An explanation for the low proportion of tuberculosis that results from transmission between household and known social contacts

Abstract

We currently have little idea where Mycobacterium tuberculosis (Mtb) transmission occurs in high incidence settings. Molecular studies suggest that only around 8-19% of transmission to adults occurs within-household, or between known social-contacts. This contrasts with findings from social-contact studies, which show that substantial proportions of contact time occur in households, workplaces and schools. A mathematical model of social-contact behaviour and Mtb transmission was developed, incorporating variation in susceptibility and infectiousness. Three types of contact were simulated: household, repeated (individuals outside household contacted repeatedly with daily-monthly frequency) and non-repeated. The model was parameterised using data from Cape Town, South Africa, on mean and variance in contact numbers and contact durations, by contact type, and fitted to an estimate of overdispersion in numbers of secondary cases ('superspreading') in Cape Town. Household, repeated, and non-repeated contacts contributed 36%, 13%, and 51% of contact time, and 13%, 8%, and 79% of disease, respectively. Results suggest contact saturation, exacerbated by long disease durations and superspreading, cause the high proportion of transmission between non-repeated contacts. Household and social-contact tracing is therefore unlikely to reach most tuberculosis cases. A better understanding of transmission locations, and methods to identify superspreaders, are urgently required to improve tuberculosis prevention strategies.

Figure 1

Model fit to data. The solid horizontal lines indicate the best estimates of the output values. The dashed horizontal lines indicate the minimum and maximum of the output plausible ranges. Circles, triangles, and squares indicate the high, medium, and low non-repeated casual transmission risk scenarios respectively. Red indicates scenarios with no additional variation in infectiousness or susceptibility simulated, and purple, turquoise, and green indicate the best estimate of k, and lower and upper bounds of the 95% confidence interval for k respectively. Full details of each output and plausible range justification are given in the supplementary information.

Figure 2

Proportion of contact time, and proportion of disease resulting from transmission between household, repeated, and non-repeated contacts with no additional variation in susceptibility and infectiousness, or with ‘superspreading’, in the high, medium and low casual transmission scenarios. For the ‘superspreading’ scenarios, coloured bars show results for the best estimate of the dispersion parameter k. Error bars indicate results for the upper and lower bounds of the 95% confidence intervals for k. Horizontal dotted lines show the range of the proportion of tuberculosis estimated to result from household transmission in empirical studies in sub-Saharan Africa^–.

Figure 3

The proportion of disease resulting from transmission between household, regular, and non-regular contacts, at different values of the dispersion parameter k. Dots indicate points where model runs were carried out. The solid and dashed vertical lines show the best empirical estimate of k, and the 95% confidence intervals for k, respectively.

Figure 4

Proportion of tuberculosis cases resulting from transmission by the most highly transmitting 20%, 10%, 5%, 2% or 1% of people with pulmonary tuberculosis. Bars on the left show the results from scenarios with no additional variation in infectiousness or susceptibility. Bars on the right show the results from scenarios where the model was fitted to empirical estimates of k, in the high casual transmission risk scenarios. The numbers below the bars show the dispersion parameter, k, from fitting a negative binomial distribution to the number of tuberculosis cases resulting from transmission by each person with pulmonary tuberculosis. Results for the medium and low casual transmission risk scenarios where the model was fitted to estimates of k were very similar to those for the high casual transmission risk scenario (right bars), and were therefore not shown.