Validation of postmortem interviews to ascertain selected causes of death in children

In developing countries, diagnoses of diseases associated with deaths in children are frequently derived from retrospective maternal interviews. To determine the validity of this methodology, and to define sensitive and specific diagnostic algorithms, we compared symptoms and signs reported by mothers using structured questionnaires, with selected physician diagnoses for 164 deaths among hospitalized children on the Philippine island of Cebu. The 164 decreased children had 256 physician diagnoses of acute lower respiratory infections (ALRI) (100), diarrhoeas (92), measles (48), and neonatal tetanus cases (16). Forty-three per cent of children had multiple illnesses. An algorithm for tetanus (age at death less than or equal to 30 days with convulsion or spasm) was 100% sensitive, but specificity could not be estimated due to the small number of comparison neonatal deaths. An algorithm for measles (age greater than or equal to 120 days, with rash and fever for at least three days) had 98% sensitivity and 90% specificity. Diagnosis of ALRI was more difficult, cough and dyspnoea alone yielding 86% sensitivity but low specificity, whereas prolonged cough and dyspnoea provided 93% specificity but low sensitivity (41%). Diarrhoea diagnoses based on frequent loose or liquid stools had high sensitivity (78-84%) and specificity (79%), irrespective of whether the child died with diarrhoea alone or in combination with other illnesses. However, maternal reports of moderate/severe dehydration had low specificity. We conclude that, in this setting, verbal autopsies can diagnose major illnesses contributing to death in children with acceptable sensitivity and specificity.