Miller, Phil
(2012)
The use of conspicuity aids by cyclists and the risk of crashes involving other road users: a population based case-control study.
PhD thesis, University of Nottingham.
Abstract
Introduction
Regular cycling has been shown to improve health and well-being and has a role in tackling obesity and inactivity. Cycle collisions, particularly those involving motorised vehicles, can lead to significant mortality and morbidity and are currently a barrier to wider uptake of cycling. There is evidence that the conspicuity of cyclists may be a factor in some injury collisions. Low-cost, easy to use retro-reflective and fluorescent clothing and accessories (’conspicuity aids’) are widely available. Their effectiveness in reducing the risk of cycling collisions is currently unknown. This study was designed to investigate the relationship between the use of conspicuity aids and risk of collision or evasion crashes for utility and commuter cyclists in an urban setting in the UK.
Methods
A matched case-control study was undertaken. Cases were adult commuter and utility cyclists who were involved in a crash resulting from a collision or attempted evasion of a collision with another road user. Cases were recruited at a large UK emergency department. Controls were commuter and utility cyclists matched by time and day of travel, season and geographical area of cycling. Controls were recruited at public and private cycle parking sites. Data on the use of conspicuity aids, crash circumstances, participant demographics, cycling experience, safety equipment use and journey characteristics including an estimate of the bicycle crash risk for each chosen route (the number of previous crashes per 100 million kilometres travelled by bicycle calculated for each participant route) were collected using self-completed questionnaires and maps. Conditional logistic regression was used to calculate crude and adjusted odds ratios and 95% confidence intervals of the risk of a crash involving a collision or evasion of a collision with another road user when cyclists reported they were using any item of fluorescent or retro-reflective clothing or equipment vs. none. Unconditional logistic regression was used to analyse associations between participant characteristics and conspicuity aid use. Continuous variables were dichotomised where there was a non-linear relationship to the bicycle crash outcome variable or the primary exposure variable.
The sensitivity of the study models to selection, recall and information biases and the effect of missing data was assessed using independent records of conspicuity aid use by potential participants during recruitment. Observations of conspicuity aid use within the study source population at sites across the study catchment area were also conducted by the researcher during the recruitment phase.
Results
There were 76 cases and 272 controls cyclists who were eligible for inclusion in the primary analysis (response rate of 13% and 54% respectively). The proportion of cases who reported using any item of fluorescent or reflective materials on their clothing or equipment (excluding bicycle mounted reflectors) was higher than for matched controls (cases users 69.7%; 95% CI 58.1% to 79.8% vs. control users 65.4%; 95% CI 59.5% to 79.1%). The unadjusted odds ratio for a collision or evasion crash when using conspicuity aids, was 1.2 (95% CI 0.66 to 2.17). Two alternative modelling strategies were employed. After adjustment for confounding from age, gender, socio-economic deprivation, number of years of cycling experience, bicycle crash risk along each route and cycle helmet use the odds ratio was 1.77 (95% CI 0.74 to 4.25). After adjustment for confounding from age, gender, socio-economic deprivation, bicycle crash risk along each route and history of previous cycle crash involvement the odds ratio was 2.4 (95% CI 1.06 to 5.7). The odds ratio was not significantly affected by adjustment for possession of a driving licence, reported bicycle safety training in childhood, psychometric associates of risk taking behaviour, cycle helmet wearing, years of experience of cycling, distance or number of trips cycled in the previous seven days, type of bicycle, the use of bike-mounted lights or reflectors, weather or lighting conditions, familiarity with the route or alcohol consumption within 8 hours prior to the recorded journey.
There was a significant difference between the measure of bicycle crash risk along each route for cases and controls with controls reporting travelling on routes with lower objective bicycle crash risk (median (IQR); cases 378.5 (232.4 to 548.3) vs. controls 268.5 (192.6 to 464.5); p= 0.006). There were no significant differences in route risk for users vs. non-users of conspicuity aids (route risk median (IQR) for conspicuity aid users vs. non-users; 308.1 (198.0 to 504.3) vs. 272.3 (203.7 to 413.4; p= 0.22). Conspicuity aid use was associated with increased length of participant route (unadjusted OR 3.25 for reported route greater than median; 95% CI 2.04 to 5.17 p<0.001), higher numbers of police-recorded bicycle crashes (unadjusted OR 2.26 for greater than median; 95% CI 1.43 to 3.55; p<0.001) and lower numbers of observed cyclists on each route (unadjusted OR 0.999; 95% CI 0.998 to 1.000 p=0.015). Route risk data were missing for 50 participants (15 cases and 35 controls).
Validation of the primary exposure showed that there was moderate agreement between participants’ self-reports and independently collected data (kappa 0.42; 95% CI 0.32 to 0.51) but independent data were collected on only 4 eligible cases. Self-reported use of conspicuity aids was higher amongst cases and controls in this study than that observed for cyclists in the study area during the recruitment period (23%; 95% CI 22% to 24%).
Discussion
The results of this study show a non-significant increase in the odds of a crash for users compared to non-users of conspicuity aids whilst cycling. This association was increased after adjustment for confounders but most models generated to adjust for confounding remained insignificant. No reduction in crash risk could be demonstrated. This is not consistent with the large body of evidence suggesting that conspicuity aids increase the distances from which wearers can be detected and recognised by drivers in a variety of settings.
There was evidence that cases were cycling along routes with greater exposure to traffic danger than controls although there were many participants with missing data for this variable potentially introducing a further source of bias. The route risk estimates did not vary significantly between conspicuity aid users and non-users. Residual confounding may have occurred if conspicuity aid users were taking more risks when encountering similar traffic conditions to non-users. This could not be measured but may go some way to explaining these results. If cyclists over-estimate the likely effect of their conspicuity aid use this could result in over compensation and a net increase in crash risk. Adjustment for route risk may have introduced bias by the loss of some participants from the analysis or by acting as a positive suppressor variable increasing the influence of uncontrolled confounding if conspicuity aid use were leading to risky riding over and above the objective risk arising from differing road and traffic conditions. The association between the odds of crash and travelling on roads with higher incidences of previous cycle crashes and fewer cyclists provides support for the “safety in numbers” effect reported in other studies.
Differential selection and misclassification biases may also have resulted in over representation of conspicuity aid users amongst cases compared to controls. Social expectation from involvement in a collision crash may have resulted in cases who were not using conspicuity aids being less likely to participate than controls who were non-users. For similar reasons cases may have been more inclined to over-estimate their conspicuity aid use than controls. Validation data were available for only a small number of cases preventing quantification of exposure-related selection or outcome related misclassification biases and meaning that presence or otherwise of differential bias could not be confirmed. The study was also not able to accurately measure relative conspicuity arising from differences in performance of the conspicuity aids chosen.
The differences observed in traffic danger estimates may also be the result of selection bias as recruitment was restricted to public and private cycle parking which may have led to over-representation of controls from areas with greater numbers of cyclists and better infrastructure which are both thought to reduce crash risk. Failure to recruit the required sample size led to low precision in the estimates of odds ratios and an increase in the risk of incorrectly accepting the null hypothesis.
Conclusion
This study was designed to assess the effect of conspicuity aid use on the risk of crash for commuter and utility cyclists. A slightly greater proportion of cases than controls reported using conspicuity aids. There was therefore a raised odds ratio of collision crash involvement for those using conspicuity aids even after adjustment for a large number of important confounders. The study results do not demonstrate a protective effect as expected given previous work testing the effects of such aids on drivers’ awareness of cyclists and pedestrians. This study demonstrates the importance of understanding why many cyclists remain at risk of collision crash resulting in injury despite the use of conspicuity aids.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
