Appendix B: Technical Notes
These notes discuss the sources of data used in creating the TBI surveillance system, case definition, data
limitations, and procedures for case ascertainment.
A. TBI SURVEILLANCE CASE DEFINITION AND METHODOLOGY
The TBI Surveillance System accounts for all traumatic brain injury-related in-state hospitalizations
to New Jersey residents and non-residents, and resident deaths regardless of place of occurrence
during 1997. Fatal and non-fatal cases were identified through New Jersey's statewide hospital
discharge reporting system (UB-92) and death files using the CDC case definition.2
The specific ICD-98 codes used in identifying TBI cases are:
N800-N801: Fractures of vault or base of skull
N803-N804: Other and unqualified skull fractures and multiple fractures of the skull
N850-N854: Intracranial injuries including concussion, contusion, laceration, and hemorrhage
without skull fractures
Additional cases of TBI are identified from death certificates with a diagnosis of N873 (other open wounds of head) accompanied by an E (external cause-of-injury) code.
The case definition excludes hospitalizations with V (supplementary classification of factors influencing health status) codes as the primary diagnosis; newborn admissions; medical complications (N996-N999); late effects of skull fractures (N905) or intracranial injury (N907). Out-of-state hospitalizations of New Jersey residents are excluded due to unavailability of this information.
The population from which statistics are derived consists of records classified into one of the three
A statewide TBI surveillance system was created by using the following methodology:
- Non-fatal TBIs;
- TBI-related hospitalized fatalities; and
- Non-hospitalized fatal TBIs
- TBI-related hospitalizations were selected from the 1997 UB-92 data file of about 1.2 million hospitalizations, using the TBI case definition. These cases include in-state TBI-related hospitalizations regardless of residence. Deaths can occur among the hospitalized cases. All hospitalizations with a disposition status of 'non-deceased' at the time of discharge were counted as non-fatal TBIs. The remaining records were classified as TBI-related hospitalized fatalities.
- The software package Automatch9 was used to locate matching death records in the single cause-
of-death (SCD) file for each of the hospitalized fatal TBI records, using probabilistic linkage.
The matching variables used in this process were Soundex of last name, name, date of birth,
residence code and gender. These hospitalized fatal TBI records have information recorded on
both the UB and death file to allow some comparisons of demographic and diagnostic variables
reported from two independent sources of information.
- The multiple cause-of-death (MCD) file10 has both nature and external cause of injury information, but no identifiers. The SCD file has patient identifiers, underlying cause-of-death and demographic information, but no nature of injury information. The MCD and SCD files for the calendar year 1997 were merged using the death certificate number. TBI-related mortality cases were then identified from one or more of the nature of injury codes listed in the merged file using the TBI case definition. These cases include hospitalized deaths and other deaths which
occurred at the scene of injury, emergency rooms, nursing homes, and while being transported.
After eliminating the hospitalized fatal TBIs, the remaining TBI-related mortality records were
identified as non-hospitalized fatal TBIs.
- All TBI-related morbidity and mortality records for the calendar year 1997 were then merged into
one statewide TBI system. The morbidity records have information on demographics, nature and
cause of injuries, disposition status, source of admission, and financial information. Mortality
records have demographic, geographic and nature and cause of injury information.
- The software package ICDMAP-90 11 was used to map ICD-9-CM codes recorded on the UB-92
record to a six level severity score. Assumptions made during the process of computing the
ICD/AIS scores included: 1. High severity option: If a record contained two or more specific
injuries of different severity levels, the severity of the highest ICD-CM code was assigned;
2. Ignore unknowns: With this option, if there are no valid ICD-9-CM codes, it results in the
ICD/AIS score of zero or nine.
After computing the ICD/AIS score, the TBI patients were classified into four levels of severity;
- Mild: Patients with ICD/AIS scores of 1 or 2
- Moderate: Patients with ICD/AIS score of 3
- Severe: Patients with ICD/AIS scores of 4 through 6
- Unknown: ICD/AIS scores of 0 or 9.
B.. DATA LIMITATIONS, PROBLEMS, AND RESOLUTION
It should be noted that the TBI surveillance system is based on multiple sources of information,
and each of the data files is subject to a certain degree of unknown error. The hospital discharge file may overstate the incidence of TBI and the death file may understate the incidence due to differences in administrative and coding practices. Efforts related to data validation are conducted whenever possible to quantify errors. Although some discrepancies exist among the various sources of information, the impact on overall TBI incidence rates is expected to be minimal.
A validation study12 was conducted by the NJDOH to evaluate non-fatal and fatal TBI-related
hospitalizations in the calendar year 1994. As part of this study, a sample of medical charts was
reviewed. Based on sample review and analysis, the study concluded that the UB-92 data set seems to be complete and accurate with respect to hospitalized cases. For non-fatal TBIs, the UB-92 data are the only centrally reported source of information for identifying TBI cases. As a result of reviewing medical charts, a majority (96.7%) of sample records was confirmed as TBI cases. This indicates that the UB-92 data set is a good source of basic information for TBI-related injury surveillance of more severe cases. Discrepancies were noted in coding of racial/ethnic classification, admission source, discharge status and length of stay. Also, in nearly half of the sampled UB-92 records the place of injury (Z code) was coded as unknown. Z codes were implemented in 1994, and Z coding improved in 1997, with 70% completed coding.
- For TBI-related in-hospital deaths, the 1994 UB-92 file seems to be a more reliable source for all injury conditions than the death file. According to the INJURY MORTALITY ATLAS13, New Jersey has a relatively low TBI-related death rate compared to the nation as a whole (13.2 vs. 21.2 per 100,000). This may be related to coding practices and under-reporting of TBI diagnoses on the death certificate in New Jersey, a situation that needs further investigation. If the death file
under-reports injury conditions as a cause of death for hospitalized deaths, especially among the
elderly population, this raises the possibility of an under-estimation of non-hospitalized TBI
deaths based on TBI surveillance data. An expanded case definition to include similar codes, and
review of additional death certificates as well as medical examiners' reports, would facilitate
accurate case counting of TBI-related non-hospitalized deaths. A study to investigate death and
medical examiners' reports is underway.
- The current report does not include analysis by race and ethnicity due to lack of completeness of race/ethnicity information.
- All resident death certificates are included on the SCD file regardless of the state of occurrence.
However, the system does not include out-of-state hospitalizations of New Jersey residents. The
effect on overall incidence rates is expected to be minimal assuming an equal number of out-of-
state hospitalizations to New Jersey residents and out-of-state resident hospitalizations in New
Jersey (approximately 4.3% of total TBIs are out-of-state residents and 1.5% are of unknown
- The TBI surveillance system does not currently include information such as disability level at discharge due to unavailability of this information. Also, both the mortality and hospitalization files lack information on Abbreviated Injury Score (AIS), Glasgow coma scores, risk factors such as use of protective devices, and alcohol involvement. Collecting information on these optional variables is beyond the scope of this project. This information is being collected by the designated trauma hospitals, but is not reported to the NJDOH under the current regulations.
- The term TBI incidence used in this report does not include TBI cases which were treated and
released from hospital emergency departments or which receive some form of superficial
attention. Also, strictly speaking, incidence rates should reflect the same at-risk population in the
numerator and denominator, which these do not. The numerator is all incidents which occurred in
| TBI incidence rate for a specified population
||Number of TBI incidents * 100,000
Population at risk specific to the event
The estimated population figures (Appendix A) for New Jersey for the calendar year 1997
developed by the U.S. Bureau of the Census were used as denominators for computing approximate incidence rates. All the rates are per 100,000 population. 1940 standard populations were used to compute age-adjusted rates to facilitate comparisons over time and geographical areas.