June 2013
Volume 54, Issue 15
ARVO Annual Meeting Abstract  |   June 2013
Estimation of design effect, sample size and number of clusters needed for trachoma prevalence surveys
Author Affiliations & Notes
  • Beatriz Munoz
    Ophthalmology, Johns Hopkins Wilmer Eye Inst, Baltimore, MD
  • Jonathan King
    Carter Center, Atlanta, GA
  • Jeremiah Ngondi
    Public Health and Primary Care, Institute of Public Health, University of Cambridge, Cambrige, United Kingdom
  • Paul Emerson
    Carter Center, Atlanta, GA
  • Sheila West
    Ophthalmology, Johns Hopkins Wilmer Eye Inst, Baltimore, MD
  • Footnotes
    Commercial Relationships Beatriz Munoz, None; Jonathan King, None; Jeremiah Ngondi, None; Paul Emerson, None; Sheila West, None
  • Footnotes
    Support None
Investigative Ophthalmology & Visual Science June 2013, Vol.54, 5728. doi:
  • Views
  • Share
  • Tools
    • Alerts
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Beatriz Munoz, Jonathan King, Jeremiah Ngondi, Paul Emerson, Sheila West; Estimation of design effect, sample size and number of clusters needed for trachoma prevalence surveys. Invest. Ophthalmol. Vis. Sci. 2013;54(15):5728.

      Download citation file:

      © ARVO (1962-2015); The Authors (2016-present)

  • Supplements

Purpose: Trachoma is the leading infectious cause of blindness world-wide. The World Health Organization (WHO) has the goal of eliminating blinding trachoma by 2020. Identifying the areas in need of trachoma control programs is imperative. District level surveys using cluster random samples (CRS) are recommended to estimate the prevalence. Because of the clustering nature of trachoma establishing the magnitude of the within-cluster association and its influence on the design effect (DE) is crucial to determine the number of clusters needed for the survey as this number drives cost and feasibility

Methods: Using data from 127 cluster surveys for trachoma carried out in districts where trachoma was suspected to be present, the within-cluster association was estimated for each survey using the pairwise odds ratio (POR) and design effects were then estimated as a function of cluster size, expected prevalence, and POR. Finally, number of required clusters of size 50, 75, 100 was calculated for expected prevalences of 5%, 10% and 15% and precisions between one third and one fourth of the prevalence.

Results: The magnitude of the POR was associated with the trachoma prevalence level. Higher levels of clustering (PORs) were observed at less than 5% prevalence (median(IQR)) 1.72 (1.29, 3.32) and decline as the prevalence increased. The current recommendation of 30 clusters of size 50 is suboptimal when the expected prevalence of trachoma is below 5%, even if the level of precision is a third of the prevalence. When the prevalence was greater than 10% the median POR was 1.43 (1.22,1.70) allowing 18 clusters of size 50 to be sufficient to estimate a prevalence of 10% with a precision of 3%.

Conclusions: For expected prevalence between 10% and 20%, depending on the desired level of precision, less than 20 clusters is enough. The reduction in number of clusters needed results in substantial savings for the trachoma control programs.

Keywords: 736 trachoma • 459 clinical (human) or epidemiologic studies: biostatistics/epidemiology methodology • 463 clinical (human) or epidemiologic studies: prevalence/incidence  

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.