Hierarchical spatial modelling of pneumonia prevalence when response outcome has misclassification error: Applications to household data from Malawi. (February 2016)

• Record Type: Journal Article
• Authors: Kazembe, Lawrence N.; Kamndaya, Mphatso S.
• Is Part Of: Spatial and spatio-temporal epidemiology. Volume 16(2016)
• Page Start: 35
• Page End: 42
• Publication Date: 2016-02
• Subjects: Misclassification error -- Pneumonia -- Hierarchical spatial modelling -- Malawi; Epidemiology -- Statistical methods -- Periodicals; Epidemiology -- Periodicals; 614.4072
• DOI: 10.1016/j.sste.2015.11.002
• Languages: English
• ISSNs: 1877-5845

Abstract

Highlights: Self-reported pneumonia data has under-reporting and over-reporting errors. Hierarchical disease mapping models are used to permit different sensitivity and specificity. Sensitivity and specificity of greater or equal to 0.75 provided reliable and stable prevalence estimates. A lower specificity underestimated the true prevalence. Misclassification of areas as high risk was visible for incorrect specificity. Abstract: Pneumonia remains a major cause of child mortality in less developed countries. However, the accuracy of its prevalence and burden remains a challenge because disease data is often based on self-reports, resulting in measurement error in a form of under- and over-reporting. We propose hierarchical disease mapping approaches that permit measurement error, through different prior distributions of sensitivity and specificity. Proposed models were used to evaluate spatial variation of risk of pneumonia in children in Malawi. Results show that the true prevalence was 0.50 (95 CI: 0.4–0.66), however, estimates were dependent on sensitivity and specificity parameters. The estimated sensitivity was 0.76 (95% CI: 0.68–0.95), whereas specificity was 0.84 (95% CI: 0.72–0.93). A lower specificity underestimated the true prevalence, while sensitivity and specificity of greater or equal to 0.75 provided reliable and stable prevalence estimates. The spatial variation in disease risk changed little; however, misclassification of areas as high risk was visible.