III. Methodology

    NHTSA’s study of 1997 death certificates was successful in locating a small number of certain types of non-traffic or non-crash motor vehicle-related deaths. That study also confirmed death certificates’ value as a source for identifying non-traffic and non-crash motor vehicle-related deaths. There were inherent limitations of the data in the study 1997 death certificates, however. Although the criteria for selecting death certificates for review were carefully chosen, there was no way of knowing for certain whether or not all of the death certificates reflecting non-traffic or non-crash deaths had been identified and if not, what percentage of the total was represented by those found. Even among death certificates received, there were some that suggested that a non-traffic or non-crash incident was involved, but this could not be confirmed based on the information in the death certificate, and in some cases even after the appropriate coroner or other official was contacted. Finally, there was no way of knowing whether or not the number of incidents found for just one year was indicative of the ongoing magnitude and scope of a hazard or whether the snapshot of data from one year might be contradicted by data derived from another year or years, or from another source or sources.
    In the broadest and simplest sense then, the methodology for the research reflected in this report was to collect whatever data was available on the non-traffic and non-crash motor vehicle-related hazards of interest from whatever sources were available. The sources in which relevant information was found for each of the issues are indicated below.

    The totals of non-traffic and non-crash motor vehicle-related incidents located in death certificates, LexisNexis™ and some of the other sources reviewed for this study represent simple counts of relevant incidents. While there were no hard and fast rules that applied to identifying relevant cases, incidents involving particularly unusual events were excluded from the counts. As an example, if a person were backed over by a vehicle that was left running, in reverse, and unattended, this case was excluded from the count of backing incidents.
    The specific methodology that applied to each of the types of sources investigated is described below.

A. 1998 Death Certificates
    State laws require death certificates to be completed for all deaths. Furthermore, federal law mandates national collection and publication of death and other vital statistics data. As a result, and as confirmed by the study of 1997 death certificates conducted by NHTSA, death certificates represent a reliable and comprehensive source of information regarding non-traffic and non-crash motor vehicle-related deaths, particularly if information is entered for all of the key elements of a death certificate and if the certificate includes at least some description of how the injury that resulted in death occurred.
    As it did in researching 1997 death certificates, NHTSA’s Office of Crash Avoidance Standards contacted and met with representatives of the NCHS to solicit their help in researching 1998 death certificates. NCHS publishes annual reports of all deaths in the United States using information derived from death certificates. Among other information, each death is assigned various codes that identify the disease or condition directly leading to death, antecedent causes, and other significant medical conditions involved.
    For the year 1998, the underlying cause of death, and the other specific injuries, diseases, and conditions related to a death are classified and coded using the International Classification of Diseases, 9th Revision (ICD-9). ICD-9 is designed for the classification of morbidity and mortality information for statistical purposes, for the indexing of hospital records by disease and operations, and for data storage and retrieval. ICD-9 also contains a supplementary classification of external causes of injury (E-Codes) that permits the classification of environmental events, circumstances, and conditions as the cause of injury and other adverse effects.
    NHTSA identified what it considered to be the ICD-9 codes, primarily E-Codes, most likely to be indicated on death certificates reflecting one of the non-traffic or non-crash motor vehicle issues included in this study. Clearly, not all deaths assigned ICD-9 codes suggestive of these conditions involve non-traffic or non-crash motor vehicle-related events. In addition, deaths of interest may have been coded, for various reasons, with codes other than those on which this research focused. However, deaths associated with the ICD-9 codes identified for this research represented a universe of deaths that would most likely contain deaths resulting from the non-traffic or non-crash injuries being studied.
    In June of 2002, NHTSA met with NCHS officials to discuss NHTSA’s review of the public use Multiple Cause of Death (MCOD) data, the annual data file containing information derived from all U.S. death certificates, and how to proceed with the study of 1998 death certificates. It was agreed that the same protocol followed for the study of 1997 death certificates would be followed for purposes of obtaining 1998 death certificates. The seven steps of the protocol are:

    Research involving death certificates necessarily creates a substantial gap between the year in which the deaths involved occurred and the time when the research is completed. This gap is largely the product of the time it takes for NCHS to receive and assemble data from the states and finalize its annual MCOD file. Typically, the most recently completed MCOD file is two years or more older than the year in which it becomes publicly available. In addition, considerable time is required to complete the process described above. Some states add another step to this process by requiring detailed paperwork and state review board approvals, which are above and beyond the initial approvals required to obtain death certificate numbers.
    NHTSA initiated and followed the recommended protocol. NCHS staff reviewed a NHTSA suggested list of selection criteria for the non-traffic and non-crash motor vehicle issues to be studied and proposed some additional selection criteria. Table I provides the E-Codes, code descriptions, associated issues and any other data filters that were used in identifying the more than 15,000 death certificates of initial interest to the study.

Table I: E-Codes Used in Locating Death Certificates
For Use in National Highway Traffic Safety Administration Research of
Non-Traffic and Non-Crash Deaths in 1998

Issues:
1. Children left in a vehicle’s passenger compartment in hot weather or who lock themselves in the trunk of a vehicle,
(Vehicle heat/trunk)
2. Children strangled by a vehicle’s power window or sunroof, (Vehicle window)
3. Persons killed or injured as a result of a vehicle backing up, (Backing) and
4. Persons killed or injured as a result of vehicle-generated carbon monoxide. (CO)

    The MCOD file includes a variable for the “Underlying Cause of Death” and variables for up to 20 injuries or diseases that may have contributed to the death. In the language of the MCOD file, these are called “Record Axis Conditions.” NHTSA expected that deaths caused by non-traffic and non-crash motor vehicle-related injuries would be a subset of those with codes for the underlying cause of death, but records were also examined according to the 20 record axis conditions. In addition to searching the underlying cause of death and other conditions using the ICD-9 codes in Table I, certain data filters, also reflected in Table I, were imposed. A search of 1998 death certificate data was limited to children less than 9 years of age for E-Code 913.8, 913.9 and 918. These E-Codes were intended to identify victims of strangulation by power window. The search using E-Code 962.2 was limited to those records also containing an N-Code (nature of injury code) of 986. Finally, for E-Code 968.4, the search was limited to those records also containing an N code of either 986 or 992.
    Using the public use version of the MCOD file, NHTSA made some preliminary calculations as to the number of death certificates that might include people who died in a non-traffic or non-crash motor vehicle-related event. Depending on the E-Codes and other criteria used, it was determined that the number of death certificates that might reflect incidents of the type under study exceeded 15,000. In contrast, only 1,792 death certificates were identified as being of interest for the study of 1997 death certificates. The bulk of the increase in the 1998 death certificates resulted from the inclusion of backing incidents as an area of study. The E-Codes that were likely to reflect backing incidents were sufficiently broadly defined that it was clear they included a wide range of other, more common motor vehicle crashes as well.
    At an average price per death certificate on the order of $8 to $9, it was strongly felt, and budgetary constraints required, that every effort be made to minimize the number of death certificates that ultimately would be requested from the 50 states, the District of Columbia and New York City.
    A large percentage of the deaths included in this number were likely from conventional traffic crashes because the E-Codes 814-825 are used in death certificates for a wide spectrum of motor vehicle crashes. For example, “E-Code 822.7 – Other motor vehicle non-traffic accident involving collision with moving object (pedestrian)” – would apply to non-traffic vehicle and pedestrian accidents in which a vehicle was moving either forward or backward. Those involving the forward motion of a vehicle were not part of this study.
    NHTSA representatives on several occasions discussed with NCHS representatives possible ways of reducing the number of death certificates to be obtained. The focus of those discussions was on how to eliminate from the death certificates that would be requested from the states those that would clearly not be of interest to this study because they resulted from traffic crashes that did not involve backing. In most cases, motor vehicle-related deaths from the MCOD file that were initially identified as being of possible interest to the study would actually not be of interest if they were also contained in FARS because the vast majority of deaths in FARS do not involve backing.
    Combining the electronic version of the MCOD file with other data is not permitted by NCHS. So that NHTSA could eliminate deaths identified in the NCHS MCOD file that were clearly the result of crashes that did not involve backing, the Office of Rulemaking printed out some basic data from this death certificate file. Death certificate numbers were not in the file used and therefore were not included in the data printout. The Office of Rulemaking also printed out some basic data from the more than 40,000 deaths contained in FARS for 1998. Using these printouts, data were manually compared and, as a result, a large number of deaths from the list of death certificates of interest were eliminated. Most deaths in the MCOD file that also appeared to be in FARS were eliminated from further consideration. However, if FARS data indicated that the crash involved a pedestrian and the first point of impact was the rear of the vehicle, these deaths were included in those for which death certificates were requested since a backing incident relevant to the study was strongly suggested. Some deaths were retained among those of interest, particularly if a pedestrian was involved, even though there was no data to suggest a backing incident was involved. The researchers wanted to eliminate those deaths that clearly were not of interest, but also wanted to be sure to leave in those with even a remote possibility of being relevant.
    Some states (a small number) were delayed in granting NCHS permission to allow NHTSA to receive death certificate numbers. As a result, some death certificates were not requested from some of the states in time for those certificates to be included in the data presented in this report. The exact number of death certificates that would have been requested from those states is not known. However, it was estimated that if all certificates of interest had been received from all of the jurisdictions, the total number of certificates would have been approximately 5,500. As of the date of this report, 4,729 death certificates have been requested from 44 states and the District of Columbia.
    Counts of incidents of any type from death certificates represent a floor as to the number of incidents that actually occurred. Deaths of interest may not have been coded according to the selection criteria chosen. Also, death certificates are often very sketchy in the descriptive information they provide. Limited follow up with medical examiner’s and other offices thus far has uncovered backing incidents, for example, that were not clearly identifiable from death certificates alone. In some cases, information on a death certificate strongly suggests it is an incident of interest, but the medical examiner’s office and even the police agency involved were unable to clarify the situation.
    Not all death certificates requested have been received, nor is it anticipated that death certificates from every jurisdiction will be received. As of the writing of this report, NHTSA has received and reviewed 4,046 death certificates from 36 jurisdictions, 35 states and the District of Columbia. Data from these death certificates, excluding personal identifiers and certificate numbers, has been entered into a database.
    The 36 jurisdictions that provided death certificates for this study represent a good cross section of states. They reflect a balanced mix of characteristics, such as urban and rural, cold climate and warm climate, and geographic regions of the country.
    Straight-line projections of anticipated national totals are made in this report using the number of deaths found in 1998 death certificates. A simple ratio was used to account for the missing death certificates. The ratio was determined from the number of certificates selected and received (5,500/4,046=1.36).
    The researchers believe that projections from the available data are reasonable estimates and reflect the magnitude and scope of the hazards studied because of the balanced mix of the states that provided death certificates. Had those states been skewed toward warmer climates, for example, this might have had an effect on the projected national totals of vehicle-related carbon monoxide deaths since these incidents tend to happen in colder climates during the winter months.

B. LexisNexis™
To support this research, NHTSA subscribed to LexisNexis™ so that it could search for news articles related to the four types of motor vehicle-related hazards under study. A variety of words and phrases likely to be associated with these hazards were used to conduct searches of LexisNexis™.
    LexisNexis™ served several purposes. When a death certificate was suggestive of a type of incident under study, but not conclusive, an article in LexisNexis™ sometimes confirmed the nature of the incident so that it could be included or excluded from this study. Sometimes, articles in LexisNexis™ identified incidents from 1998 that were not located in 1998 death certificates. Also, LexisNexis™ identified incidents from years beyond 1998.
    References to LexisNexis™ derived data or information appear in various contexts in the data and information presented in this report. In the case of some death certificates, an article found in LexisNexis™ either confirmed the nature of the incident or provided additional information concerning that incident. In such cases, this is indicated. A count of the incidents that have occurred in the five completed calendar years, 1998-2002, and that were located in LexisNexis™ is presented in this report. For the year 1998, the number of LexisNexis™ identified deaths that were also found in death certificates is indicated. For the years 1999-2002, data presented are derived solely from LexisNexis™.
    It should be emphasized again here that the cases identified through LexisNexis™ only represent a count of cases. It is very clear from experience with LexisNexis™ that many cases of interest to this research are either not reported in news outlets or not included in LexisNexis™ because the news outlets in which such cases might be reported are not included in the universe of outlets from which LexisNexis™ draws.
    Unlike either death certificates or FARS, as described immediately following, articles found in LexisNexis™ report on both deaths and injuries. Therefore, LexisNexis™ derived data presented in this report include injuries.

C. Fatality Analysis Reporting System
    NHTSA’s Fatality Analysis Reporting System (FARS) focuses on fatalities from vehicle crashes that occur on public roads. FARS is a count of the annual national total of fatalities resulting from vehicle crashes on public roadways. FARS represented a potential source of backing incidents since backing incidents may occur off-road or on a public road.
    Among the various bits of data noted and entered into the FARS database is the point on the vehicle at which the first impact relating to the incident occurred. An examination of FARS data relating to pedestrian fatalities that occurred in 1998 identified a number of incidents in which a pedestrian was first struck by the rear of the vehicle involved. In other words, cases were identified in FARS with a high probability of being backing incidents of the sort of interest to this research.
    As described previously, this was helpful in being sure to select for review certain 1998 death certificates whose data matched a FARS record in which the rear of the vehicle was the first point of impact. It also demonstrated that FARS contains backing incidents of the sort under study.
    Since backing incidents were found in FARS for 1998, FARS data was examined to identify potentially relevant backing incidents that occurred in years later than 1998. FARS is created from data that are derived from Police Accident Reports (PARs) that are reviewed to identify deaths from vehicle crashes that are appropriate to include in FARS. The rules governing FARS require that PARs from the two most recently completed calendar years plus the current year be retained. Older PARs are typically destroyed. FARS data was examined for possible backing incidents in late 2002. At that time, the years for which PARs were still available were 2000 and 2001.
    A search of FARS data for the years 2000 and 2001 located 91 apparent backing fatalities in 2000 and 67 in 2001. States were requested to provide the PARs for these incidents. A total of 138 for the two years were received. Thirty-six of the incidents reflected in the PARs received and reviewed were not entered into the NHTSA database of non-traffic and non-crash cases because they involved situations outside the scope of the research, such as an incident involving an unusual vehicle like a small front end loader. A few of the 36 were not entered because either the quality of the photocopy or the limited information provided was such that what exactly happened could not be determined.
    Fourteen (14) backing deaths identified in FARS had been previously identified through other sources, usually LexisNexis™. The remaining 88 deaths had not been found in other sources. Even though FARS focuses on deaths from vehicle crashes that occur on public roads, many of the backing incidents in FARS occurred in driveways or other off-road locations similar to those found in 1998 death certificates and in other sources. These incidents may have been included in FARS because the circumstances were ambiguous as to exactly where the incident took place or they ultimately involved a public roadway in that the vehicle and/or the victim ended up in a public roadway.

D. Peer-Reviewed Scientific, Medical and Public Health Journals
    A review of scientific, medical and public health research literature dealing with the issues being investigated was begun using PubMed, an on-line service of the National Library of Medicine that provides access to the library’s MEDLINE. MEDLINE contains more than 12 million life science and public health article citations dating back to the mid-1960's. In most cases, only abstracts are provided so several trips to the National Library of Medicine in Bethesda, MD were made to examine and, when necessary and appropriate, make copies of full articles relevant to the issues covered in this report.

E. The National Electronic Injury Surveillance System (NEISS)
    For nearly 30 years, the U.S. Consumer Product Safety Commission (CPSC) has operated this statistically valid injury surveillance and follow-back system. The primary purpose of NEISS has been to provide timely data on consumer product-related injuries occurring in the U.S. Beginning in 2000, NEISS was expanded to collect data on all injuries, including those involving a motor vehicle. Collection of information on this more broadly defined universe of injuries began as of July 1, 2000. NEISS provided valuable information on backing injuries.
    NEISS injury data are gathered from the emergency departments of 100 hospitals. These hospitals are selected as a probability sample of all 5,300+ U.S. hospitals with emergency departments. A “trauma weight” is determined for each case in NEISS. According to the CPSC, when there are 20 or more cases of a particular type of incident, one may add the trauma weights for those incidents to come up with a projection as to the number of such incidents that occur nationally in a given year. When there are fewer than 20 incidents of a given type, the CPSC indicates that the national projections that one derives from these trauma weights are less reliable statistically.
    As part of this round of research, NHTSA obtained finalized NEISS data for the last six months of 2000 and preliminary, although close to finalized data for the first six months of 2001. This was the most recent data available when this research was conducted.
    This year’s worth of NEISS data was searched for young children who were left in a hot vehicle, trapped in a trunk, or caught by a power window. It was also searched for backing incidents. Backing was the only area in which multiple cases of injuries were found.
    A total of 265 possible backing incidents were located in NEISS. The text fields in each record were reviewed to determine which of the incidents actually involved incidents of the type under study.

F. NHTSA’s General Estimates System (GES)
    Data for the General Estimates Systems (GES) come from a nationally representative sample of police reported motor vehicle crashes of all types, from minor to fatal. The system began operation in 1988. It was created to identify traffic safety problem areas, provide a basis for regulatory and consumer initiatives, and support cost benefit analyses of traffic safety programs. The information is used to estimate how many motor vehicle crashes of different kinds take place, and what happens when they occur. Although various sources suggest that about half the motor vehicle crashes in the country are not reported to the police, the majority of these unreported crashes involve only minor property damage and no significant personal injury. By restricting attention to police-reported crashes, GES concentrates on those crashes of greatest concern to the highway safety community and the general public.
    A search of GES data for the years 1996 to 2000, the most recent data available when the search was done, was conducted to identify GES records for which non-occupant impact was the first harmful event in the crash, with damage to the rear of the vehicle as the initial impact and with vehicle role and maneuver consistent with backing. Incidents were identified that allowed for national projections of backing incidents. There was no language describing what occurred in the incidents that could be reviewed. So the national projections may include backing incidents of a type, such as vehicle rollaways, that are outside the scope of this study.

G. Health Care Utilization Databases of the Centers for Disease Control and Prevention
    Several databases related to health care utilization maintained by the National Center for Health Statistics (NCHS) of the Centers for Disease Control and Prevention seemed like strong candidates for providing good information on the issues under study. They were not. Information derived from these databases applicable to the non-traffic and non-crash injuries considered for this report was extremely limited.
    The CDC databases exist to measure large indicators of how health care resources in the United States are utilized, such as the extent to which health care resources are used by various age groups, the payers involved in obtaining health care, or the types of treatments, therapies and drugs that are employed nationally. In short, the databases are aimed at big picture issues, not more narrow subjects such as non-traffic and non-crash motor vehicle-related safety issues that involve small numbers of occurrences annually.
    The databases do offer a limited opportunity to identify incidents involving non-traffic and non-crash motor vehicle-related safety concerns and for getting a very generalized (non-statistical) sense of the relative magnitude of certain non-traffic and non-crash motor vehicle-related hazards. That opportunity exists in the text fields for each record in the databases. Unfortunately, information is not always entered in these text fields and the information that is entered is not always descriptive enough to determine exactly what happened to cause an injury. For example “MVA,” referring to a motor vehicle accident, sometimes appears as the only thing entered in the text field of a record.
    Unintentional injuries, which are how each of the types of incidents in this study would be characterized, are a subset of the data contained in each of the CDC databases. By searching the text fields of unintentional injuries in the databases for certain words likely to be used in describing an injury resulting from the issues under study, several relevant incidents were located. However, because the number of incidents found was so few, reliable national projections of the numbers of these types of incidents could not be made. The CDC databases generally require at least 30 incidents of a given type before statistically reliable national projections about the incident type can be made.
    Each of the CDC databases examined is briefly described below.

1. National Ambulatory Medical Care Survey
    The National Ambulatory Medical Care Survey (NMACS) is a national sample of patient visits to office-based physicians who are not employed through the federal government and who are primarily engaged in direct patient care. The survey, which includes specialists, is conducted annually. The opportunity to gain some insights into how an injury occurred is in a small text field on the survey form called “cause of injury.”
    The NAMCS determines a “patient visit weight” for each record in this database. This represents the projected total number of office visits that occurred across the country and that are similar to the particular record involved. The patient-visit weights for all of the records in the NAMCS sample indicate that there were a total of 823,541,999 physician office visits in the year 2000 involving physicians of the type included in the NAMCS. National projections for subcategories of office visits, by age group for example, may also be tallied with varying degrees of statistical confidence. If there are less than 30 records in any subgroup, national projections are considered less reliable statistically.
    NHTSA examined the NAMCS data file for the calendar year 2000, the latest year for which survey results were available when research for this report was conducted. That file contains 27,369 records, of which 3,042 relate to an unintentional injury. Of the 3,042 unintentional injury records, 994 contained no text in the “cause of injury” field. A review of the 2,148 records with entries in the “cause of injury” field located one record of interest to this study. That was an incident in which the “cause of injury” was given as “mother backed car over child,” in this case a three year old. With only one record of a backing incident identified, the “patient visit weight” of 3,983 for that incident was of little or no predictive value.

2. National Hospital Ambulatory Medical Care Survey
    The National Hospital Ambulatory Medical Care Survey (NHAMCS) collects data on the utilization and provision of ambulatory care services in hospital emergency and outpatient departments. Findings are based on a national sample of visits to the emergency and outpatient departments of non-institutional general and short-stay hospitals, exclusive of federal, military, and Veterans Administration hospitals, located in the 50 States and the District of Columbia. Annual data collection began in 1992.
    As with the other CDC databases, the text fields of each record were searched to locate incidents of interest that are included in the database.
Emergency Departments
    The 2000 NHAMCS file for emergency department visits contains 25,622 records, 8,791 of which are for unintentional injuries. The text fields of the unintentional injury records were searched for certain words that would likely be used in describing one of the types of non-traffic or non-crash motor vehicle-related incidents under study. A total of five incidents involving a vehicle backing up were found in the NHAMCS emergency department file. Details of those incidents are provided elsewhere in this report.
Outpatient Departments
    The outpatient portion of the NHAMCS for the calendar year 2000 file has 27,510 records, 3,002 of which relate to an unintentional injury. As with the emergency department records, the text fields of the outpatient department unintentional injury records were searched for certain words that would likely be used in describing one of the types of non-traffic or non-crash motor vehicle-related incidents under study. No records of injuries reflecting the types of hazards under study were found in the search of this file.

3. National Hospital Discharge Survey
    The National Hospital Discharge Survey (NHDS) has been conducted annually since 1965. It is a national probability survey designed to meet the need for information on characteristics of inpatients discharged from non-Federal short-stay hospitals in the United States. The NHDS collects data from a sample of approximately 270,000 inpatient records acquired from a national sample of about 500 hospitals. Only hospitals with an average length of stay of fewer than 30 days for all patients, general hospitals, or children’s general hospitals are included in the survey. Federal, military, and Department of Veterans Affairs hospitals, as well as hospital units of institutions (such as prison hospitals), and hospitals with fewer than six beds staffed for patient use, are excluded.
    This database proved to be of no use since it does not contain a text field for describing the circumstances that led to the patient being hospitalized in the first place.