Accurate coding of injury event information is critical in developing targeted injury prevention strategies. However, little is known about the validity of the most universally used coding system, the International Classification of Diseases (ICD-10), in characterising crash counterparts in pedal cycling events. This study aimed to determine the agreement between hospital-coded ICD-10-AM (Australian modification) external cause codes with self-reported crash characteristics in a sample of pedal cyclists admitted to hospital following bicycle crashes. Interview responses from 141 injured cyclists were mapped to a single ICD-10-AM external cause code for comparison with ICD-10-AM external cause codes from hospital administrative data. The percentage of agreement was 77.3% with a κ value of 0.68 (95% CI 0.61 to 0.77), indicating substantial agreement. Nevertheless, studies reliant on ICD-10 codes from administrative data should consider the 23% level of disagreement when characterising crash counterparts in cycling crashes.
Statistics from Altmetric.com
Accurate coding of injury event information is critical in understanding injury patterns and developing targeted injury countermeasures. Injury event data are routinely coded from information in hospital records in Australia. The most universally used coding system is the International Classification of Diseases (ICD-10),1 which includes injury event codes. In 1998, a modified version of ICD-10 was developed for use in Australia for greater specificity in areas of interest relevant to the Australian clinical and epidemiological setting (the ICD 10th Revision—Australian modification (ICD-10-AM)).
Previous studies have shown increasing rates of serious injury in pedal cyclists,5 ,6 leading to an increased focus on injury prevention strategies for this vulnerable road user group. While ICD-10 coding is commonly used to characterise bicycle crashes,7–9 the validity of ICD-10-AM external cause codes in characterising pedal cycling events is unknown. This study assessed the agreement between hospital-coded ICD-10-AM external injury causes and self-reported crash characteristics in pedal cyclists admitted to hospital following bicycle crashes.
Cycling-related trauma patients were prospectively recruited from two hospitals in the state of Victoria, Australia; the two adult major trauma services (level 1 trauma centre equivalent). The study methods have previously been described,10 and are summarised here.
Patients were invited to participate if they met the following criteria: emergency admission >24 hours for management of a cycling-related injury; and eligible for registration on the Victorian State Trauma Registry (VSTR) or the Victorian Orthopaedic Trauma Outcomes Registry (VOTOR). As patients were directly interviewed about their crash circumstances, patients unable to consent to participate in the study due to pre-existing conditions, or due to the injuries sustained, were excluded. Ethics approval for the study was obtained from the human research ethics committees at the participating hospitals and Monash University.
Participants completed a structured interview during their hospital stay. Trained research nurses conducted the interviews, which included questions about demographic details, crash circumstances, potential crash risk factors identified from the literature and the events leading to the crash. Relevant interview questions to this study were:
If in a single vehicle crash, what objects did you (the rider) hit?
If in a multiple vehicle crash, what did you (the rider) collide with?
The response options are provided in online supplementary material.
VSTR and VOTOR data linkage
Interview data were linked with data from the VSTR and VOTOR. The VSTR is a population-based registry that collects data about all hospitalised major trauma patients in Victoria.11 VOTOR is a sentinel site registry collecting data about all adult (>15 years) orthopaedic trauma patients with a length of stay >24 hours, and admitted to four participating hospitals. Data extracted from the registries included ICD-10-AM coded injury event details (V10–V19). In Australia, ICD-10-AM coding is routinely undertaken by the hospital clinical coders to capture diagnoses related to the admission. For injury admissions, the external cause, activity and place are routinely coded using ICD-10-AM.
Responses from the interview questions were mapped to three-character level ICD-10-AM external cause codes (V10–V19) (see online supplementary appendix) by the lead author (BB) and a qualified ICD-10-AM coder. Discrepancies were resolved through discussion. Some participants provided more than one response, resulting in more than one ICD-10-AM external cause code. These participants were excluded from the main analysis and their responses reported separately. Additional subanalyses were performed to understand the level of agreement when crash counterparts were collapsed into specific subgroups: motor vehicles (V12–V15) versus non-motor vehicles (V10, V11, V16–V19) and single vehicle collisions (V17–V19) versus multivehicle collisions (V10–V16). Agreement between ICD-10-AM codes and self-reported crash information was assessed using percentage of agreement and Cohen's kappa (κ) statistic,12 with κ scores interpreted as fair: (κ=0.21–0.40), moderate (κ=0.41–0.60), substantial (κ=0.61–0.80) and almost perfect (κ=0.81–1.00).13
The responses of 144 cyclists were mapped to ICD-10-AM external cause codes; 3 (2.1%) had responses that were mapped to multiple ICD-10-AM external cause codes and were excluded from the main analysis.
Most crashes were non-collisions (44%), and collisions with motor vehicles (30%) or other cyclists (18%). When comparing self-reported crash counterparts with ICD-10-AM external cause codes (table 1), the percentage of agreement was 77.3% (κ=0.68, 95% CI 0.61 to 0.77), indicating ‘substantial’ agreement. The agreement between the participating hospitals was similar (78.5%, κ=0.71 vs 76.3%, κ=0.65).
Most disagreements occurred when the ICD-10-AM code was either a non-collision (n=17 disagreements), or in other and unspecified transport accidents (n=7 disagreements) (table 1). For the non-collision events in disagreement, self-reported counterparts were identified as other pedal cyclists, fixed or stationary objects or other and unspecified events. For the other and unspecified transport accidents in disagreement, these were commonly self-reported as non-collisions.
Cases that were mapped to multiple ICD-10-AM external cause codes
In the self-reported two single vehicle collisions (cases 1 and 2), ICD-10 coding preferred the non-collision event (case 1) or there were coding discrepancies (case 2) (table 2). In the self-reported multivehicle collision (case 3), the fixed or stationary object (presumably the power/light pole) was preferred over the vehicle counterpart in ICD-10-AM coding (table 2).
When crash counterparts were grouped as motor vehicle versus non-motor vehicle collisions, the percentage agreement was 97.2% (κ=0.94, 95% CI 0.87 to 1.00) (table 3). When grouped as single vehicle collisions versus multivehicle collisions, the percentage agreement was 92.2% (κ=0.84, 95% CI 0.76 to 0.93).
This study evaluated the level of agreement between responses to detailed interview questions and ICD-10-AM external cause codes in a cohort of hospitalised injured cyclists. Our results demonstrated substantial agreement with a percentage of agreement of 77% at the three-character level.
While previous studies have evaluated the inter-rater reliability of ICD-10 external cause codes,3 ,14 our study is the first to validate ICD-10-AM external cause codes for cycling crashes against self-reported crash characteristics. Despite substantial agreement between these measurements, 31 external causes were misclassified. These were predominantly collisions with other pedal cyclists, fixed or stationary objects or non-collisions. Specifically, ICD-10-AM coded data overestimated the proportion of cycling crashes that were non-collision events, while the self-reported data indicated that these crashes were commonly collisions with another cyclist or a fixed or stationary object. As primary prevention efforts differ between these crash types, these findings should be considered when using ICD-10-AM external cause codes to inform interventions. Additionally, these are likely the most challenging scenarios to code, such as when other cyclists may precipitate non-collision events or non-collision events may result in the cyclist impacting with a fixed or stationary object. A lack of information available at the crash scene or inadequate information contained in ambulance or police notes may contribute to the challenges associated with coding these cases. Collapsing crash counterparts into broader crash types resulted in improved levels of agreement. Our results demonstrated that ICD-10-AM external cause codes provided a reliable basis for classifying bicycle crashes with and without motor vehicles, particularly when ‘other and unspecified’ crash counterparts were classified as ‘non-motor vehicle collisions’. Classifying crashes as single or multivehicle events also demonstrated excellent agreement, although the percentage of agreement was lower than when crashes were classified as with and without motor vehicles (92% vs 97%). These findings suggest that these broader classifications of crash types may be more appropriate than individual ICD-10-AM codes at the three-character level when using ICD-10-AM codes to classify cyclist crash counterparts.
There are other sources of variation that need to be considered when reviewing the results of this study. First, there is a degree of subjectivity in translating interview responses to ICD-10-AM external cause codes. However, we attempted to reduce the potential for error by using two independent coders, one an experienced ICD-10-AM coder, to translate the interview responses. We were unable to review medical records to augment the level of information about the crash counterpart provided in the interview; such information would be important to collect in order to understand the sources of error in coding. Second, recall bias may impact on the accuracy of postcrash interview responses in determining crash counterparts. Concurrently, there may be variation in the information sources and level of information provided in medical records that ICD-10-AM coders use to classify crash counterparts. We were not able to control for these potential sources of variation. Additionally, participants were recruited from two major trauma services with a high degree of scrutiny of trauma coding. The results may not be representative of non-trauma hospitals and the patient population may not be representative of the broader injured cyclist population.
Given the widespread utility of ICD-10 coding in understanding injury patterns and informing injury prevention strategies, there is a need for greater evaluation of the accuracy and reliability of coding. ICD-10 coding is commonly used to compare crash counterparts and crash characteristics in pedal cyclists8 ,9 and while our results demonstrated substantial agreement between self-reported crash characteristics and ICD-10 external cause codes, further research is required to understand coding errors and improve accuracy.
This study compared self-reported crash characteristics with ICD-10-AM external cause codes in a cohort of hospitalised injured cyclists. We demonstrated substantial agreement between these two data sources, however we observed a 23% level of disagreement which should be considered when using ICD-10 coding to inform injury prevention strategies.
What is already known on the subject
Accurate and reliable coding of injury event information is critical in developing targeted injury prevention strategies.
There is variability in the reliability of International Classification of Diseases (ICD-10) external cause coding across all causes of injury.
What this study adds
Validation of ICD-10 external cause codes from self-report is rarely performed. This study adds to the limited literature on the validity of these codes in bicycle crashes.
We demonstrated substantial agreement between self-reported crash characteristics and crash counterparts as defined by the third character of ICD-10 external cause codes (κ value of 0.68).
However, ICD-10 coded data overestimated the proportion of cycling crashes that were non-collision events. This finding should be considered when using ICD-10 data to inform injury prevention strategies.
The authors would like to acknowledge the contributions of Mandy Brown, Melissa Hart, Carol Roberts, Jasmine Fischer and Jane Ford.
Twitter Follow Ben Beck @DrBenBeck
Contributors BB, CLE, PC, MS, RJ, AB, EE and BG contributed to the planning of the study. BB, CLE and BG contributed to the analysis. BB drafted the manuscript and CLE, PC, MS, RJ, AB, EE and BG provided critical review. BB and BG are responsible for the overall content.
Funding The cycling study from which the interview responses were derived was specifically supported by a Monash University, Faculty of Medicine, Nursing and Health Sciences Strategic Grant. The Safer Cycling in the Urban Road Environment Study15 is supported by an Australian Research Council Grant (number: LP130100380). The Victorian Orthopaedic Trauma Outcomes Registry is funded by the Transport Accident Commission via the Institute for Safety Compensation and Recovery Research. BB received salary support from the National Health and Medical Research Council (NHMRC) Australian Resuscitation Outcomes Consortium Centre of Research Excellence (#1029983). CLE is supported by a NHMRC of Australia Early Career Fellowship (GNT1106633). PC, BG and MS were supported by a Practitioner Fellowship (#545926), Career Development Fellowship (GNT1048731) and a Research Fellowship (#1043091) from the NHMRC, respectively.
Competing interests None declared.
Ethics approval Human Research Ethics Committees at The Alfred Hospital and Royal Melbourne Hospital, and the Monash University Human Research Ethics Committee.
Provenance and peer review Not commissioned; externally peer reviewed.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.