• Catherine Corbell, Ishmael Katjitae, Assegid Mengistu, Francis Kalemeera, Evans Sagwa, David Mabirizi, Jennie Lates, Jude Nwokike, Sherrilyne Fuller, and Andy Stergachis.
• Department of Pharmacy, School of Pharmacy, University of Washington, Seattle, WA 98195, USA. cww@u.washington.edu
• Pharmacoepidemiol Drug Saf. 2012 Apr 1; 21 (4): 407-14.

PurposeIn 2009, the Ministry of Health and Social Services in Namibia decided to conduct a confirmatory assessment of the risk of anemia associated with zidovudine (AZT)-based highly active antiretroviral therapy (HAART) using records contained in three electronic databases. These records did not share a unique identifying number. The first step was to apply probabilistic record linkage methods to link records in the three databases.MethodsRecords of persons, aged 19-65 years, newly initiated on HAART between January 2007 and June 2008, were selected from a pharmacy electronic dispensing tool (EDT) and linked to an electronic medical records database (ePMS) and a laboratory database (MEDITECH). Using the paper-based clinical record as the gold standard, we measured the sensitivity of the starting HAART regimen, that is, proportion of AZT users in the clinical record correctly identified in electronic record, and specificity of severe anemia, that is, proportion of non-cases of severe anemia in the clinical records correctly identified in the electronic record. Kappa and intraclass correlation coefficients were used to determine reliability.ResultsA total of 12 358 records were selected from EDT. Seventy-six percent and 58% of EDT records were linked to ePMS and MEDITECH, respectively. The sensitivity of the starting HAART regimen was 98%, whereas specificity of severe anemia was 100%. The reliability scores for variables including weight, hemoglobin, and CD4 counts were moderate to perfect and ranged from 0.59 to 0.99.ConclusionProbabilistic record linkage methods were effective for records linkage in this sub-Saharan African setting.Copyright © 2011 John Wiley & Sons, Ltd.

35 keywords...

hide…