A validation study of verbal autopsy as a means of determining the cause of adult Dealth.

Abstract

To validate a verbal autopsy form by comparing diagnoses derived from verbal autopsy with the hospital diagnoses of patients admitted to the medical wards of Muhimbili Medical Centre. Design: Setting: Patients: Prospective study of patients who were accompanied by relatives during their admission to two medical wards of Muhimbili Medical Centre during the period 1st July to 31st December 1993. Muhimbili Medical Centre, Dar-es-Salaam. 835 patients: 508 males and 327 female. Results: Among the 835 patients 510 (61.1 %) were correctly diagnosed by verbal autopsy. Of the 508 male patients, 326 (64.2%) were accurately diagnosed by verbal autopsy while of 327 females 184 (56.3%) were correctly diagnosed. The number and proportion of patients accurately diagnosed .by verbal autopsy among diagnoses made in the wards were as follows: Tuberculosis associated with Human immunodeficiency virus disease (TB/HIV disease) 48 (62.5%), tuberculosis 79 (74.5%), HIV disease (AIDS) 27 (37.5%), acute gastroenteritis 38 (80.9%), malaria 49 (62.6%%), meningitis 22 (88.0%), cerebral malaria 22 (66.7%), heart failure 23 (63.9%), diabetes mellitus 20 (62.5%), suicide (intentional self harm) 17 (94.4%), liver diseases 11 (47.8%), cancer 4 (30.8%)., anaemia 11 (45.8%), acute lower respiratory tract infection 11 (37.9%). Sensitivities, specificities, positive and negative predictive values of common diagnoses derived from verbal autopsy compared with hospital diagnoses were calculated. Meningitis, acute gastroenteritis, external causes, suicide and tetanus had sensitivities ~ 80.0% and specifities ~90.0%. Their positvc predictive values were> 65.0%; while negative predictive values were> 95.0%. HIV disease, pulmonary tuberculosis, HfV disease associated with tuberculosis, malaria, cerebral malaria, stroke and diabetes mellitus had sensitivities between 60.0-75.0% and spccificities > 94%. Their positive predictive values were between 58.0% - 90.0% and negative predictive values> 90.0%. The adult verbal autopsy performed poorly in patients with cancer, abscesses and cellulitis, acute lower respiratory tract infection (pneumonia), liver diseases, abdominal conditions. Their sensitivities were less than 50.0%. Of the 835 respondents 715 (85.6%) had cared for the patients during the illness which led to admission. The probability of a correct diagnosis using verbal autopsy was greater when patient had been cared for by the respondent (P=O.OOOl). The gender of respondents was not associated with delivery of a more accurate history. Conclusion: This study has shown that for certain conditions verbal autopsy is a useful diagnostic tool, but for others its sensitivity is low. Awareness of the strength and limitation of verbal autopsy is important for those conducting community based studies to determine cause-specific mortality rates.