Physician's manual reporting underestimates mortality: evidence from a population-based HIV/AIDS treatment program

BACKGROUND: In clinical and cohort research, mortality estimates are often derived from manual reports generated by physicians or electronic reports from vital event registries. We examined the rate of underreporting of deaths by manual methods as compared with electronic reports from a vital event registry. METHODS: The retrospective analyses included deaths among participants registered in an observational cohort who initiated highly-active antiretroviral therapy (HAART) between August 1, 1996 and June 30, 2006. Deaths were routinely reported manually by physicians and through annual electronic record linkages with a population-based vital event registry. Multivariate logistic regression was carried out to assess independent predictors of death reporting by manual methods. RESULTS: Of the 3,116 individuals included in the analyses, 622 (20.0%) died during follow-up. Manual reporting by physicians only identified 377 (60.6%), while electronic linkages captured 598 (96.1%) of all deaths. Multivariate analysis indicated that deaths among individuals with lower CD4 cell count, higher HIV plasma viral load, a history of injection drug use, and under the care of an HIV-experienced physicians were more likely to be reported manually. Furthermore, non-accidental deaths were more likely to be reported manually, and manual reporting of deaths increased over time. CONCLUSIONS: Relying only on manual reports to ascertain deaths significantly underestimates the total number of deaths in the population. This can generate important biases when evaluating the impact of therapeutic interventions in the populational setting.