The unformatted htm text below is from
|Copyright © 2004 by Leonard Evans|
|The text below is © Leonard Evans. Portions totaling not more than 400 words my be freely used with appropriate citation.|
|For more extensive use, permission in writing must be first obtained by emailing the author|
6 Gender, age, and alcohol effects on survival
This html version contains only the text (no figures, tables equations, or summary and conclusions). To check printed book appearance see pdf version of Chapter 1 or pdf version of Chapter 16.
This chapter has nothing to do with the risk of being involved in a crash. Instead, it deals with the risk of surviving, given that a crash has occurred. Drivers involved in identical crashes can have different risks of dying due to factors that affect the human body's ability to survive a given impact. The chapter is therefore devoted to examining the fragility of the human body when subject to blunt trauma from physical impacts, also called physical insults.
We investigate how the risk of death from the same impact depends on gender, age, and alcohol consumption. Although such fragility effects originate in fundamental human physiology, and apply to blunt trauma insults from sources other than traffic crashes, it is only traffic crash data sets that provide sufficient numbers of fatalities to determine quantitative relationships.
Gender and survivability
If a female and a male suffer similar potentially lethal physical impacts, which of them (other factors being equal) is more likely to die? This question cannot be answered by standard epidemiological methods because adequate samples of sufficiently similar cases are unavailable, and likely to remain unavailable. Even though the FARS data set (Chapter 2) codes hundreds of thousands of female and male fatalities, such data do not immediately answer the question. To illustrate the problem, consider that the most common type of US fatal crash involves one vehicle (Table 3-3, p. 49) containing one person, the driver. Examining such crashes when the driver is female reveals that 100% of them are killed. If they were not killed, the case would not be in FARS. The corresponding male case similarly shows 100% of male drivers killed. The FARS data do show about three times as many male deaths as female deaths in single-vehicle, single-occupancy crashes. This provides no information on how gender affects outcome, given that a crash occurs. The question was answered using a technique that extracts the required information from FARS data.
Double pair comparison method
The double pair comparison method was devised specifically to make inferences from FARS data. The method effectively isolates the influence of a particular factor of interest (in the present case, gender) from the multitude of other influences that affect fatality risk in a crash. The method focuses on vehicles containing two specific occupants, at least one being killed. We refer to one as the subject occupant, and aim to discover how some characteristic of the subject occupant affects that person's fatality risk. The other, the control occupant, serves to standardize conditions in order to estimate risk to the subject occupant.
The method is described below for the specific case in which the subject occupant is a car driver and the control occupant is a male passenger seated in the right-front seat. The aim is to determine how the driver's gender influences the driver's fatality risk in a crash.
Two sets of crashes are selected. The first contains cars with a female driver and a male passenger, at least one being killed. This first set of crashes provides
It might seem that r1 immediately measures how risk depends on gender. As driver and passenger are involved in the same crash, factors like impact speed, or type and properties of object struck (tree, vehicle, etc.), apply equally to both occupants. Factors that influence the risk of crashing, such as driver behavior, change sample sizes but should not systematically affect r1. However, a factor that could contribute to differences in risk between drivers and passengers is the different risk associated with different vehicle seats (p. 53-57).
To correct for this, a second set of crashes uses male driver subjects accompanied by male control passengers, at least one being killed. That is, the subject gender is different from the first set of crashes, but the control characteristics are the same. These crashes provide
Dividing the two ratios gives
Subject to assumptions that are likely to be more than adequately satisfied, the quantity R measures the risk of death to a female driver compared to the risk of death to a male driver, other factors being essentially the same.2 The crash conditions are effectively standardized because the female and male drivers experienced their injuries in a mix of crashes that posed similar risks to accompanying male passengers.
The control occupant does not enter directly into the result. Because of this, many separate estimates can be calculated using various control occupants. Combining estimates based on many controls helps diminish potential confounding due to differences that may exist between subject and control occupant with regard to such factors as age and safety belt use. The basic assumptions of the method require that the probability of a passenger death should not depend (in the present example) on the gender of the driver. This assumption would be violated if, for example, the same physical impact was
less likely to kill a passenger traveling with a male driver than one traveling
with a female driver. Departures from this assumption could arise if, for example, passengers traveling with male drivers tended to be younger than those traveling with female drivers. The biasing influences of such potential confounding can be reduced by dividing control subjects into gender
and age categories, thus insuring that passengers of similar age and gender accompany the female and male drivers being compared. As the use of
such occupant-protection devices as safety belts or helmets affects fatality risk, the control occupant should have the same use in the first and second set
The calculations are described below using the example of comparing unbelted car-driver fatality risk for females aged 38-42 to that for males in the same age interval (call them 40-year-old drivers). For the control occupant we first choose unbelted male right-front passengers aged 16-24, hereafter referred to as 20-year-old male passengers. The 1975-1998 FARS data give A = 90 female drivers aged 40 were killed while traveling with 20-year-old male passengers, while B = 36 male passengers aged 20 were killed while traveling with female drivers aged 40. These give a 40-year-old female driver to 20-year-old male passenger fatality risk ratio r1 = 2.500. This departs substantially from unity because, as we quantify later, fatality risk from the same impact depends strongly on age. For the second set of crashes, C = 244 and D = 133, giving r2 = 1.835. The ratio of r1 to r2 gives R = 1.363, so this combination of subject and control estimates that females are 36% more likely to die than males from the same physical impact.
Deriving relationship from large numbers of data
The above example provides the first of the eight estimates for 40-year-old subject drivers shown in Table 6-1. Summing the appropriate columns in Table 6-1 shows 3,038 driver fatalities (930 female and 2,108 male) and 2,870 right-front passenger fatalities. The conclusion from Table 6-1 is that 40-year-old female car drivers are (19.9 ± 7.8)% more likely to be killed than 40-year-old male drivers in similar severity crashes. This value provides the point for age 40 (plotted at age = 40.5) in the top left graph in Fig. 6-1.
Table 6-1. Female to male fatality risk, R, for 40-year-old unbelted car drivers.
The other points plotted in the top-left graph in Fig. 6-1 are based on extracting data in the same form as Table 6-1 for other driver ages, using a total of 14,873 female and 47,989 male driver fatalities. The total number of subject fatalities is given on this and subsequent graphs. All errors are standard errors. - For driver subjects, there are no estimates at ages below the age of licensure due to too few cases (but far from zero cases, see Table 9-5, p. 228). When passengers are subjects and drivers serve as controls, estimates at younger ages are available, as shown in the other five graphs in Fig. 6-1. Belted includes the use of any restraint system, such as a baby or infant seat.
A parallel process produces corresponding results for occupants of light trucks (Fig. 6-2) and for motorcycle passengers (Fig. 6-3). There were insufficient female motorcycle-driver fatalities to perform the analysis for motorcycle drivers.
If there were no systematic differences between male and female risk, then all of the data in the 14 graphs in Figs 6-1 through 6-3 would distribute randomly around the value R = 1.0 (marked by a dashed line). Instead, clear systematic departures are apparent in every graph, with the departures being similar from graph to graph.
None of the six individual graphs in Fig. 6-1 departs systematically from their collective trend. It is therefore appropriate to combine all these data to obtain a best estimate for car occupants (top graph in Fig. 6-4). The values plotted are the weighted averages of the values plotted at the indicated ages in Fig. 6-1. The other two graphs in Fig. 6-4 show corresponding information for light trucks and motorcycles. Weighted average values of R at age 20 are 1.285 ± 0.027 for cars, 1.241 ± 0.052 for light trucks, and 1.312 ± 0.078 for motorcycle passengers. Female risk exceeds male risk by amounts that are not systematically different depending on which of the three vehicles provides the data. The R values for occupants of each of the three vehicle types are, to within their error limits, consistent with the weighted average of 1.279 ± 0.023
As results for individual vehicle categories do not depart systematically from their collective trend, it is appropriate to combine the data in Fig. 6-4 to produce Fig. 6-5. Each point plotted can be considered the weighted average of the (up to 14) values from the 14 occupant categories, or the mathematically identical weighted average of the values from each of the three vehicle categories. The relationship in Fig. 6-5 is the best estimate of how the risk of death from the same impact depends on gender.
The values in Fig. 6-5 at ages 20, 25, 30, and 35 are 1.279 ± 0.023, 1.301 ± 0.028, 1.291 ± 0.033, and 1.287 ± 0.038. These values show consistently that, between ages 20 and 35, female risk exceeds male risk by (28 ± 3)%. From about age 10 to the late 50s, female risk exceeds male risk by amounts that depend on age. For ages below 10 or above 60, the data provide no indication of clear differences, although there is a weak suggestion that older men might be more vulnerable than older women.
A physiological gender-dependent difference
Subjects in the 14 categories are killed by a wide variety of impact mechanisms. For example, fatalities to belted vehicle occupants usually result from impacts with the vehicle interior, while motorcyclist fatalities result from impacts into objects external to the motorcycle. The absence or presence of steering wheels, safety belts, helmets, cushioning effects of occupants in front, car interiors compared to truck interiors, etc. all affect injury mechanisms. Yet the results obtained for the 14 occupant categories are similar. In particular, for ages 20-35 female risk exceeds male risk by the same (28 ± 3)% for occupants of cars, trucks, and motorcycles.
Further evidence supporting how robust these findings are is provided by a study focused on examining if the relationships found in an earlier study, which used FARS 1975 to 1983 data, were similarly revealed in FARS 1984-1996 data. No distinguishable differences were found dependent on which of the independent data sets was used. This supports that the effects remain unchanged in time, and apply for different vintage vehicles.
The finding that female risk exceeds male risk by amounts that do not appear different for different time periods, different types or vintages of vehicles, seating positions, or use of occupant restraints, supports the interpretation that females are intrinsically more likely to die from physical impacts in general. This is a finding somewhat parallel to other findings of gender-dependent physiological differences. For example, females live longer than males and are more likely to survive infancy. In each case the findings are phenomenological in nature - they are unambiguous inferences from large data sets. Explanations of why such phenomena occur await different types of investigations than the investigations that established that they do occur. However, greater risk of blunt trauma fatality, greater longevity, and higher survivability in infancy all likely reflect basic physiological differences between females and males.
Although traffic-crash data and the method used provided a laboratory to discover and quantify gender-dependent differences, these differences are interpreted to apply beyond the laboratory in which they were investigated. Thus Fig. 6-5 is interpreted to apply not just to risks from traffic crashes, but from other sources such as falling from a roof or down stairs. This interpretation is consistent with cadaver tests using fixed impacts, which find that females have a 20% greater risk of injury to the thorax than males.
The age range in which female risk from blunt trauma exceeds male risk (pre-teens to late fifties) is similar to the child-bearing years, thus inviting speculation that biological factors associated with the potential to have children could increase risk from physical impacts.
Males and females involved in identical crashes are subject to similar decelerations rather than similar forces. As force is the product of body mass times deceleration, heavier subjects experience proportionally higher forces, just as they do when they fall or walk into a fixed object. The results have been derived for forces that generally increase with body mass. For an impact with a fixed amount of energy, say being struck by a falling object or an inflating airbag, even larger gender-dependent effects are expected. This is because the gender effects found are based on risks in identical crashes in which females are subject to smaller forces on account of their smaller masses.
Inferring involvement rates from fatality data
FARS data for 2001 show 3.4 times as many 20-year-old male driver deaths as 20-year-old female driver deaths. Such ratios are often interpreted to mean that males are 3.4 times as likely as females to be involved in lethal crashes. The results here show that this interpretation should be modified. If 20-year-old males and females had equal involvement rates, females would experience 28% more deaths. The driver fatality ratio should be multiplied by 1.28 to take account of the different risk from the same impact, so a 20-year-old male is 4.4 times, not 3.4 times, as likely to be involved in a potentially lethal crash.
Male deaths far exceed female deaths from all types of injuries, including interpersonal violence, suicide, drowning, fire, falls and poisoning. Male
road-user deaths far exceed female road-user deaths - and not just drivers (Fig. 3-10, p. 46). The results here show that male involvement risk exceeds female involvement risk by even larger amounts than captured by the numbers of casualties for cases in which death is due to blunt trauma.
The estimates of differences between male and female fatality risk from the same physical impact use the double pair comparison method. It is therefore appropriate to examine if the findings could be due to biases in the method or data, or if the results could have an explanation based on the method rather than intrinsic gender differences.
The fact that male drivers have higher crash involvement rates does not systematically affect R for drivers. Higher crash rates generate additional crashes which increase subject and control fatalities by similar proportions, but do not change ratios systematically. The data show no systematic differences in estimates of R for passengers dependent on whether the driver is male or female (male controls provide about three times as many data).
Let us assume that males not only have more crashes than females, but when they do crash they also have crashes of higher severity. Formal mathematical reasoning shows that plausibly different distributions by severity can have, at most, only small influences on R.2 This result is more immediately apparent from the following data-based examples.
Safety belts reduce fatality risk for drivers by 42% (Chapter 11) and by a similar percent for right-front passengers. Yet values of R are not systematically different for belted and unbelted occupants (compare rows 1 and 2 in Figs 6-1 and 6-2). Could incorrect coding of belt use bias results- Perhaps about 10% of surviving occupants coded as belted were unbelted. This is a major problem in estimating belt effectiveness, but will influence the gender effect only if miscoding rates are highly gender dependent. If R does not much depend on belt use, which appears to be the case in Figs 6-1 and 6-2, then males and females being miscoded in similar proportions will not systematically affect R. Plausible departures from these assumptions could lead to no more than small changes in R. It is widely accepted that when occupants are coded as unbelted they are very likely unbelted. Cases with unknown belt use were excluded.
Rear-seat unbelted occupants have fatality risks 26% lower than unbelted front-seat occupants, yet R values for rear-seat occupants do not differ systematically from those for front-seat occupants (compare rows 1 and 3 in Figs 6-1 and 6-2). Motorcycle helmets reduce passenger fatality risk by 28%, yet values of R do not systematically depend on helmet use (Fig. 6-3). Fatality risk differs between cars, trucks and, particularly, motorcycles, yet values of R do not systematically differ (compare the 3 graphs in Fig. 6-4).
The study was replicated using only rural, and then using only urban, crashes.1 A typical rural crash is about four times as likely to be fatal as a typical urban crash. Despite such a large difference in severity, the rural and urban replications produced values of R that are in good agreement with each other and with Fig. 6-5.
Possible alternate explanations
Could the results reflect merely differences in stature? One could certainly speculate that risk might be greater for smaller drivers because of differences in the details of their interaction with the steering wheel during a crash (cases with airbag deployment were excluded). However, it seems implausible that the same explanation could apply to occupants whether or not they were belted, as belts alter the mix of injury mechanisms. Rear-seat occupants strike different parts of the vehicle from those struck by front-seat occupants. There does not seem to be any plausible mechanism that would favor taller individuals in all seats by amounts approaching the 28% found here. As motorcyclists are typically killed by striking objects external to the motorcycle, characteristics
of the interaction between occupant and vehicle can have little bearing
An additional reason why the effects cannot plausibly be attributed to differences in stature is because at older ages female risk is, if anything, less than male risk (Fig. 6-5), yet at all ages females remain about the same percent shorter than males. International anthropometric data show consistently that at age 20, females are 7.5% shorter than males compared to 7.4% shorter at
age 70. ,
The discussion above showed that factors known to have large influences on traffic fatality risk do not substantially affect R. The results, in common with results from any study using real-world data, may still be influenced by an essentially unlimited list of possible biases. However, it seems difficult to posit any plausible bias in the data that could change R values by amounts that would materially change the values shown in Fig. 6-5.
Helps explain observed fatality risk differences
The finding that females are more likely than males to die from the same severity impact helps explain two much-reported traffic safety topics.
Increased female risk from airbags. When injuries result from airbag deploy-ments, they are of a different nature from other crash injuries in that the device provides its own source of energy. Unlike crash forces which are proportional to body mass, the impact delivered to an occupant in the deployment envelope of an airbag is independent of the occupant's body mass. Thus the increased risk airbags pose to females is expected to be greater than the 28% found (for ages 20-35) for crash forces. This may be an important part of the explanation why it is found that, while airbags reduce net risk to males, they increase net risk to females. The females, being shorter, are more likely to be in the deployment envelope thereby increasing the risk of being struck by the airbag, and if they are struck, the blow is more likely to be fatal.
Suicide seat. In the 1960s the right-front seat was commonly referred to as the suicide seat because police officers observed that its occupants were more likely to be killed than drivers, or occupants of rear seats. At that time a male driver and a female right-front passenger was an even more common combination than today. The 28% higher risk to females would generate a sufficient disparity to be noticed and attributed to the risk in the seat. When both occupants are of the same gender and age, the risk in driver and right-front passenger seats are not distinguishably different (Fig. 3-15, p. 54).
Age and survivability
It is common knowledge that, as people age, their injury risk from the same physical impact, as might occur in falling, increases. A study comparing fatality rates to crash rates for the same distance of travel showed an increase in the risk of death per crash as drivers age. Detailed quantification of the relationship between the risk of death from the same severity impact and age proves elusive because it can rarely be concluded that subjects received similar physical insults. An approach parallel to that used to investigate gender effects was adopted to compare risk for either gender at any age to the risk for a 20-year-old male. An important difference in method from the gender investigation was required because of the age and gender mix of passengers who accompany drivers.
Most commonly, the right-front passenger accompanying a driver is of similar age and opposite gender. This facilitated the investigation of gender effects, but is a difficulty for the age analysis because there are few cases in which, say, a 70-year-old driver will be accompanied by a passenger of similar age to a passenger accompanying a 20-year-old driver. This makes it infeasible to compare risks to 70-year-old and 20-year-old drivers directly. Instead, the risk at 70 was compared to the risk at 65, the risk at 65 compared to the risk at 60, and so on through comparing the risk at 25 to the risk at 20. The risk at age 70 was compared to the risk at age 20 by multiplying the series of risk factors for each of the 5-year steps. Because each step has an associated error, the error in the risk at any age increases the further this age is from the reference age 20.
For ages below 20, comparisons are direct between the younger age categories and 20-year-old males.
Male age effect
Figure 6-6 shows results for male car occupants, based on 112,736 fatalities (all male). Values, plotted on a logarithmic scale, are relative to the risk to 20-year-old males, marked by diamond symbol on each graph. The parameter b is the slope of a least squared fit to the data constrained to pass through the point (age = 20, R = 1). The interpretation is (using the top-left graph for unbelted
drivers) that after age 20, the risk of death from the same physical impact increases at a compound rate of 2.47% per year.
Corresponding graphs are given in the source paper for light-truck occupants and motorcyclists (drivers and passengers), leading to 16 graphs (6 for cars,
6 for light trucks and 4 for motorcyclists).15 There are no indications of systematic differences between the 16 graphs, thus supporting the same interpretation as in the gender case that the effects are of a basic physiological nature and apply in general, not just to crashes.
The composite graphs for the data from each vehicle in Fig. 6-7 additionally support that the effect is relatively independent of vehicle type. The slopes are all consistent with a 2.5% annual compound increase in risk after age 20.
The summary result for males is given in Fig. 6-8. This can be viewed as either the weighted average of the values for the individual vehicles in Fig. 6-7, or the mathematically identical weighted average of the (up to) 16 values from each of the 16 graphs.
Before age 20, risk increases with decreasing age. After age 20, risk increases at (2.52 ± 0.08)% per year. The risk at a given age is computed as
Female age effect
The analysis proceeds in parallel with the male case. All risk values are relative to the same reference value used in the male analysis, namely a 20-year-old male. Hence the fit to the female data is not constrained to pass through any point, but rather, the value at age 20 provides a separate estimate of the risk to 20-year-old females compared to the risk to 20-year-old males.
The summary graph in Fig. 6-9 is the average for the three vehicles, or, equivalently, the weighted average for 14 individual graphs (two less than the male analysis because there were insufficient female motorcycle driver fatalities). The parameter a is the value of the fit with age = 20. The interpretation is that this analysis gives that female risk at age 20 is (31.1 ± 2.2)% higher than male risk at age 20, an estimate in good agreement with the (27.9 ± 2.3)% value obtained for age 20 in the gender comparison.
As for the male case, risk increases with decreasing age for ages younger than 20. This is not due simply to increased risk to infants seated on the laps of adults being at increased risk due to loading from the adult. Increasing risk with decreasing age is similarly present for belted occupants (belted here means that some type of restraint, including a baby or infant seat, was used). It is also present for rear-seat occupants. It thus appears that effects due to infants on the laps of adults can make no more than a small contribution to the observed risk increase. The risk for one-year-old babies of either gender is about twice the risk at age 20.
After age 20, female risk increases at a compound rate of (2.16 ± 0.10)% per year, somewhat lower than the (2.52 ± 0.08)% yearly increase for males. The risk at a given age is computed as
Following the same reasoning applied to the gender examination, the age dependence of the risk of death from similar physical impacts is interpreted to reflect fundamental physiological processes. The relationships in Eqn 6-8 and Eqn 6-9 apply to physical impacts in general, and not just to those resulting
After age 20 risk increases at compound rates of more than 2% per year for males and females. So anyone who believed that life was straight downhill after age 20 was being far too optimistic - it is downhill at an exponentially increasing rate!
Gender and age effects determined using two-car crashes
All the inferences above relating to gender and age were derived using the double pair comparison method. All vehicles used contained at least one passenger. Results were interpreted to apply to blunt trauma in general. Such a universal interpretation would receive additional support if similar effects were revealed in studies using different methods and data.
This was pursued using outcomes of two-car crashes in which at least one driver was killed. The analysis was confined to cars containing only one occupant, the driver, thereby assuring that no crash used in double pair comparison analyses contributed to the two-car crash analyses. Each method therefore used independent data.
The two-car crash method is conceptually very simple. Measure the ratio, R, of female to male fatalities when cars with female drivers and cars with male drivers of similar age crash into each other. It is, however, not quite that simple because of the presence of another large confounding factor, namely, car mass. This effect of mass was removed by analyzing R versus the ratio of the car masses, and inferring the value of R if the cars had equal mass using the same method that produced Fig. 4-8, p. 74 and Fig. 4-11, p. 81.
Gender effect estimated using two-car crashes
Results are plotted using the bold symbols in Fig. 6-10. The R values measure female fatality risk divided by male fatality risk when drivers of similar age traveling in cars of the same mass crash into each other. The double pair comparison method results in Fig. 6-5 are shown again in Fig. 6-10 using smaller symbols. As the two-car crash method provides far fewer data, the data were divided into just four broad age categories centered at ages 20, 30, 45, and 70 years. The risk ratios for ages 20, 30, and 45 (which together included ages in the range 16-56 years) are (22 ± 14)%, (23 ± 19)%, and (21 ± 14)%. The weighted average of these indicates that females older than about 20 but not older than the mid fifties are (22 ± 9)% more likely to die than males of the same age when their cars crash into each other, a result in good agreement with the (28 ± 3)% value expected based on the double pair comparison method.
The finding for age 70 (which included ages in the range 56-97) that females are (15% ± 12) less likely to die than males corroborates the trend towards values of R less than one found in the double pair comparison study. The consistency of the findings supports the interpretation that both methods are measuring the same fundamental difference between female and male risk of death from the same impact.
Age effects estimated using two-car crashes
For the age investigation, one car is always driven by a 20-year-old male driver (age 16-24), and the other by a male driver (for the male analysis) or a female driver (for the female analysis). Thus risks for both genders are relative to risks to 20-year-old males, as before.
Age effect for males. The data in Fig. 6-11 for males show good agreement between the two-car method and the double pair comparison method except at older ages. The line is a least square fit to an equation of the same form as Eqn 6-8 constrained to pass through the point R = 1 at age = 20 (and ignoring the outlier point at age 80). The fit gives b = (2.86 ± 0.32), meaning that risk increases at a compound rate of 2.86% per year. This may be compared to b = (2.52 ± 0.08)% obtained using the double pair comparison method.
Age effect for females. In Fig. 6-12 the straight line is a fit to the data (excluding the outlier point at age 80) yielding two parameters (Eqn 6-9).
The first, a = (20.6 ± 11.3)% estimates that at age 20 female risk exceeds
male risk by 20.6%. The second, b = (2.66 ± 0.37)%, indicates that female
risk increases at a compound rate of 2.66% per year. These values may be compared to the double pair comparison values a = (31.1 ± 2.2)% and b = (2.16 ± 0.10)%. Both methods show higher rates of increase per year for males than for females.
Figure 6-12. The ratio, R, of the risk of death for a female of the indicated age to the risk of death to a 20-year-old male. The bold symbols are results from two-car crashes in which one driver is a 20-year-old male and the other is a female of the indicated age. Based on 1768 female and 877 male fatalities in FARS 1975-1998.16 The smaller symbols reproduce the data in Fig. 6-9.
Comments on results from the two methods
In the age analyses, the higher than trend values of R at the oldest ages likely reflects that when older drivers are involved in two-car crashes, their vehicles are more likely to be struck on the side (Fig. 7-20, p. 165). A driver in a car struck on the side is at much higher risk than a driver of a frontally-impacted car (Fig. 4-8, p. 74). Thus, the data at the oldest ages reflect that average impact severity is greater for older drivers than for the 20-year-old comparison drivers. There are insufficient data to restrict this study to frontal crashes only, which would avoid this problem.
The relatively close quantitative agreement between the two-car and double-pair-comparison estimates for the gender and for the age analyses increases confidence in the validity of both methods, in the results derived from them, and in the interpretation given to these results.
Alcohol consumption and survivability
There is a common impression that the presence of alcohol reduces the likelihood of injury, given an impact of specific severity. This fits a common notion that, by being more relaxed, drunks are more likely to "roll with the punches." More importantly, some clinical studies seemed to support this notion. In general, these studies monitored the progress of sets of drunk and sober patients admitted to hospitals with injuries of similar severity. It was generally observed that the drunks exhibited higher rates of recovery or survivability. These studies were methodologically flawed in that the agent being studied, namely alcohol, played a crucial role in subject selection. If alcohol increases the probability of dying at the scene of a crash, then subjects whose injuries proved fatal because of alcohol use were excluded from the comparison in the hospital tests. Similarly, if being sober compared to being drunk were to reduce injury to below that requiring hospitalization, this would similarly negate any conclusions based exclusively on those admitted to hospital. Indeed, instead of examining how alcohol influences injury risk, such studies examine secondary and unimportant details of the non-normalized distributions of injury versus recovery curves for drunk and sober drivers.
The first study to really address how alcohol affected survivability in a crash compared injuries to drunk and sober drivers involved in crashes matched in a sufficient number of important characteristics that they could be judged to be of similar severity. It concluded, based on data on 1,126,507 drivers involved in 1979-1983 North Carolina crashes, that alcohol-impaired drivers were 3.85 times as likely to die as alcohol-free drivers in crashes of comparable severity.
Being overweight was found to increase an occupant's risk of death and serious injury in traffic crashes. The authors comment that co-morbid factors could have contributed to the effect. Interactive effects between alcohol consumption and being overweight could be one of those.
Additional evidence that alcohol increases injury risk is provided by findings that an intoxicated person might be at greater risk of immediate death due to increased vulnerability to shock and therefore decreased time available for emergency medical intervention. , Alcohol was found to increase the severity of traumatic brain injury in motor vehicle crash victims controlling for crash severity characteristics.
Addressing alcohol effect using FARS data
FARS contains a variable Alcohol Test Result presenting measured levels of Blood Alcohol Concentration (BAC) (Chapter 10). It might therefore appear that survivability could be addressed using the double pair comparison method or the two-car crash method. Three problems preclude using either method:
1. BAC is not measured for all drivers. In FARS 2002, 35% of fatally injured drivers had no BAC level coded (Table 10-3, p. 249).
2. The probability that BAC is measured is substantially lower for surviving than for fatally-injured drivers. In FARS 2002, 75% of the drivers who were not killed had no BAC level coded.
3. The probability that BAC is measured increases with BAC for surviving and for fatally-injured drivers.
FARS advises "Alcohol Test Results from this database should be interpreted with caution." Because of the need to estimate the role of alcohol in the nation's fatal crashes, procedures to impute the missing BAC values based on relationships between such factors as nighttime driving and single-vehicle crashes that are known to correlate with alcohol use have been developed and refined over the years.
Any attempt to use the double pair comparison method runs into yet another problem. Only 12% of passengers involved in fatal crashes have BAC values coded in FARS. There is rarely a reason to measure the BAC of a surviving passenger, so that most values are from autopsies. If used to investigate how alcohol affected risk of survival, control occupants in the first and second comparisons must have similar alcohol use. If driver and passenger BAC were identical (and they tend to be similar), then no effect would be measured regardless of its magnitude.
In view of these problems, an approach was adopted that used only fatally injured drivers with measured BAC. Although the approach used two-car crashes, it was described in terms of the following non-traffic analogy. Assume that elevators are dramatically less safe than they are, and that they are prone to come crashing freely to the ground. Assume that a sober person has a 1% probability of being killed in a low severity crash in which the elevator falls from the second floor. Assume that being drunk doubles that probability to 2%. From many such crashes a treatment data set is formed consisting exclusively of fatally-injured elevator riders with known BAC. In order to extract from such data the assumed doubling of risk associated with alcohol, we need to know the mix of drunk and sober people who ride elevators. This is obtained from an exposure data set consisting of fatalities in elevators that fell from, say, the 40th or higher floors. Essentially everyone is such a crash will be killed, so the mix of drunk and sober fatalities provides the required exposure.
If two cars crash head on into each other and one car is more than 25% heavier than the other, the driver of the heavier car is far less likely to die than the driver of the lighter car (Fig. 4-5, p. 69). A treatment set can therefore be formed from drivers who died in heavier cars, given that drivers in lighter cars survived. All these treatment drivers died in crashes in which the probability of death was low, so that any factor that increased risk of death would increase their numbers. The exposure set is formed from drivers in lighter cars in two-car crashes in which the driver of the heavier (by ³ 25%) car was killed. The probability that sober drivers are killed in the lighter car is so high that any additional risk-increasing factor has little opportunity to influence the outcome.
Additional two-car crash configurations were included, including side impact compared to frontal impact (Fig. 4-8, p. 74) to augment the small sample sizes resulting from the strict criteria for data inclusion. The probability of death in the treatment sample was 9.2% (higher than ideal), and in the exposure set 76.2% (substantially lower than ideal). Correction factors were applied to extrapolate these to the more extreme values of near zero and 100% assumed by the method.25
This same method was applied to investigate how age affected risk of death from the same physical impact. The results, based on much larger samples than available for the alcohol analysis, agreed with those reported above, providing additional validation for the age effects, and more importantly for the present method.25
A comparison of the distributions of BAC in the treatment and exposure sets led to the result plotted in Fig. 6-13. The straight line
is a least-squares fit to the data reflecting the definition that R = 1 at BAC = 0. Thus R gives the risk at a given BAC relative to a value of unity for a driver with BAC = 0. The value derived for the parameter from the fit is
Equation 6-10 with k = 9.1 estimates that, given involvement in a crash, the risk of death is increased by 73% by a BAC of 0.08%, the legal limit for driving in most US states. The average BAC in the bodies of fatally injured drivers who have a non-zero BAC in 2001 FARS is 0.17%, at which level the risk of death in a crash is 2.5 times that for a zero BAC driver.
While the effect of alcohol on increasing risk of death in a given crash is substantial, it is much smaller that the effect of alcohol on increasing a driver's risk of crashing. However, the increasing effect on fatality risk in a crash is present whether the person drives or travels as a passenger. Thus, a taxi passenger with BAC = 0.17% is 2.5 times as likely to die as a taxi passenger with BAC = 0 if the taxi crashes.
Summary and conclusions (see printed text)
There is old adage that God protects drunks and babies. The detailed analyses in this chapter show it is false on both counts - from the same severity impact babies and drunks are more likely to die.
References for Chapter 6 - Numbers in [ ] refer to superscript references in book that do not correctly show in this html version. To see how they appear in book see the pdf version of Chapter 1.
 Evans L. Female
compared to male fatality risk from similar physical
insults. J Trauma. 2001; 50: 281-288.
 Evans L. Double pair comparison – a new method to determine how occupant characteristics affect fatality risk in traffic crashes. Accid Anal Prev. 1986; 18: 217-27.
 Fleiss JL.
Statistical Methods for Rates and Proportions. New
York, NY: Wiley; 1973.
 Schlesselman JJ.
Case-Control Studies: Design, Conduct, Analysis. New
York, NY: Oxford University Press; 1982.
 Young HH.
Statistical Treatment of Experimental Data. New York,
NY: McGraw-Hill; 1962.
 Evans L. Rear seat restraint system effectiveness in preventing fatalities. Accid Anal Prev. 1988; 20: 129-36.
 Evans L. Age
dependence of female to male fatality risk in the same
crash: An independent reexamination. Crash Prev Inj
Control. 2000; 2: 111-21.
 Evans L. Risk
of fatality from physical trauma versus sex and age.
J. Trauma. 1988; 28: 368-78.
 Foret-Bruno JY,
Faverjon G, Brun-Cassan F, et al. Females more
vulnerable than males in road accidents. Paper no.
905122, Proceedings of the XXIII FISITA Congress, Torino,
Italy; 7-11 May 1990. Volume 1, p. 941-950.
 World Health Organization.
The Injury Chartbook, p. 14. Geneva; 2002.
 Pheasant S. Body Space:
Anthropometry, Ergonomics and Design. London: Taylor
& Francis; 1986.
 Najjar MF, Rowland M.
Anthropometric reference data and prevalence of overweight,
United States, 1976-1980. Hyattsville, MD: US
Department of Health and Human Services, Public Health
Service, National Center for Health Statistics; 1987.
 Dalmotas DJ, Hurley J, German A, Digges K. Air bag deployment crashes in Canada. Paper 96-S1O-05. 15th Enhanced Safety of Vehicles Conference, Melbourne, Australia; 13‑17 May 1996.
 Li G, Braver ER, Chen LH.
Fragility versus excessive crash involvement as determinants
of high death rates per vehicle-mile of travel among older
drivers. Accid Anal Prev. 2003; 35: 227-235.
 Evans L. Age and fatality
risk from similar severity impacts. J Traf Med. 2001; 29:
 Evans L, Gerrish PH.
Gender and age influence on fatality risk from the same
physical impact determined using two-car crashes. SAE
paper 011174. Warrendale, PA: Society of Automotive
Engineers; March 2001. (Also included in: Vehicle
Aggressivity and Compatibility in Automotive Crashes.
SAE special publication SP-1601, 2001).
 Waller JA. Methodologic
issues in hospital based injury research. American
Association for Automotive Medicine, 31st Annual
Proceedings, New Orleans, LA, p. 95-108; 1987.
 Waller PF, Stewart JR, Hansen
AR, Stutts JC, Popkin CL, Rodgman EA. The potentiating
effects of alcohol on driver injury. J Am Med Assoc.
1986; 256: 1461-1466.
 Mock CN, Grossman DC, Kaufman
RP, Mack CD, Rivara FP. The relationship between body
weight and risk of death and serious injury in motor vehicle
crashes. Accid Anal Prev. 2002; 34: 221-228.
 Anderson TE, Viano DC. Effect of acute alcohol intoxication on injury tolerance and outcome. In: Noordzij P, Roszbach R, editors. Alcohol, drugs and traffic safety – T86, p. 251-254. Amsterdam, Netherlands: Excerpta Medical Elsevier Science Publisher; 1987.
 Dischinger PC, Soderstrom CA,
Shankar BS, Cowley RA, Smialek JE. The relationship
between use of alcohol and place of death in vehicular
fatalities. Association for the Advancement of
Automotive Medicine, 32nd Annual Proceedings, Seattle, WA,
p. 299-311; 12-14 September 1988.
 Cunningham RM, Maio RF, Hill EM,
Zink BJ. The effects of alcohol on head injury in the
motor vehicle crash victim. Alcohol Alcoholism. 2002;
 Fatality Analysis Reporting
System (FARS) Web-Based Encyclopedia. Create a query,
and choose “Alcohol Test Result” as a variable to obtain
caution regarding use of BAC values in site.
 Subramanian R. Transitioning to Multiple Imputation – A New Method to Estimate Missing Blood Alcohol Concentration (BAC) Values in FARS. Report DOT HS 809 403. Washington, DC: US Dept of Transportation; January 2002 (Revised October 2002). http://www-nrd.nhtsa.dot.gov/pdf/nrd-30/NCSA/Rpts/2002/809-403.pdf
 Evans L, Frick MC. Alcohol’s effect on fatality risk from a physical insult . J Stud Alcohol. 1993; 54: 441-449.