Validation of a Hierarchical Deterministic Record-Linkage Algorithm Using Data From 2 Different Cohorts of Human Immunodeficiency Virus-Infected Persons and Mortality Databases in Brazil

• Published in: American journal of epidemiology 2008-12, Vol.168 (11), p.1326-1332
• Main Authors: Pacheco, Antonio G., Saraceni, Valeria, Tuboi, Suely H., Moulton, Lawrence H., Chaisson, Richard E., Cavalcante, Solange C., Durovni, Betina, Faulhaber, José C., Golub, Jonathan E., King, Bonnie, Schechter, Mauro, Harrison, Lee H.
• Format: Article
• Language: English
• Subjects: Adolescent; Adult; Algorithms; Analysis. Health state; Biological and medical sciences; Brazil - epidemiology; Cohort Studies; data collection; Databases, Factual - statistics & numerical data; Death Certificates; Epidemiology; Female; General aspects; HIV; HIV Infections - complications; HIV Infections - mortality; Human immunodeficiency virus; Human viral diseases; Humans; Immunodeficiencies; Immunodeficiencies. Immunoglobulinopathies; Immunopathology; Infectious diseases; Male; Medical Record Linkage; Medical sciences; Miscellaneous; Mortality; Practice of Epidemiology; Public health. Hygiene; Public health. Hygiene-occupational medicine; Research methodology; Sensitivity and Specificity; software validation; Survival Rate; Tuberculosis, Pulmonary - complications; Tuberculosis, Pulmonary - mortality; Viral diseases; Viral diseases of the lymphoid tissue and the blood. Aids
• ISSN: 0002-9262; 1476-6256
• DOI: 10.1093/aje/kwn249

Loss to follow-up is a major source of bias in cohorts of patients with human immunodeficiency virus (HIV) and could lead to underestimation of mortality. The authors developed a hierarchical deterministic linkage algorithm to be used primarily with cohorts of HIV-infected persons to recover vital status information for patients lost to follow-up. Data from patients known to be deceased in 2 cohorts in Rio de Janeiro, Brazil, and data from the Rio de Janeiro State mortality database for 1999–2006 were used to validate the algorithm. A fully automated procedure yielded a sensitivity of 92.9% and specificity of 100% when no information was missing. When the automated procedure was combined with clerical review, in a scenario of 5% death prevalence and 20% missing mothers’ names, sensitivity reached 96.5% and specificity 100%. In a practical application, the algorithm significantly increased death rates and decreased the rate of loss to follow-up in the cohorts. The finding that 23.9% of matched records did not give HIV or acquired immunodeficiency syndrome as the cause of death reinforces the need to search all-cause mortality databases and alerts for possible underestimation of death rates. These results indicate that the algorithm is accurate enough to recover vital status information on patients lost to follow-up in cohort studies.