Analysis of deaths with an unknown cause in epidemiologic analyses of mortality burden

To describe options for analysing deaths with an unknown cause, which often occur in community-based studies that are used to estimate disease-specific mortality burden and trends in low-income countries.

Mathematical formulae were derived that accommodate deaths with an unknown cause for the disease-specific mortality rate, proportion of deaths attributable to the disease and all-cause mortality rate. Seven specific options are presented, including example calculations from a study of childhood malaria mortality in The Gambia. An algorithm is proposed to help make decisions on analysing deaths with an unknown cause.

In the Gambian study, 25.2% of deaths had an unknown cause. Three options would result in 23.6% (minimum), 48.8% (maximum) and 28.7% (probably the best estimate) of deaths attributed to malaria. The best analysis option depends on the disease of interest: diseases for which the diagnostic method has high sensitivity and specificity (e.g., measles, neonatal tetanus) are best analysed assuming that deaths with an unknown cause never have this cause, while diseases for which specificity and/or sensitivity is low (e.g., malaria) are likely to account for some proportion of deaths with an unknown cause.

The most important aspects of analysing deaths with unknown cause are choosing appropriate assumptions, describing them explicitly and performing a sensitivity analysis. Studies of causes of death should report several key pieces of information on deaths with unknown cause to aid interpretation.