Artificial intelligence – the future of antibiotic resistance prevention?

2022-01-26

Artificial intelligence, a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. Could this be an asset in addressing antibiotic resistance? In recent years, artificial intelligence has proven to be a potential tool for managing antibiotic resistance. More specifically, it has been employed as aid for clinicians in antibiotic therapy optimization, for example by monitoring trends in resistance and improving use of antibiotics. Could artificial intelligence be the future of antibiotic resistance prevention?

WHAT is artificial intelligence?

Artificial Intelligence (AI) is an expanding branch of computer science that studies and develops machines capable of learning and predicting certain outcomes – this by using a large amount of data. The hope is to emulate natural intelligence.

According to the Council on Artificial Intelligence of the OECD, an AI system is:

”…a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. AI systems are designed to operate with varying levels of autonomy”.

Infrastructure needed for AI is a challenge in low- and middle-income countries

Given the required infrastructure – such as electricity grid, internet, wireless and mobile networks – AI has been mostly implemented in high-income countries. Only a few pilot studies have recently been initiated in low- and middle-income countries.

For example, in India, Kenya, Malawi, Rwanda, South Africa and Zambia the 2019-2021 Unitaid’s $33 million project with Clinton Health Access Initiative (CHAI) had planned to adopt AI in the field of cervical cancer. The aim was to introduce artificial intelligence-based portable devices for early detection of cancer. Nonetheless, the COVID-19 pandemic disrupted the entire health system, reshuffling priorities, and slowing down the implementation of non-pandemic-related projects, including the Unitaid-CHAI initiative.

HOW has AI been applied in the field of antibiotic resistance?

In recent years, artificial intelligence has proven to be a potential asset for managing antibiotic resistance. For example to aid clinicians in antibiotic therapy optimization, for drug-development, as well as for containing resistant infections. Below are specific examples of how AI has been implemented in the field of antibiotic resistance, as well as some of the challenges connected to each application.

AI to improve diagnostics and treatment

Standard methods to diagnose antibiotic resistance are neither fast nor intuitive. For example, standard antimicrobial susceptibility testing takes more than 24 hours, whereas whole-genome sequencing for antimicrobial susceptibility testing requires the expertise of a bioinformatician and the processing of a large amount of data.

Learn more in this article: Diagnostics: Antibiotic susceptibility  

Several studies have employed AI to shorten the diagnostic time to as little as three hours such as by applying flowcytometer antimicrobial susceptibility testing and supervised machine learning. Similarly, AI could aid improve genome data management in a more efficient and easy-interpretable manner (Wattman et al., 2013).

Another AI data-driven model has been conducive towards establishing optimal antibiotic use strategies in sepsis treatment. More specifically, AI positively identified favorable actions, predicted mortality, and length of stay with high accuracy, hence improved patient outcomes.

However, in order to assess the AI added value, the advancements seen in single studies need to be tested and confirmed more systematically, thus improving reproducibility and scalability.

Prediction of new antibiotic molecules

AI applications have also been widely used for in silico prediction of new antibiotic molecules and synergistic drug combination investigations. Considering that between 2014 and 2019, only 14 new antibiotics were developed and approved, the implementation of AI algorithms could accelerate the discovery and production of new antibiotics.

It remains to be seen if these efforts translate into effective antibiotic medicines, considering the scientific challenges related to in vivo studies (such as safety and efficiency testing, behavioural studies, animal model testing). Additional challenges lie in the limited cooperation between academic institutions and drug developing industries, as well as the need for a broader concept of “open-science” inclusive of sharing algorithms.

AI in response to water crisis

AI has been recently harnessed in response to the water crisis. This in work to provide access to clean water and sanitation for all – relevant for reducing infectious diseases and the spread of antibiotic resistant bacteria. Major fields of application have been:

• management of water resources
• detection of contaminants
• improved effluent quality
• overall data monitoring

Specifically, the adoption of AI in wastewater solutions promises to reduce the number of infections, hence the need of antibiotics, and consequently the development of antibiotic resistance.

Worldwide databases could benefit from AI

Additionally, data deriving from monitoring and surveillance systems could largely benefit from an AI approach, including worldwide antibiotic resistance databases like:

• The Global Antimicrobial Resistance and Use Surveillance System (GLASS)
• The National DB of Antibiotic-Resistant Organisms
• The Antimicrobial DB for High-Throughput Sequencing

WHICH are the challenges connected to the implementation of AI?

Whilst AI has the potential to help advance quality of care and contain antibiotic resistance, it also comes with a wide spectrum of challenges that goes from individual health professionals to the whole health system.

Challenge: Training of the AI with limited data

One major challenge is connected to training of the AI with limited available data. In order for the predictions to be accurate the underlying data needs to be of good quality. For example, an unbiased algorithm should be inclusive of all ethnicities. Also, the implementation of AI entails access to electricity, technology, and capacity building for health professionals.

Implementation of AI in low resource settings would require a major systematic challenge

Given the required infrastructure and need of data, most AI initiatives are currently limited to high income countries (largely US and Europe), leaving a huge gap in low- and middle-income countries. Considering the current frameworks, implementation of AI in low resource settings represents a major systematic challenge. Other country-dependent challenges have been identified, such as transparency in the data acquisition, confidentiality, and liability.

Full understanding of patients’ needs requires clinician expertise

It is behoved to emphasize that understanding the patient’s needs cannot solely rely on an algorithm. It requires the clinician’s expertise and insightfulness, as well as the patient’s trust, which are still traits built on the human relationship between patient and clinician.

AI – potential tool for containing antibiotic resistance, if rightfully applied

In summary, AI promises to be a unique tool for modern medicine and a potential asset for curbing antibiotic resistance, provided that ethics and human rights are embraced in both design and implementation.

WHO guidelines for AI in health care sector

As for the latter, the World Health Organization (WHO) has recently published the first guidelines outlining the key principles adopting AI in the health care sector, including, among others, “Protecting human autonomy” and “Ensuring inclusiveness and equity”.

An ethical, transparent and responsible adoption of AI – coupled with the unbiased expansion of patient-related databases – could improve the identification of antibiotic resistance determinants and contain the spreading of drug-resistant infectious diseases.