[Proxy informant reliability and bias in epidemiological research: analysis of a screening questionnaire for mental disorders].

OBJECTIVES To evaluate the reliability, magnitude and direction of the resulting bias in the application of a screening instrument for mental disorders by considering proxy informants in comparison to primary informants. METHODS Data are taken from a general morbidity community-based survey carried out in 520 randomly selected households of an industrial area of the Metropolitan Region of Salvador, the capital of Bahia state, Brazil. During the pilot phase, the first 70 families of the total sample were asked to participate in the evaluation of research instruments. The Questionnaire of Adult Psychiatric Morbidity, QAPM, consists of 44 questions about psychiatric symptoms widely used in Brazil. The husbands and wives of the selected families answered QAPM questions regarding themselves and their respective partners. One family refused to participate. The Kappa index was estimated for each QAPM question. To assess the magnitude and direction of bias, the proportional variation of prevalence was estimated from proxy and primary respondents. Each informant was analyzed as a primary informant when answering about his/her own symptoms and as a proxy informant when answering those about his/her partner. RESULTS Proxy informants as compared to primary informants show weak reliability, as measured by the Kappa Index, particularly when husbands reported on their wives' symptoms. An overall underestimation of prevalence estimates was found, which reveals the potential negative bias with the use of proxy informants for psychological symptoms. No bias was found for only two questions (lack of appetite and globus hystericus) when women were taken as proxy informants for their husbands. In addition, departures of proxy informants from primary informant-based estimates were greater among men than to women. CONCLUSIONS Proxy informants underestimate the occurrence of psychological symptoms in this community-based study. When the feasibility of a research project, based on the QAPM depends on the use of proxies, wives may be recommended as better informants than their husbands.