LAUSR

Understanding pharmacology and drug resistance patterns plus appropriate use of laboratory testing is vital for managing treatment-experienced patients with new agents. While we acknowledge that patients with extensive drug resistance now have multiple options for suppressive therapy, and expert care is essential to avoid the rapid emergence of resistance to these new agents, clinicians are unaware of the inherent (hidden) patterns created by combined drug regimens that could trigger adverse drug reactions. This paper proposes a novel hybrid system framework that combines soft computing techniques, for drugs interaction modelling and precise patient response optimisation. A Fuzzy Logic system was developed to address the uncertainty in treatment change episodes (TCEs). A weighted least-squares cost function was then employed to auto-tune hyperparameters for training the neural network. After acceptable tuning, the final hyperparameters served the neural network—to efficiently learn the ensuing patterns for precice drug interaction classification. The proposed framework was experimented with clinical data of TCEs from two disparate sources: a publicly available HIV database (the Stanford HIV database: https://hivdb.stanford.edu ), and clinical data collected from 13 health centers managing HIV cases in Akwa Ibom State of Nigeria (the Akwa-Ibom HIV database). In both databases, a correlation of prognostic markers suggests strong association between first line CD4 and follow-up CD4 counts; while a moderately weak association was observed for first line and follow-up viral loads. Correlation of physiological feature gave very strong association between first line and follow-up body mass index in Akwa-Ibom database. Analysis of the patients progress explains the decreased potency of CD4 count and body mass index as HIV predictors. The root mean square error (RMSE) and classification accuracy were used as performance metrics for measuring the precision of our hybrid framework. Results obtained showed improved RMSE and classification accuracy for both databases, when compared with existing works.