Generalization Challenges in Drug-Resistant Tuberculosis Detection from Chest X-rays

Classification of drug-resistant tuberculosis (DR-TB) and drug-sensitive tuberculosis (DS-TB) from chest radiographs remains an open problem. Our previous cross validation performance on publicly available chest X-ray (CXR) data combined with image augmentation, the addition of synthetically generated and publicly available images achieved a performance of 85% AUC with a deep convolutional neural network (CNN). However, when we evaluated the CNN model trained to classify DR-TB and DS-TB on unseen data, significant performance degradation was observed (65% AUC). Hence, in this paper, we investigate the generalizability of our models on images from a held out country’s dataset. We explore the extent of the problem and the possible reasons behind the lack of good generalization. A comparison of radiologist-annotated lesion locations in the lung and the trained model’s localization of areas of interest, using GradCAM, did not show much overlap. Using the same network architecture, a multi-country classifier was able to identify the country of origin of the X-ray with high accuracy (86%), suggesting that image acquisition differences and the distribution of non-pathological and non-anatomical aspects of the images are affecting the generalization and localization of the drug resistance classification model as well. When CXR images were severely corrupted, the performance on the validation set was still better than 60% AUC. The model overfitted to the data from countries in the cross validation set but did not generalize to the held out country. Finally, we applied a multi-task based approach that uses prior TB lesions location information to guide the classifier network to focus its attention on improving the generalization performance on the held out set from another country to 68% AUC.

Optimized loading dose strategies for bedaquiline when restarting interrupted drug-resistant tuberculosis treatment

Interruption of treatment is common in drug-resistant tuberculosis patients. Bedaquiline has a long terminal half-life therefore, restarting after an interruption without a loading dose could increase the risk of suboptimal treatment outcome and resistance development. We aimed to identify the most suitable loading dose strategies for bedaquiline restart after an interruption. A model-based simulation study was performed. Pharmacokinetic profiles of bedaquiline and its metabolite M2 (associated with QT-prolongation) were simulated for 5000 virtual patients for different durations and starting points of treatment interruption. Weekly bedaquiline area under the concentration-time curve (AUC) and M2 maximum concentration (Cmax) deviation before interruption and after reloading were assessed to evaluate the efficacy and safety respectively of the reloading strategies. Bedaquiline weekly AUC and M2 Cmax deviation were mainly driven by the duration of interruption and only marginally by the starting point of interruption. For interruptions with a duration shorter than two weeks, no new loading dose is needed. For interruptions with durations between two weeks and one month, one month and one year, and longer than one year, reloading periods of three days, one week, and two weeks, respectively, are recommended. This reloading strategy results in an average bedaquiline AUC deviation of 1.88% to 5.98% compared with -16.4% to -59.8% without reloading for interruptions of two weeks and one year respectively, without increasing M2 Cmax. This study presents easy-to-implement reloading strategies for restarting a patient on bedaquiline treatment after an interruption.

Budgetary impact of using BPaL for treating extensively drug-resistant tuberculosis

IntroductionBedaquiline, pretomanid and linezolid (BPaL) is a new all oral, 6-month regimen comprised of bedaquiline, the new drug pretomanid and linezolid, endorsed by the WHO for use under operational research conditions in patients with extensively drug-resistant tuberculosis (XDR-TB). We quantified per-patient treatment costs and the 5-year budgetary impact of introducing BPaL in Indonesia, Kyrgyzstan and Nigeria.MethodsPer-patient treatment cost of BPaL regimen was compared head-to-head with the conventional XDR-TB treatment regimen for respective countries based on cost estimates primarily assessed using microcosting method and expected frequency of each TB service. The 5-year budget impact of gradual introduction of BPaL against the status quo was assessed using a Markov model that represented patient’s treatment management and outcome pathways.ResultsThe cost per patient completing treatment with BPaL was US$7142 in Indonesia, US$4782 in Kyrgyzstan and US$7152 in Nigeria – 57%, 78% and 68% lower than the conventional regimens in the respective countries. A gradual adoption of the BPaL regimen over 5 years would result in an 5-year average national TB service budget reduction of 17% (US$128 780) in XDR-TB treatment-related expenditure in Indonesia, 15% (US$700 247) in Kyrgyzstan and 32% (US$1 543 047) in Nigeria.ConclusionOur study demonstrates that the BPaL regimen can be highly cost-saving compared with the conventional regimens to treat patients with XDR-TB in high drug-resistant TB burden settings. This supports the rapid adoption of the BPaL regimen to address the significant programmatic and clinical challenges in managing patients with XDR-TB in high DR-TB burden countries.