Abstract:

Objective To understand the prevalence of latent tuberculosis infection (LTBI) among TB contacts who were institutionalized for drugs in a compulsory detoxification centers; further, to analyze the factors associated with risks of TB infection and establish predictive models for LTBI.Methods A total of 13 TB cases were detected during October 2016 to November 2018. Tuberculosis skin test (TST) for TB infection were provided to all the institutionalized persons who had contacted the TB cases at the same period. Follow-up TSTs were given every 6 months to these contacts for two years until July 2019. TST positivity was defined as an induration ≥10 mm for BCG-vaccinated (with a BCG scar), or ≥5 mm for people without a BCG scar, according to the WS 288-2017 for diagnosis of tuberculosis. Prevalence of LTBI were measured. Cox regression and conditional logistic regression were applied to develop risk prediction models with and without site characteristics (drug-using time, frequency of detoxification, and no previous admission to the detoxification centers). Independent variables include age, BMI, BCG vaccination history, tuberculosis history, suspected symptoms of TB, contact degree and contacts with etiology-positive patients. Model performances were evaluated based on the area under the ROC curve (AUC) to find the optimal predictive model.Results This study included 2062 eligible subjects who had accepted TST and had contacted the 13 TB cases at the same period in detoxification centers. The prevalence of LTBI at the first TST examination was 51.4% (1060/2062). During the follow-up, another 267 contacts who were negative at first became TST positive, led to an 26.6% (267/1002) increment. Furthermore, 173 cases (64.8%, 173/267) had a TST positive conversion after those new TB cases were diagnosed in detoxification centers. In total, the 2-year positivity was 64.4% (1327/2062). Using logistic regression, if the site characteristics were adapted, the model showed that TB infection risk=0.041×age+0.373×exposure degree+0.046×BMI+0.028×drug-using time -2.285; if not, the model was that TB infection risk=0.050×age+0.372×contact degree+0.041×BMI-2.282. The AUC of these two models were 0.584 (95%CI: 0.558-0.609) and 0.625 (95%CI: 0.600-0.650), P<0.001. The predictive accuracy of TB infection were 93.6% (1242/1327) and 94.1% (1249/1327), respectively. Using Cox regression, with the site characteristics, the model was presented as that TB infection risk=0.020×age+0.133×contact degree+0.030×BMI+0.013×drug-using time, the AUC of which was 0.633 (95%CI: 0.608-0.658), P<0.001; without site characteristics, the predictive model was that TB infection risk=0.025×age+0.135×contact degree+0.028×BMI, the AUC of which was 0.625 (95%CI: 0.600-0.650), P<0.001.Conclusion Three-factor models have been constructed without site characteristics, including age, BMI, and exposure degree. Cox regression and conditional logistic regression led to similar effects. Considering site characteristics, adding the drug-using time, Cox regression will be better if the follow-up time is available. These models have a high level of accuracy in predicting LTBI risk in institutionalized population, although more studies are needed to identify potential risk factors associated with TB infection.

Key words: Tuberculosis,pulmonary, Contact tracing, Tuberculin test, Latent tuberculosis, Population surveillance, Models,statistical, Forecasting, Drug users