In a recent study that was published in Diagnostics, artificial intelligence (AI) was utilized to forecast the outcome and detrimental effects of tuberculosis (TB).

Physicians must assess the risk of hepatitis and monitor liver enzymes because the majority of TB medications have the potential to be toxic to the liver. Recently, AI and machine learning (ML) models have been used to diagnose tuberculosis (TB), but less research has been done on their application to forecast unfavorable outcomes.

4,018 cases were found by the authors during the research period. 2,248 patients were chosen for the model building process after exclusions. The majority of the subjects (71.7%) were men, and the average age was 67.7. The serum levels of alanine aminotransferase, aspartate aminotransferase, and total bilirubin were identified by Spearman correlation analysis as relevant features for acute hepatitis, and blood urea nitrogen, age, and white blood cell (WBC) count were identified as relevant features for acute respiratory failure and mortality.

The sensitivity, specificity, and precision of the MLP technique were 0.722, 0.736 and 0.735 respectively and it had the greatest AUROC value of 0.834 for predicting mortality. For predicting acute respiratory failure, random forest achieved the highest score of 0.884, with sensitivity of 0.812, specificity of 0.82, and accuracy of 0.819. For acute hepatitis, XGBoost had the greatest AUROC value (0.92); its sensitivity, specificity, and accuracy were each 0.77, 0.92, and 0.86.

In order to apply AI/ML models, the researchers combined easily accessible clinical and demographic data, for early diagnosis of respiratory failure, hepatitis, and death in TB patients. Notably, the sample only included patients from southern Taiwan, which limited the finding′s representativeness. Also, due to the retrospective nature of the data collection, the alcohol/smoking status was not provided.

(Source: https://www.news-medical.net/news/20230316/Forecasting-adverse-effects-and-prognosis-in-TB-patients-using-AI.aspx)