An assessment of nucleic acid amplification testing for active mycobacterial infection

Sensitivity analyses

The tornado analysis indicates that the ICER is most sensitive to changes in the prevalence of TB in the tested population, the specificity of NAAT (ICERs exceeding $200,000 for the three NAAT specificity estimates) and the changes in the specificity of AFB.

There is considerable uncertainty in the prevalence of TB in the tested population (those with clinical signs and symptoms of TB). A best-guess estimate was applied in the model (22%) provided by the applicant. This estimate was similar to the prevalence of TB reported in the diagnostic accuracy of studies conducted in low-incidence countries (24%). A range of 1030% was applied in the tornado analysis. Increasing the prevalence was observed to increase the cost-effectiveness of NAAT; therefore, conversely, decreasing the prevalence decreased the cost-effectiveness of NAAT, increasing the ICER to $967,000.

As previously described, NAAT may lead to an increase in false-positive results (and so false initiation of treatment) in those with an AFB-negative result based on the specificity of NAAT in AFB-negative patients, and the model is highly sensitive to this change. Any reduction in the specificity of this parameter (from 100%) increases the number of false-positive patients that receive detrimental treatment, leading to increases in cost and poorer quality of life. However, it should be noted that, as culture is an imperfect reference standard, some proportion of NAAT false-positive patients may truly have clinical disease (see ‘Comparison of NAAT and culture, using clinical diagnosis as a reference standard’).

Figure 36 Tornado sensitivity analysis

AE = adverse event; AFB = acid-fast bacilli test; ICER = incremental cost-effectiveness ratio; LTBI = latent tuberculosis infection; MDR = multidrug-resistant; NAAT = nucleic acid amplification test; RIF = rifampicin; TB = tuberculosis

As false-positive rifampicin resistance (in patients without TB and those with susceptible TB) is driven only by the specificity of NAAT for rifampicin resistance; reductions in this parameter increase the number of patients falsely treated with a more costly and more toxic treatment regimen.

The model is sensitive to the 95%CI of the specificity of NAAT in AFB-positive results, but this is more reflective of the imprecision of the pooled estimate (95%CI 1, 100) rather than the model being highly sensitive to changes in this variable alone.

Increases in the specificity of AFB (from 98% to 100%) increase the ICER substantially (to $203,000). As NAAT may correctly identify no TB in patients with an initial AFB (false)-positive result, reducing AFB false-positive results reduces the benefits of NAAT; in contrast, reducing the specificity of AFB (to 94%) increases the number of AFB false-positive results, substantially increasing the benefits of NAAT (dominant ICER).

Other variables that the ICER was seen to be sensitive to include:

• 
  the probability of hospitalisation with susceptible TB—decreasing from 35% to 0% increases the ICER by 50% due to the net reduction in false-positive results associated with the introduction of NAAT
  
• 
  the infectivity of MDR-TB relative to susceptible TB—increasing the infectivity coefficient from 30% to 100% (i.e. assuming the same relative infectivity as susceptible TB) decreases the ICER to $4,000, due to an increase in the number of MDR-TB transmissions, which disproportionally affects the comparator (as AFB cannot identify drug resistance)
  
• 
  the median number of contacts screened per susceptible TB case—increasing the number from 3 to 6 decreased the ICER by one-third, due to the net reduction in false-negative results associated with the introduction of NAAT.
  

Table 64 Sensitivity analysis using test accuracy results of AFB, NAAT from all studies identified in the systematic review