Population prevalence estimates from complex samples

J Clin Epidemiol. 1992 Apr;45(4):393-402. doi: 10.1016/0895-4356(92)90040-t.

Abstract

Epidemiologic studies are often designed to severe several purposes. The complex sampling plans necessary to ensure an adequate number of cases, a similar age distribution among cases and controls, or other important design constraints for comparative studies may make the additional goal of estimating prevalence in the population or in important subgroups difficult to attain with existing computer software, which typically assumes simple random sample selection. We consider here various methods for estimating overall and subgroup prevalence from complex samples, including crude prevalences, direct standardization of prevalences, and standardization using logistic regression to smooth the sampling group prevalences. We illustrate these methods using a complex sample to estimate the prevalence of Alzheimer's disease in an urban community. A simulation study under various models in this setting is also described. We conclude that the use of logistic regression to smooth sampling group prevalences before standardization is an effective method for estimation of overall prevalence, provided that the adequacy of fit of a logistic model is carefully checked.

Publication types

  • Research Support, U.S. Gov't, P.H.S.

MeSH terms

  • Aged
  • Aged, 80 and over
  • Alzheimer Disease / epidemiology
  • Boston / epidemiology
  • Confidence Intervals
  • Epidemiologic Methods*
  • Female
  • Humans
  • Logistic Models
  • Male
  • Models, Statistical
  • Prevalence*
  • Research Design
  • Sampling Studies