are not tested. Insofar as the reported percentages reflect only those tested, they would tend to exaggerate the actual rates of alcohol use by motorcyclists. Therefore, all of these reported percentages of alcohol use and intoxication must be interpreted as upper bounds.

In addition, it should be noted that rates of both alcohol use and helmet use among motorcyclists vary over time and between states, depending on state laws and their enforcement, as well as cultural attitudes towards alcohol use and drunk driving. Analysis of 1990-98 FARS data shows trends towards lower rates of alcohol use and higher rates of helmet use among fatally injured motorcycle operators during the 1990s.

The unique contributions of Hell & Lob
Wolfram Hell and Günter Lob, two doctors from Munich, Germany, took up a number of questions that were not dealt with in any other articles we encountered in the course of this project. Working with a sample of 210 injury victims from 173 motorcycle crashes, they analyzed each crash and each injury in minute detail. They not only examined the severity and body region of each injury, but they also looked at the relationship between these injury dimensions and crash types -- i.e., the motorcycle’s relationship to its crash opponent (if any) and what happened to the rider(s). In addition, they considered the benefits of protective leather clothing and helmet performance in crashes. It should be noted that their sample is probably not a representative sample of all motorcycle crashes. In their sample, 50 victims (24 percent) died, while on average only about 2 percent of police-reported crashes in Germany result in death.

Of all crashes involving injuries of moderate or greater severity (AISŽ2), 43 percent involved injuries to the head, followed by lower extremities (37 percent), upper extremities (30 percent), thorax (25 percent), abdomen (16 percent), spine (12 percent), and pelvis (8 percent). In a US study based on 1992 GES data, Miller and Galbraith (1994) found a much higher share of injuries to the lower extremities and a lower share of injuries to the trunk.

Severity and body region vary greatly depending on the crash type. Three of the eleven crash types accounted for 41 of the 50 fatalities: head-on collisions 1) with the front of an opposing vehicle, 2) with the side of an opposing vehicle (without fly-over), and 3) with a stationary object (e.g., tree, crash barrier). These same three crash types also had the highest rates of head injury and the highest ISS.

Two-thirds of crashes involved an opponent; the other one-third were solo. Two-thirds of crash opponents were cars, and one-sixth were commercial vehicles. Most collisions with cars (90 percent) struck the front or side of the car, while a plurality of collisions with commercial vehicles (48 percent) struck the rear end. The motorcycle’s first impact point was in front in 67 percent of cases, and in the side in 29 percent.

Of 147 riders whose clothing was analyzed, 45 percent wore a leather jacket and leather trousers, 29 percent wore only a leather jacket, and 26 percent wore no leather safety clothing. Wearing leather was found to be correlated with lower severity of injuries to the extremities, particularly the legs.

6. What Don't We Know Well Enough?
Some topics in motorcycle safety, such as the benefits of helmet use and acute medical costs, have been studied extensively. Many other interesting topics, however, have received little coverage. The primary culprit for these research gaps is the absence of necessary data.

As in so many areas of safety research, a particularly troublesome gap occurs with data on the long-term consequences of injury, such as disability, rehabilitation, and long-term care. These long-term effects are particularly important for analyzing the impact of helmet non-use, since head injury can result in permanent brain damage. But data on both the frequency and the cost of such injury consequences are scarce. The few sources that could provide such data are not E-coded, and hence not easily associated with motorcycle crashes (or, indeed, with any particular cause of injury).

This raises the broader issue of the need for more comprehensive use of E-codes in medical datasets. Many national datasets available now employ E-coding only partially, if at all. In addition, it might be helpful if the E-codes used for identifying motorcyclists were modified to include motorcyclists whose status as driver or passenger is unknown. The ICD-9-CM coding system as presently structured does not accommodate the identification of motorcyclists if their driver/passenger status is unknown.

Data on other factors that might affect crash probability or injury severity are also lacking. For instance, some motorcycling groups have claimed that rider training is responsible for reductions in injury rates, but it is impossible to determine from current datasets which motorcyclists underwent rider training. The issue, therefore, can be addressed only at a very aggregate level. Data on the motorcycle operator’s riding experience and riding patterns are similarly unavailable. It would be useful, among other things, to know how many miles per year each motorcycle operator rides, in order to have a good measure of exposure. Data on the use of protective clothing are also not reported in US datasets.

All of the problems listed so far involve individual items of data. Two other data shortcomings involve the structure and composition of datasets.

Given that the questions typically addressed in analysis of motorcycle injuries relate to interventions, they necessarily involve changes over time between when the intervention did and did not exist. For example, in comparing the impact of a helmet law, one needs to look at data before and after implementation of the law. Hence, it is necessary to have longitudinal data on crash costs, which are presently lacking.

Most studies examine only cases in which motorcyclists died or received medical treatment. No single-source datasets contain all the information needed to do a comprehensive study encompassing all motorcycle crash victims. It is therefore difficult to estimate how many crash victims are not injured. Insofar as an intervention might prevent injury, part of its success will not be captured by the medical data. Unless more comprehensive datasets become available, the only solution is linkage of data from different sources. The CODES project has provided an initial glimpse of how such comprehensive linked datasets might be created and used effectively.

7. Conclusions
Most of the studies reviewed in the course of this project examined the impact of safety helmets or helmet laws on motorcycle injuries. These studies consistently found that helmet use reduced the fatality rate, the probability and severity of head injuries, the cost of medical treatment, the length of hospital stay, the necessity for special medical treatments (including ventilation, intubation, and follow-up care), and the probability of long-term disability. This work reinforces similar conclusions from earlier studies.

A number of the reviewed studies examined the question of who pays for medical costs. Only slightly more than half of motorcycle crash victims have private health insurance coverage. For patients without private insurance, a majority of medical costs are paid by the government. Some crash patients are covered directly through Medicaid or another government program. Others, who are listed by the hospital as “self-pay” status, might eventually become indigent and qualify for Medicaid when their costs reach a certain level.

While the literature has widely explored acute medical costs, research is sparse in the areas of long-term medical and work-loss costs. For victims of serious head injury, acute hospital care might be only the first stage of a long and costly treatment program. For many crash victims, lost wages from missed work days will outweigh medical costs. And for victims who are permanently disabled, their earnings might be reduced for the rest of their lives. More research is needed on these subjects to provide a more comprehensive pictu re of the full cost of motorcycle crash injuries.

1. A companion report will provide data that fill some knowledge gaps on issues surrounding motorcycle insurance. (BACK)

2. A handfull of state hospital datasets claim 100 percent E-coding, but they achive it by making extensive use of "unspecified codes, which are of little help to the researcher. (BACK)