Category: Spinal Cord Injury; Quality Improvement and Implementation Science; Measurement
To develop weights to extend the generalizbility of the SCIMS to the target U.S. adults and provide estimates to describe the patterns of key variables associated with spinal cord injuries.
Design : De-identified, case-level data on adults aged 18 years and above, discharged alive from the SCIMS and auxiliary dataset (UDSMR and eRehabData) between January 2001 and December 2008 will be obtained.
Setting : The SCIMS, established with a goal to prospectively capture continuous and provide comprehensive-high accuracy data, is not population based and its representativeness to the U.S. population emains unestablished. This research , proposes to use UDSMR and eRehabData to weight the SCIMS data such that it would permit inferences to the target population.
Participants (or Animals, Specimens, Cadavers) : NA
Interventions : NA
Main Outcome Measure(s) : Three sets of weights that will be generated are described below:
Inverse probability of selection and non-reponse weights: To model each individual's potential differential selection and non-response probabilities.
Raking weights : To improve the quality of design-weighted estimates obtained from the two weights described above using Raking.
This effort will also provide population characteristics for variables that have previously not been available (e.g. mental health outcomes) since they were not a part of the auxiliary dataset. Descriptive characteristics (socio-demographic, mental health-related, behavioral, and injury-related) associated with U.S. adults under rehabilitation will be also estimated using weights described above.
Results : This study is currently in progress; expected results are discussed below:
Case-level sampling weights obtained in this research efforts will permit unbiased estimates taking into account that the individual selection, non-response probabilities, and that marginal variable distributions differ between the sample and auxiliary data. The weights will be able to bring up the SCIMS data to the population dimensions obtained from auxiliary data. The weighted data will then be used to provide descriptive characteristics for key variables associated with SCI through pre-injury and discharge phase, and projected incidence, stratified by age-group will also be provided.
Conclusions : This research effort will generate weights using which researchers can extend the generalizbility of the estimated generated using SCIMS data to the target U.S. population. The descriptive summary of representative estimates provided by this effort may also help formulate and test previously un-researched questions associated with this population.