Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
14 4 Data Linkage in States Since using a medical-diagnosis-based definition of serious injury such as MAIS 3+ calls for linkage between crash and medical outcome data, it is important to first assess the condition of state databases and linkage programs. In collaboration with the NCHRP 20-24 (37K) project, conducted by Cambridge Systematics, we surveyed state TRCCs to learn about their datasets and linkage systems (Cambridge Systematics, 2013). Fifty-three respondents from 40 states plus the District of Columbia and Puerto Rico responded to the survey including those that responded only after follow-up phone calls were made. States responding are shown in Figure 1. Figure 1. Map of U.S. showing states originally returning surveys (light blue) and states responding to follow-up phone calls (dark blue). 4.1 Definition of Serious Injury Of the 42 states responding, all except Florida reported that they measure and report on serious injuries as part of their transportation safety improvement efforts. The majority specified four ways in which injury severity is used: research (86%), safety/program planning and management (93%), generating reports (98%), and evaluating and refining existing policy and regulation (83%). States were also asked to provide the definition of serious injury that they use in reporting. Each state gave a unique answer, and all of the answers are compiled in Appendix A. In reviewing the responses, 33 of the 42 respondents gave definitions that either referenced KABCO âAâ injuries or used language from the definition of âAâ injury. The remaining states provided unique definitions. Some, such as Texas and Louisiana, use a more inclusive definition of injury in reporting (e.g., KAB). Original Followup Missing
15 4.2 Linkage Activities For each database type, states were asked whether they were linking or planning to link from that database to crash data, and if so, was linkage probabilistic or deterministic, direct or indirect, and whether the linkage was related to a previous CODES program. The CODES program, for which federal funding ended in 2012, promoted probabilistic linkage between state crash and medical outcome databases in participating states. Direct linkage was defined as linkage directly with crash data, whereas indirect linkage was defined as linkage via a different database. Probabilistic and deterministic were defined for the respondents as follows: Deterministic linkage is based on the number of individual identifiers that can be matched among the combined data sets. When using a deterministic record linkage procedure, two records are considered to match if all or some identifiers above a certain statistical threshold are identical. Probabilistic linkage involves a wider range of potential identifiers and computing weights for each identifier based on its estimated ability to correctly identify a match or a non-match. The weights are used to calculate the probability that two given records refer to the same entity. Pairs with probabilities above a certain statistical threshold are considered matches, pairs with probabilities below another threshold are considered non-matches; and those in between the two thresholds are considered âpossible matchesâ. In the section on data linkage, there were often conflicts between responses when a state sent more than one response. Thus, the percent of state responses is not strictly the percent of states since it was not possible to determine which respondent most authoritatively spoke for each state. In Tables 3-8, for the questions about details of linkage, we indicate the number of unique states for which at least one respondent indicated that linkage was being done or planned. Table 3 shows the results of questions related to linkage between crash data and state Emergency Medical Services (EMS) data. About two-thirds of states are linking or planning to link crash-to-EMS databases. About half considered their linkages to be probabilistic and half deterministic. Almost two-thirds were considered direct. One-third of linkages were associated with CODES.
16 Table 3. Percent of States Engaging in or Planning Linkage Between Crash and State EMS Databases Question Response Percent of States Data Linkage Yes 65 % No 30 % Unknown 5 % Type of Linkage (among states that are linking) Unique states=24 Probabilistic 50 % Deterministic 50 % Direct 62 % Indirect 38 % CODES-Related (among states that are linking) Unique states=24 Yes 33 % No 60 % Unknown 7 % Table 4 shows the results of questions related to linkage between crash data and state Emergency Department (ED) data. Fewer than half of states are linking or planning to link crash to ED databases, though a substantial portion (14%) did not know. Of those who are linking, most considered their linkages to be probabilistic and about half were considered direct. One- quarter of linkages were associated with CODES. Table 4. Percent of States Engaging in or Planning Linkage Between Crash and State ED Discharge Databases Question Response Percent of States Data Linkage Yes 41 % No 46 % Unknown 14 % Type of Linkage (among states that are linking) Unique states=17 Probabilistic 72 % Deterministic 28 % Direct 50 % Indirect 50 % CODES-Related (among states that are linking) Unique states=17 Yes 26 % No 67 % Unknown 7 % Table 5 shows the results of questions related to linkage between crash data and state hospital discharge data. Sixty percent of states are linking or planning to link crash to hospital discharge databases. Of those who are linking, most considered their linkages to be probabilistic and about half were considered direct. Over one-third of linkages were associated with CODES.
17 Table 5. Percent of States Engaging in or Planning Linkage Between Crash and State Hospital Discharge Databases Question Response Percent of States Data Linkage Yes 60 % No 32 % Unknown 8 % Type of Linkage (among states that are linking) Unique states=22 Probabilistic 76 % Deterministic 24 % Direct 48 % Indirect 52 % CODES-Related (among states that are linking) Unique states=22 Yes 38 % No 50 % Unknown 12 % Table 6 shows the results of questions related to linkage between crash data and state trauma data. Just over half of states are linking or planning to link crash to state trauma registries. Of those who are linking, 62% considered their linkages to be probabilistic and half were considered direct. About one-quarter of linkages were associated with CODES. Table 6. Percent of States Engaging in or Planning Linkage Between Crash and State Trauma Registry Databases Question Response Percent of States Data Linkage Yes 54 % No 43 % Unknown 3 % Type of Linkage (among states that are linking) Unique states=21 Probabilistic 62 % Deterministic 38 % Direct 50 % Indirect 50 % CODES-Related (among states that are linking) Unique states=21 Yes 28 % No 62 % Unknown 10 % Table 7 shows the results of questions related to linkage between crash data and vital records data. About half of states are linking or planning to link crashes to vital records databases, though 10% of respondents did not know the answer to this question. Of those who are linking, half considered their linkages to be probabilistic and just over half were considered direct. About one-quarter of linkages were associated with CODES.
18 Table 7. Percent of States Engaging in or Planning Linkage Between Crash and Vital Records Databases Question Response Percent of States Data Linkage Yes 46 % No 43 % Unknown 10 % Type of Linkage (among states that are linking) Unique states=17 Probabilistic 50 % Deterministic 50 % Direct 53 % Indirect 47 % CODES-Related (among states that are linking) Unique states=17 Yes 22 % No 70 % Unknown 8 % Table 8 shows the results of questions related to linkage between crash data and roadway inventory data. Almost 90% of states are linking or planning to link crashes to roadway inventory databases. Of those who are linking, all are deterministic and three-quarters are considered direct. Only 12% of roadway inventory linkages were associated with CODES. Table 8. Percent of States Engaging in or Planning Linkage Between Crash and Roadway Inventory Databases Question Response Percent of States Data Linkage Yes 89 % No 8 % Unknown 3 % Type of Linkage (among states that are linking) Unique states=33 Probabilistic 0 % Deterministic 100 % Direct 76 % Indirect 24 % CODES-Related (among states that are linking) Unique states=33 Yes 12 % No 82 % Unknown 6 % States were also asked about state laws related to linkage. Specifically, 75% of states indicated that state law did not set conditions for data linkage, while 11% said the law did set conditions (14% were unknown). The same 11% also indicated that state law required records linkage, while 75% of states did not. Regarding state law that established access to linked records for research, 14% of states had such a law, 64% did not, 17% were unknown, and 3% indicated that the law may allow access with approval. Finally, respondents were asked to provide a list of identifiers being used for all databases that are being linked to crash. The complete list of identifiers is given in Appendix B.
19 There is enormous variety in identifiers used across states, even for the same type of database. States using probabilistic linkage to medical outcome databases tend to use date of birth, age, gender, date of crash and date of admission. Several states use name for these linkages and a few use some type of patient identifier or other numeric identifier (e.g., last four digits of social security number). 4.3 Database Coverage A key issue in implementing statewide data linkage is the coverage of the state databases. States were asked to indicate the percent of all included cases that are captured in each of the various databases in the survey. Table 9 shows the percent of states with complete, partial or unknown/no coverage for each of seven datasets. In the âcompleteâ coverage category in Table 9, we included all responses of âcompleteâ and any âpartialâ where the respondent reported 90% or greater coverage. Most states have complete crash datasets. However, closer to half have complete coverage in statewide medical databases, including EMS, hospital, emergency, and trauma databases. Roadway inventory is widely available, but many states do not have complete coverage. Table 9. Percent of Respondents with Complete and Partial Coverage of State Databases (n=42) State Database Complete (90-100%) Coverage Partial (<90%) Coverage Unknown Crash 81% (34) 12% 7% EMS 45% (19) 17% 32% Emergency Department 38% (16) 2% 47% Hospital Discharge 50% (21) 2% 41% Trauma Registry 40% (17) 7% 47% Roadway Inventory 48% (20) 24% 25% Vital Records 57% (24) 2% 38% In looking at results for medical-outcome databases coverage from our survey, we were concerned at the high percent of unknown responses. The most recent national assessment of state trauma registry attributes was published in 2006 (Mann et al.). Based upon this published information, 32 states maintain some form of a centralized trauma registry. The majority of state data collection efforts require hospitals to report data (27 states [84%; 95% CI: 71.8%, 96.9%]). However, variability exists in the type of hospital required to submit data to the centralized registry. Thirteen states require data submission from only designated/accredited trauma centers. Another 11 states collect injury data from all acute care facilities. States requesting submission of trauma data combine information from a subset of trauma centers with existing registries. Coverage of trauma registries available within individual states vary. It is, however, interesting to note that a significant proportion of the hospitalized trauma occurring in the U.S. is considered to be captured in a trauma registry at the hospital and/or state level. Based upon individual state estimates, Mann et al. (2006) estimated that approximately 66% of registry- eligible trauma occurring in the country is captured in a state, regional or hospital-specific trauma registry.
20 4.4 Challenges, Priority and Timing The remaining survey questions cover general issues about the challenges and timing of data linkage. The first of these questions asked which of several challenges states face in implementing data linkage. Respondents could check more than one option, and the percent of respondents choosing each option is shown in Figure 2, ordered from most common to least common answer selected. Funding, confidentiality, data usage issues, and hardware/software issues were identified as a challenge for linkage by more than 50% of respondents. Figure 2. Percent of respondents who indicated challenges they faced in implementing data linkage in their state. Figure 3 shows the percent of state respondents who indicate each item that would facilitate data linkage in their state. Not surprisingly, over 80% of respondents selected increased funding. Updated equipment/software, willingness of partners to collaborate and enabling legislation were all selected by at least 50% of respondents. These responses mirror the list of challenges selected by respondents and identify the key hurdles to data linkage for most states. 0 20 40 60 80 100State law prohibits data linkage Other Interest/Willingness of partners tocollaborate Inadequate access to technical expertiseJurisdictional issues Software/Hardware infrastructureData usage restrictions Confidentiality issuesFunding Percent of States Ch al le ng es to L in ka ge
21 Figure 3. Percent of respondents who indicated changes that would facilitate implementing data linkage in their state. The final two questions were about the importance and timing of linkage in each state. Of the respondents, 4 (11%) considered linkage to be mission critical, 26 (68%) considered it to be important, and 5 (13%) considered it to be somewhat important. On the issue of timing, ten respondents (27%) estimated that linkage would be implemented within two years, nine (24%) estimated 3-5 years, 11 (30%) estimated 6-10 years, and the remaining seven said linkage would never happen in their state. In summary, linkage that is being done in states is most often between crash and roadway databases, and such linkage is deterministic. Among medical outcome databases, states using probabilistic linkage often had CODES programs, which were designed to promote probabilistic linkage between state crash and medical outcome databases. EMS is the medical outcome database being linked (or planned to be linked) to crash data in the most unique states. However, almost as many states are linking or planning linkage to hospital discharge and state trauma databases. Probabilistic linkage is common for these databases. The identifiers used for linkage are virtually unique to each state, though name, age, birthdate, gender, and crash time/location are seen in many lists. One of the challenges of this survey was that questions covered a broad range of topics, and the original survey recipient generally had to find other people to answer specific questions. This was especially true for questions about EMS, hospital, and other medical outcome databases. As a result, the findings of this survey that pertained to these topics seemed to be inconsistent with a previous survey of state trauma databases (Mann et al., 2006). Since the latter survey was likely more accurate on questions about these databases, we present key results of that study here. For present purposes, the results of the Mann et al. (2006) survey plus our own survey suggest that the majority of states have sufficient coverage and completeness in their trauma registries to facilitate data linkage. Note that most of a decade has passed since the Mann et al. (2006) trauma registry survey, so the current state of trauma registries is expected to be substantially better. Similarly, a majority of states at this time should have state EMS datasets with sufficient completeness and coverage to facilitate data linkage to crash datasets. 0 20 40 60 80 100Other Regulatory changesGreater access to technical⦠Enabling legislationInterest/Willingness of partners⦠Updated equipment/softwareIncreased funding Percentage of States Ch an ge s F ac ili ta ti ng L in ka ge
