Influence of Sampling on Estimates of Clustering and Recent Transmission of Mycobacterium tuberculosis Derived from DNA Fingerprinting Techniques

• Published in: American journal of epidemiology Vol. 149; no. 4; pp. 366 - 371
• Main Authors: Glynn, J. R., Vyonycky, E., Fine, P. E. M.
• Format: Journal Article
• Language: English
• Published: Cary, NC Oxford University Press 15.02.1999
• Subjects: Bacterial diseases; Bias; bias (epidemiology); Biological and medical sciences; Cluster Analysis; DNA Fingerprinting; Epidemiologic Methods; epidemiology; Human bacterial diseases; Humans; Infectious diseases; Medical sciences; Models, Statistical; molecular; Mycobacterium tuberculosis - genetics; Sampling Studies; San Francisco; South Africa; Stochastic Processes; study design; tuberculosis; Tuberculosis and atypical mycobacterial infections; Tuberculosis, Pulmonary - transmission; United Kingdom; Europe; Cluster analysis; Human; Evaluation; Statistical analysis; Methodology; Respiratory disease; Mycobacterial infection; Medical screening; Epidemiology; Infection; Tuberculosis; Statistical model; Fingerprint method; Bacteriosis; Transmission from man to man; Sampling; Public health
• ISSN: 0002-9262; 1476-6256
• DOI: 10.1093/oxfordjournals.aje.a009822

The availability of DNA fingerprinting techniques for Mycobacterium tuberculosis has led to attempts to estimate the extent of recent transmission in populations, using the assumption that groups of tuberculosis patients with identical isolates (“clusters”) are likely to reflect recently acquired infections. It is never possible to include all cases of tuberculosis in a given population in a study, and the proportion of isolates found to be clustered will depend on the completeness of the sampling. Using stochastic simulation models based on real and hypothetical populations, the authors demonstrate the influence of incomplete sampling on the estimates of clustering obtained. The results show that as the sampling fraction increases, the proportion of isolates identified as clustered also increases and the variance of the estimated proportion clustered decreases. Cluster size is also important: the underestimation of clustering for any given sampling fraction is greater, and the variability in the results obtained is larger, for populations with small clusters than for those with the same number of individuals arranged in large clusters. A considerable amount of caution should be used in interpreting the results of studies on clustering of M. tuberculosis isolates, particularly when sampling fractions are small. Am J Epidemiol 1999;149:366–71.