Artificial intelligence (AI) is truly enjoying its zeitgeisty heyday: it's all over our news feeds, social media ads, and (increasingly) in the academic literature. As a medic and aspiring scientist, I'm constantly faced with claims that AI will improve every aspect of my clinical and research career, alongside apocalyptic predictions that AI will soon be stealing my job. Looking past the hype and hyperbole, how much will AI really impact clinical and academic microbiology?
Donning my hat as Associate Editor for Journal of Infection, I recently reviewed the literature to address the following:
What exactly is artificial intelligence, and where do machine learning, deep learning, and generative AI fit in?
How can we measure the accuracy and usefulness of AI applications in medical research and clinical practice?
What are the potential pitfalls of and barriers to AI adoption in clinical and academic medicine?
Are any AI-based tools currently being used in the diagnosis and management of human infections?
Check out my review in the Journal of Infection to find answers to each of these questions!
Trends in Pubmed search results (search field: Title/Abstract)
Here are some highlights to get you started:
There has been an exponential rise in publications on AI, machine learning and deep learning in recent years, with over 40,000 articles in 2022 alone featuring one or more of these terms in the title or abstract.
The first AI-based clinical microbiology system to receive FDA approval (as a Class II Medical Device) uses automated plate reading to facilitate urine and MRSA swab culture.
Other laboratory-based AI applications include analysis of microscope slides to diagnose infections like tuberculosis and malaria, and detection of antimicrobial resistant organisms using culture, whole genome sequencing and MALDI-TOF.
Outside the laboratory, AI-based applications have been applied in clinical practice to facilitate early sepsis detection, antimicrobial prescribing, and predicting antimicrobial resistance from clinical risk factors.
An exciting application of AI is outbreak management and biopreparedness, which has already been used in the international response to the SARS-CoV2 pandemic, including automated outbreak alerts, targeting testing, and predicting disease trajectory and vaccine efficacy.
In medical research, AI has been applied to novel antibiotic development, such as halicin, and designing vaccines. It may also offer an avenue for incorporating cutting-edge microbiome research into clinical practice.
That all sounds pretty promising, right? Well, don't hang up your stethoscope just yet...
The vast majority of papers we reviewed lacked any assessment of clinical utility, such as impact on patient care or efficiency; instead, they reported only statistical algorithm performance metrics, such as sensitivity and specificity. Often, these metrics were based on retrospective data analysis only, without any real-world or prospective evaluation.
There is significant heterogeneity in study design, data sources and handling, outcome reporting, and algorithm transparency. This limits comparability, reproducibility and meta-analysis.
Especially for deep learning, hidden processing layers give rise to the so-called "black box", making it impossible to determine how a particular output was produced. This limits transparency and could represent a significant barrier to clinical uptake.
AI may introduce and amplify existing biases based on factors such as race, gender, age, ethnicity and weight.
Recent advances in generative AI (including ChatGPT and GPT-4) have raised concerns of computer-assisted scientific fraud, and consensus guidelines on the use of AI to produce and publish scientific research are still lacking.
So, which is it? Is AI the boon or bane of the infection clinician? Fascinating work is undoubtedly being done, with scope to improve diagnosis, management and research of human infections. However, evidence of real-world clinical utility is still sorely lacking, and social, ethical and practical issues must be resolved before widespread uptake can be expected. As an early adopter who is already using ChatGPT to help me with everything from writing computer code to planning dinner, I'm genuinely excited about the prospects of AI to improve efficiency (especially if it can reduce our administrative workload and help us focus on patient care and research). But I can't say that I'm worried about losing my job to a computer any time soon...
Please check out the full review for details, and share with your colleagues.
And sign up to Hello Micro now to receive emails about new blog posts and video tutorials!