The study enrolled 120 index cases with tuberculosis disease and 1,179 household and community contacts to generate social networks. Complete social network information was obtained for 1,154 of 1,179 contacts (98%). The majority of index cases was men (83%), young adults (57%) and had a microbiologically confirmed diagnosis (Table S4). Each case had a median of nine contacts (IQR 8, 12). For male cases, the median number of household contacts was 4 (IQR 2,6) and extra-household contacts was 7 (IQR 4,9), whereas for women, the median number of household contacts was also 4 (IQR 2,6), and extra-household contacts was 5.5 (IQR 4,8).
We initially considered 15 variables from the social network questionnaire for the exploratory factor analysis (Table 1 and Supplementary material, Figure S1). Of these 15 variables, 13 variables had an overall KMO of 0.72, with individual KMO measurements of >0.60. Two variables (“Frequency of meeting since onset of cough” and “Number of other people met in addition to contact”) were excluded from the factor analysis because of a low individual KMO (Table 1).
Of the 13 factors that were identified, only two factors met our a priori criteria for selection with eigenvalues greater than one (Supplementary material, Figure S2). The first factor had a loading of 5.7 and the second factor a loading of 1.9. These two factors captured 82% of the variance (61% and 21% respectively). Variable “Contact has cough” produced low factor loadings in both factors, implying that this variable might not contribute particularly to either of them.
The first factor grouped together variables related to the setting and environment of the contact between the index case and his/her contact; we named this factor the “setting” factor. The six variables in this category had factor loadings of 0.60 or more (Table 2). The second factor grouped together variables that corresponded to the intimacy and social relationship of the index case and contact; we named the factor the “relationship” factor. The six variables in this category had factor loadings of 0.50 or more (Table 2).
Among 940 contacts with a full complement of information, we generated setting and relationship scores. In this subset of contacts, the social network findings corresponded well to the findings obtained from the full set of 1,154 contacts (Table S5), thereby reducing the likelihood of selection bias. Both the setting factor and relationship factor followed a multimodal distribution (p=0.02 and P<0.0001, respectively; unimodality test) (Figure 1). Scores for the setting factor had a median of 10.2 (IQR 7.0, 13.6) with a range of 5.3-18.8; the relationship scores had a median of 7.7 (IQR 6.4, 10.1) with a range of 4.0 to 14.8. Setting and relationship scores varied according to the nature of the relationship among a tuberculosis case and their contact (Figure 2). Spouses had the highest setting score, followed by children and siblings. Altogether, family members had a higher median setting score (13.8, IQR 11.6, 15.7) than non-family members (7.2, IQR 6.2, 9.4). In the case of the relationship factor, spouses, siblings and other relatives had the highest score. The median relationship score in family members (9.9, IQR 7.6, 11.5) was also higher than in non-family members (6.9, IQR 5.6, 8.1).
For both the setting and relationship scores, household contacts (N = 350) had higher scores than extra-household contacts (N = 590, Figure 3, p <.0001). For the setting factor, household contacts had a greater median score (14.6, IQR 12.8,16.2) as compared to extra-household contacts (7.4, IQR 6.4, 9.8); the difference in medians was 7.2 (95% CI: 6.9, 7.6). For the relationship factor, household contacts again had a greater median score (9.8, IQR 7.7, 11.6) as compared to extra-household contacts (7.0, IQR 5.8, 8.6). The difference in medians was 2.8 (95% CI 2.2-3.5). Extra-household family members had a higher median setting score (9.9, IQR 7.3, 11.8) than extra-household non-family members (7.1, IQR 6.2, 9.1). Similarly, extra-household family members had a higher median relationship score (9.4, IQR 7.0,11.2) than extra-household non-family members (6.8, IQR 5.6, 8.1).
When stratifying the analysis by the sex or age of the index case, the median values of the factors scores among contacts revealed differences according to age and sex of the participants (Table 3). Contacts of male cases had a lower setting score as opposed to contacts of female cases (difference in medians = 2.2 (95% CI 1.4 -3.3)). In contrast, contacts of male and female cases had similar relationship scores.
Contacts ≤ 4 years old had the highest setting score, followed by contacts aged 5-14 years and contacts ≥15 years old. The difference in medians among contacts ≤ 15 years of age and contacts ≥ 15 years of age was 4.6 (95% CI 3.7, 5.8). Interactions of contacts ≥15 years old with cases aged 25-44 years were the ones with the lowest setting score. In contrast, interactions of contacts ≤ 15 years with cases 15-24 years proved to be the highest. For the relationship score, an inverse association with the age of contact was found.