This study was an analysis of differences in casualty rates of pedestrians per 100 million miles of E-HE car travel and rates per 100 million miles of ICE car travel.
SettingThis study was set in Great Britain between 2013 and 2017.
ParticipantsThe study participants were all pedestrians reported to have been injured in a collision with a car or a taxi.
ExposureThe exposure was the type of propulsion of the colliding vehicle, E-HE or ICE. E-HE vehicles were treated as a single powertrain type, regardless of the mode of operation that a hybrid vehicle was in at the time of collision (hybrid vehicles typically start in electric mode and change from battery to combustion engine at higher speeds).13
OutcomeThe outcome of interest was a pedestrian casualty.
Effect modification by road environmentWe used the urban–rural classification14 of the roads on which the collisions occurred to investigate whether casualty rate ratios comparing E-HE with ICE vehicles differed between rural and urban environments.
Data sources/measurementNumerator data (numbers of pedestrians injured in collisions) were extracted from the Road Safety Data (STATS19) datasets.15
Denominator data (100 million miles of car travel per year) were estimated by multiplying average annual mileage by numbers of vehicle registrations.16 Average annual mileage for E-HE and ICE vehicles was estimated separately for urban and rural environments using data obtained under special licence from the National Travel Survey (NTS) datasets.17 We estimated average annual mileage for the years 2013–2017 because the NTS variable for the vehicle fuel type did not include ‘hybrid’ prior to 2013 and data from 2018 had not been uploaded to the UK data service due to problems with the archiving process (Andrew Kelly, Database Manager, NTS, Department for Transport, 23 March 2020, personal communication). Denominators were thus available for the years 2013–2017.
Data preparationThe datasets for collisions, casualties and vehicles from the STATS19 database were merged using a unique identification number for each collision.
Statistical methodsWe calculated annual casualty rates for E-HE and ICE vehicles separately and we compared these by calculating a rate ratio. We used Poisson regression models to estimate rate ratios with 95% CIs and to investigate any modifying effects of the road environment in which the collisions occurred. For this analysis, our regression model included explanatory terms for the main effects of the road environment, plus terms for the interaction between type of propulsion and the road environment. The assumptions for Poisson regression were met in our study: we modelled count data (counts of pedestrians injured), traffic collisions were independent of each other, occurring in different places over time, and never occurring simultaneously. Data preparation, management and analyses were carried out using Microsoft Access 2019 and Stata V.16.18
Sensitivity analysisWe conducted an extreme case analysis where all missing propulsion codes were assumed to be ICE vehicles (there were over a 100 times more ICE vehicles than E-HE vehicles on the roads in Great Britain during our study period,16 so missing propulsion is more likely to have been ICE).
Study sizeThe sample size for this study included all available recorded road traffic collisions in Great Britain during the study period. We estimated that for our study to have 80% power at the 5% significance level to show a difference in casualty rates of 2 per 100 miles versus 5.5 per 100 miles, we would require 481 million miles of vehicle travel in each group (E-HE and ICE); whereas to have 90% power at the 1% significance level to show this difference, 911 million miles of vehicle travel would be required in each group. Our study includes 32 000 million miles of E-HE vehicle travel and 3 000 000 million miles of ICE vehicle travel and therefore our study was sufficiently powered to detect differences in casualty rates of these magnitudes.
ResultsParticipantsBetween 2013 and 2017, there were 916 713 casualties from reported road traffic collisions in Great Britain. 120 197 casualties were pedestrians. Of these pedestrians, 96 285 had been hit by a car or taxi. Most pedestrians—71 666 (74%) were hit by an ICE car or taxi. 1652 (2%) casualties were hit by an E-HE car or taxi. For 22 829 (24%) casualties, the vehicle propulsion code was missing. Most collisions occurred in urban environments and a greater proportion of the collisions with E-HE vehicles occurred in an urban environment (94%) than did collisions with ICE vehicles (88%) (figure 1).
Flow chart of pedestrian casualties in collisions with E-HE or ICE cars or taxis from reported road traffic collisions in Great Britain 2013–2017. E-HE, electric and hybrid-electric; ICE, internal combustion engine.
Main resultsDuring the period 2013 to 2017, the average annual casualty rates of pedestrians per 100 million miles were 5.16 (95% CI 4.92 to 5.42) for E-HE vehicles and 2.40 (95% CI 2.38 to 2.41) for ICE vehicles, which indicates that collisions with pedestrians were on average twice as likely (RR 2.15 (95% CI 2.05 to 2.26), p<0.001) with E-HE vehicles as with ICE vehicles (table 1).
Table 1Pedestrian casualties due to collisions with cars or taxis from reported road traffic collisions in Great Britain 2013–2017—by vehicle propulsion type
Sensitivity analysisIn our extreme case analysis, the 22 829 pedestrian casualties where vehicle propulsion was missing were all assumed to have been struck by ICE vehicles. In this case, average casualty rates of pedestrians per 100 million miles were 3.16 (95% CI 3.14 to 3.18) for ICE vehicles, which would indicate that collisions with pedestrians were on average 63% more likely (RR 1.63 (95% CI 1.56 to 1.71), p<0.001) with E-HE vehicles than with ICE vehicles (table 2).
Table 2Extreme case sensitivity analysis—pedestrian casualties due to collisions with cars or taxis from reported road traffic collisions in Great Britain 2013–2017 by vehicle propulsion type where 22 829 missing vehicle propulsion codes are assumed to be ICE vehicles
Relative risks according to road environmentCasualty rates were higher in urban than rural environments (tables 3 and 4).
Table 3Pedestrian casualties due to collisions with cars or taxis from reported road traffic collisions in Great Britain 2013–2017—by vehicle propulsion type in urban road environments
Table 4Pedestrian casualties due to collisions with cars or taxis from reported road traffic collisions in Great Britain 2013–2017—by vehicle propulsion type in rural road environments
Urban environmentsCollisions with pedestrians in urban environments were on average over two and a half times as likely (RR 2.69 (95% CI 2.56 to 2.83, p<0.001) with E-HE vehicles as with ICE vehicles (table 3).
Sensitivity analysisThe extreme case sensitivity analysis showed collisions with pedestrians in urban environments were more likely with E-HE vehicles (RR 2.05; 95% CI 1.95 to 2.15).
Rural environmentsCollisions with pedestrians in rural environments were equally likely (RR 0.91; 95% CI 0.74 to 1.11) with E-HE vehicles as with ICE vehicles (table 4).
Sensitivity analysisThe extreme case sensitivity analysis found evidence that collisions with pedestrians in rural environments were less likely with E-HE vehicles (RR 0.68; 95% CI 0.55 to 0.83).
Results of Poisson regression analysisOur Poisson regression model results (table 5) showed that pedestrian injury rates were on average 9.28 (95% CI 9.07 to 9.49) times greater in urban than in rural environments. There was no evidence that E-HE vehicles were more dangerous than ICE vehicles in rural environments (RR 0.91; 95% CI 0.74 to 1.11), consistent with our finding in table 4. There was strong evidence that E-HE vehicles were on average three times more dangerous than ICE vehicles in urban environments (RR 2.97; 95% CI 2.41 to 3.67).
Table 5Results of Poisson regression analysis of annual casualty rates of pedestrians per 100 million miles by road environment and the interaction between vehicle propulsion type and environment
DiscussionStatement of principal findingsThis study found that in Great Britain between 2013 and 2017, casualty rates of pedestrians due to collisions with E-HE cars and taxis were higher than those due to collisions with ICE cars and taxis. Our best estimate is that such collisions are on average twice as likely, and in urban areas E-HE vehicles are on average three times more dangerous than ICE vehicles, consistent with the theory that E-HE vehicles are less audible to pedestrians in urban areas where background ambient noise levels are higher.
Strengths and weaknesses of the studyThere are several limitations to this study which are discussed below.
The data used were not very recent. However, ours is the most current analysis of E-HE vehicle collisions using the STATS19 dataset.
Before we can infer that E-HE vehicles pose a greater risk to pedestrians than ICE vehicles, we must consider whether our study is free from confounding and selection bias. Confounding occurs when the exposure and outcome share a common cause.19 Confounders in this study would be factors that may both cause a traffic collision and also cause the exposure (use of an E-HE car). Younger, less experienced drivers (ie, ages 16–24) are more likely to be involved in a road traffic collision20 and are also more likely to own an electric car.21 Some of the observed increased risk of electric cars may therefore be due to younger drivers preferring electric cars. This would cause positive confounding, meaning that the true relative risk of electric cars is less than we have estimated in our study. Regarding selection bias, it is known that the STATS19 dataset does not include every road traffic casualty in Great Britain, as some non-fatal casualties are not reported to the police.22 If casualties from collisions are reported to the police differentially according to the type of vehicle propulsion, this may have biased our results; however, there is no reason to suspect that a pedestrian struck by a petrol or diesel car is any more or less likely to report the collision to the police than one struck by an electric car.
We must also address two additional concerns as ours is a cross-sectional study: The accuracy of exposure assignment (including the potential for recall bias) and the adequacy of prevalence as a proxy for incidence.23 First, the accuracy of exposure assignment and the potential for recall bias are not issues for this study, as the exposure (type of propulsion of the colliding vehicle, E-HE or ICE), is assigned independently of the casualties by the UK Department for Transport who link the vehicle registration number (VRN) of each colliding vehicle to vehicle data held by the UK Driver Vehicle and Licensing Agency (DVLA).10 Second, we have not used prevalence as a proxy for incidence but have estimated incidence using total distance travelled by cars as the measure of exposure.
We may therefore reasonably infer from our study results that E-HE vehicles pose a greater risk to pedestrians than ICE vehicles in urban environments, and that part of the risk may be due to younger people’s preference for E-HE cars.
A major limitation of the STATS19 road safety dataset used in this study was that it did not contain a vehicle propulsion code for all vehicles in collisions with pedestrians. We excluded these vehicles from our primary analysis (a complete case analysis) and we also conducted an extreme case sensitivity analysis. We will now argue why imputation of missing vehicle propulsion codes would not have added value to this study. Vehicle propulsion data are obtained for the STATS19 dataset by the UK Department for Transport who link the VRN of each colliding vehicle recorded in STATS19 to vehicles data held by the UK DVLA. The STATS19 data on reported collisions and casualties are collected by a Police Officer when an injury road accident is reported to them; Most police officers write details of the casualties and the vehicles involved in their notebooks for transcription onto the STATS19 form later at the Police station.24 The VRN is one of 18 items recorded on each vehicle involved in a collision. Items may occasionally be missed due to human error during this process. Where a VRN is missing, vehicle propulsion will be missing in the STATS19 dataset. The chance that any vehicle-related item is missing will be independent of any characteristics of the casualties involved and so the vehicle propulsion codes are missing completely at random (MCAR). As the missing propulsion data are very likely MCAR, the set of pedestrians with no missing data is a random sample from the source population and hence our complete case analysis for handling the missing data gives unbiased results. The extreme case sensitivity analysis we performed shows a possible result that could occur, and it demonstrates our conclusions in urban environments are robust to the missing data. Lastly, to impute the missing data would require additional variables which are related to the likelihood of a VRN being missing. Such variables were not available and therefore we do not believe a useful multiple imputation analysis could have been performed.
Strengths and weaknesses in relation to other studiesOur study uses hundreds of millions of miles of car travel as the denominators in our estimates of annual pedestrian casualty rates which is a more accurate measure of exposure to road hazards than the number of registered vehicles, which was used as the denominator in a previous study in the UK.10 Our results differ to this previous study which found that pedestrian casualty rates from collisions with E-HE vehicles during 2005–2007 were lower than those from ICE vehicles. Our study has updated this previous analysis and shows that casualty rates due to E-HE vehicle collisions exceed those due to ICE vehicle collisions. Similarly, our study uses a more robust measure of risk (casualty rates per miles of car travel) than that used in a US study.9 Our study results are consistent with this US study that found that the odds of an E-HE vehicle causing a pedestrian injury were 35% greater than an ICE vehicle. Brand et al 8 hypothesised, without any supporting data, that “hybrid and electric low-noise cars cause an increase in traffic collisions involving vulnerable road users in urban areas” and recommended that “further investigations have to be done with the increase of low-noise cars to prove our hypothesis right.”8 We believe that our study is the first to provide empirical evidence in support of this hypothesis.
Meaning of the study: possible explanations and implications for clinicians and policymakersMore pedestrians are injured in Great Britain by petrol and diesel cars than by electric cars, but compared with petrol and diesel cars, electric cars pose a greater risk to pedestrians and the risk is greater in urban environments. One plausible explanation for our results is that background ambient noise levels differ between urban and rural areas, causing electric vehicles to be less audible to pedestrians in urban areas. Such differences may impact on safety because pedestrians usually hear traffic approaching and take care to avoid any collision, which is more difficult if they do not hear electric vehicles. This is consistent with audio-testing evidence in a small study of vision-impaired participants.10 From a Public Health perspective, our results should not discourage active forms of transport beneficial to health, such as walking and cycling, rather they can be used to ensure that any potential increased traffic injury risks are understood and safeguarded against. A better transport policy response to the climate emergency might be the provision of safe, affordable, accessible and integrated public transport systems for all.25
Unanswered questions and future researchIt will be of interest to investigate the extent to which younger drivers are involved in collisions of E-HE cars with pedestrians.
If the braking distance of electric cars is longer,26 and electric cars are heavier than their petrol and diesel counterparts,27 these factors may increase the risks and the severity of injuries sustained by pedestrians and require investigation.
As car manufacturers continue to develop and equip new electric cars with Collision Avoidance Systems and Autonomous Emergency Braking to ensure automatic braking in cases where pedestrians or cyclists move into the path of an oncoming car, future research can repeat the analyses presented in this study to evaluate whether the risks of E-HE cars to pedestrians in urban areas have been sufficiently mitigated.
AcknowledgmentsWe thank Rebecca Steinbach for her advice on analysis of National Travel Survey data, Jonathan Bartlett for his advice on missing data, and Ben Armstrong for his advice on Poisson regression. We are grateful to the reviewers and to Dr C Mary Schooling, Associate Editor, whose comments helped us improve the manuscript. We are grateful to Jim Edwards and Graham Try for their comments on earlier versions of this manuscript.
留言 (0)