Assessing data quality: A computerized approach

With the growing reliance on large health care data bases, the need to verify data quality increases as well. Because of the considerable costs involved in checks using primary data collection, a computerized methodology for performing such checks is suggested. The technique seems appropriate for any situation where two data collection systems (i.e. hospital discharge abstracts and physician claims for payment) relate to the same event, such as a patient's hospitalization. After reviewing other approaches, this paper suggests linking physician claims for performing particular surgical procedures with hospital discharge abstracts for the stay in which the surgery took place. Physician and hospital data for adults age 25 and over in Manitoba from 1 April, 1979 to 31 March, 1984 were used to address the questions: 1. 1. How well can the two data sets be linked? 2. 2. Given linkage of the two data sets, how much agreement is there as to procedure and diagnosis? Linkage between hospital and physician data was excellent (over 95%) for 5 out of 11 surgical procedures (hysterectomy, prostatectomy, total hip replacement, coronary artery bypass surgery, and heart valve replacement); there was over 90% perfect agreement for three other procedures (cholecystectomy, cataract surgery and total knee replacement). Problems with matching the Manitoba Health Services Commission tariffs (on physician claims) with ICD-9-CM operation codes (on hospital data) led to only 77% perfect agreement for vascular surgery and 84% for gallbladder and biliary tract operations other than cholecystectomy; over 10% of the cases linked on surgeon and date but not on the designated procedures. Diagnostic agreement between physician claim and hospital discharge abstract averaged 75% using 3-digit ICD-9-CM codes. Agreement varied considerably by type of surgery; further analysis at the 2-digit ICD-9-CM level showed considerable improvement.