Individualizing Tuberculosis (TB) Treatment: Are TB Programs in High Burden Settings Ready for Prime Time Therapeutic Drug Monitoring?

(See the Major Article by Sekaggya-Wiltshire et al on pages 708–16.) One of the most well-recognized principles in the care of infectious diseases is that pharmacokinetic variability can account for failure of therapy on one hand and toxicity on the other. Thus, we routinely measure concentrations of aminoglycosides, vancomycin, and some antifungal drugs, and respond by changing dose in routine patient care. It is a well-known principle that when a similar dose of any drug is given to patients, different concentration-time profiles are expected to be achieved for each patient and thus could result in different efficacy rates. This was originally reported by Paul Ehrlich at the dawn of the age of chemotherapy in 1908, in the seminal “Ueber moderne Chemotherapie,” whereby some outbred mice treated with prototype antibiotic arsanil for trypanosomiasis became sick at a low dose of the drug, whereas some mice tolerated higher doses of the same: mice that did not tolerate the low doses were cured of the infection as if they had been treated by high doses. The causes of this pharmacokinetic variability are many, but the most dominant factor is evolution-driven variability in xenobiotic metabolism, which likely developed to protect the human species (and other animals) from chemical “poisoning” by plant food in what has been termed “animal-plant warfare” [1]. Yet these simple infectious diseases lessons have been lost when it comes to chemotherapy for, arguably, the most important infectious disease, tuberculosis (TB). Dosing has been “one size fits all” ostensibly to make it easier for TB programs, for which measuring drug concentrations in order to respond with optimized doses was considered out of the question. Pharmacokinetic variability is one of the most important drivers of chemotherapy outcomes in TB: cure, acquired drug resistance, relapse, and death [2–11]. This has been consistent, especially in TB patients with relatively low human immunodeficiency virus (HIV) coinfection rates in South Africa and India, where subtherapeutic drug concentrations arising from pharmacokinetic variability accounted for up to 91% of poor outcomes [2–4]. In the case of patients with the dual HIV-TB coinfection, patients from Botswana and the United States reported in the Journal 10–15 years ago had higher rates of poor outcomes with low pyrazinamide, rifabutin, or isoniazid concentrations [5, 6]. However, one of these studies involved only 7 patients with poor outcome in the pharmacokinetic analysis [6], whereas the second only found an association for pyrazinamide and outcome [5]. The study by Sekaggya-Wiltshire et al in the current issues of the Journal reports that intensive pharmacokinetic sampling of anti-TB drugs was performed in 268 HIV-TB Ugandan patients, who were then closely followed up [7]. The study was well designed, prospective, and well powered to interrogate the question of the role of pharmacokinetic variability, and the researchers employed well-executed pharmacometric analyses plus standard well-tested statistical approaches. Of the 227 patients evaluated for clinical outcomes 158 (70%) were cured, 10 (4.4%) failed therapy, whereas 17 (7.5%) withdrew from the study. Low isoniazid and rifampin concentrations were associated delayed time-to-sputum conversion; the adjusted hazards ratio for delayed time-to-sputum conversion was 0.26 (95% confidence interval, 0.14–0.47) if both drugs were below threshold concentrations. Indeed, based on their observations the authors argue that a 1 mg*h/L increase in isoniazid or rifampin increases the likelihood of sputum conversion by 2–4%. These findings show that pharmacokinetic variability drives time-to-sputum conversion, and that drug concentrations can be reliably measured in low- and middle-income countries with high HIV-TB burden. All clinical studies of this size have their own challenges. As an example, the Sekaggya-Wiltshire study does not explain whether the same findings extended to smear negative but culture positive patients. In addition, the rifampin and isoniazid minimum inhibitory concentrations (MIC) were not reported. MICs are also major determinants of outcome and interact with drug concentrations, hence standard pharmacokinetics/pharmacodynamics values are expressed as indices of drug concentration divided by MIC [8, 9]. Another interesting finding is that Sekaggya-Wiltshire et al utilized concentration thresholds identified as being associated with poor outcomes that differ from those identified by others in the past [2, 8, 9]. The most likely reason is that the Ugandan study utilized standard frequentist statistics based on linear analyses, although the effect of biological phenomena such as pharmacokinetic variability and interaction of several drugs in a regimen are inherently nonlinear and occur only within certain limited specific data spaces, and often in a discontinuous fashion [2, 4]. These properties are best characterized using nonlinear science analyses such as artificial intelligence-based machine learning [2, 4, 8, 9]. Nonetheless, the study by Sekaggya-Wiltshire demonstrates that patients taking recommended doses, but with low drug concentrations, are at risk of slower sputum conversion and thus at risk for longer duration of transmission of infection. The study has practical consequences: therapeutic drug monitoring (TDM) to identify those with low drug concentrations and iterative dose increases to reduce therapy failure, acquired drug resistance, and time-to-sputum conversion [2–9]. One of the major excuses, and in our opinion patronizing, offered by naysayers has been that this approach would not be practical and be expensive in high TB burden countries. Clearly, they had no difficulty, technically or intellectually, to measure drug concentrations and interpret them in Kampala, Uganda. Indeed, nearby countries such as Tanzania also employ this approach, and also include MIC identification [10]. In Virginia, in the United States, TDM and dose individualization are used routinely for diabetes patients with drug- susceptible TB in the state TB program, and have shown reduced time-to-sputum conversion by 33% and improved 2-month sputum conversion rates by 60% compared to non-TDM [11]. In any case, appropriate health interventions should not be determined by level of income: if a therapeutic maneuver leads to good clinical outcomes among the wealthy patients in Dallas, Texas, it will likely be as useful even in middle-income patients in Kampala, Uganda. Cost can also be lowered by volume and use of economics of scale, as well as technologies such as dried blood spots that are in use for TB TDM programs in places such as the Netherlands [12]. These can be applied in places such as Uganda, in place of the more expensive research methods used in the current study. Overall the Sekaggya-Wiltshire study demonstrates that pharmacokinetic variability affects outcomes everywhere, and that it is already prime time to respond to this biological phenomenon, by measuring TB drug concentrations and using the results in clinical care. Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.