Publications

Sampling Serious Injuries in Traffic Crashes at the State Level

The Moving Ahead for Progress in the 21st Century Act requires a new emphasis on measurement of serious injuries in crashes. Current practice in states that measure serious injuries is to use the police-reported injury severity of A (suspected serious injury) to define serious injury. However, serious injury is more appropriately defined as a medical diagnosis. The biggest challenge to using a medical diagnosis as the definition of serious injury is that medical outcome data are typically not part of a state crash data set. The most comprehensive solution to this challenge is to link state crash databases with state trauma registries. However, for most states, data linkage is still many years away. This paper discusses sampling medical records as an interim solution. Sampling approaches are discussed, and a stratified sampling approach is recommended. Optimal allocation of a sample provides a framework for determining the proportion of cases to sample from each subgroup, and the desired precision of the estimate and the cost per sample determine the number of cases that should be sampled. A roadmap for developing a stratified sampling plan is presented as is a hypothetical numerical example that uses data from Michigan. Sampling as an interim solution has several advantages, including comparability, progress toward comprehensive linkage, and scalability. Logistical challenges, costs of sampling, and privacy considerations are also discussed. Although sampling is not free, it represents an important way to move ahead with the necessary step of enabling measurement of serious injuries by using a diagnosis-based definition.; The Moving Ahead for Progress in the 21st Century Act requires a new emphasis on measurement of serious injuries in crashes. Current practice in states that measure serious injuries is to use the police-reported injury severity of A (suspected serious injury) to define serious injury. However, serious injury is more appropriately defined as a medical diagnosis. The biggest challenge to using a medical diagnosis as the definition of serious injury is that medical outcome data are typically not part of a state crash data set. The most comprehensive solution to this challenge is to link state crash databases with state trauma registries. However, for most states, data linkage is still many years away. This paper discusses sampling medical records as an interim solution. Sampling approaches are discussed, and a stratified sampling approach is recommended. Optimal allocation of a sample provides a framework for determining the proportion of cases to sample from each subgroup, and the desired precision of the estimate and the cost per sample determine the number of cases that should be sampled. A roadmap for developing a stratified sampling plan is presented as is a hypothetical numerical example that uses data from Michigan. Sampling as an interim solution has several advantages, including comparability, progress toward comprehensive linkage, and scalability. Logistical challenges, costs of sampling, and privacy considerations are also discussed. Although sampling is not free, it represents an important way to move ahead with the necessary step of enabling measurement of serious injuries by using a diagnosis-based definition.