Re-defining Multidrug Resistant Tuberculosis Based on Clinical Response to 1

Combination Therapy 2

Tawanda Gumbo1,2, Jotam G Pasipanodya1, Peter Wash, André Burger3, Helen McIlleron4. 3

4

1Office of Global Health and 2Department of Medicine, UT Southwestern Medical Center, Dallas, 5

Texas, USA, 3Brewelskloof Hospital, Worcester, South Africa, and the 1Division of 6

Pharmacology, Department of Medicine, University of Cape Town, Observatory, South Africa 7

8

Corresponding author: Helen McIlleron, MBChB, PhD 9

Division of Clinical Pharmacology 10

Department of Medicine, University of Cape Town 11

Observatory, Cape Town 7925, South Africa 12

E-mail: helen.mcIlleron@uct.ac.za 13

Phone: +27-21-406-6292 14

Fax: +27-21-448-1989 15

16

Key words: multi-drug resistant tuberculosis, isoniazid, rifampin, critical concentrations, 17

classification and regression tree analysis. 18

Running title: A new definition of MDR-TB 19

20

21

22

23

24

AAC Accepts, published online ahead of print on 4 August 2014Antimicrob. Agents Chemother. doi:10.1128/AAC.03549-14Copyright © 2014, American Society for Microbiology. All Rights Reserved.

ABSTRACT 25

In tuberculosis treatment, susceptibility is defined by a critical concentration of 1.0 mg/L for 26

rifampin and 0.2 mg/L or 1.0 mg/L for low- and high-level isoniazid resistance, based on 27

epidemiologic cut-off method of the distribution of isolates’ MICs. However, 28

pharmacokinetics/pharmacodynamics based clinical trial simulations suggested that the 29

breakpoints should be 0.0625 mg/L for rifampin and 0.0312 mg/L or 0.125 mg/L for isoniazid. 30

We examined outcomes in 36 patients with drug-susceptible tuberculosis, who had rifampin and 31

isoniazid MICs performed and also had plasma drug concentrations measured, who were part of 32

a prospective cohort study in Western Cape, South Africa. We performed classification and 33

regression tree analysis to identify clinical and laboratory factors which predicted 2-month 34

sputum conversion rates and long-term clinical outcomes. Poor long-term clinical outcomes 35

were defined as microbiological failure, relapse or death, with 2 years of follow up. Peak drug 36

concentrations and area under the concentration-time curve were most predictive of outcomes, 37

and constituted the primary node, similar to our findings with the larger cohort. However, 38

rifampin MIC and isoniazid MIC improved the predictive capacity of the primary decision node 39

by 20% and 17%, respectively in these 36 patients. The rifampin MIC cut-off above which there 40

was therapy failure was 0.125 mg/L, while that for isoniazid was 0.0312 mg/L, which are similar 41

to those derived in clinical trial simulations. The critical concentrations used to define multi-42

drug resistance for clinical decision-making should take into account clinical outcomes. 43

44

45

46

47

48

Patients with drug resistant tuberculosis, especially multi-drug resistant- tuberculosis (MDR-49

TB; defined as tuberculosis resistant to both isoniazid and rifampin) and extremely drug-50

resistant tuberculosis (XDR-TB; MDR-TB with additional resistance to fluoroquinolones and 51

injectable compounds) need to be accurately categorized from the outset as they need specific 52

management with second-line regimens (1-5). Hence it is crucial that resistance testing 53

identifies infections with a susceptibility threshold beyond which patients receiving the first-line 54

regimen with standard doses of rifampicin and isoniazid will respond poorly. Resistance to 55

isoniazid and rifampin is said to be present when ≥1% of Mycobacterium tuberculosis culture 56

grows in media supplemented with the critical concentration of each drug. The critical 57

concentrations of rifampin and isoniazid were derived from the MICs of ≥95% of wild-type 58

isolates of M. tuberculosis, termed epidemiologic cut-off values (6-8). While extensively used, as 59

evidenced by the regular release of MDR-TB worldwide figures by the World Health 60

Organization (3), it is nevertheless unclear if the epidemiologic cut-off values are predictive of 61

clinical responses, and if they are the most accurate. Here, we investigated this possibility based 62

on data from a prospective cohort study. 63

64

Another approach that has been used to identify susceptibility breakpoints is pharmacokinetic-65

pharmacodynamic (PK/PD) based derivation that takes into account pharmacokinetic 66

variability encountered for each drug (9). PK/PD studies in pre-clinical models such as the 67

hollow fiber model as well as in clinical studies of tuberculosis patients have identified peak-to-68

MIC and area under the concentration-time curve (AUC)-to-MIC ratios below which there is 69

poor outcome (10-14). In addition, pharmacokinetic variability has been shown to be one of the 70

important determinants of therapy failure in patients, and has delineated optimal AUCs and 71

peak concentrations below which patients fail therapy (15,16). Therefore, we have proposed that 72

the critical concentration should be defined as the MIC above which maximum tolerated doses 73

fail to effectively kill M. tuberculosis at site of infection (17). If the recommended doses in the 74

treatment regimen are unable to effectively kill bacilli in patients because of their MIC, then that 75

MIC defines clinically meaningful drug resistance. Monte Carlo simulations, which utilized 76

PK/PD exposures associated with optimal outcomes and antibiotic pharmacokinetic variability 77

encountered in patients, led to the proposal that rifampin critical concentrations should be 78

lowered from the epidemiologic cut-off derived value of 1.0 mg/L to 0.0625 mg/L, and isoniazid 79

low- and high-level resistance concentrations from 0.2/1.0 mg/L to 0.0312/0.125 mg/L in 80

Middlebrook media (17). Here, we used an agnostic machine learning method to identify the 81

rifampin and isoniazid MICs above which patients fail therapy during treatment with short-82

course chemotherapy in the clinic. 83

84

METHODS 85

86

One hundred and forty two patients with sputum culture positive tuberculosis were enrolled in a 87

pharmacokinetic and pharmacodynamic study at Brewelskloof Hospital in Worcester, Western 88

Cape, South Africa, as described in prior studies (15,18). Patients were recruited between August 89

1999 and February 2002. Thirty six of these 142 patients, randomly chosen, also had isoniazid 90

and rifampin MICs identified in the isolates at the time of diagnosis. Patients were treated using 91

the following daily doses during the intensive phase of therapy: 300 mg of isoniazid, 20-35 92

mg/kg pyrazinamide, 15 mg/kg ethambutol, and 600mg of rifampicin daily if they weighed 93

>50kg or 450 mg if they weighed less. Patients with prior tuberculosis also received intra-94

muscular streptomycin (1 g if they weighed more than 55 kg, 0.75 g if they weighed 38-54kg, or 95

0.5g if they weighed≤37kg). In the continuation phase all patients received the same isoniazid 96

and rifampin doses but for 5 in 7 days and patients with prior tuberculosis were continued on 97

ethambutol. All patients were hospitalized during the first two months of therapy to ensure 98

directly observed therapy (DOTS), for reasons outlined in the primary pharmacokinetic study 99

publication (18). Pharmacokinetic studies were performed over 24 hours during the eighth week 100

of therapy. The compartmental pharmacokinetic analyses have been published elsewhere (15). 101

Sputum microscopy and culture (using liquid cultures in the BACTEC 460 instrument) were 102

performed after eight weeks of treatment. Patients were then followed prospectively for sputum 103

conversion and long-term outcomes over 2 years 104

105

All patients’ isolates had direct sensitivity testing performed using BACTEC. In 36 patients, 106

isoniazid and rifampin MIC assays were performed on the initial isolate at diagnosis. Isoniazid 107

susceptibility was determined in the BACTEC system using the following concentrations: 108

0.0125, 0.025, 0.05, 0.1 and 0.2 mg/L. For rifampin the following concentrations were 109

examined: 0.06, 0.125, 0.25, 0.5, 1.0 and 2.0 mg/L. MICs of pyrazinamide or ethambutol were 110

not identified. 111

112

Classification and regression tree (CART) analysis is a non-parametric method utilizing 113

recursive partitioning to classify data (19-21), which we and others have used for clinical 114

decision making to examine outcomes in several infectious diseases studies (15,22-25). We were 115

interested in classification of the 36 patients into two categories: therapy failure versus success. 116

Two clinical outcome measures were utilized: sputum microbiology status at two months as well 117

as long term outcomes. Poor long-term outcomes were defined as either relapse or 118

microbiologic failure, or death, for up to 2 tears of follow up since the start of therapy (15). 119

Several predictors of outcome were examined, including AUC, peak concentrations, trough 120

concentrations, isoniazid MIC, rifampin MIC, age, gender, weight, HIV status, and treatment 121

with streptomycin. CART analysis examines all these potential predictors of outcome, and 122

examines all possible threshold values of the predictors, to create a tree. The step-by-step 123

description of this analytic method for tuberculosis therapy were described in detail in a prior 124

publication (15). Briefly, the upside tree created starts with a root node, which is the most 125

important potential predictor or decision node. Daughter nodes are added to the tree, and a 126

score of how much they improve the primary decision node (as a % score of the primary node) is 127

computed. For each outcome, maximal trees were generated by split each daughter node so that 128

each class was homogenous in the examined outcome until further splits were not possible. 129

Maximal trees are important in identifying data structure as well clinically meaningful 130

interactions between covariates, particularly among fewer patients (small-sized daughter 131

nodes). We then pruned the maximal trees based on relative misclassification costs, complexity 132

and maximization of parsimony. Next, we performed a 10-fold validation. In this process, the 133

dataset is randomly split 10 times in order to construct optimal trees, to identify how predictive 134

the primary analysis was with these randomly created “test” data sets. This process obviates the 135

need for a validation data set. 136

137

RESULTS 138

139

The full clinical characteristics of the 36 patients in whom isoniazid and rifampin MICs were 140

performed, are shown in Table 1. The characteristics are similar to the entire data set of 142 141

patients (15), confirming that this subset of patients was randomly chosen from the larger data 142

set. The distribution of isoniazid MICs is shown in Figure 1. The mean isoniazid MIC was 143

0.07±0.04 mg/L. The epidemiologic cut-off value was 0.2 mg/L. The distribution of rifampin 144

MICs is shown in Figure 2. The mean rifampin MIC was 0.28±0.13 mg/L. The rifampin 145

epidemiologic cut-off value was 0.5 mg/L. 146

147

As regards to the two month sputum conversion rates, 18/36 (50%) patients had a positive 148

culture or smear at 8 weeks. The factors most predictive of 2-month sputum conversion 149

included peak pyrazinamide, rifampin and isoniazid concentrations, as identified in the larger 150

142 patient study (15). However, in the 36 patients, isoniazid and rifampin MICs were also 151

predictive of 2-month sputum conversion, with MIC thresholds shown in Table 2. While 2-152

month sputum conversion is an important surrogate, the gold standard of antituberculosis 153

therapy efficacy is still long-term outcomes. As regards to the long term outcomes, pyrazinamide 154

AUC was the primary root node, followed by rifampin AUC, and the isoniazid AUC, consistent 155

with findings from the entire dataset of 142 patients. However, the next root node was of 156

rifampin MIC, followed by isoniazid MIC. Rifampin and isoniazid MICs improved the primary 157

decision node by 20% and 17%, respectively. The rifampin and isoniazid cut-off points 158

associated with outcome are shown in table 2. However, in standard assays, MICs are performed 159

based on two-fold dilution steps, so that the nearest observed MIC that fulfills the non-strict 160

inequality values shown in Table 2 was chosen as the breakpoint MIC (Table 2). The results 161

held in the 10-fold validation process. Since predictive power is defined as performance of the 162

training set–derived tree on the test dataset, our results suggest a good predictive role for 163

therapy failure for these rifampin and isoniazid MICs. As can be seen in the table, the isoniazid 164

MIC breakpoints for patients who failed were 0.0312 mg/L while that for rifampin was 0.0625 165

mg/L. 166

167

DISCUSSION 168

169

Isoniazid and rifampin MICs were predictive of clinical outcome, both at the 2-month time point 170

and for long-term outcomes. These data are consistent with classic antimicrobial 171

pharmacokinetic/pharmacodynamics theory, which is that the MIC of a bacterial species is an 172

important determinant of outcome (26-29). Therefore, similar to many other pathogens such as 173

standard gram negative and gram positive bacteria, MICs of anti-tuberculosis drugs affect 174

clinical outcome. The MIC’s effect on microbial kill is considered relative to the drug 175

concentration achieved at site of infection. The drug concentrations achieved in patients have a 176

ceiling at the maximum tolerated dose, and are driven mainly by between-patient 177

pharmacokinetic variability. As the MIC rises in the face of that ceiling concentration, the ratio 178

of AUC/MIC or peak/MIC fall, leading to less and less kill. Eventually, an MIC is reached above 179

which microbial killing is not effective in most patients and which defines clinical resistance. 180

This MIC above which therapeutic failure occurs is not necessarily linked to the 95% 181

epidemiologic cut-off value derived from the MIC distribution. Indeed, there should be no 182

mathematical or physiological reason to expect patient response to be linked to Gaussian 183

parameters of M. tuberculosis isolates’ drug MICs. On the other hand, a shift in the 95% 184

epidemiologic cut-off value at different time periods indicates that the population of isolates 185

from the locale is moving more and more towards drug resistance, making it an excellent index 186

for epidemiologic and public health work tracking of MDR-TB and XDR-TB. Our current study 187

suggests that for clinical decision making during combination therapy, however, susceptibility 188

breakpoint values should be lowered to 0.0312 mg/L for isoniazid, and 0.125 mg/L for rifampin, 189

and do not coincide with the 95% epidemiologic cut-off value. The implication is that MDR-TB 190

rates in the world are likely multiple-fold higher than currently assumed. 191

192

Our results redefine MDR-TB, and by extension XDR-TB, given that these definitions are 193

dependent on critical concentrations of isoniazid and rifampin. In the case of rifampin, four sets 194

of case reports from New Zealand, Australia and the Netherlands, in which a total of 11 clinical 195

isolates from these 3 centers had mutations in the β subunit of RNA polymerase (rpoB) flagged 196

as rifampin-resistant by Gene-Xpert® technology, were found to have rifampin MICs of 0.125-197

1.0 mg/L and failed rifampin containing multiple drug therapy (30-32). The authors termed this 198

“phenotypically occult” resistance. We propose, instead, that the rifampin breakpoint should in 199

fact be lower than current standards, at ≤0.125 mg/L. Moreover, our proposed breakpoint is 200

based on failure of patients to respond to therapy, and is therefore not defined by chromosomal 201

mutations as is the case with Gene-Xpert. That is because not all drug-resistance is due to 202

mutations; some M. tuberculosis isolates have naturally high MICs, while other mechanisms of 203

drug resistance such as efflux pump induction could also lead to drug-resistance (33-36). There 204

have not been any similar case reports for isoniazid as the 11 cases for rifampin since isoniazid 205

does not as of yet have a rapid molecular test to identify resistant mutants. Nevertheless, 206

mechanisms such as efflux pumps are also known to play a role in isoniazid resistance, which 207

means the final definition of isoniazid resistance will ultimately have to be based on phenotypic 208

tests such as MICs (11,13,37,38). Our current findings suggest the critical concentration for 209

isoniazid should also be lowered for clinical decision making. In essence, there is now clinical 210

support to change these critical concentrations that define MDR-TB. These new concentrations 211

should be considered for clinical decision making by the several bodies important in the clinical 212

care of tuberculosis around the world such as the WHO and STOP TB, as well as by groups 213

designing assays for diagnosis of MDR-TB. 214

215

The approach using hollow fiber PK/PD and pharmacokinetic variability in Monte Carlo 216

simulations to identify provisional and definitive susceptibility breakpoints of antituberculosis 217

agents (17) identified virtually the same breakpoints as those identified in our current clinical 218

study. Similarly, the breakpoint of pyrazinamide was correctly identified using this method, and 219

was also recently identified using CART to be identical (25). This is noteworthy because CART 220

analyses did not pre-specify the susceptibility breakpoints that should be selected but were 221

“agnostic” in both identifying MIC as a predictor of clinical outcome from among several clinical 222

factors, but also calculated threshold MICs that classifies patients into those who fail or succeed 223

therapy without relying on a prior choice. Therefore, the pharmacometric pathway consisting of 224

(a) pre-clinical PK/PD studies of monotherapy to identify optimal drug exposures in the hollow 225

fiber or other pre-clinical models, followed by (b) use of population pharmacokinetics in Monte 226

Carlo experiments, and (c) the choice of a susceptibility breakpoint based on >10% of patients 227

failing to achieve optimal exposures, is validated for setting susceptibility breakpoints of anti-TB 228

drugs. We propose its use for the new and experimental antituberculosis therapies currently 229

being studied. 230

Our study has several limitations. First is the size of the clinical study, which could limit 231

generalizability of the findings. However, CART has been able to correctly identify thresholds 232

with similarly small populations in the past (24,25,39). Second, several other clinical factors also 233

determine clinical outcome, including drug concentrations, cavitary disease, and bacterial 234

burden. However, these factors need not exclude a role for MICs in outcome. Indeed, our CART 235

analysis also examined some of these as possible predictors, and with the exception of drug 236

concentrations, they were outranked by MICs. Third, one potential limitation of CART is over 237

fitting and biasing toward covariates with many possible splits. Thus our findings should be 238

taken with these three factors in context. Nevertheless, cross-validation identified the same MIC 239

threshold values, which were virtually identical to Monte Carlo simulations results published 5 240

years earlier (17). This means that the same breakpoints have now been identified using two 241

completely independent methods. In the case of rifampin, retrospective case reports lead to the 242

same conclusion, adding a third independent method. Thus, critical concentrations of 0.125 243

mg/L for rifampicin, and 0.0312 mg/L for isoniazid should be used to define MDR-TB. Such 244

PK/PD evaluation should form the basis for accurate susceptibility breakpoints. The continued 245

use of breakpoints which disregard the drug exposures in patients administered recommended 246

doses, will lead to incorrect categorization of patients and treatment with inappropriate 247

regimens (40). 248

249

250

251

252

253

254

FUNDING 255

This work was supported by the National Institutes of Health via the NIH Director New 256

Innovator Award (National Institutes of General Medical Sciences DP2 OD001886 for JGP and 257

TG, and the National Institute of Allergy and Infectious Diseases R01AI079497 for JGP and TG). 258

Funding for recruitment of patients and for Helen McIlleron was provided by the Division of 259

Pharmacology of the University of Cape Town, and the Medical Research Council of South 260

Africa. 261

262

CONFLICTS OF INTEREST 263

TG founded Jacaranda Biomed. JGP and HM have no conflicts of interest. 264

265

ACKNOWLEDGEMENTS. 266

We would like to acknowledge the patients who participated in the study, the nursing staff of 267

Brewelskloof Hospital, Western Cape, for taking care of patients, and Dr Andrew Whitelaw for 268

MIC determination. We would also like to acknowledge the University of Cape Town’s 269

Department of Medicine and UT Southwestern Medical Center’s Office of Global Health for 270

making this collaboration possible. 271

272

273

274

275

REFERENCES 276

1. Migliori, G. B., G. Sotgiu, N. R. Gandhi, D. Falzon, K. DeRiemer, R. Centis, 277

M. G. Hollm-Delgado, D. Palmero, C. Perez-Guzman, M. H. Vargas, L. 278

D'Ambrosio, A. Spanevello, M. Bauer, E. D. Chan, H. S. Schaaf, S. Keshavjee, 279

T. H. Holtz, and D. Menzies. 2012. Drug resistance beyond XDR-TB: results from a 280

large individual patient data meta-analysis. Eur.Respir.J. 42:169-79. 281

2. 2006. Emergence of Mycobacterium tuberculosis with extensive resistance to second-282

line drugs--worldwide, 2000-2004. MMWR Morb.Mortal.Wkly.Rep. 55:301-305. 283

doi:mm5511a2 [pii]. 284

3. World Health Organization. Multidrug and extensively drug-resistant TB (M/XDR-TB): 285

2010 global report on surveillance and response. WHO/HTM/TB/2010.3. 2010. Geneva, 286

World Health Organization. 287

288

4. Gandhi, N. R., A. Moll, A. W. Sturm, R. Pawinski, T. Govender, U. Lalloo, K. 289

Zeller, J. Andrews, and G. Friedland. 2006. Extensively drug-resistant tuberculosis 290

as a cause of death in patients co-infected with tuberculosis and HIV in a rural area of 291

South Africa. Lancet 368:1575-1580. 292

5. Dheda, K., T. Gumbo, N. R. Gandhi, M. Murray, G. Theron, Z. Udwadia, G. B. 293

Migliori, and R. Warren. 2014. Global control of tuberculosis: from extensively-drug 294

resistant to untreatable tuberculosis. Lancet Respir Med 2:321-338. 295

6. Canetti, G., S. Froman, J. Grosset, P. Hauduroy, M. Langerova, H. T. Mahler, 296

G. Meissner, D. A. Mitchison, and L. Sula. 1963. Mycobacteria: Laboratory 297

methods for testing drug sensitivity and resistance. Bull.World Health Organ 29:565-298

578. 299

7. Ambrose, P. G. 2006. Monte Carlo simulation in the evaluation of susceptibility 300

breakpoints: predicting the future: insights from the society of infectious diseases 301

pharmacists. Pharmacotherapy 26:129-134. 302

8. Clinical and Laboratory Standards Institute. Susceptibility testing of Mycobacteria, 303

Nocardiae, and other aerobic Actinomycetes; Approved Standard. 2003. Wayne, PA, 304

Clinical and Laboratory Standards Institute. 305

306

9. Drusano, G. L., S. L. Preston, C. Hardalo, R. Hare, C. Banfield, D. Andes, O. 307

Vesga, and W. A. Craig. 2001. Use of preclinical data for selection of a phase II/III 308

dose for evernimicin and identification of a preclinical MIC breakpoint. 309

Antimicrob.Agents Chemother. 45:13-22. 310

10. Gumbo, T., A. Louie, M. R. Deziel, W. Liu, L. M. Parsons, M. Salfinger, and 311

G. L. Drusano. 2007. Concentration-dependent Mycobacterium tuberculosis killing 312

and prevention of resistance by rifampin. Antimicrob.Agents Chemother. 51:3781-3788. 313

11. Gumbo, T., A. Louie, W. Liu, D. Brown, P. G. Ambrose, S. M. Bhavnani, and 314

G. L. Drusano. 2007. Isoniazid bactericidal activity and resistance emergence: 315

integrating pharmacodynamics and pharmacogenomics to predict efficacy in different 316

ethnic populations. Antimicrob.Agents Chemother. 51:2329-2336. 317

12. Gumbo, T., C. S. Siyambalapitiyage Dona, C. Meek, and R. Leff. 2009. 318

Pharmacokinetics-Pharmacodynamics of pyrazinamide in a novel in vitro model of 319

tuberculosis for sterilizing effect: A paradigm for faster assessment of new 320

antituberculosis drugs. Antimicrob.Agents Chemother. 53:3197-3204. 321

13. Srivastava, S., S. Musuka, C. Sherman, C. Meek, R. Leff, and T. Gumbo. 2010. 322

Efflux-pump-derived multiple drug resistance to ethambutol monotherapy in 323

Mycobacterium tuberculosis and the pharmacokinetics and pharmacodynamics of 324

ethambutol. J Infect Dis 201:1225-1231. 325

14. Chigutsa, E., J. G. Pasipanodya, F. A. Sirgel, T. Gumbo, and H. McIlleron. 326

Multivariate adaptive regression splines analysis of the effect of drug concentration and 327

MIC on sterilizing activity in patients on multidrug therapy. In: Abstracts of the Sixth 328

International Workshop on Clinical Pharmacology of Tuberculosis Drugs, Denver, CO, 329

2013. Abstract 14, p. 15. 2013. 330

15. Pasipanodya, J. G., H. McIlleron, A. Burger, P. A. Wash, P. Smith, and T. 331

Gumbo. 2013. Serum Drug Concentrations Predictive of Pulmonary Tuberculosis 332

Outcomes. J.Infect.Dis. 208:1464-1473. 333

16. Pasipanodya, J. G., S. Srivastava, and T. Gumbo. 2012. Meta-analysis of clinical 334

studies supports the pharmacokinetic variability hypothesis for acquired drug resistance 335

and failure of antituberculosis therapy. Clin Infect.Dis 55:169-177. 336

17. Gumbo, T. 2010. New susceptibility breakpoints for first-line antituberculosis drugs 337

based on antimicrobial pharmacokinetic/pharmacodynamic science and population 338

pharmacokinetic variability. Antimicrob.Agents Chemother. 54:1484-1491. 339

18. McIlleron, H., P. Wash, A. Burger, J. Norman, P. I. Folb, and P. Smith. 2006. 340

Determinants of rifampin, isoniazid, pyrazinamide, and ethambutol pharmacokinetics in 341

a cohort of tuberculosis patients. Antimicrob.Agents Chemother. 50:1170-1177. 342

19. Breiman, L., J. Friedman, C. J. Stone, and R. A. Olshen. 1984. Classification and 343

Regression Trees. Chapman and Hall/CRC, Boca Raton. 344

20. Kim, H. and W. Y. Loh. 2001. Classification trees with unbiased multiway splits. J 345

Am.Stat.Assoc. 88:457-467. 346

21. Steinberg, D. and P. Colla. 1995. CART: Tree-Structured Non-parametric Data 347

Analysis. Salford Systems, San Diego. 348

22. Lewis, R. J. An introduction to classification and regression tree (CART) analysis. 349

2000 Annual Meeting of the Society for Academic Emergency Medicine . 2000. Des 350

Plaines, IL, Society for Academic Emergency Medicine. 2000. 351

352

23. Andes, D., P. G. Ambrose, J. P. Hammel, S. A. Van Wart, V. Iyer, D. K. 353

Reynolds, D. N. Buell, L. L. Kovanda, and S. M. Bhavnani. 2011. Use of 354

pharmacokinetic-pharmacodynamic analyses to optimize therapy with the systemic 355

antifungal micafungin for invasive candidiasis or candidemia. Antimicrob.Agents 356

Chemother. 55:2113-2121. 357

24. Jain, M. K., J. G. Pasipanodya, L. Alder, W. M. Lee, and T. Gumbo. 2013. 358

Pegylated interferon fractal pharmacokinetics: individualized dosing for hepatitis C virus 359

infection. Antimicrob.Agents Chemother. 57:1115-1120. 360

25. Gumbo, T., E. Chigutsa, J. G. Pasipanodya, M. Visser, H. McIlleron, and F. A. 361

Sirgel. 2014. The pyrazinamide susceptibility breakpoint above which combination 362

therapy fails. J Antimicrob.Chemother; doi: 10.1093/jac/dkt524. 363

26. Drusano, G. L., D. E. Johnson, M. Rosen, and H. C. Standiford. 1993. 364

Pharmacodynamics of a fluoroquinolone antimicrobial agent in a neutropenic rat model 365

of Pseudomonas sepsis. Antimicrob.Agents Chemother. 37:483-490. 366

27. Craig, W. A. 1998. Pharmacokinetic/pharmacodynamic parameters: rationale for 367

antibacterial dosing of mice and men. Clin.Infect Dis 26:1-10. 368

28. Craig, W. A. 2007. Pharmacodynamics of antimicrobials: General concepts and 369

applications, p. 1-19. In: C. H. Nightangle, P. G. Ambrose, Drusano G.L, and T. 370

Murakawa (eds.), Antimicrobial pharmacodynamics in theory and practice. 2 ed., vol. 44. 371

Informa Healthcare USA, Inc., New York. 372

29. Ambrose, P. G., S. M. Bhavnani, C. M. Rubino, A. Louie, T. Gumbo, A. 373

Forrest, and G. L. Drusano. 2007. Pharmacokinetics-pharmacodynamics of 374

antimicrobial therapy: it's not just for mice anymore. Clin.Infect.Dis. 44:79-86. 375

30. Williamson, D. A., S. A. Roberts, J. E. Bower, R. Vaughan, S. Newton, O. 376

Lowe, C. A. Lewis, and J. T. Freeman. 2012. Clinical failures associated with rpoB 377

mutations in phenotypically occult multidrug-resistant Mycobacterium tuberculosis. 378

Int.J.Tuberc.Lung Dis. 16:216-20 379

31. van Ingen, J., R. Aarnoutse, V. G. de, M. J. Boeree, and S. D. van. 2011. Low-380

level rifampicin-resistant Mycobacterium tuberculosis strains raise a new therapeutic 381

challenge. Int.J.Tuberc.Lung Dis. 15:990-992. 382

32. Ho, J., P. Jelfs, and V. Sintchencko. 2013. Phenotypically occult multidrug-383

resistant Mycobacterium tuberculosis: dilemmas in diagnosis and treatment. J 384

Antimicrob.Chemother. 68:2915-20 385

33. Pasipanodya, J. G. and T. Gumbo. 2011. A new evolutionary and pharmacokinetic-386

pharmacodynamic scenario for rapid emergence of resistance to single and multiple anti-387

tuberculosis drugs. Curr.Opin.Pharmacol. 11:457-463. 388

34. Schmalstieg, A. M., S. Srivastava, S. Belkaya, D. Deshpande, C. Meek, R. 389

Leff, N. S. van Oers, and T. Gumbo. 2012. The antibiotic resistance arrow of time: 390

efflux pump induction is a general first step in the evolution of mycobacterial drug 391

resistance. Antimicrob.Agents Chemother. 56:4806-4815. 392

35. Zimic, M., P. Fuentes, R. H. Gilman, A. H. Gutierrez, D. Kirwan, and P. 393

Sheen. 2012. Pyrazinoic acid efflux rate in Mycobacterium tuberculosis is a better proxy 394

of pyrazinamide resistance. Tuberculosis (Edinb.) 92:84-91. 395

36. Louw, G. E., R. M. Warren, N. C. Gey van Pittius, R. Leon, A. Jimenez, R. H. 396

Pando, C. R. McEvoy, M. Grobbelaar, M. Murray, P. D. van Helden, and T. C. 397

Victor. 2011. Rifampicin Reduces Susceptibility to Ofloxacin in Rifampicin Resistant 398

Mycobacterium tuberculosis through Efflux. Am.J.Respir.Crit Care Med. 184:269-76. 399

37. Gumbo, T., A. Louie, W. Liu, P. G. Ambrose, S. M. Bhavnani, D. Brown, and 400

G. L. Drusano. 2007. Isoniazid's bactericidal activity ceases because of the emergence 401

of resistance, not depletion of Mycobacterium tuberculosis in the log phase of growth. 402

J.Infect.Dis. 195:194-201. 403

38. Machado, D., I. Couto, J. Perdigao, L. Rodrigues, I. Portugal, P. Baptista, B. 404

Veigas, L. Amaral, and M. Viveiros. 2012. Contribution of Efflux to the Emergence 405

of Isoniazid and Multidrug Resistance in Mycobacterium tuberculosis. PLoS.One. 406

7:e34538. 407

39. Gumbo, T., J. Hiemenz, L. Ma, J. J. Keirns, D. N. Buell, and G. L. Drusano. 408

2008. Population pharmacokinetics of micafungin in adult patients. 409

Diagn.Microbiol.Infect.Dis. 60:329-331. 410

40. Pasipanodya, J. G., S. Srivastava, and T. Gumbo. 2012. Scientific and patient care 411

evidence to change susceptibility breakpoints for first-line anti-tuberculosis drugs. Int J 412

Tuberc Lung Dis 16:706-707. 413

414

415

416

417

418

419

Figure 1. Isoniazid MIC distribution in isolates from 36 patients. 420

Isoniazid MIC (mg/L)

Per

cent

of c

linic

al is

olat

es

0.0125 0.025 0.05 0.1 0.20

20

40

60

421

The Gaussian distribution has a right skew, and a slightly higher epidemiologic cut-off value 422

than in the literature (6). 423

424

425

426

427

428

429

430

431

432

433

434

435

436

Figure 2. Rifampin MIC distribution in isolates from 36 patients. 437

438

Rifampin MIC (mg/L)

Per

cent

of c

linic

al is

olat

es

0.03125 0.0625 0.125 0.25 0.5 1 20

10

20

30

40

50

60

439

Gaussian distribution of rifampin MICs in patients, means that there is no one “MIC” to a drug 440

in clinical isolates. Thus the standard notion of stating that rifampin clinical isolates have an 441

MIC of say 0.125 mg/L is meaningless. 442

443

444

445

446

447

448

449

450

451

452

Table 1. Clinical and demographic factors in 35 patients treated for tuberculosis in 453

Cape Town, South Africa. 454

Characteristic Estimate or

median

Range or

percentage

Sex-female (%) 23 65.71%

Age in years 35 20-71

Self-identified “race” or ethnic group

Mixed Race South African 31 88.57%

Black South African 4 11.43%

Weight (kg)

Weight 47.3 30-68

HIV infection (%) 3 8.57%

Dose and range in mg/kg

Rifampin 10.71 8.38-15.00

Isoniazid 6.39 4.41-10.00

Pyrazinamide 35.14 19.69-50.00

Ethambutol 23.87 12.88-30.77

Prior tuberculosis (%) 23 65.71%

Hemoglobin g/dL (range) 12.00 7.40-14.00

White cell count x109 cells/mL 8.60 3.80-25.70

Platelet count x109 cells/mL 410.0 66.0- 813.0

ESR in mm/hr 58 12-136

Total protein in g/L 77 62-92

Albumin in g/L 34.00 21-43

455

Table 2. Classification and regression tree analysis derived MIC breakpoints in 36 456 patients. 457

Two month sputum conversion Cut-off point 2 log scale

Long term outcomes Cut-off point 2 log scale

Isoniazid MIC (mg/L) ≤0.150 0.125 ≤0.038 0.0312 Rifampin MIC (mg/L) ≤0.188 0.125 ≤0.38 0.25 458

459

460

461

462

463