Correction of self-reported prevalence in epidemiological studies with large samples

Disease prevalence rates are useful when formulating and evaluating public policies. Self-reported measurement is commonly used, since it is easy to collect and does not require specific health training or additional cost. However, this measurement process can produce a biased measure. This study aimed to present the existing methods to adjust prevalence, based on self-report, focusing on computational problems in the case of large samples and proposing an alternative solution. The methods were classified as: algebraic, simple to perform, but not applicable to any combination of self-reported prevalence, specificity, and sensitivity; and Bayesian, which does not have the previous strategy limitations, but displays computational problems when applied to large samples in personal computers. These problems impede the existing method's direct implementation, raising the need to present an approximate strategy to make estimation possible. The empirical method proposed here for application to large samples consists of reducing the sample as far as possible to calculate with the statistical package, maintaining the proportion of patients. We found the method adequate, since it converges with the true value. In the example, a self-reported prevalence of 5% with sensitivity = 0.4 and specificity = 0.9 was corrected to 0.17% (95% CI: 0.10-0.24). The study presented the existing methods for adjusting prevalence rates and a new strategy for prevalence rates in large samples, allowing estimates closer to the true values without the need to directly measure all the individuals.