Optimising Antibiotic Treatments using Evolutionary Algorithms

Abstract

Antimicrobial resistance is one of the biggest threats to global health, food security, and development. Antibiotic overuse and misuse are the main drivers for the emergence of resistance. Studies in the medical sphere have indicated that shortened antibiotic treatments can be as effective as standard fixed-dose ones and have shown that an initial higher dose followed by a lower maintenance dose are more beneficial to patients with critical illnesses. It is crucial to optimise the use of existing antibiotics in order to improve medical outcomes, decrease toxicity and reduce the emergence of resistance. We formulate the design of antibiotic dosing regimens as a continuous optimisation problem and use several evolutionary algorithms as the search technique. Regimens are represented as vectors of real numbers encoding daily doses, which can vary across the treatment duration. A stochastic mathematical model of bacterial infections with tuneable resistance levels is used to evaluate the effectiveness of evolved regimens. The main objective is to minimise the treatment failure rate, subject to a constraint on the maximum total antibiotic used. We consider simulations with different levels of bacterial resistance; two ways of administering the drug (orally and intravenously); as well as coinfections with two strains of bacteria. The approach produced effective dosing regimens, with an average improvement in lowering the failure rate 30%, when compared with standard fixed-daily-dose regimens with the same total amount of antibiotic. A general pattern of an optimised treatment is found, where if 2x is the standard daily dose then the optimised treatment follows the 3x mg, followed by several 2x mg with a last dose of x mg. A noise handling technique is used to minimise the runtime of the experiments while maintaining the quality of treatments. The results of this work indicate that clinical studies confirming the effectiveness of this approach could be highly beneficial to future of antibiotic treatments.