Estimation of therapeutic threshold for tuberculosis using adapted nominal group technique and clinical vignettes in clinical and community settings in Southern Africa

Abstract

When confronted with diagnostic uncertainty and a decision on whether to start treatment or not, clinicians consider the potential harm and benefit of offering versus withholding treatment. Treatment can be offered if the probability of tuberculosis (TB) in the patient is above the “therapeutic threshold” (ThT): the probability of disease at which the expected utility of treating and not treating is the same. We estimated ThT for TB in clinical and community settings in Southern Africa using two methods: an adapted nominal group technique (aNGT), and decisions made based on clinical vignettes (CVs). We enrolled health professionals involved in the routine management of TB patients in South Africa and Lesotho. The participants elicited, discussed and refined the harms of false positive (FP) and false negative (FN) treatment decisions for stable ambulatory patients in the clinical and community settings. They weighed all harms according to their importance in treatment decisions by distributing 100 points. ThT, calculated as the sum of the weights of the harms of the FP decision divided by the total weight, was estimated using a hierarchical Beta regression model. For the CVs, participants were presented with ten hypothetical TB cases in each setting and asked to indicate whether they would offer TB treatment or not. ThT was estimated using the generalized linear model for binary outcomes. We enrolled 138 health professionals (aNGT: 123, CVs: 130 and 115 in both). Using aNGT, the overall ThT was 37.7% (95% credible intervals (95% CrI): 35.8–39.8) and 38.2% (95% CrI: 35.9–40.6) in the clinical and community settings, respectively. Compared to aNGT, CVs produced a significantly lower estimate in the clinical setting (27.7%; 95% CrI: 23.8–31.3) but similar in the community setting (37.7%; 95% CrI: 33.1–41.7). We did not find significant differences across the subgroups defined by the measured covariates. The aNGT produced a reliable estimate of ThT. The difference in the estimates of ThT between the aNGT and CVs may have a limited impact on clinical decisions. Factors influencing ThT and the acceptability of results by healthcare workers will be explored in focus group discussions and in-depth interviews.