Performance of a Score to Characterize Adequate Contact among the Social Network of Persons with Tuberculosis

Background. Exposure to an individual with tuberculosis is necessary for transmission to occur. Previously, we developed a score that measures contact between tuberculosis cases and their social networks in an African urban context. This score was built using exploratory factor analysis and identied contact as the conjunction of two domains – setting and relationship. Now, our aim is to determine whether this score covaries with the presence of tuberculous infection among social contacts of tuberculosis cases. Methods. This was a large cross-sectional study conducted in Kampala, Uganda from 2012-2016. Tuberculous infection was assessed in social contacts of adult tuberculosis cases. We estimated the prevalence of tuberculous infection in this population, overall and according to the setting and relationship domains. We calculated the prevalence ratio (PR) for the association between increasing scores in the setting and relationship domains and tuberculous infection, adjusted by other covariates, using modied Poisson regression models.

Household contact studies have provided a useful design for discovering factors from the index case or from the contact that can increase the risk of tuberculosis transmission, as measured by tuberculous infection. The age of the contact, sputum smear grade of the index case, cavitary lung disease, the number of people living in a household are just some of the factors that have been identi ed [3][4][5]. These factors by themselves, however, will not cause a transmission event unless there is adequate contact between the infectious case and a susceptible host. Previous studies have shown that the nature of the contact of an index case with a household contact can increase the likelihood of tuberculosis transmission [6,7]. For example, spouses are at higher risk of tuberculous infection than other household members and daily contact can increase the risk of this infection in the household [8,9].
Outside of the household, less is known about the transmission of M. tuberculosis. From epidemiologic and molecular studies conducted in diverse settings, it appears that over 80% of M. tuberculosis transmission occurs outside of the households of index cases [3,10,11]. From anecdotal reports of community outbreaks of tuberculosis [12][13][14], we know that some congregate settings are associated with transmission and the risk for tuberculosis disease, but the nature of interactions between infectious case and susceptible contact that de ne adequate contact for transmission in these settings are still not well understood. For this reason, there remains uncertainty about how best to implement public health strategies that detect detection and prevent transmission of tuberculosis in community-settings.

STUDY POPULATION
We enrolled tuberculosis index cases aged ≥15 years who were residents of Kampala, Uganda from 2012-2016. Index cases were microbiologically con rmed by a positive sputum smear and had signs and symptoms of pulmonary tuberculosis. We then ascertained their social networks including contacts within and outside of the household. These contacts were traced and enrolled; demographic and clinical information were collected from them using standardized interviews. More details of this study have been previously provided [11,15].

DERIVED RISK SCORE
The study exposure was a derived score between tuberculosis cases and their contacts. The development of this score have been previously described [15]. Brie y, index cases answered questions related to the social mixing between them and each of their social networks. A factor analysis was conducted among these variables, which identi ed two main domains. The setting domain comprised six variables: nature of ventilation at usual place of meeting, frequency of sleeping in same room/same bed, most recent meeting was indoors or outdoors, frequency of shared meals since onset of cough, place of usual meeting (home versus other location) and frequency and duration of contact over the past month. The relationship domain also included six variables: case trusted contact, case shared tuberculosis diagnosis with contact, case was provided care by the contact in the past 3 months, length of knowing contact, how well does the case knows the contact, means of transportation used most often with contactnone/walking versus a type of transportation. Factor analyses results provided weights to each of these variables. We then took the sum of these weights to obtain a setting and a relationship score for each interaction case-contact. We have previously shown that these domain scores reliably measure the extent and nature of the contact between an infectious case and susceptible contact [15].

OUTCOME
The study outcome was tuberculous infection in contacts, either latent tuberculous infection or active disease. Latent tuberculous infection was determined using the tuberculin skin test (TST). A positive TST result was de ned as an induration ≥ 10 millimetres as it has shown to be an adequate cut-off in the Ugandan setting [6]. Active tuberculosis was de ned either as: the presence of at least one smear positive for acid fast bacilli, positive culture for Mycobacterium tuberculosis, positive molecular result for Mycobacterium tuberculosis and history of previous tuberculosis disease, informed by the social contact.

ANALYTICAL STRATEGY
The main objective of this analysis was to estimate and compare the prevalence of tuberculous infection according to the contact scores. We included in the analysis contacts with complete exposure and outcome data.
We explored the probability of tuberculous infection against setting and relationship scores, using a loess (locally weighted scatterplot smoothing) model to obtain a nonparametric smoothed curve [16,17]. Next, we created a matrix of infection prevalence according to setting and relationship score combinations. We rounded scores to the nearest integer for presentation.
We explored the association of the twelve individual variables that comprised the setting and relationship factors with the prevalence of tuberculous infection in the social contacts of tuberculosis cases. The prevalence of tuberculous infection according to each of the responses is shown with 95% con dence intervals (CI).
We categorized the setting and relationship scores as very low, low, moderate, and high contact according to quartiles and estimated the prevalence and 95% CI of tuberculous infection in each category, overall and stratifying by household status (household or extra-household) and smear grade. We used the Cochran-Armitage Trend Test to assess for trends across setting and relationship quartiles.
We used Poisson regression models with a robust variance to estimate the prevalence ratio for the association between increasing scores in setting and relationship domains and tuberculous infection [18][19][20]. This technique allows the analysis of clustered data [21]. Final models included the setting or relationship score and pre-selected potential confounders: age, sex, HIV status, body mass index (BMI) and smear grade of the index case, and the age, sex, BCG, and HIV status of contacts [6,[22][23][24][25][26][27][28][29]. We also included independent social factors that were shown to be associated with tuberculous infection. The nal models are presented strati ed by age of contact: 0-4 years, 5-14 years, and ≥15 years, as the strati ed analysis and regression models suggested age of contact was an effect modi er of the association.
In the Appendix, we showed the results of an overall combined score, considering both the setting and relationship score.

Results
Contacts of 123 index cases were enrolled in the study. Complete data for this analysis were available for 955 of 1006 contacts (95%), including 119 index cases ( Figure S1). The median age of index cases was 28 years (IQR, 23-36) while 82% were men. Almost one in ve index cases were living with HIV (17%).
Contacts were similarly distributed in terms of sex (52% female). The median age was 23 years (IQR, 13-31) ( Table 1). Almost two-thirds of contacts were from outside the household (61%). Sex assortment with cases differed by sex; one-third of female contacts were exposed to a female index case whereas 73% of male contacts were exposed to a male index case.
The overall prevalence of tuberculous infection in social networks was 52% (95% CI, 48-55). Both setting and relationship scores were positively associated with tuberculous infection in contacts ( Figure S2).
When we categorized contact scores into quartiles, we found a rise in the risk of infection with each increasing quartile. In very low, low, medium, and high setting-contact quartiles, the prevalence of tuberculous infection was 44%, 40%, 53% and 70%, respectively (  The relationship between the 12 component variables of the two factors and tuberculous infection prevalence followed a similar pattern as the overall factor scores ( Table 3). In the setting score, indoor meetings with reduced ventilation resulted in higher prevalence of infection for contacts of index cases as compared with outdoor meetings or rooms with ventilation. Similarly, the prevalence of infection was higher among contacts spending more than 66.5 hours/week with the case (74%) and individuals who shared meals daily with the index case (64%). Contacts who slept in the same room and/or same bed as the index case had a higher prevalence of tuberculous infection (62-75%) than contacts who did not sleep in the same room as the index case (48%) ( Table 3). For each variable in the relationship score, there was a monotonic increase in proportion of infection as the extent of exposure increased within each category, except in care of contact ( Table 3). The highest prevalence of tuberculous infection occurred in contacts who discuss and con de with the case (62%), knew the tuberculosis diagnosis of the case (59%), provided daily care to the case (76%), known each other for more than 6 years (61%) and were known very well by their index case (58%).
Since the nature of exposure may differ according to household or extra-household exposure ( Figure S4), we strati ed the analysis by each category. Among household contacts, the majority had setting scores in the medium and high quartiles and the prevalence of tuberculous infection was highest in the highest quartile (70%, P trend <0.0001). Similarly, for the relationship score, the prevalence of tuberculous infection increased with increasing quartile in the contact score, from 41% in the very low quartile to 68% in the high quartile (P trend =0.0052). Among extra-household contacts, the prevalence of tuberculous infection tended to increase across quartiles from very low (44%), low (42%) medium (54%), to high (70%) (P trend =0.0713). For the relationship score, the prevalence of tuberculous infection in extra-household contacts was 41%, 42%, 50 and 60% in the very low, low, middle, and high relationship-contact quartiles, respectively (P trend =0.0048). Among contacts exposed to index cases with a high-smear grade, the prevalence of tuberculous infection increased from 42% in the lowest quartile of the setting score to 69% in the highest quartile (P trend <0.0001) ( Figure S5). Among contacts of index cases with a low smear grade, this pattern was not found.
Instead, the prevalence of tuberculous infection did not differ markedly among contacts in the lowest three quartiles (40%, 46%, 32%, respectively), and was highest among contacts in the highest quartile of the setting score (75%) (P trend =0.0716). For the relationship score, both contacts of index cases with low or high smear grade results showed an increase in the prevalence of tuberculous infection according to the relationship quartiles ( Figure S5).
After adjustment for covariates, the setting and relationship scores continued to be associated with the prevalence of tuberculous infection in contacts (Fig. 2). This association was most pronounced in children. In the contacts 0-4 years old, the prevalence ratio for infection was 1.11 (95% CI, 1.04-1.19) for the setting score, meaning that with each increment of one unit in the score, the risk of infection was 11% higher. Similarly, for the relationship score where the prevalence ratio for infection was 1.42 (95% CI, 1.10-1.82). For contacts 5-14 years old, the adjusted prevalence ratio was 1.26 (95% CI, 1.15-1.39) for setting score and 1.14 (95% CI, 1.06-1.23) for the relationship score. Among adults, both scores were associated with infection, but magnitude of effect of the score was less. When analysed by quartiles, the results were like the ones of the crude prevalence ratio. For the setting score, the adjusted prevalence ratios were 1.00 (95% CI 0.82-1.

Discussion
Ongoing transmission of M. tuberculosis is the central reason for persistence of tuberculosis in the world today. Transmission of any pathogen is di cult to observe and therefore measure but de ning adequate contact for transmission is more feasible. To this end, in a large observational study from an East African city with endemic tuberculosis, we developed a scoring method that used exploratory factor analysis as an agnostic approach to identify ways in which tuberculosis index cases interact with their social network contacts [15]. The factor analysis identi ed two factors that described the setting and relationship between index cases and their contacts.
Here, we found that scores from each factor were consistently associated with tuberculous infection, especially in young children. We found consistent performance of the factors scores across different scenarios, such as household or extra-household exposure, and infectiousness of the index case as measured by sputum microscopy. These integrated analyses provide strong support for the overall validity of the scores in assessing adequate contact for transmission.
The validity of the factor scores is further supported by the component variables that describe the settings of the interactions and the relationships between the cases and contacts. In the setting factor, three component variables have been identi ed as risk factors for tuberculous infection in prior studies [30][31][32]. These variable included ventilation in the usual meeting place, sleeping with the index case, and interaction predominantly indoors or outdoors For instance, it is known that higher ventilation rates decrease quanta concentration in inspired air, which in turn reduces the number of new incident cases [33], and prolonged and repeated exposures as happens during sleep is associated with new infections and disease [8,30,34]. Thus, the inclusion of these variables in our setting domain supports its strong content validity.
Importantly, our setting-speci c score disaggregated contacts with and without M. tuberculosis infection both in household and community contacts. For example, in community contacts, the prevalence of tuberculous infection was greatest (70%) among the quartile with the highest setting score. The prevalence of tuberculous infection was almost 30% lower in the lowest quartile (44%). Most transmission in high-burden settings occurs outside the household [10]. However, e cient tools to identify high-risk individuals in the community are lacking. Potentially, an approach targeting community members with high setting or relationship scores may be a viable approach to both improve e ciency when screening the community and improve cost-effectiveness.
For the relationship factor, we found a strong association between the score and tuberculous infection in children under 15 years of age. This effect was especially pronounced among young children less than ve years because the component variables measure the close relationship with household members, especially their parents who must care for them. The relationship domain likely re ected this intimate relationship and showed its utility in establishing M. tuberculosis transmission. Previous studies have used similar methods [8,35] to measure exposure to tuberculosis and the risk of tuberculous infection and disease among child household contacts [35]. Acuna-Villaorduna and colleagues modi ed this methodology to include adult household contacts in Brazil [8]. Both studies found an association between the score and infection.
Our approach re nes these approaches in important ways. First, we included both household and extrahousehold contacts in developing the factors that underly the score, whereas the earlier studies included only household contacts. Furthermore, we validated the performance of our scores separately for household and extra-household contacts. It has been established for a long time that the household is an environment for likely transmission of M. tuberculosis, especially to resident children. Our approach takes a step further to validate partially the use of the factors scores in the community. Moreover, the results from these prior studies may not be directly comparable because they did not use social science methods to ascertain networks and social networks are distinct from place-to-place [36]. In addition, the background tuberculosis burden in Brazil is much lower than Uganda while South Africa has a higher prevalence of tuberculous infection and disease [1]. Local transmission dynamics should be considered when evaluating such scores and validation of our methods should be performed in other settings.
Our study has limitations. First, in our de nition of tuberculous infection we included contacts with latent and active tuberculosis. As this is a cross-sectional study, we cannot establish the directionality of the transmission for the latter category. Nevertheless, the large majority (87%) of our contacts had latent tuberculosis suggesting it is unlikely that our ndings would differ if we excluded contacts with active tuberculosis. Second, setting and relationship characteristics were reported by the index case, so recall bias is possible. Third, social mixing and behaviour is context-dependent [26]; our ndings might not be applicable in other settings. Nevertheless, we developed the factors through and agnostic and unbiased approach to describe contact between the index case and social network member. We further validated the relationship between the factor scores and infection among contacts. As with any predictive tool, further development and testing should be done in other locations. This score was developed to quantitively measure one of the components that drives tuberculosis transmission, adequate contact.
Based on our analyses, we propose that the setting and relationship domains can contribute to characterize adequate contact among tuberculosis cases and their contacts. All infectious tuberculosis cases have a contact network that comprises both known (household members, friends, workmates) and unknown members with whom the index case may have a single contact. A previous study has suggested that causal contacts might represent more than 60% of the total contacts of the index cases [37] indicating that the household network is only partially de ned through the social network.

Conclusions
In conclusion, relationship and setting-speci c grouped characteristics identi ed contacts with M. tuberculosis infection. These factor scores were especially apparent in children, a group at high-risk to develop tuberculosis once infected. These factors may be useful in prioritizing contact investigations of tuberculosis index cases, not only in the households of cases but in the community as well.

Declarations
Ethics approval and consent to participate.
Written informed consent was obtained from all participants prior to study inclusion. For all participants 18 years or younger we obtained written parental or guardian informed consent. In Uganda, the age of assent is eight years, so for all participants between the ages of 8 and 17 (inclusive), we obtained an assent. Institutional review board clearance was obtained from the Higher Degrees Research and Ethics Committee at Makerere University School of Public Health and approved by the Uganda National Council for Science and Technology in Uganda and the Institutional Review Board at University of Georgia. All methods were carried out in accordance with relevant guidelines and regulations.

Consent for publication
Not applicable

Availability of data and materials
The data that support the ndings of this study are available on request from the senior author, CCW. The data are not publicly available due to containing information that could compromise the privacy of research participants.