Artificial neural network for pulmonary rifampicin resistant tuberculosis screening Indonesia : a study in accuracy and cost-effectiveness analysis of the model

Abstract

Background: Diagnosis of Pulmonary Rifampicin Resistant Tuberculosis (RR-TB) with the Drug-Susceptibility Test (DST) is costly and time-consuming, and the GeneXpert for rapid diagnosis is not widely available in Indonesia. This study aims to develop an Artificial Neural Network model and evaluate the deployed model performance for RR-TB screening.

Methods: A cross-sectional study involved suspected RR-TB patients with complete sputum Lowenstein Jensen DST (reference) and 19 clinical, laboratory, and radiology parameter results, retrieved from medical records in hospitals under the Faculty of Medicine, Hasanuddin University Indonesia, from January 2015-December 2019. The Artificial Neural Network (ANN) models were built along with other classifiers. The model was tested on participants recruited from January 2020-October 2020 and deployed into CUHAS-ROBUST (index test) application. Sensitivity, specificity, and accuracy were obtained for assessment. A qualitative approach with content analysis was performed from September 2020 to October 2020. Medical staff from the primary care center were invited online for application trials and in-depth video call interviews. Transcripts were derived as a data source. An inductive thematic data saturation technique was conducted. Descriptive data of participants, user experience, and impact on the health service was summarized. Cost-effectiveness analysis of direct cost was made using the data of 330 participants who underwent Genexpert and Model, confirmed by DST. The Quality Adjusted Life years of TB being untreated was used as the approximation of the undiagnosed TB (acute morbidity, chronic morbidity, and mortality). The Incremental Cost-Effectiveness Ratio (ICER) was calculated.

Results: A total of 487 participants (32 Multidrug-Resistant/MDR 57 RR-TB, 398 drug-sensitive) were recruited for model building and 157 participants (23 MDR and 21 RR) in prospective testing. The ANN full model yields 88% (95% CI 85-91) accuracy, 84% (95% CI 76-89) sensitivity, and 90% specificity (95% CI 86-93). This ANN model outperforms other classifiers and selected for the CUHAS-ROBUST application. A total of 33 participants (an average of 33.12 years old) were recruited from all parts of Indonesia. The findings show that DR-TB is a new threat, and its diagnosis faces obstacles particularly prolonged waiting time and inevitable delayed treatment. Despite overcoming the RR-TB screening problems with fast prediction, the dubious screening performance, and the reliability of data collection for input parameters were the main concerns of CUHAS-ROBUST. Nevertheless, this application increases confidence in decision making, promotes medical procedure compliance, active surveillance, and enhancing a low-cost screening approach. A cost-effectiveness analysis was made. The ICER Mortality value is -3601.706137. The ICER Acute Morbidity value is -17225.55 and The ICER Chronic Morbidity value is -825.391. The very minimum sensitivity of the model to not surpass the willingness to pay (WTP) of 100 USD per QALYs gained is 80.6%​. The ideal prevalence of RR-TB according to the screening using model is 14.8% to 23.3%​. Using the average cost, the results still consistent, showing the model as the dominant intervention​

Conclusions: Despite showing lower sensitivity than global GeneXpert results. The ANN-CUHAS ROBUST outperforms other AI classifier models, and by deploying it into the application, the health staff can utilize the tool for screening purposes particularly at the primary care level. Moreover, this study demonstrates AI's roles in enhancing healthcare quality and boost public health efforts against tuberculosis. The advantage of this device is cost-effective although it should need a bigger test expansion.