Measuring the burden of common morbidities: sampling disease experience versus continuous surveillance

Longitudinal prevalence, the proportion of all days of observation that a given individual manifests symptoms of illness, is a measure of disease frequency that is easy to generate from daily morbidity data and has been shown to be strongly related to subsequent health outcome. It is hypothesized that this measure could be derived using a representative sample of days of observation rather than continuous surveillance. The authors use 1990-1991 data from a Brazilian supplementation trial comprising a year's daily records of the occurrence of diarrhea, fever, and cough in 906 children under 5 years of age to examine how many days of morbidity data need to be observed to rank subjects into quintiles of illness frequency. Systematic samples of the full data set, based on every 2nd, 3rd, 5th, 10th, 15th, 20th, and 30th day of data, are compared with the continuous record. For diarrhea and fever, estimates based on less than 72 days of observation result in over one fourth of individuals who should have been in the extreme quintiles of the morbidity distribution being misclassified, and over one fifth of all subjects appear (falsely) to suffer no morbidity. Estimates of longitudinal prevalence should be based on at least 72 days of observation.