TY - JOUR
T1 - The validity of using multiple imputation for missing out-of-hospital data in a state trauma registry
AU - Newgard, Craig D.
N1 - Funding Information:
Supported by a career development grant from the Emergency Medicine Foundation.
PY - 2006/3
Y1 - 2006/3
N2 - Objectives: To assess 1) the agreement of multiply imputed out-of-hospital values previously missing in a state trauma registry compared with known ambulance values and 2) the potential impact of using multiple imputation versus a commonly used method for handling missing data (i.e., complete case analysis) in a typical multivariable injury analysis. Methods: This was a retrospective cohort analysis. Multiply imputed out-of-hospital data from 1998 to 2003 for four variables (intubation attempt, Glasgow Coma Scale score, systolic blood pressure, and respiratory rate) were compared with known values from probabilistically linked ambulance records using measures of agreement (κ, weighted κ, and Bland-Altman plots). Ambulance values were assumed to represent the "true" values for all analyses. A hypothetical multivariable regression model was used to demonstrate the impact (i.e., bias and precision of model results) of handling missing out-of-hospital data with multiple imputation versus complete case analysis. Results: A total of 6,150 matched ambulance and trauma registry records were available for comparison. Multiply imputed values for the four out-of-hospital variables demonstrated fair to good agreement with known ambulance values. When included in typical multivariable analyses, multiple imputation increased precision and reduced bias compared with using complete case analysis for the same data set. Conclusions: Multiply imputed out-of-hospital values for intubation attempt, Glasgow Coma Scale score, systolic blood pressure, and respiratory rate have fair to good agreement with known ambulance values. Multiple imputation also increased precision and reduced bias compared with complete case analysis in a typical multivariable injury model, and it should be considered for studies using out-of-hospital data from a trauma registry, particularly when substantial portions of data are missing.
AB - Objectives: To assess 1) the agreement of multiply imputed out-of-hospital values previously missing in a state trauma registry compared with known ambulance values and 2) the potential impact of using multiple imputation versus a commonly used method for handling missing data (i.e., complete case analysis) in a typical multivariable injury analysis. Methods: This was a retrospective cohort analysis. Multiply imputed out-of-hospital data from 1998 to 2003 for four variables (intubation attempt, Glasgow Coma Scale score, systolic blood pressure, and respiratory rate) were compared with known values from probabilistically linked ambulance records using measures of agreement (κ, weighted κ, and Bland-Altman plots). Ambulance values were assumed to represent the "true" values for all analyses. A hypothetical multivariable regression model was used to demonstrate the impact (i.e., bias and precision of model results) of handling missing out-of-hospital data with multiple imputation versus complete case analysis. Results: A total of 6,150 matched ambulance and trauma registry records were available for comparison. Multiply imputed values for the four out-of-hospital variables demonstrated fair to good agreement with known ambulance values. When included in typical multivariable analyses, multiple imputation increased precision and reduced bias compared with using complete case analysis for the same data set. Conclusions: Multiply imputed out-of-hospital values for intubation attempt, Glasgow Coma Scale score, systolic blood pressure, and respiratory rate have fair to good agreement with known ambulance values. Multiple imputation also increased precision and reduced bias compared with complete case analysis in a typical multivariable injury model, and it should be considered for studies using out-of-hospital data from a trauma registry, particularly when substantial portions of data are missing.
KW - Agreement
KW - EMS
KW - Imputation
KW - Out-of-hospital
KW - Prehospital
KW - Validity
UR - http://www.scopus.com/inward/record.url?scp=33644502610&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=33644502610&partnerID=8YFLogxK
U2 - 10.1197/j.aem.2005.09.011
DO - 10.1197/j.aem.2005.09.011
M3 - Article
C2 - 16495420
AN - SCOPUS:33644502610
SN - 1069-6563
VL - 13
SP - 314
EP - 324
JO - Academic Emergency Medicine
JF - Academic Emergency Medicine
IS - 3
ER -