Artificial intelligence in Antimicrobial stewardship

Home
Blog
Artificial intelligence in Antimicrobial stewardship

Artificial intelligence in Antimicrobial stewardship

Artificial Intelligence (AI) rapidly transforms the healthcare industry and pharmacy practice. This is Artificial intelligence in Antimicrobial stewardship.

However, many clinicians lack a foundational understanding of AI technology and may have difficulty applying it in their practice setting.

Machine Learning is a way to train a model to make accurate predictions.

This is a simple example, yet it highlights the countless possibilities used in machine learning.

The basic steps to developing a machine learning algorithm are to:

Formulate a question
Collect data
Select, train, and evaluate the model
Make predictions

What is NLP

Natural Language Processing (NLP) is a way to decipher complicated human language and produce a simple, understandable output.
For instance, there are many ways tan brilliant smart speakers about the weather.
You could utilise voice command to activate the intelligent speaker and ask, “What is the high for today? Will it rain?” and so forth.
NLP analyses the language, generates a human response, and provides an outcome such as “The high today is 65 degrees Fahrenheit and bring your umbrella because there is a 100% chance of rain.”
Computer vision uses pattern recognition and deep learning to recognise what’s in a picture or video.
Have you ever wondered how social media pages like Facebook could tell you who precisely tagged your picture?
This is because computer vision technology can process and then “recognise” the facial features of individuals in the photo.

Applying artificial intelligence to healthcare delivery

Now that we have covered the basics of AI, how can these advanced technologies be used in healthcare delivery? Several goals include:

Increase efficiency
Digital task shifting and managing staff shortages
Population health applications to enable targeted and differentiated services
Earlier detection of diseases
Improve the quality of clinical decision making
Continuous patient monitoring

AI applications used in pharmacy practice can be broken down into prevention, diagnosis, and treatment. Examples include:

Nguyen et al. utilised machine learning to predict minimum inhibitory concentrations (MICs) of Salmonella strains and their susceptibility to 15 antibiotics.
Doctors Without Borders developed a microbiology tool for resource-limited settings that uses computer vision to read “zones of inhibition” to advance patient care.

Key factors to consider when evaluating an AI solution for healthcare:

Feature engineering – data elements that go into the model
Clinical expertise – did the data science team collaborate with clinicians in building the model?
Validation – How well does the model predict the outcome of interest using validation data?
Application and clinical benefits – how is it being used? A decent model that makes accurate predictions yet has no practical clinical benefits has no place in the real world.
Interpretation – Is the application easy to interpret and understand?
User education and training – clinicians should be educated on the model output and how it may impact patient care decisions.
Workflow integration – how and when will the clinician receive the information to impact clinical decision-making?

What are antibiotics and antibiotic stewardship?

Antibiotics are one of the most significant advances in modern medicine, increasing g average human lifespan by over 20 years.
Yet, increasing bacterial resistance is a natural and inevitable consequence of antibiotic use, requiring essential strategies to sustain these gains in health.
The driving concept of antibiotic stewardship is using the right antibiotic at the correct dose and time.
This collides with data suggesting that up to half of all antibiotic prescriptions are unnecessary or inappropriate.

Why is the appropriate use of antibiotics so hard?

The choice of antibiotics when first encountering a patient, known as “empiric” antibiotic prescribing, is essentially educated guesswork.
Antibiotic appropriateness is determined by the infectious pathogen and its susceptibility to various antibiotics.
Unfortunately, the necessary critical tests providing this data, microbial cultures, often take days to result.
This is far too long to delay empiric treatment that could minimise the chance of patient death from sepsis.
If doctors don’t know what we’re treating, but we know we need something that works immediately or the patient might die, we often give an antibiotic that treats everything possible.
This mentality leads to physicians picking a “shotgun” broad-spectrum antibiotic when a narrow spectrum precision “scalpel” would have been more effective.
Moreover, many non-infectious diseases can look similar to an infection, making decisions even more complicated as to whether any antibiotic is needed.
National guidelines on specific syndromes (e.g. pneumonia, skin/soft tissue infection) provide general suggestions on when to treat disease and what antibiotics to use.
Still, they can take years to develop and can only offer overly general one-size-fits-all guidance that providers are often not even aware of.
Local hospital antibiograms document resistance patterns within a specific healthcare facility but are still unclear to particular patients.
We need better tools at the point of care to guide precision antibiotic prescribing, personalised and rapidly updated to new data streams.

Artificial intelligence (AI) and machine learning for antibiotic stewardship?

Existing tools and standards to support antibiotic prescribing provide essential guidance but are too broad for personalised recommendations that must balance immediate risks of undertreatment against the nebulous risks of overtreatment.
Artificial intelligence (AI) is often depicted as science fiction of a distant future, yet modern algorithms are already shaping our lives daily.
Do internet advertisements seem uncannily specific to your interests? This is the power of predictive analytics: Using machine learning predictive models on large-scale data to generate individualised predictions and suggestions.
Imagine if we were to use this same power, except in choosing the right antibiotic, individualised towards a single patient.
With similar technology, we can use the vast amounts of data from electronic medical records to create predictive models to optimise the accuracy and consistency of the currently educated guesswork of empiric antibiotic prescribing.
Electronic patient charts provide ample information that models can utilise, from the history of past infections and antibiotic susceptibility data to a patient presenting symptoms, medical history, laboratory results, and imaging.
We currently use this data in large extensive scientific studies to better determine which variables independently predict the need for specific types of antibiotics in general.
Does this particular patient have an infection or not?
If so, is it safer to use the shotgun or the scalpel?

The AI models:

If we give our AI the correct variables and training.
then, models can estimate probabilities of infection with drug-resistant bacteria for specific patients quantitatively in a manner that humans can only intuit qualitatively.
Just as we expect from competent clinicians, AI models should not be stagnant.
Continuously learning algorithms should adapt to incoming new patients and epidemiologic data streams.
With additional training, predictions can become better and more relevant, allowing for a more dynamic guidance tool than static guidelines or yearly antibiograms.
There have already been breakthroughs in this area
This indicates the potential to improve both safety and stewardship simultaneously.

More similar points:

Similar work is emerging as locally trained and optimised tools in multiple sites.
Impact on population scales will require advances in exchanging data across diverse health systems, the essential fuel to power modern AI systems.
While substantial inertia tends to lock such data into local health system silos.
the shock of the COVID pandemic finally motivated many to pool their data, resources, and effort to face a more significant threat.
Laws on improving interoperability between health systems and international standards like Fast Healthcare Interoperability Resources (FHIR) are advancing efforts in this area.
Hopefully, we will collectively recognise the even more significant but slowly boiling problem of antimicrobial resistance to taking action.
Within the next decade, we envision providers routinely using AI models as ls.
which provide practical recommendations for treating immediate infections and combating the otherwise relentless development of antimicrobial resistance.

Antibiotic Resistance in Pediatrics and how AI can help

Antimicrobial resistance (AMR) is defined as the ability of a microorganism (bacterium, virus, or fungi) to prevent an antimicrobial agent from acting against it.
AMR is considered a global public health emergency for epidemiological and economic reasons to the extent that the World Health Organization has published an action plan.
About paediatrics, antibiotics are among the most widely prescribed drugs for children in hospitals and the community.
However, many factors can affect the use of these drugs.
The result is that over the last 15 years, there have been significant deficiencies in the development and availability of new antibiotics to combat emerging resistance cases.
Implementation of containment strategies to address this rapidly growing problem, an effort called antimicrobial stewardship (AS), is therefore essential.
These strategies have had a positive impact on adult patients, but only recently have they been used in the pediatric field, where targeted interventions are needed considering the heterogeneity in the age and weight of the patients.
As a consequence, the next section will report and discuss possible applications of artificial intelligence against AMR.

Application Strategies for Artificial Intelligence (AI) Against Antibiotic Resistance – Prediction, Assessment and Diagnosis of Pediatric Infectious Diseases

an essential aspect of fighting antibiotic resistance is the early recognition of the infectious pathology.
the distinction between pathologies on a transferable or non-infectious basis and the proper management of complications.
Children have higher infection rates than adults and often exhibit non-specific symptoms, which increases diagnostic uncertainty. AI is, in this sense, a potentially powerful weapon.
In 2017, Komorowski et al. presented a tool based on reinforcement learning.
Its use has resulted in lower mortality in patients for whom thewhose doctors’ actual decisions matched those of AI.
showing the clinically reliable ability of this tool to customise sepsis treatment and assist doctors in making real-time decisions.
A trial conducted in a German pediatric tertiary intensive care unit aimed to distinguish and diagnose.
infectious sepsis from non-infectious forms of SIRS at an early stage based on the concept that similar symptoms characterise the two entities.
To this end, a diagnostic model based on ML, specifically on a random forest approach, was developed, taking into account 44 variables available at the time of patient admission (baseline characteristics, clinical/laboratory parameters, and technical/medical support)

Diagnostic models:

The model allowed for early recognition of all sepsis cases, and a potential reduction of 30% in the use of antibiotics in patients with non-infectious SIRS was calculated.
A further pediatric study was presented in 2019 by Liang et al., in which 101.6 million data points from 567,498 outpatients were analysed.
The primary diagnoses considered 55 diagnostic codes that are deemed common pediatric diseases.
Among the most frequently found diagnoses were acute upper respiratory tract infections, bronchitis, bronchopneumonia, and acute tonsillitis.
However, the system also showed strong performance in diagnosing potentially life-threatening conditions such as meningitis.
The analysis was carried out using logistic regression classifiers to establish a hierarchical diagnostic system that achieved excellent performance in all organ systems and subsystems.
demonstrating a high accuracy of the expected diagnoses compared to the initial diagnoses made by a medical examiner.
A relevant advantage is the reduction in inappropriate testing and cost.

Appropriate Prescription of Antibiotics

In general, appropriate prescription of antimicrobials is a complex challenge, as it involves selecting the proper therapy for the suspected pathogen.
regulating the antimicrobial agent concentration and frequency of administration and identifying the appropriate route to ensure that the actual drug levels reach the site of infection.
One of the difficulties in prescribing antimicrobials is adjusting a patient’s treatment sequentially as new clinical data becomes available.
therefore, hospitals increasingly rely on automated decision support systems to review antimicrobial prescriptions.
Most prescription monitoring systems use a rule base acquired from published and expert guidelines to identify inappropriate prescriptions and prevent potential adverse events.
In 2014, Beaudoin et al. described APSS, an antimicrobial monitoring system that, unlike the previous methods, could learn new rules for prescription surveillance.
This feature, combined with user feedback supervision, was designed to enable APSS to self-improve its knowledge base in the long term.
The learning module extracted clinically relevant rules by identifying inappropriate prescriptions not recognised by the baseline system.

Researchers way in pharmacokinetic variability

pyrazinamide and rifampicin, within specific concentration ranges.
The problem of health care costs and staff shortages in managing the appropriate antibiotic prescription is significant in developing countries.
For this reason, in 2018, a group of researchers hypothesised applying machine learning approaches to collect patient data readily.
it would be possible to obtain low-cost individualised predictions for targeted empirical choices of antibiotics.
The results showed that modern ML algorithms could amplify widely used logistic regression models by predicting antibiotic susceptibility.
In this case, the random forest approach worked remarkably well, especially for predicting resistance to ceftriaxone, the study’s most widely used empirical antibiotic in patients.
In addition, automated systems can play an essential role in the real-time surveillance of the adverse effects of antibiotic therapies, thus contributing to their appropriate prescription.
The study found that this system can effectively detect adverse effects in hospitalised children.
When used as part of a decision support system, the best approaches are based on AI and, more recently.
on ML should substantially increase the percentage of patients receiving effective empirical antibiotic therapy while minimising the risks of expanded resistance selection.

Predicting Antibiotic Resistance

An area where AI is proving useful is in predicting antibiotic resistance.
thus, it is a valuable aid to physicians in the care of their patients, considering that diagnostic tests and antibiotic resistance assays often require prolonged periods.
Moreover, empirical therapy is more complex in the pediatric field than in adult medicine, as susceptibility differs with age.
Yelin et al. analysed a 10-year longitudinal dataset of over 700,000 community-acquired urinary tract infections.
identifying strong associations between resistance to six analysed antibiotics and demographic characteristics, past urine culture results, and the history of drug purchases by patients.
They then developed a personalised ML-based antibiotic resistance prediction model, which identified a higher peak risk for some antibiotics (e.g., nitrofurantoin) in infancy and childhood.
the presumed site of infection, the Gram stain result of the pathogen, and previous susceptibility data.
An example of this application is VAriant Mapping and Prediction of antibiotic resistance.
The researchers detected 14,615 variant genotypes and constructed 93 association and prediction models that confirmed the mechanisms of genetic resistance to known antibiotics.
with an average accuracy of 91.1% for all antibiotic-pathogenic combinations.
accurately predicting the point mutations associated with the emergence of antibiotic resistance to first-line drugs.

Limitations of Artificial Intelligence (AI)

AI has reached a level of accuracy in healthcare that was unimaginable until a few years ago. However, numerous limitations still make it difficult to translate into care pathways.
First, the lack of randomised clinical trials demonstrated, especially in the pediatric field.
the reliability and improved effectiveness of AI systems compared to traditional methods in diagnosing infectious diseases or suggesting appropriate.
therapies create a certain mistrust of physicians toward using techniques based on AI.
The AI culture is lacking in health care personnel: many doctors—in our case, paediatricians—have never heard of AI.
Another limitation is the methodological bias that these systems may present, as they are often based on studies, databases, and guidelines from other countries that may not represent all patients.
Another relevant limitation relies on the need for a large amount of data.
Depending on the complexity of the AI/ML architecture, the demand for data might increase
(e.g., deep neural networks require extensive data).
There are also specific applications within the pediatric field.
collecting a large amount of clean, certified, and variable data might be complex, if not unfeasible.
Another topic concerns the protection of privacy and security: we can consider, for example, the need for consent to processing personal health data by artificial intelligence systems.
It may be necessary for these systems to provide the evidence behind their reasoning so that health professionals can decide whether to follow the suggestion.
This is Artificial intelligence in Antimicrobial stewardship

Conclusion

Artificial Intelligence (AI) is rapidly transforming the healthcare industry and pharmacy practice.
Recent years have witnessed the rise of artificial intelligence (AI) in antimicrobial resistance (AMR) management, implying a positive signal in the fight against antibiotic-resistant microbes.
This is Artificial intelligence in Antimicrobial stewardship

Copyright ©: All content on FADIC Website, including medical opinion and any other health-related information, and drug Informtation is for informational purposes only