Influence of sampling on clustering and associations with risk factors in the molecular epidemiology of tuberculosis

Molecular epidemiologic studies may use genotypic clustering of isolates as an indicator of recent transmission. It has been shown that missing cases lead to underestimating clustering, and modelling studies suggested that they may also lead to underestimating odds ratios for clustering. Using a national, comprehensive database from the Netherlands covering 15 years between 1993 and 2007 and including over 12,000 patients and their isolates, the authors determined the effects of sampling at random, in time, and by geographic area. As expected, sampling reduced the observed clustering percentages. However, sampling did not reduce the observed odds ratios for clustering. The main explanations for this discrepancy with model outcomes were that a substantial proportion of clustered cases were found in large clusters and that risk factors for clustering tended to be-among clustered cases-also risk factors for large clusters. The authors conclude that, in settings where risk factors for clustering may be interpreted as risk factors for recent transmission, these risk factors are also associated with larger cluster sizes. As a result, odds ratios would show limited sampling bias.