Mohamed G Qayad1, Hui Zhang. 1. MCH Epidemiology Section, Epidemiology Branch, Georgia Division of Public Health, Georgia Department of Human Resources, Atlanta, GA, USA. MQayad@cdc.gov
Abstract
OBJECTIVE: To evaluate the accuracy of computer-matched records for maternal and child health epidemiologic investigations. METHODS: Using Automatch probabilistic record linkage software, we linked electronic records in Georgia for the 2001 Medicaid paid delivery claims and infant deaths of the 2001 birth cohort to 2001 births, and the 2002 newborn screening and hospital discharge data to 2002 births, using multiple variables for matching. We evaluated the accuracy of matches using a manual inspection of a subset of linked pairs and validated with external data. We assessed the agreement between two programmers who linked the same data independently to examine the reliability of the manual inspection technique, and estimated the percent matches to examine further the accuracy of linked data. RESULTS: The percent of records matched in the data linked ranged from 90% to 100% on the measures used. The positive and negative predictive values of the computer-matched records were both 100%. The agreement between the computer-matched data and the external data on the method of payment for the delivery was 90%. The reliability of the manual inspection technique was 95%. CONCLUSION: The computer-matched records evaluated were sufficiently accurate. This provides a unique source to public health practitioners and researchers to perform epidemiological studies that could not be possible with single sources data.
OBJECTIVE: To evaluate the accuracy of computer-matched records for maternal and child health epidemiologic investigations. METHODS: Using Automatch probabilistic record linkage software, we linked electronic records in Georgia for the 2001 Medicaid paid delivery claims and infant deaths of the 2001 birth cohort to 2001 births, and the 2002 newborn screening and hospital discharge data to 2002 births, using multiple variables for matching. We evaluated the accuracy of matches using a manual inspection of a subset of linked pairs and validated with external data. We assessed the agreement between two programmers who linked the same data independently to examine the reliability of the manual inspection technique, and estimated the percent matches to examine further the accuracy of linked data. RESULTS: The percent of records matched in the data linked ranged from 90% to 100% on the measures used. The positive and negative predictive values of the computer-matched records were both 100%. The agreement between the computer-matched data and the external data on the method of payment for the delivery was 90%. The reliability of the manual inspection technique was 95%. CONCLUSION: The computer-matched records evaluated were sufficiently accurate. This provides a unique source to public health practitioners and researchers to perform epidemiological studies that could not be possible with single sources data.
Authors: Cate M Cameron; David M Purdie; Erich V Kliewer; Roderick J McClure; Andre Wajda Journal: Aust N Z J Public Health Date: 2007-10 Impact factor: 2.939
Authors: Dorothea Nitsch; Susan Morton; Bianca L DeStavola; Heather Clark; David A Leon Journal: BMC Med Res Methodol Date: 2006-03-22 Impact factor: 4.615
Authors: Veralice Maria Gonçalves; Rosemeri Pedroso; Antônio Marcos dos Santos; Lisia Von Diemen; Flavio Pechansky Journal: Biomed Res Int Date: 2015-09-06 Impact factor: 3.411
Authors: Jared W Parrish; Meghan E Shanahan; Patricia G Schnitzer; Paul Lanier; Julie L Daniels; Stephen W Marshall Journal: Inj Epidemiol Date: 2017-08-07