Use of primary diagnosis during hospitalization in the Unified Health System (Sistema Único de Saúde) to qualify information regarding the underlying cause of natural deaths among the elderly.

Introduction:: Ill-defined causes of death are common among the elderly owing to the high frequency of comorbidities and, consequently, to the difficulty in defining the underlying cause of death.

Objective:: To analyze the validity and reliability of the "primary diagnosis" in hospitalization to recover the information on the underlying cause of death in natural deaths among the elderly whose deaths were originally assigned to "ill-defined cause" in their Death Certificate. The hospitalizations occurred in the state of Rio de Janeiro, in 2006.

Methods:: The databases obtained in the Information Systems on Mortality and Hospitalization were probabilistically linked. The following data were calculated for hospitalizations of the elderly that evolved into deaths with a natural cause: concordance percentages, Kappa coefficient, sensitivity, specificity, and the positive predictive value of the primary diagnosis. Deaths related to "ill-defined causes" were assigned to a new cause, which was defined based on the primary diagnosis.

Results:: The reliability of the primary diagnosis was good, according to the total percentage of consistency (50.2%), and fair, according to the Kappa coefficient (k = 0.4; p < 0.0001). Diseases related to the circulatory system and neoplasia occurred with the highest frequency among the deaths and the hospitalizations and presented a higher consistency of positive predictive values per chapter and grouping of the International Classification of Diseases. The recovery of the information on the primary cause occurred in 22.6% of the deaths with ill-defined causes (n = 14).

Conclusion:: The methodology developed and applied for the recovery of the information on the natural cause of death among the elderly in this study had the advantage of effectiveness and the reduction of costs compared to an investigation of the death that is recommended in situations of non-linked and low positive predictive values. Monitoring the mortality profile by the cause of death is necessary to periodically update the predictive values.