Global health policies that affect millions of patients should be subject to hypothesis testing in the laboratory. We tested the nonadherence hypothesis in the HFS. The HFS has no immune system and can achieve large bacillary loads, which are 2 important risk factors for emergence of drug resistance [36
]. Drug resistance is encountered within 1–3 weeks when each of the drugs is administered as monotherapy in the HFS [7
]. Thus, it was surprising that there was no emergence or amplification of either rifampin- or isoniazid-resistant subpopulations with nonadherence in multiple experiments. The “pyrazinamide-resistant” subpopulation that was encountered early during therapy merely reflects the fact that critical concentrations of pyrazinamide do not kill >90% of bacilli in liquid culture [37
]. In any case, this subpopulation was transient. Thus, nonadherence was not a sufficient condition for MDR-tuberculosis emergence, or even monoresistance, during the initial phase of therapy.
We propose pharmacokinetic variability as a working hypothesis for the emergence of MDR-tuberculosis. We used Monte Carlo simulations, which have been shown to correctly predict efficacy of anti-tuberculosis drugs [7
], and achieved microbial kill indices and patterns similar to those encountered in the clinic [33
]. The simulations demonstrated that between-patient pharmacokinetic variability will lead to a considerable proportion of patients being on monotherapy throughout the entire initial phase of therapy, leading to drug resistance. However, this hypothesis needs further testing in the clinic. Nevertheless, if proven to be correct, this problem lends itself to a scientific solution of either optimizing doses for local populations by taking into account pharmacokinetic variability or, better still, individualization of each patient’s doses if resources are available.
Our findings also have immediate clinical and public health implications. Although poor adherence did not lead to MDR-tuberculosis, it nevertheless led to therapy failure in the HFS. Failure was encountered at >60% nonadherence with the daily therapy regimen. On the other hand, the 5/7 regimen is equivalent to our start-stop-start-stop regimen, with 29% nonadherence. Thus, if patients took all 40 doses in that regimen, there would be no failure of therapy. Ours is the first evidence to demonstrate that the regimens have equivalent efficacy. However, the 5/7 regimen is less forgiving if nonadherence occurs. For the thrice-weekly regimen, practically any nonadherence leads to therapeutic failure. Thus, it should be questioned whether it is wise to recommend this regimen.
Our experiments also answered another practical question. Based on expert opinion, if a patient misses 2 weeks of cumulative doses during the initial phase of therapy, then standard therapy should be started again as if the patient had not taken any therapy [11
]. Here, we provide the first experimental evidence that standard therapy will work even after nonadherence (). However, the therapy should be restarted from the beginning only if the extents of nonadherence associated with failure were reached.
In summary, therapy failure occurred only at high rates of nonadherence. No MDR-tuberculosis emerged with nonadherence in repeated experiments in the HFS. Instead, we propose pharmacokinetic variability as a more likely cause of MDR-tuberculosis emergence.