Design, setting and participants

The sampling frame comprised adult cyclists aged 16 years and over who enrolled online in the Lake Taupo Cycle Challenge, New Zealand's largest mass cycling event held each November. Participants have varying degrees of cycling experience ranging from competitive sports cyclists and experienced social riders to relative novices of all ages.

Recruitment was undertaken at the time of the 2006 event for the majority of participants, as described, in detail, elsewhere [16]. In brief, email invitations, containing a hyperlink to the study information page, were sent to 5653 contestants who provided their email addresses at registration for the event. Those who agreed to participate in the study were asked about demographic characteristics, general cycling activity in the past twelve months and habitual risk behaviours with options ranging from ‘never’ to ‘always’. They were also asked about the number of bicycle crashes they had experienced in the past twelve months as well as the number of crashes that required consulting a doctor or other health professional (e.g., physiotherapist, chiropractor). The questionnaire was completed and submitted by 2438 cyclists (43.1% response rate). Another 190 cyclists were recruited from the 2008 event by including a short description about the study in the event newsletter.

Crash outcome data

Crash outcome data were collected through record linkage to four administrative databases, covering the period from the date of recruitment to 30 June 2011.

In New Zealand, the Accident Compensation Corporation (ACC) provides personal injury cover for all residents and temporary visitors to New Zealand no matter who is at fault. The claims database is a major source of information on relatively minor injuries with over 80% of the claims relating to primary care (e.g., GPs, emergency room treatment) only [17].

The hospital discharge data contains information about inpatients and day patients discharged after a minimum stay of three hours from all public hospitals and over 90% of private hospitals in New Zealand [18]. The mortality data contains information about all deaths registered in the country [19]. Diagnoses in each hospital visit and underlying causes of death are coded under ICD-10-AM. Bicycle crashes were identified using the E codes V10-V19. Readmissions were identified as described previously [20] and excluded.

In New Zealand, it is mandatory that any fatal or injury crash involving a collision with a motor vehicle on a public road be reported to the police. The crash analysis system data contains information on all police-reported bicycle collisions involving a motor vehicle.

For each participant, bicycle crashes identified across different databases were matched based on the date of crash allowing for a two-day difference, so as to avoid double-counting of the same crash.

Analyses

The study sample was restricted to 2590 participants who were resident in New Zealand at recruitment. The participants were identified as having previous crash experience if they reported in the baseline survey that they had had one or more bicycle crashes in the past twelve months. Five participants with missing data on this variable were excluded.

All analyses were performed using SAS (release 9.2, SAS Institute Inc., Cary, North Carolina). Baseline data were presented as means with standard deviations and medians with interquartile ranges for continuous variables and percentages for categorical variables. All the data were complete for 2438 participants (94.3%). Missing values were computed using multiple imputation with 25 complete datasets created by the Markov chain Monte Carlo method [21], incorporating all baseline covariates and injury outcomes. Crude and adjusted differences in baseline characteristics between the participants who reported having previous crash experience and the rest of the cohort were assessed using PROC GLM. Differences were adjusted for quintiles of the propensity score. The propensity score or expected probability of having experienced a crash was computed with a multivariate logistic regression model in which previous crash experience was the dependent variable and all the baseline covariates presented in Table S1 (including age, gender, ethnicity, level of education, body mass index, the amount of cycling in general, off-road, in the dark and in a bunch, cycling to work, type of bicycle, use of helmet and conspicuity aids, distraction, neighbourhood deprivation, urban-rural status and region of residence) were the independent variables. Detailed information on the associations between the baseline covariates and crash experience at baseline and during follow-up was reported in previous publications from this study [15], [16]. The participants were then ranked by their estimated propensity score and grouped within quintiles as suggested previously [22]. The propensity score was then evaluated with: (1) a reasonable Nagelkerke's r² statistic as a measure of fit, (2) a c-statistic between 0.65 and 0.85 as a measure of discriminative power, (3) similarity between the predicted and observed proportion of participants with previous crash experience within quintiles as a measure of good calibration, and (4) balanced covariates within quintiles [23], [24]. It has been demonstrated that five strata are sufficient to eliminate approximately 90% of the bias in each covariate [25].

Using bicycle crash data extracted through record linkage, incidence rates of repeated events were calculated for the participants with previous crash experience and the rest of the cohort using the person-years approach. Confidence intervals (CI) were based on the Poisson distribution. The rates were then presented by quintiles of the propensity score. The participants were censored on 30 June 2011 or date of death.

Cox proportional hazards regression modelling for repeated events was carried out using a counting process approach to assess hazards of subsequent crash involvement associated with previous crash experience. The model took into account all crash events experienced during follow-up. For the first event, the survival time was the time to the first event (or time until censored). For all later events, the survival time was the time from the previous event to the next event (or time until censored). Hazard ratios (HR) were adjusted for all the baseline covariates mentioned in Table S1, for the propensity scores and for quintiles of the propensity scores. Analyses were then stratified by quintiles and crude and propensity score adjusted HRs were calculated. In addition, previous crashes experienced were categorised by whether medical treatment was required, and similar Cox regression analyses were undertaken.

Ethics statement

As the study used a web-based questionnaire for recruitment, informed consent was obtained electronically. All eligible participants were provided with detailed information about the study including the record linkage procedure. At the end of the information sheet, they could click a Yes/No button that says “I have read the information sheet and consent to take part in the Taupo Bicycle Study”. Clicking the “Yes” button was regarded as consent and only those who did so were taken to the next page containing the study questionnaire. This information was stored electronically for all records. Informed consent from a third party was not sought as the study sample was restricted to adult cyclists. Age of potential participants was checked at the beginning of the questionnaire by asking for their birth date. Only those aged 16 and over at the time of the survey could continue to complete the questionnaire. Ethical approval for the study including the consent procedure was obtained from the University of Auckland Human Participants' Ethics Committee. Ethical approval for linkage to the claims data was also obtained from the ACC Research Ethics Committee.

Main findings

We found that crash involvement propensity influenced the association between previous crash experience and future crash involvement in New Zealand cyclists. In the stratified analysis, the association was significant only in the highest two quintiles of the propensity score. The association was also stronger for previous crashes that had received medical care compared to those that had not.

Strengths and limitations

In this prospective cohort study, baseline data were near-complete as mandatory fields and validation checks were incorporated in the web questionnaire. Crash outcome data were collected from four administrative databases, thereby minimising potential biases associated with loss to follow-up and self-reports.

The outcome data, however, exclude minor crashes not coming to the attention of the police or medical personnel but these events are captured in self-reported data collected at baseline. Although we were not able to estimate the association for future minor crashes, we have reported differences in the associations by the need for medical care in the previous crash. Ascertainment of crash outcomes may be affected by personal, social and health service factors [26] as well as the quality of individual data sources and record linkage [27]. Self-reported crash experience at baseline may also be affected by failure to recall [28]and socially desirable responses [29]. Likewise, self-reported covariates may not be accurate and may change over time. Nevertheless, potential misclassifications of crash outcomes and exposures tend to be non-differential in a prospective cohort study [30]. While the propensity score method was used to balance baseline covariates between exposure groups, the effect of unmeasured or unknown confounders may still be present. Finally, our participants cannot be considered representative of all New Zealand cyclists; however, this may have minimal impact on the risk estimates [31]. Importantly, the participants represented a wide variation with regard to demographics, cycling exposure and experience and the results will be valid for all cyclists and traffic environments similar to New Zealand.

Interpretation

Our finding showing that cyclists with previous crash experience had a 43% higher risk of future crash involvement is in accordance with the existing literature [6]. However, unlike previous research, this study was able to account for (stratify over) a range of covariates including demographics, risk exposure and residential characteristics by using the propensity score approach.

In the stratified analysis, the association was significant only in quintiles 4 and 5 (where the likelihood of having experienced a crash was more than 33%), indicating that the accident (or injury) proneness phenomenon applied only to the high-risk group in this study. Cyclists in this group were younger, more likely to be males and university graduates, to cycle off-road, in the dark, in a bunch and for commuting, to engage in distracting activities such as listening to music while riding, and to reside in an urban area, and less likely to always use conspicuity aids. Explanations for “crash proneness” have included personality maladjustments, cognitive failures, stress and other mental and physical health problems [32], [33], [34], [35]. Another possibility is that crash repeaters may be more frequently exposed to environmental hazards, for example, poor road surface, poor cycling facilities and bad weather. This is beyond the scope of this paper but is worthy of further investigation.

This study found that previous experience with medically attended crashes carried a higher risk of future crash involvement than experience with other crashes, particularly in quintiles 3, 4 and 5. Crashes requiring medical care tend to be more severe, and may result in physical, psychosocial and possibly cognitive and behavioural sequelae [36], [37], [38] that affect the risk of re-injury of the same type and location [39], [40]. On the other hand, the observed association may simply reflect a higher rate of service utilisation among cyclists with previous crashes attended medically (note that crash outcome data collected during follow-up covered only those crashes that came to the attention of medical personnel or police).

Our findings suggest that Emergency Medical Services (EMS) and other health care professionals may have a role in identifying crash-prone high-risk cyclists, in delivering targeted interventions to prevent future crashes and in ensuring adequate rehabilitation before patients resume riding a bicycle. The need for EMS involvement in such activities has been recognised [41], [42], [43], [44], [45] and the success of EMS-based programs has been documented, for example, in increasing helmet use among children and adolescents [46], [47], [48]. Bicycle safety was one of the most important injury prevention activities in US emergency departments and trauma centres [49], [50]. Other studies reported that few EMS providers practise injury prevention activities [51] and some were sceptical about the appropriateness and potential impact of such activities [52]. Such information is not currently available in New Zealand.