U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-RD-99-078
Injuries to Pedestrians and Bicyclists: An Analysis Based on Hospital Emergency Department Data
CHAPTER 6 - INJURY ESTIMATES FROM THE COMBINED DATA
In this chapter, information from the emergency department database is combined with statewide hospital discharge and motor vehicle crash data to estimate overall numbers of pedestrians and bicyclists being injured. Two approaches are examined, the first relying on the percentage of emergency department cases hospitalized and overall number hospitalized, and the second on the percentage of emergency department cases identified in police crash files and overall number of police-reported cases. Both approaches are described below. A final discussion section highlights some of the constraints of trying to project overall numbers of injured pedestrians and bicyclists.
Estimates Based on Hospitalization Data
This first approach is the one that was followed in Stutts et al. (1990) to estimate the total number of bicyclists being treated annually in North Carolina hospital emergency departments. For this study, data on injured bicyclists were collected from 10-15 hospital emergency departments spread out geographically across the State during the late spring and summer of 1985 and again in 1986. Of the 649 cases identified, 17.3 percent were injured in collisions with motor vehicles and 6.1 percent were hospitalized. Based on a 1980 survey of pediatric hospital discharges that showed 800 children hospitalized in the state for bicycle-related injuries, it was estimated that more than 13,000 children received treatment in North Carolina hospital emergency departments for injuries incurred while bicycling.
For the current study, emergency department data were collected from hospitals in three States (California, New York, and North Carolina) over a 1-year time period. E-coded hospital discharge data were also obtained from these same States. Table 56 summarizes information on the percentage of bicycle and pedestrian emergency department cases in each State that were hospitalized as a result of their injuries.
Table 56. Percentage of emergency department cases requiring hospitalization, by type of injury.
The percentages vary greatly among the States, with North Carolina showing by far the highest percentage hospitalized. This probably reflects the fact that both of the two North Carolina hospitals that participated in the study were located in smaller communities serving large rural areas. Pitt County Memorial Hospital, which contributed about three-fourths of the cases, is located in a community with a population of about 50,000, but the hospital is also a major trauma center, drawing patients from a 10-county region in the rural northeastern part of the State. Crashes occurring in rural areas are more likely to involve higher vehicle speeds, to occur at nighttime, and to involve alcohol--all factors that can exacerbate the level of injury for the parties involved. In contrast, the New York State cases all came from a single large urban area (Buffalo), while the California cases came from three hospitals that served smaller metropolitan and suburban areas.
Along with this emergency department data, Table 57 summarizes available information on the number of hospitalized pedestrians and bicyclists in each of the three States. The data reflect the most recent year available from each State. The California and New York State totals represent the actual number of hospitalizations as recorded in statewide hospital discharge datafiles (see chapter 2 and appendix D). For North Carolina, the numbers represent a weighted estimate based on available data from the North Carolina Medical Database Commission for fiscal year 1994 (October 1, 1993 - September 30, 1994). The numbers reported by the Commission were weighted by a factor of 2.3 to adjust for statewide underreporting of E-codes (i.e., only 43.4 percent of the cases on file that had an injury diagnosis also had a recorded E-code), and by a factor of 2.2 to adjust for underreporting by the hospitals (i.e., only 68 of the State's 152 accredited hospitals contributed to the file). Due to problems in identifying pedestrian-only events with the available E-coded data, no counts are presented for this category.
Table 57. Number of injured bicyclists and pedestrians hospitalized, based on available statewide hospital discharge data.
Based on the information shown in these two tables, it was possible to estimate the total number of bicyclists and pedestrians receiving emergency department treatment in each State by dividing the number of hospitalized cases (from table 57) by the proportion of such cases hospitalized (from the corresponding cell in table 56). For example, to estimate the total number of bicycle-motor vehicle cases treated in California hospital emergency departments, 1,272 was divided by .197, producing the annual estimate of 6,457 cases. Table 58 shows the results of these calculations for each State.
Table 58. Estimated number of bicyclists and pedestrians receiving hospital emergency department treatment, based on data in tables 56 and 57.
These numbers probably underestimate the total numbers of pedestrians and bicyclists treated in hospital emergency departments, because the percentages hospitalized (from table 56) are higher than expected. Baker et al. (1993), for example, have reported that nationally, 10.4 percent of bicycle-motor vehicle cases, 2.0 percent of bicycle-only cases, and 4.0 percent of all bicycle cases treated in emergency departments are hospitalized. If the first two percentages are applied to the counts of bicycle hospitalizations in table 57, the bicycle projections are increased as follows:
These numbers appear unrealistically high, however, at least in the case of bicycle-only events. Based on these numbers, our three States alone would account for 266,000, or nearly half, of the widely accepted figure of 550,000 annual bicycle-related injuries estimated by the Consumer Product Safety Commission's NEISS data.
It is likely that the true percentages of hospitalized cases lie somewhere between the estimates produced by the current study and those reported by Baker et al. (1993). Two factors are crucial in deriving such estimates: (1) a representative sampling of hospital emergency departments, and (2) a complete count (or representative sampling) of cases attending that hospital. For the current study, the participating hospital emergency departments represented a cross-section of larger and smaller hospitals that were located in different regions of the country and that served various sizes of communities. However, they were not chosen to typify their respective States. To produce more valid statewide estimates, a better approach would have been to select a larger number of hospitals from only one or two States (as was done for the Stutts et al. (1990) study).
The second requirement for valid estimates of the percentage of hospitalized cases is complete reporting within a hospital. In particular, it is important that hospital emergency departments capture less severe injury cases at the same rate as more severe injury cases, and that they capture "bicycle only" and "pedestrian only" events to the same extent as their more readily identified motor vehicle counterparts. How well the current study accomplished this goal is difficult to assess. The New York and California hospital emergency departments had similar ratios of bicycle-only to bicycle-motor vehicle cases (2.4 for New York and 2.8 for California), and although New York had a higher ratio of pedestrian-only cases (2.9 versus 1.9 for California), this might be expected due to the large number of "slips on ice" cases. With these cases omitted, the New York ratio drops to 1.9, the same as for California. The ratios for North Carolina are considerably lower, 1.4 for bicycle only versus bicycle-motor vehicle, and 0.22 for pedestrian only versus pedestrian-motor vehicle. Since in North Carolina the majority of cases were identified retrospectively from E-coded emergency department records, it is unlikely that bicycle-only cases were missed unless they were incorrectly E-coded. However, it seems clear that the New York State and California hospital emergency departments captured a broader array of pedestrian-only events than what could be identified using the E-coded hospital records.
Given this lack of certainty in the study data, a different approach that does not rely on estimates of the proportion of cases hospitalized was examined. The results of applying this second approach to the study data are described below.
Estimates Based on Matched Emergency department and Crash Data
The second approach examined for making statewide projections of injured bicyclists and pedestrians utilized the emergency department data collected in conjunction with police-reported motor vehicle crash data. Since the police-reported data contains virtually no crashes not involving a motor vehicle, the initial focus of this exercise was on pedestrian- and bicycle-motor vehicle injury cases.
A key piece of information needed with this approach was the percentages of emergency department cases that were also identified in the State motor vehicle crash files. Since no names, addresses, social security numbers, or other unique identifiers were available either for the emergency department data or the State crash data, the process of "matching" cases was carried out based on the following variables that were available in both files:
Using these variables, the goal was to identify those cases in the emergency department files that were also found in the State motor vehicle crash files. To carry out the matching, a listing was generated of pedestrian and bicyclist cases identified in each of the State crash files, containing the values for the above variables and ordered by the date of the injury event. To reduce the number of potential matches needing to be checked, the crash file listings were restricted to cases occurring in counties that might reasonably be serviced by one of the participating hospital emergency departments. For example, for North Carolina, only counties in the eastern part of the state were included in the crash file listing, while for New York State, the counties were restricted to those in the Buffalo area (Erie, Cattaraugus, and Niagara).
The process then became one of checking case-by-case to determine whether each reported emergency department case was duplicated in the State crash file listing. For those cases where a match was uncertain, the hard copy of the emergency case report was checked for additional information that might facilitate a decision. For example, in some cases, the narrative might provide an approximate age if no exact age was available, or note that the time of the injury event was only approximate. Cases were coded at four levels: match, probable match, possible match, and no match. For a definite match to occur, the crash date, victim age and sex, county or city, and approximate time would all need to be in agreement. A probable match might have one of these items (usually the time or location, but not age) in disagreement or missing, while possible matches would generally have two or more "disagreements" or missing pieces of information.
Table 59 summarizes the results of the matching process for the emergency department data collected in each of the three participating States. The percentages shown are based on the number of actual plus probable matches, but exclude the very small number of "maybe" matches recorded. Since probable matches are included, these percentages probably overestimate the percentage of emergency department cases that were also reported in the State motor vehicle crash files. The table shows that 43-45 percent of the California cases, 43-56 percent of the New York State cases, and 67-68 percent of the North Carolina cases were matched. Pedestrians had a somewhat higher match rate than bicyclists for the New York State data.
Table 59. Percentage of cases reported by participating hospital emergency departments identified in State crash files.
* Includes seven cases identified by police as bicyclists.
The higher matching rate for the North Carolina cases may, to some extent, reflect the more rural nature of the sample and the generally more serious level of injuries sustained. However, it should be noted that even in the earlier study of bicycle emergency department injuries reported in Stutts et al. (1990), 60 percent of the bicycle-motor vehicle cases were matched. Another interesting aspect of the North Carolina matching was that 7 of the 91 pedestrian-motor vehicle cases identified in the emergency department file were matched to bicycle-motor vehicle crashes in the crash file. Thus, it appears that at least in some instances, hospital emergency department personnel may not be aware that the patient being treated is a bicyclist. If these cases had been counted as "non-matches" (i.e., if only the pedestrian crash listing had been checked), the percentage matched would have dropped from 67.9 to 62.7 percent. (The "cross-file" matching was possible with the North Carolina cases since date of birth was available as a matching variable, providing added confidence in the matching process. No attempt was made to match the California and New York pedestrian cases to their State's bicycle crash listing.)
The second piece of information needed to produce statewide emergency department estimates was the actual number of police-reported bicyclists and pedestrians injured in motor vehicle crashes. These data are reported in table 60 for 1995 only. Almost all of the cases involved some level of injury.
Table 60. Number of police-reported pedestrians and bicyclists injured, based on 1995 State motor vehicle crash file data.
Finally, table 61 presents the projected statewide hospital emergency department visits, calculated by dividing the total number of police-reported cases by the proportion of emergency department cases matched to the State crash files. For example, to obtain the estimate of 34,134 bicyclists treated in California hospital emergency departments, 14,780 police-reported cases (from table 60) was divided by .433, the proportion of emergency department cases matched (from table 59).
Table 61. Estimated number of bicyclists and pedestrians receiving emergency department treatment, based on data in tables 59 and 60.
To also obtain an estimate of the total number of injured bicyclists and pedestrians, including those injured in falls and other non-collision events, the numbers in table 61 can be divided by the proportion of bicycle (or pedestrian) events reported by the participating hospital emergency departments in each State that involved a motor vehicle collision. For example, 26.0 percent of the bicycle injury cases reported by the three participating California hospitals involved a motor vehicle, so that the projected number of all bicycle injury cases would be 34,134 / .260, or 131,285, and the number of bicycle-only injuries would be 131,285 minus 34,134, or 97,151. Table 62 shows the percentages of emergency department reported cases involving a motor vehicle, and table 63 the projected total bicycle and pedestrian injuries treated in hospital emergency departments in each State.
Table 62. Percentage of emergency department bicycle and pedestrian cases that involved a collision with a motor vehicle.
The estimates reported in table 63 are substantially higher than their counterparts in table 58. They may also be closer to reality, at least for California and New York State. (The North Carolina projections are too low, due to the higher severity of the cases reported and the higher frequency of motor vehicle involvement.) Comparing numbers of actual reported hospital cases in table 57 with the estimated emergency department cases in table 63, if one assumes that the emergency department estimates are correct, then the numbers hospitalized represent 2-3 percent of the bicycle-only cases, approximately 4 percent of the bicycle-motor vehicle cases, and 14-16 percent of the pedestrian-motor vehicle cases for California and New York State. These numbers are well within expectations. The New York data shows an unusually high number of estimated pedestrian-only cases, but again this is probably due to the prominent role that "slips on ice" played in the Buffalo emergency department database. In contrast, the North Carolina emergency department data contained very few pedestrian-only cases, and this is reflected in the statewide estimates.
Table 63. Overall estimates of bicycle and pedestrian emergency department cases.
Estimating the total number of injured pedestrians and bicyclists is clearly no simple task, whether at the local, State, or national level. For the current analyses, use was made of both hospital discharge and statewide motor vehicle crash data. Ultimately, however, the success of either approach depends on the quality of the emergency department data obtained and, in particular, the completeness of case ascertainment. To the extent that less serious injury cases or events not involving a motor vehicle are missed, the resulting estimates will underestimate the true extent of the problem. This may have occurred with the North Carolina data, which had higher percentages hospitalized and lower percentages of pedestrian-only (but not bicycle-only) cases.
Other factors, however, may help to explain the results. In the original study by Stutts et al. (1990) based on North Carolina hospital emergency department data, the estimating procedure based on the percentage of cases hospitalized and overall numbers hospitalized (the first of the two approaches described in this chapter) appears to have "worked," producing bicycle injury estimates in line with available national data from the National Electronic Injury Surveillance System (NEISS). There were two key differences between that study and the current study. One is that, at the time of the original study, there happened to be available in North Carolina a reliable source of statewide hospital discharge data (albeit only for children under 20 years of age). In contrast, the hospital data available for the current study was incomplete, and a variety of assumptions had to be made to produce adjusted estimates that could be used in the analyses. (The New York State and California hospital data, in contrast, required no such adjustments.)
Perhaps a more important difference between the two studies, however, rests in the selection of hospital emergency departments and, in particular, on the number of hospitals participating. In the original study, a total of 10 hospitals participated in the emergency department data collection in 1985 and 15 hospitals participated the following year. Even though this was a convenience sample of hospitals interested in and willing to participate in the study, the end result was a quite varied sampling of small, medium, and large hospitals spread out geographically across the State and representing urban as well as more rural areas. In contrast, 75 percent of the North Carolina cases for the current study came from one hospital, which happens to be a large Level I trauma center serving a largely rural section of the State. Thus, it is not surprising that the data might capture a larger proportion of serious injury cases, or that it might not represent the State in other characteristics as well.
To a lesser extent, these weaknesses in the North Carolina emergency department data may be present in the California and New York State data as well. Without some reliable data for comparison, however, it is not possible to draw conclusions. In retrospect, it would have been desirable to have had a larger and more representative sampling of hospital emergency departments participating in each of the States. However, available funds were not sufficient for such a large-scale study.
Finally, it is interesting to draw perspective from numbers that have been reported nationally concerning injured pedestrians and bicyclists. Two obvious sources for information on events that involve a motor vehicle are the Fatal Accident Reporting System (FARS) and the General Estimates System (GES). According to FARS data, 830 bicyclists were killed in collisions with motor vehicles in 1995, including 136 in California, 50 in New York, and 35 in North Carolina (NHTSA, 1996). For pedestrians, the corresponding numbers were 5,585 overall, 825 in California, 412 in New York, and 188 in North Carolina. Information on injured bicyclists and pedestrians, based on GES data, is only available at the national level: an estimated 61,000 bicyclists and 84,000 pedestrians were injured in collisions with motor vehicles (NHTSA, 1996).
These numbers, based on police crash reports, only reflect the "tip of the iceberg" as far as injuries to pedestrians and bicyclists are concerned. However, there are few alternative sources of information to draw upon. Bicyclist fatalities based on death certificate data from the National Center for Health Statistics are typically about 8 percent higher than the FARS counts, due primarily to the inclusion of non-motor vehicle-related bicyclist deaths (Baker et al., 1993). The NEISS data, based on a stratified sample of hospital emergency departments, produces annual estimates of approximately 580,000 injured bicyclists. There are no comparably broad data sources for information on injured pedestrians, in part because pedestrian injuries not involving a motor vehicle are typically lumped into the immense category of "falls."
In summary, very little data exist for providing even a broad framework for interpreting the results of attempts such as reported here to estimate the overall magnitude and scope of the bicyclist and pedestrian injury problem. The final chapter provides a brief overview of key findings from the study and offers some recommendations for future research.