Premium
Testing for clustering in record linkage databases
Author(s) -
Hammerstrom Thomas
Publication year - 1995
Publication title -
pharmacoepidemiology and drug safety
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.023
H-Index - 96
eISSN - 1099-1557
pISSN - 1053-8569
DOI - 10.1002/pds.2630040402
Subject(s) - medicine , cluster analysis , suspect , medical record , pharmacy , medicaid , record linkage , pharmacoepidemiology , event (particle physics) , medical emergency , database , data mining , family medicine , health care , statistics , computer science , environmental health , medical prescription , pharmacology , population , physics , mathematics , quantum mechanics , political science , law , economics , radiology , economic growth
One way to identify an excess risk of an adverse event during drug exposure is to use administrative, medical, and pharmacy records data, such as Medicaid or HMO (Health Maintenance Organization) records, to compare the relative frequency of the suspect event in exposed and unexposed periods. Such estimates can be seriously biased if there is clustering in patient visits to physicians and pharmacies. This paper presents two simple tests for determining whether the patient visit dates in such a database are Poisson processes of visits or clustered processes. The paper shows that this test is quite powerful at detecting both clustered sequences and nearly regular sequences of visits. Finally, it gives a straightforward SAS program which will carry out the tests.