METHODS

For the purpose of this study a serious cyclist crash was defined as any crash which required admission for treatment in a public hospital in New Zealand. Public hospitals provide acute care for the overwhelming majority of admitted injury cases in New Zealand.⁶ Cases were selected from the New Zealand Health Information Service (NZHIS) public hospital morbidity data. The NZHIS records data on all public hospital discharges in New Zealand.

To obtain reliable estimates we used five years of national public hospital inpatient data, namely 1995–99.

The circumstances of injury are coded according to the “Supplementary Classification of External Cause of Injury and Poisoning” (E code) of the International Classification of Diseases (ICD).⁷ The E codes permit the identification and enumeration of victims classified under a variety of categories relating to “environmental events, circumstances and conditions”. Only admissions with a primary diagnosis of injury were included in the analysis. Readmissions for the treatment of the same injury, were excluded. The rationale for case selection is provided elsewhere.⁸

The potential population of interest was identified by two strategies. All crashes that involved a motor vehicle and occurred on a public road are coded in the range E810–E819: “Motor vehicle traffic accidents”. For these cases we identified all the cyclists by reference to the fourth digit of the E code which identifies the type of road user.

All crashes that involve other road vehicles and which may be on a public road are coded in the range E826–E829: “Other road vehicle accidents”. For these cases we also identified all the cyclists by reference to the fourth digit of the E code. We then sought to identify the subpopulation of incidents that occurred on a public road by reference to the place of occurrence code.

Injury severity in terms of threat to life was measured according to the abbreviated injury scale (AIS), an anatomical scale ranging from minor (1) to virtually unsurvivable (6).⁹ Scores were derived from the first four diagnoses using ICDMAP-90.¹⁰ We used the convention of ICD/AIS for referring to the indirectly derived scores.

For the purposes of estimating cumulative length of stay we traced each inpatient discharge for a period of 12 months. Initial investigations on tracing a population of cyclists injured in 1995 for six years revealed that all readmissions for further treatment occurred within a 12 month period. We used data from 2000 for calculating cumulative length of stay for the 1999 cases

Records of cyclists seriously injured on a public road where there was a motor vehicle involved were linked to the TCR database. The two datasets were linked on a variety of common fields (for example, surname, initials, gender, age, date of injury) using the probabilistic record linkage software AUTOMATCH¹¹ with a manual review of a sample of matched pairs and manual linkage of unlinked or “orphaned” records.

In summary we sought to identify three categories of crashes involving cyclists:

1. Involving a motor vehicle linked to a TCR.

2. Involving a motor vehicle not linked to a TCR.

3. Cyclist only (that is, where a motor vehicle was not involved).

The degree to which under-reporting of serious crashes involving motor vehicles was associated with various demographic, environmental, and injury factors was examined by using univariate and multivariate logistic regression analyses, similar to those employed by Alsop and Langley.¹ Based on this earlier work we sought to determine the effects of age, gender, ethnicity, threat to life, cumulative length of stay, region, day, month, and year of injury on the probability of a crash being recorded in the TCR database. For ethnicity the population was classified as Maori (the indigenous people of New Zealand) or non-Maori. The age groups we used in our analysis are based on those given in the New Zealand Travel Survey with those groups covering the 25–39 age range combined into one due to relatively low exposure for the age groups within this range.¹²

RESULTS

For the period 1995–99 there were 1033 cases identified from the hospital records as “motor vehicle traffic accidents” (E810–E819). There were an additional 5462 cycle crashes identified as “other road vehicle accidents” (E826–E829). Of the latter group 1892 were positively identified as having occurred on a public road (hereafter referred to as cyclist only crashes). A further 896 were positively identified as having occurred off a public road. For the remaining 2674 it was not possible to determine the location. In summary we were able to identify 2925 cyclist crashes on public roads, 35% of which involved motor vehicles. This estimate is conservative given the large number of cyclist-involved crashes in which the location of the crash was unspecified.

Of the cyclist only crashes there were 163 cases (9%) with an injury of ICD/AIS of 3+. The comparable number of crashes involving a motor vehicle was 164 (16%) (table 1). Table 2 shows that cyclist only crashes comprised 46% of cumulative length of stay for all cyclists injured on a public road. The average length of stay, for cyclist only crashes was approximately half that for motor vehicle involved cyclist crashes. Similarly crashes involving a motor vehicle were approximately three times more likely to result in hospital stays of more than seven days. Nevertheless victims of cyclist only crashes accounted for 39% of these longer stays.

Of the 2925 crashes, 652 (22%) were linked to a TCR (table 1). Of the crashes involving motor vehicles (n = 1033), 562 (54%), could be linked to the LTSA database. Ninety (5%) of the 1892 cyclist only crashes linked.

There were significant univariate associations between the probability of being recorded in the LTSA database and gender, ethnicity (table 3), ICD/AIS (χ² = 26.252, p<0.001), and cumulative length of stay (χ² = 34.006, p<0.001). There were no significant variations in reporting rates by day, month, or year of injury, region, or age.

Following the recommendation of Hosmer and Lemeshow,¹³ all factors moderately significant from the univariate analyses were included in the multivariate logistic regression model. The model provided an adequate fit to the data, indicated by the Hosmer and Lemeshow goodness-of-fit test statistic (χ² = 4.97 with 8 df, p = 0.76).¹³ With the exception of age the results from the multiple logistic regression were consistent with those found in univariate analyses (table 4).

DISCUSSION

In this study the NZHIS hospital discharge classification served as our gold standard. This database is likely to contain some coding errors. For example, 90 of the cyclist only crashes linked to the LTSA database. By definition these crashes were subject to a detailed crash report by a police officer, they are thus likely to have been incorrectly coded in the NZHIS database as non-motor vehicle traffic crashes. Similarly, while in law it may be clear what is classified as a public road crash, victims may adopt a more liberal definition of what is a road and what is public in their reports to police. Finally, the ICD coding instructions state: “A motor vehicle accident is assumed to have occurred on the highway unless another place is specified, except in the case of accidents involving only off-road motor vehicles, which are classified as non-traffic accidents unless the contrary is stated”. Consequently, hospital derived estimates may over-state the incidence of motor vehicle crashes on a public road, and under-state the incidence of off-road crashes and on-road crashes not involving a motor vehicle. We are unable to arrive at an estimate of the size of these respective errors, however, our estimates for the cyclist only crashes are likely to be conservative given the large number of events (n = 2674) for which place of occurrence could not be ascertained. The size of this category is cause for concern, since it seems likely that a significant portion of these cases may have occurred on a public road. Accordingly our estimate of 22% reporting for all cyclist crashes on public roads is likely to be low.

While the extent of under-reporting of motor vehicle crashes on public roads in New Zealand has been documented for users of motor vehicles,¹ this report represents the first attempt to determine the extent for cyclists in New Zealand. The under-reporting is substantial at 46% for cyclist crashes involving a motor vehicle but is within the range previously reported for other road users. Alsop and Langley demonstrated that under-reporting ranged from a high of 59% for motor cyclists injured in single vehicle crashes to a low of 21% for passengers injured in multiple vehicle crashes.¹

A large proportion of serious motor vehicle involved cases were linked to TCRs, however, relatively large numbers of the non-reported cases were also serious in terms of threat to life and cumulative length of stay in hospital. Furthermore, there was significant variation in reporting by age, severity, ethnicity, and length of stay. Consequently LTSA estimates cannot be corrected by a simple adjustment.

The relatively large number of cyclist only crashes that are not recorded in the TCR database was not unexpected but it is nevertheless cause for concern, given that a significant proportion of them resulted in serious threat to life and length of stay in hospital. There are a number of possible reasons why the public do not report such crashes to the police. First, by law they are not required to do so. Secondly, in most cases only the cyclist was injured, and many may consider that they would be inviting a prosecution for illegal road behaviour by reporting an incident. Thirdly, in contrast to motor vehicle crashes, property damage is usually not significant in absolute terms and consequently there is no incentive to report to comply with insurance requirements. While the law could be changed to make reporting mandatory, we believe this is unlikely to result in a significant improvement in reporting. We endorse the recommendation of Hvoslef that crashes involving non-motorised vehicle road users should be extracted from hospital records,¹⁴ as we have done here for cyclists, and published separately. Failure to do this runs the risk of these important crashes not receiving prevention resources commensurate with the size of the problem.

The LTSA has sought to address the generic issue of under-reporting in its annual publication Motor accidents in New Zealand by providing an estimate of crashes resulting in hospital inpatient treatment. The relevant section of the report typically contains estimates for cyclists. Unfortunately the estimates are for all non-motor vehicle crashes irrespective of whether they occurred on the road. As the results presented here demonstrate, this approach results in an over estimation of the size of the problem, since many of these incidents would have occurred off-road.

In summary, we estimated that each year there were an average of 584 cyclist crashes on public roads which require public hospital inpatient treatment. LTSA has records for only: 22% of these crashes, 40% of the most life threatening cases (those with ICD/AIS 3+), and 43% of the most costly cases requiring a length of stay in hospital of 7+ days. Greater effort and precision needs to be applied by LTSA to documenting the burden of cyclist-involved crashes, especially cyclist only crashes. This would be greatly facilitated if NZHIS sought to encourage hospitals to improve the place of occurrence coding for the non-motor vehicle incidents resulting in inpatient treatment. Hospital admission data, however, cannot provide the information on the circumstances surrounding a crash which is found in the typical TCR. Since this information is critical to the development of effective countermeasures LTSA should be actively seeking to obtain it for cycle only crashes, since the circumstances of crashes are different to those involving cars and as such different prevention strategies may be required.