UTAH CODES FINAL REPORT
Submitted April 1994

I. Introduction/Description of Problem

A. Utah seat belt and helmet legislation

Mandatory Seatbelt Law
Child Passenger Safety Law
Motorcycle Helmet Law

B. Observed and Reported Safety Device Usage Rates

Table 1: Observational Study: In-City Belt Usage by Gender and Site (% usage rates)
Table 2: Observational Study: Highway/Freeway Safety Belt Usage by Gender and Site (% usage rates)

CODES 1991 Belt Use Rates:

Table 3: Seat Belt Coding Classification
Table 4: CODES In-City Belt Usage by Gender and Site (in percent usage rates)
Table 5: CODES Highway/Freeway Safety Belt Usage by Gender and Site (% usage rates)
Table 6: CODES Safety Belt Usage Passenger Car and Light Truck Occupants Age 5 and Over
Table 7: CODES Helmet Usage for All Motorcycle Occupants

C. Type motor vehicle insurance system
D. Sociodemographic characteristics of state
E. Incidence of injuries from motor vehicle crashes

Table 8: Utah Crash Population Injury Rates (1991)

II. Methodology

A. Crash and injury data sources

Crash characteristics
EMS response
Emergency Department (Outpatient)
Hospital Inpatient
Rehabilitation

B. Description of state data sources

Data quality

C. Outline record linkage process

1. Implementation aspects affecting mandated models.

Ancillary file preparation

EMERGENCY MEDICAL SERVICES FILES
ALCOHOL CONVICTION FILE

2. Validation of linked data for the mandated model

Review of false positive
Review of false negatives

I. Introduction/Description of Problem

A. History, current status and proposed changes of belt/helmet legislation

Mandatory Seatbelt Law:

Effective April 28, 1986, this Utah law requires all drivers and front seat passengers to be buckled up when traveling in a motor vehicle. Visitors from outside the state of Utah are also required to buckle up regardless of safety belt laws in their place of residence. The law is a secondary law which means that a person may be issued a citation only when the police officer has stopped the vehicle for another reason. Any person who violates this law is subject to a fine of $10. Evidence of increased injury due to non-use of safety belts is not admissible in civil suits.

Current exceptions to the law:

• delivery personnel;
• rural letter carriers;
• persons driving vehicles used for farm purposes;
• individuals driving or riding in motor vehicles manufactured before July 1, 1966; and
• drivers or passengers with a physically disabling condition or medical condition which would prevent appropriate use of a safety belt.

Proposed Change: Seat belts would be required of all occupants of a vehicle.
Proposed Change: Violation would be a primary offense.
Proposed Change: The fine would be increased to $50.
Proposed Change: Safety belt use evidence would be admissible in civil suits.

Child Passenger Safety Law:

Effective February 1990, all children under age eight must be properly restrained in a motor vehicle driven by any driver. This does not apply when all seating positions are occupied by other passengers. Children under age two must ride in an approved car safety seat. Children two to eight years of age must ride in an approved car safety seat or use a safety belt. The law exempts authorized emergency vehicles, mopeds, campers, sleepers, motorcycles, motor homes, school buses, and such vehicles that offer transportation for hire. Violators are subject to a fine of not more than $20. This is a primary law which means that a person may be stopped and a fine issued solely for not restraining a child under age eight. The first offense is dismissed if the driver shows proof of acquiring a car safety seat or safety belt during or before any court appearance.

Proposed Change: All passengers under the age of eight would be required to be restrained.

Legislative Action (1994): House Bill 430 defeated in the Rules Committee.

Motorcycle Helmet Law:

Utah has no motorcycle helmet law for adults, but requires helmet use for motor cyclists under age 18.

Proposed Change: Motor cycle helmet legislation for all drivers and passengers of motor cycles.

Legislative Actions: Utah Public Safety officials have backed a proposed law three times in recent years (1991, 1993, and 1994) but each time measures were introduced, heavy opposition from motorcyclists killed the bills in committee. Opponents have used arguments about preservation of personal rights and choice in order to defeat legislative attempts.

B. Observed versus reported versus unknown usage rates

Utah Safety Belt Observation Study conducted by the Utah Department of Health, Bureau of Health Promotion/Risk Reduction:

Methods: Observation periods were selected from all daylight hours on various days of the week, including weekends, in May of 1991. All in-city observations were selected by stratification according to number of registered vehicles in the state. Only stopped or slow-moving vehicles were observed. Sites were preselected for both location and direction of traffic. High/freeway observations were conducted at entrances and exits for limited access freeways and during travel to and from in-city sites. Safety belt use was observed among all passenger cars, light trucks, and vans including convertibles, commercial vehicles, government vehicles, taxis, and police cars. The data was collected on drivers and front seat passengers and all children who appeared to be under eight years of age.

Overall observed adult seat belt use was 42% for 1991.

CODES 1991 Belt Use Rates:

Seat belt use information is obtained from police crash reports and in most instances is reported by the crash occupant. The reporting officer is to indicate which of the following types of safety equipment each driver or occupant(s) was using at the time of the crash:

The variable, BELTUSE, is coded as 'used' if the reporting officer explicitly indicated that the lap, lap and shoulder, or child restraint device was used. It is coded as 'not used' if the officer explicitly stated that belts were not used or not installed. All others were left empty. In the CODES passenger car and light truck over age 5 dataset, this field was left empty 7195 times out of a crash population of 92784 (7.75%). These cases were deleted from all analyses. Since the reporting officer usually does not witness the crash, uninjured occupants and occupants with minor injuries may be outside their vehicle when the officer is obtaining information concerning seat belt use. The police officer obtains information about seat belt use by asking these occupants. Since seat belt use is mandatory in Utah, crash occupants may report that they use seat belts in order to avoid a citation and fine. However, seat belt use by more severely injured or killed occupants can be directly assessed by police officers, particularly if extrication is required. This issue will be addressed in greater detail in our state-specific analysis.

The variable, HELMET, is coded as 'used' if the reporting officer explicitly indicated that a helmet, or helmet and eye protection were used. It is coded as 'not used or unknown' if the officer indicated any other choice or left the field empty. This variable is weakened because, until 1994, the officer was only offered the choice of helmet used or unknown. That is, there is no option on the report form to specifically indicate that a helmet was NOT used. This will result in a large possible misclassification. In the CODES cycles dataset, this field was left empty 386 times and were therefore coded as "NOT USED or UNKNOWN."

C. Type motor vehicle insurance system

All motor vehicle owners in the state of Utah must show proof of vehicle security (insurance) to obtain registration, license plates, or safety inspection and must maintain the security throughout the registration period of the motor vehicle or at any time that the motor vehicle is operated on a highway within the state. Every policy of insurance must include: motor vehicle liability coverage, uninsured motorist coverage, underinsured motorist coverage, and personal injury protection. The Utah Criminal and Traffic Code Owner's or Operator's Security Requirement 41-12a-302. states:

1. Any owner of a motor vehicle on which owner's or operator's security is required under Section 41-12a-301, who operates his vehicle or permits it to be operated on a highway in this state without owner's security being in effect is guilty of a class B misdemeanor. 2. Any other person who operates a motor vehicle upon a highway in Utah with the knowledge that the owner does not have owner's security in effect for the motor vehicle is also guilty of a class B misdemeanor, unless that person has in effect owner's security on a Utah-registered motor vehicle or its equivalent that covers the operation, by him, of the motor vehicle in question.

D. Sociodemographic characteristics of state relative to the mandated model

Utah presents as a unique area. The state is the eleventh largest in the United States, encompassing an area of 84,916 square miles, and in the study year of 1991, had a population of 1,774,241. Utah has a varied geographic distribution of its population, with large rural and frontier areas. Nearly 80% of the population are found within a 100 mile strip in the north central area of the state referred to as the Wasatch Front (Davis County, Salt Lake County, Utah County, and Weber County). The CODES crash records reflect the rural/urban distribution as expected: 73.1% of all crashes were in urban areas with the remaining 26.9% occurring in the rural areas of Utah. In 1990, Utah ranked as #1 in the nation in percentage of population under age 5 years, percentage of population between ages 5 and 17 years, number of school age children per adult, and average household size. The CODES crash record (the entire passenger car and light truck data set including children under the age of 5) has a median age of 25 years, the same median age as the state overall. The Utah birth rate is the second highest in the United States, with a rate of 31.3 births per 1,000 residents compared with 15.7 nationally. The fact that Utah has so many young people makes it particularly important that effective efforts are made to reduce risk of mortality and injury from motor vehicle crashes. While 48% of the general population in 1991 were under the age of 25, 49.5% of the entire 1991 crash population was under the age of 25. The population density and geography of Utah have offered several advantages for the CODES project. First, the 1991 Utah population at 1,774,241 was small enough that population-based data collection was workable. The CRASH population was 98,373. Second, the small number (40) of acute care hospitals in the state made data collection from hospitals feasible. 100% of the hospitals in the state agreed to share their hospital case mix files with the CODES project. Third, the varied geography permits comparison of the impact of safety restraints and motorcycle helmets in densely populated urban areas with the impact of those devices in rural areas. The geographical variation has also allowed some assessment of difficulties encountered in EMS response to motor vehicle crashes

E. Incidence of injuries from motor vehicle crashes

II. Methodology

A. Crash and injury data sources

Crash characteristics

All data fields that were electronically collected by the Department of Transportation were included in the CARS file. This database includes extensive information concerning the crash, the road site, each vehicle involved, drivers of each vehicle, and occupants of each vehicle. This database has been used to generate annual traffic reports, and a subset of this database is sent annually to NHTSA in participation with the FARS program. All of the 1991 data are completely computerized. Fields that are included in this dataset include details of crash location, which has been correlated with FIPS (Federal Information Processing Standards code) for comparison with census information. Vehicle identification (VIN info), driver information, occupant position in vehicle, injuries sustained at the scene (assessed by scene personnel using A-B-C-K system), beltuse, alcohol involvement, and other data are available (see data dictionary for all specifics).

EMS response

The EMS prehospital incident database includes fields that deal with response times, scene-assessed injuries (CRAMS scores and Glasgow scale), treatment and medication administered, disposition, and destination. These files are relatively reliable with respect to response times, scene times, etc. There were missing records from this system, in that about 20 agencies do not report electronically. However, our major agencies report via computer and nearly 75% of pre-hospital incidents were electronically transmitted. The 1991 EMS data set was completed by manual entry of data from agencies' hard copy information.

Emergency Department (Outpatient)

Emergency department information includes the date of service, diagnostic codes that are maintained by hospital medical record departments and kept in casemix files, and in some instances treatment (procedure) codes. Charge, reimbursement, and categorical payer information is also included. All of this information was available from the hospital case mix files.

Hospital Inpatient

Using hospital inpatient casemix information, all the discharge diagnoses were provided. This includes a primary diagnosis and up to 15 additional diagnostic fields, which will be ICD-9 codes. E-codes are coded in these fields. These codes are not in separate fields from other diagnostic codes. Charge, reimbursement, and categorical payer information is also included. Other fields of interest include length of stay in the hospital, status at end of hospitalization, etc. All of this information was available from hospital case mix files.

Rehabilitation

Inpatient rehabilitative services are also included in hospital inpatient casemix data files. Inpatient rehabilitation services in Utah and are not distinguished from other inpatient charges. The rehab file in the CODES data set is completely comprised from the case mix file from a single rehab source.

B. Outline detailed description of state data sources (scope/reporting thresholds, volume of records, and case selection for linkage)

The Utah Department of Transportation keeps detailed records of each traffic incident occurring in the state in which any injury occurs or in which property damage exceeds $750. During 1991, there were 47,435 crashes, of which 33,443 were described as property damage only crashes. There were 103,612 individuals involved in crashes that were reported through the Department. After eliminating all crash victims who were not in vehicles regulated by seat belt laws, the passenger car and light truck/van file contained 98,373 persons. The motorcycle file contains 863 persons. These individuals were the target population of the case selection for linkage for this study. The crash file for the State of Utah consists of three physical files of records. These files refer to the crash, vehicle, or occupant records. Thus we will speak about a crash file, which refers to the details of a motor vehicle incident in general. We will refer to vehicle file, which refers to the details about the specific vehicles which have been involved in any specific motor vehicle incident. Finally we will speak of an occupant file, which consists of all the detail information that is related to individual occupants, whether they are drivers or non drivers. For purposes of the project, it was necessary to flatten these files into one flat file structure, in order that we could export an appropriate data set for use with the probabilistic linkage software. For some purposes, it was necessary to divide these records into driver records and non driver records. We finally end up with a single file which contains all of the necessary occupants. There are several quirks to this file which need to be carefully noted. First, the data base file which has been submitted as the deliverable product of this project is an occupant based data base. Thus, for each individual occupant of a motor vehicle crash, the vehicle and crash file characteristics are repeated in the data file. This has important implications for frequency analyses of characteristics related to vehicles and crashes. For example, if one were to do a frequency analysis of a specific vehicle manufacturer by using this crash file, it might appear that there are 100,000 or more independent vehicles, as this is the number of involved occupants. A specific manufacturer, make, and series might appear for instance, ten times. However, it is possible that this vehicle actually participated in a crash only one time, but contained ten occupants. Thus it is crucial to understand the analysts focus. If the analyst is interested in occupant related statistics, then the crash file as submitted is appropriate to use. If on the other hand, the analyst is interested in vehicle characteristics using vehicles as the denominator, this could best be accomplished by selecting out records which only relate to drivers. Finally, if the analyst is interested in crash specific characteristics using crashes or incidents as the denominator, it would be important for the analyst to use incidents that have the same crash control number. For most of the analytical work conducted by the Utah CODES project, the denominator is occupants, and the file as delivered is the appropriate data file.

Data quality

There are several limitations to our data which needed to be noted. For the emergency medical services file, we do not have charge information. For that reason, we have a field which indicates the number of EMS response vehicles. This is accomplished by using the method noted here in this section, and if there was not a duplicate record the number of visits was assumed to be one if there was a linkage. The outpatient file, as well as the hospital based files, we have made the assumption that all medical services rendered to a patient following a crash were related to the crash. This is necessary because to do otherwise requires an inspection of the ICD-9 diagnosis for each of the admissions. This is not feasible. However we have carried out an analysis which suggests that this is a valid approach. A description of this analysis follows. We used the crash victims who have been linked in calendar year 1991 to medical records from the same year. For example, a crash victim from January 1991 may be linked potentially to eleven months worth of medical records. These months will follow his crash. However the victim of a crash in December 1991 has no ability to link to records which are beyond one month from the crash, but does have the possibility of linking backwards to up to eleven months of medical services. Thus the victim in January can link forward into medical records which might be ascribed to the crash, while the victim in December can link backwards to records which cannot be ascribed to the crash. We created a field in each of the crash files to indicate charges backward from the crash date. We then summated all of the charges which we could identify which occurred prior to the crash date, and placed these values in those fields. This resulted in the following. For the inpatient charges which were identified on 1991 medical records, the total was $15,863,840.00. These are charges which occurred after the motor vehicle crash. When evaluating backwards, a total of $670,684.00 medical charges were identified for the same patient population. One might conclude from these data that approximately 4.2% of our twelve month charges may be erroneously attributed to the motor vehicle crash, when in fact it is related to background medical activities of that population. We believe this is a small enough percentage of the total dollars involved to justify the simplifying assumption that all medical costs incurred after an incident inside of the six and twelve month period are related to the incident. Several other findings are of note with respect to the multiple inpatient and outpatient visits. First of all for 978 patients, there was only 1 inpatient visit. An additional 97 patients had 2 inpatient hospitalizations, and 28 patients had more than 2 hospitalizations. When one examines the inpatient costs based entirely on 1991 data, the total initial hospitalization costs were $13,518,223.00, with a total of 6,324 days of care. The total six month charges were $15,785,602.00, for a total of 8,088 days. The total for twelve months was $15,863,840.00, for a total of 8,132 days. It is apparent that the majority of the costs are incurred on the initial hospitalization. It should be noted that these figures relate to 1991 hospitalization data only, and six and twelve month totals which are included in the data base do include 1992 data. On the outpatient side, it would be expected that there would be more multiple visits and this is indeed the case. However if we take up to 4 visits, we account for 92% of the charges which were incurred on an outpatient basis. Again it should be noted that these numbers are based on 1991 hospitalization and outpatient data, while the data set which has been provided to NHTSA includes 1992 data. Of the charges which were identified, 71% were related patients who only had one outpatient visit. Such patients accounted for $2,798,011.00 in the six month period, and $2,798,302.00 for the twelve month period. Once again as has already been noted, the first four visits account for 92% of the six and twelve month charges, when one ignores the 1992 data which have been placed in the NHTSA data set. To summarize, duplicates have been handled using the Automatcher software in its match mode, undupe mode, and geocoding mode. This has enabled us to sum the six and twelve month total charges and days of care for victims who have been linked to one of these medical files period. We have made an additional assumption all medical care incurred after a motor vehicle crash are related to the crash itself rather than baseline medical activity. We have performed an analysis which demonstrates that this is probably accurate to within 5%.

C. Outline record linkage process

1. Implementation aspects which affected linked data generated used in mandated models.

a. Ancillary file preparation

PREPARATION OF THE EMERGENCY MEDICAL SERVICES FILES

The 1991 EMS files were provided by the Bureau of Emergency Medical Services at the Utah Department of Health. The final total number of records which were included in the 1991 data were 80,967. The incident year was 1991 and 80,926 of these were retained. There were additional records which had an empty incident date, but the day, month, and year were present so that these could be reconstructed. The EMS files were further processed in order to identify records which were considered likely to match with motor vehicle incidents. The inj1 -- inj5 fields are available for the EMT to encode an injury or an injury cause, and the code 29 refers to motor vehicle incident. The same coding structure is used for the dispatch code which initiates the EMS response. Thus, we simply went through all of the available injury fields as well as the dispatch code and identified all records which have a 29 in any of those locations. The final result was an EMS file consisting of 18,047 records which were related to motor vehicle crashes.

PREPARATION AND LINKAGE OF ALCOHOL CONVICTION FILE

The Utah Crash File includes several fields which are related to alcohol use. The reliability of these fields was unclear, and in order to improve the reliability of alcohol related information, we linked the crash files with the conviction file which is maintained in the state of Utah for alcohol convictions. This file included demographic information, as well as the blood alcohol level. The alcohol file, which we will entitle the DUI file, included 11,078 individuals. Linkage with the crash file resulted in 1,909 matches. The date in the conviction file was the same as the date of incident in 1,512, within 2 days of the incident in 1,566, and within 7 days of the incident in 1,581 instances. There were a very significant number of individuals for whom the conviction date was definitely prior to the crash date, indicating the high risk of this population for involvement in motor vehicle crashes. The linked information was then posted back to the crash file. The alcohol flag was then built using both police indicator of alcohol tests done as well as the conviction. When we combined all the records for which tests was either done or the DUI was convicted, this involved 3,139 occupant records. Once again, it must be remembered that the occupants have driver related information attached to them. Thus it is important to restrict certain analyses to drivers. When we analyze this by driver or non-driver, there were 2,140 drivers in whom alcohol was considered to be a problem, and 999 non-drivers who were involved in incidents in which drivers were considered to have been under the influence of alcohol. For the motorcyclist, the same approach was used. There were 57 matches to the motorcycle file, and blood alcohol level was available for 31 of these.

2. Validation of linked data for the mandated model

a. Review of false positive (records which linked but should not have)

Utah has the unique advantage of having names and birthdates on all records in all files. Thus, when conducting the linkage we were able to assess the rate of false positive linkages directly. The parameters of the linkage were set in such a manner that this rate of false linkage was kept well below 5%, and in instances where the names were not missing, the rate of false positive linkages was essentially zero.

b. Review of false negatives (records which did not link but should have)

Utah does not have comprehensive E-coding on its inpatient hospital records, and outpatient records are even worse. For that reason, we have no methodology for assessing false negative charts, but we are reassured by the fact that our numbers of linkages within each of the categories is consistent with other statistics that have been previously obtained by NHTSA in other localities by other methodologies. When E-codes are more comprehensively used, it may be feasible to examine the percentage of E-coded patients with crash indicated as cause of injury actually have a record in the CODES file. This is simplistic, however, as the Hawaii CODES project has apparently found numerous charges accrued through the insurance companies where no medical visit appears to have occurred in the first place. Thus we believe that this will always remain an unanswered question.

3. Linkage obstacles

a. Non computerized data

Approximately 3.6% of Utah hospital data are not electronically collected. Three CODES data entry technicians went to these hospitals and performed onsite data collection of all motor vehicle involved crash injuries for both inpatient and outpatient services. Although the data were collected, it was not added to the final data set. Complications concerning the formatting of the data did not allow for this hand-generated data to be added to the data set.

b. Different storage media

All data from providers was delivered on 9 track tape. We contracted to have the 1991 data downloaded, and put into readable flat file format. In order that CODES need not continue to pay to have the 9 track tapes read, we have secured a PC-based 9 track tape reader.

c. Missing/inaccurate data

As can be noted in the data dictionary, some values (CARS and EMS in particular) in the data set are missing or obviously inaccurate. The data dictionary lists each variable with comment and shows number of missing variables. In some instances, the inaccurate data were obvious, however, we chose to leave the inaccurate data as provided. The data dictionary was reviewed variable by variable at an advisory committee meeting and the various data providers were made aware of the problems. Also, the variable 'PAYOR' was not changed in the 6 month and 12 month follow-ups. Therefore, if the payor changed at any time between the initial hospitalization and the 12 month follow-up, we did not account for the change.

d. Lack of uniformity

We received the Hospital case mix files, the EMS files, and the CARS files in a variety of formats. We chose to allow each data provider to deliver the data in the format that was easiest for them. PERL scripts were written to manipulate the data for Foxpro extraction. Since we were making the request and would be inconveniencing the data provider, we felt that the burden of uniformity was not as important as receipt of the data. Most providers were quick with the requests because they did not have to write programs specific to our needs.

e. Equipment problems

The only problem that we were minimally faced with was transferring the linked data sets quickly from Mike Dean to Jim Reading and Pat Nechodom. We solved this by purchasing a BACKPACK portable printer port access tape backup drive which can store 250Mb using data compression. This unit allowed quick transfer of even a large data set.

f. Legislative/Interagency politics

The public relations of CODES could not be any better. All data providers have been invited to join the advisory committee and work well together in a community effort to link the various data sets. A representative for the US Congresswoman, Karen Shepherd has joined the advisory committee and been extremely supportive of the CODES mission. During the 1994 January-February Utah state legislative sessions, various legislators requested data from CODES to aid in their attempts to strengthen Utah's seat belt and helmet laws.

g. CODES advisory committee

The advisory committee is comprised of individuals who represent data user and data provider groups. The advisory committee has been instrumental in giving the CODES project direction and guidance. All data providers are invited to join the advisory committee, initially as a show of good faith concerning confidentiality issues. As CODES has progressed, however, advisory committee members have turned to the project to streamline their operations and disseminate data in a timely manner. Data requests have come from hospital planning boards, Health districts, UT Highway Patrol, UT Head Injury Association, Hold on to Dear Life Campaign (a children's safety program), FOX television, Salt Lake Tribune, Deseret News, state legislators, Department of Health, Department of Public Safety, etc. Following is a list of Charter members, and their affiliation:

Craig Allred, Director Department Of Public Safety, Utah Highway Safety Office Randy Baker, Director of Medical Records Legacy Health Sciences Darren Baggs Department Of Health, Bureau Of Emergency Medical Services Jan Buttrey, Director Department Of Health, Emergency Medical Services Denise Beaudoin, M.D. Department Of Health, Bureau Of Epidemiology Calvert Cazier Department Of Health, Family Health Services Kurt Bernhisel, M.D. Emergency Department, University Of Utah Medical Center Christine Chalkley Department Of Health, Health Promotion/Risk Reduction Diane Brown Western Rehabilitation Institute, Medical Records Supervisor Byron Clawson Utah Hospital Association Bradley Brown Department Of Health, Division Of Health Care Financing John Dame, Jr Utah Highway Safety Office John Brockert Department Of Health, Vital Records And Health Statistics Marty Caravati, M.D. Emergency Department, University Of Utah Med. Center Steven Erickson, Prevention Program Director Workers Compensation Fund Of Utah Robert Herr, M.D. FHP Hospital, Emergency Room Max Lauderdale, Administrator HCA St. Mark's Hospital Mike Openshaw, Controller University Of Utah Med. Center, Administrative Services Denise Love, Director Department Of Health, Health Data Analysis Marilee Gomez, Director Department Of Public Safety, FARS Floyd Morgan, Administrator Humana Hospital Davis North Susan Horn, Ph.D. Intermountain Health Care Mark Neff, Administrator Holy Cross Jordan Valley Hospital Doug Julander, Regional Director Healthtrust Incorporated Kathy Nutter Department Of Public Safety Richard Julio Department Of Transportation, Traffic And Safety Luis Piata , Statistician Department Of Health, Health Data Analysis Trish Keller, Program Director Department Of Health, Bureau of Child and Maternal Health Todd Steward, Assistant Administrator HCA St. Mark's Hospital Matt Bradford University Of Utah Med. Center, Medical Records Al Tokanaga, Director University Of Utah Med. Center, Medical Records Beverly Miller, Assistant Congresswoman Karen Shephard Joel Macey Bryner Clinic, Medical Records Cindy Wilson, Coordinator Holy Cross Jordan Valley Hospital, Emergency Room Mary Webb Worker's Compensation Fund Of Utah, Medical Records Gerry Vanorman Department Of Health, Emergency Medical Services for Children

h. Publication guidelines

The Utah CODES advisory committee formed a subcommittee for the purpose of creating a publication policy was submitted and approved by the advisory committee at the October 1993 advisory committee meeting.

D. Outline statistical procedures

1. Logistic regression

Logistic regression is a method of finding the best fitting and most parsimonious model to describe the relationship between a binary outcome and a set of explanatory variables. Like all other models logistic regression depends upon a set of assumptions including a probability model. Briefly, in logistic regression the conditional mean of the regression equation is assumed to follow a logistic distribution, the errors are assumed to follow a binomial distribution and the logit of the conditional probability that the outcome is present is modeled by a linear combination of the explanatory variables. The theory of logistic regression is quite similar to multiple regression except that it is impossible to solve the necessary equations in closed form and so iterative techniques are required. An advantage of logistic regression is that the regression coefficient of any variable exponentiated is an estimate of the odds ratio of that variable with respect to the dependent variable. The theory of logistic regression is explained in detail in a number of books, e.g. "Applied Logistic Regression" by David W. Hosmer and Stanley Lemeshow, John Wiley and Sons, New York, 1989. Logistic regression is the method used in the NHTSA mandated model. In this analysis four dependent variables (labeled A, B, C, and D) were created to specify various levels of injury (including death) and the explanatory variables were specified as core variables (BELTUSE, ROLL, SVFO, SVO, MVH, RURAL, AGE, MALE SPLIM, DRIVER, FRNTPAS) and non-core variables (WET, TIME, INTER, PC). These runs were made using SAS procedure LOGISTIC and the output from these runs and comments about the results are contained in the RESULTS section of this report.

2. Multiple Regression

Multiple regression is the method used to determine the best model for establishing a relationship between an outcome variable and a set of explanatory variables. In this case the outcome variable is assumed to be continuous, the relationship between the outcome and explanatory variables linear and the errors distributed multivariate normal. Multiple regression was used in the NHTSA mandated model when charges were the outcome variable and the explanatory variables were specified (as with logistic regression) as core variables (BELTUSE, ROLL, SVFO, SVO, MVH, RURAL, AGE, MALE SPLIM, DRIVER, FRNTPAS) and non-core variables (WET, TIME, INTER, PC). These runs were made using SAS procedure REG and the output from these runs and comments about the results are contained in the RESULTS section of this report. In our RESULTS section we not only ran the model specified by NHTSA in which only the individuals who had charges were included, but we performed the same analysis on all individuals whether or not they had charges. We feel this second analysis is the most appropriate one and is explained in detail in state specific analysis Item 6.

3. Validation tables

The validation tables were run as instructed in the model. However, the variable 'BELTUSE' was missing in many of the crash reports. When the BELTUSE variable was missing, the entire report was disallowed. We were interested in checking the impact that incomplete and subsequently discarded reports would have on validation results. Therefore, the validation tables were rerun for the passenger car -- 5 and over data set, and the passenger car -- drivers only data set. The results of the validations with missing values calculated follow the table of passenger car - age 5 and over population, and passenger car - drivers only general beltuse.

E. Outline validation process

1. Case Selection/Reliability of study population

As noted above, the Utah CODES crash population accurately reflects the general population and demographics of the state. We have reviewed the results, the individual data sets and feel confident that our final data set accurately reflects the crash population for 1991 in population density, rural/urban distribution, and age. Nearly 80% of the Utah population are found within a 100 mile strip in the north central area of the state referred to as the Wasatch Front (Davis County, Salt Lake County, Utah County, and Weber County). The CODES passenger car - light truck and van data set reflects this population density with Davis County accounting for 7.4% of crashes, Salt Lake County at 50.1%, Utah County at 13.4%, and Weber County at 8.4% of the crashes for a total of 79.3% of all 1991 crashes. Likewise, the CODES data set reflects the rural/urban distribution as expected: 73.1% of all crashes were in urban areas with the remaining 26.9% occurring in the rural areas of Utah, much the same as the statewide distribution.. In 1990, Utah ranked as #1 in the nation in percentage of population under age 5 years, percentage of population between ages 5 and 17 years, number of school age children per adult, and average household size. The CODES passenger car - light truck and van data set has a median age of 25 years, the same median age as the state overall. While 48% of the Utah general population in 1991 were under the age of 25, 49.5% of the entire 1991 crash population were under the age of 25.

III. Discussion/conclusion

A. Caveats and limitations

1. Significance of false negatives and false positives

As mentioned previously, Utah CODES has not performed any specific record review concerning false negatives or false positives.

2. Impact of unique characteristics of the site

Since Utah does not have a primary enforcement seat belt law or a helmet law, our results could be skewed. Our state-specific analysis will take a deeper look into this problem of over-reporting of seat belt use. Concerning the lack of helmet law, it is noteworthy that crash forms do not account for no helmet use. Therefore, reporting officers do not always see the need to fill out the field of 'safety device use.' The reporting form does not include a place for 'no helmet used' and therefore all missing safety device used fields for motorcycles were coded as 'helmet not used.'

B. Recommendations

1. Linkage process

MATCHING STRATEGIES FOR DUPLICATE MEDICAL ENCOUNTERS

Using probabilistic linkage, it is possible to link a crash record to an emergency medical services, outpatient, medical record. However, in its usual method of use, the software which we employ for probabilistic linkage will only link one record in file A to one record in file B. This is, in fact, the behavior one would desire. A problem immediately arises because if a patient is linked to an emergency medical service record, but there were multiple vehicles dispatched to care for the victim, only one of those multiple EMS records will be appropriately linked. Similarly, when linking the crash record to an outpatient record, only one outpatient visit will be linked to the crash before the crash will be removed from file A. A similar problem exists with all the subsequent medical record files. We have devised a successful strategy for using the Automatcher in its unduplicating and geocoding modes to handle this problem.

Home   Staff   Publications   Linkage   Data   Traffic Safety   TRCC   Other Resources

Utah CODES (Crash Outcome Data Evaluation System)
 615 Arapeen Dr, Suite 202 Salt Lake City, UT 84108-1226 
Ph: (801) 581-6410, Fax: (801) 581-8686
General Information: larry.cook@hsc.utah.edu Website: IICRC Website