Creating Microbiology Presentations Using AI

Confessions of a Microbiologist, Future of Microbiology, , ,

I have been using AI to help create a couple of microbiology-related educational presentations over the past few weeks, for the first time in a couple of years.

I shouldn’t have left it so long…

Things have clearly improved. AI technology is now evolving at an incredible rate. Pause for breath and you will be left behind. It is not perfect by any means but starting to become functional in the professional setting in many different facets, including presentations.

Here are a couple of my personal comments on using AI to generate microbiology presentations for teaching purposes.

  • I used Co-Pilot to generate the presentations: Because I am old and stuck in my ways, I still use PowerPoint for all my presentations. As a result, it makes sense for me to use Co-Pilot, as this is also a Microsoft product and is thus designed to generate PowerPoint presentations. 
  • Prompting is everything: A good initial “prompt ” is key in producing a good AI-generated presentation. How many slides do you want it to produce? Who is the target audience? What areas do you want it to focus on? Do you want it to reference the evidence for its statements? Do you want it to be humorous? Do you want it to include pictures/graphs/tables. Do you want it to include hypothetical case studies. Do you want it to produce MCQs on the topic at the end. And that is just for starters… I think in future I am going to develop a template prompt for presentations which I can then amend for each topic I need to teach on.
  • It still has a tendency to produce generic presentations: My experience so far is that AI has a tendency to produce generic presentations. As I get better at prompting, hopefully I will get better at getting AI to include interesting details in the presentations and not “genericise” so much.
  • Editing is key: The presentation it produces for you will never be exactly the way you want it, even with optimal prompting. Therefore, you will need to dedicate a significant amount of time to editing and polishing it, and indeed giving it a personal touch.
  • Watch out for Hallucinations: In the AI age, hallucinations now have a very different meaning than back in my student days! Large Language Model (LLM) AI systems still have a tendency to make things up if they don’t know the answer. This flaw is becoming less common with more sophisticated software but watch out for it nevertheless. Bias can be another issue. My AI generated presentations did not have any evidence of hallucinations, but there were a couple of clear instances of bias which required editing.
  • It is more difficult to present an AI generated presentation: Because you have not actually written the presentation yourself, it is more difficult to actually present it to an audience. So rehearsal is key. Usually, I would spend 3-4 hours writing a presentation then 1/2 hour rehearsing it. With AI, I would recommend 1/2 hour on the prompt, an hour editing and an hour rehearsing your presentation. AI does save time, but not as much as you might think.

Despite my amateurish initial attempts, I am going to persist with using AI for presentations. The technology can only continue to improve, and very quickly at that. The presentations that it can produce now are light years ahead of what it could do a couple of years ago. I am keen to get tips from anyone else experimenting with giving microbiology presentations by AI. I would also encourage all my colleagues to give it a go.

Several of my children have already been “pinged” at school for using AI in their assignments. Being the bad parent that I am, I ignore the teachers’ frustrations and let the kids embrace the technology as much as they are able. This is the future whether the teachers like it or not…

Michael

 

“Carbapenem Conundrums”

Antimicrobial Resistance, Antimicrobial Stewardship, The Science of Microbiology,

Last week, while on-call I recommended a carbapenem for three different patients within the space of 30 minutes. Yes, it happens sometimes! Most empiric antibiotic choices do not require the inclusion of a carbapenem, but key factors to consider are ESBL history, travel or hospitalisation in areas with high ESBL endemicity, and how sick the patient is.

It is interesting to look at the psychology of carbapenem prescribing. Some doctors prescribe carbapenems because they are afraid of giving treatment to their patients that might not cover all resistance profiles. Others are afraid of prescribing carbapenems because they are traditionally the top line treatment and scared of criticism from antimicrobial stewards like myself!

But if you have to use a carbapenem, which one should you use?

The main choice in New Zealand is generally between meropenem and ertapenem. Imipenem-cilastatin is rarely used now in New Zealand, mainly due to its seizure risk. (There is a little evidence that it is the optimal carbapenem for disseminated nocardiosis and a few other isoteric indications) Other carbapenems outside these three have limited availability in NZ hospitals, or at least the ones I work in. This may be different elsewhere in the world.

Choosing between meropenem and ertapenem:

Here are most of the key factors I take into account when choosing between the two

Organism coverage-If I need to empirically cover Pseudomonas or Enterococci or Acinetobacter spp., then meropenem is a better option than ertapenem due to its broader coverage.

CNS penetration-Meropenem is a better option than ertapenem due to better CNS penetration. I had a patient with E.coli meningitis recently who required meropenem until the susceptibilities were known.

Hypo-albuminaemia– Ertapenem is highly protein bound compared to meropenem, so in hypo-albuminaemic states, the free fraction of ertapenem is increased, and it is chucked out through the kidneys leading to a decreased half-life. Therefore, meropenem is preferred in hypo-albuminemia. I use 25 g/l as an arbitrary cut-off.

Dosing frequency– If reduced dosing frequency is preferred due to patient compliance/outpatient therapy etc, then once daily ertapenem is preferable to three times a day meropenem.

Penetration into biliary tissue-Ertapenem has poor penetration into biliary tissue compared to meropenem, so I prefer meropenem for biliary infections.

Duration of treatment- Meropenem is more stable than ertapenem against resistance mechanisms such as upregulation of efflux pumps or porin channel loss. These mechanisms can become an issue with prolonged treatment and should be taken into account when choosing between the two.

Although the above points might suggest otherwise, I actually recommend more ertapenem than meropenem. Ertapenem is absolutely fine for most straightforward cases of urosepsis where empiric ESBL coverage is required.

One other point. If you do need to utilise a carbapenem, then regular review and timely de-escalation based on the patient’s condition and/or susceptibility results is important to optimise antimicrobial stewardship.

Michael

“Sometimes you just have to admit that you were wrong”

Confessions of a Microbiologist, The Science of Microbiology, , ,

Many microbiology laboratories, including my own, have in place a CSF leucocyte count cut-off of 5 × 10⁶/L as a criterion for performing multiplex PCR in the investigation of meningoencephalitis. This diagnostic stewardship policy has attempted to focus testing on those most likely to have CNS infection, and to reduce unnecessary testing. However, growing evidence indicates that this approach is not appropriate when there is clinical suspicion of encephalitis…

Take this hypothetical case study…

A 58-year-old man, Mr J Bloggs, presents with fever, headache, confusion and a vague history of what could be a focal seizure. MRI was unremarkable. A lumbar puncture is performed within eight hours of presentation. The CSF shows a white cell count of 4 × 10⁶/L, normal glucose, and mildly elevated protein. Under existing laboratory policy, the CSF multiplex PCR panel is not performed because the leucocyte count is below 5 × 10⁶/L. Empirical acyclovir therapy is discontinued on the basis of the normal CSF leucocyte count. However, the diagnosis remained uncertain and persistent symptoms prompted a repeat CSF several days later. HSV-1 DNA is detected by PCR on the second CSF sample.

CSF analysis is excellent for detecting meningeal inflammation, as occurs in meningitis, but it is less reliable for parenchymal infection, which characterises encephalitis. In encephalitic processes, inflammation may be largely confined to the brain parenchyma, without a corresponding CSF pleocytosis, especially early in the disease. Leucocyte thresholds designed for meningitis are therefore poorly suited to encephalitis and potentially risk giving false reassurance.

Recent evidence has demonstrated that a substantial proportion of patients with encephalitis have normal CSF leucocyte counts. The most compelling data to date come from a recent large retrospective study by Habis et al., involving 597 adult patients with encephalitis. They found that 25.3% had no CSF pleocytosis (<5 × 10⁶/L). Among those with infectious encephalitis, 19% lacked pleocytosis, and notably, 23.7% of HSV-1 encephalitis cases had normal CSF cell counts at presentation. Patients without pleocytosis were also less likely to receive empiric acyclovir, showing how laboratory thresholds influence clinical care. These findings strongly challenge the validity of using a fixed leucocyte cut-off to determine whether PCR testing should be performed, as it would exclude roughly one in four encephalitis patients, including many with HSV infection.

In addition, and as also shown by Habis et al., patients without pleocytosis are less likely to receive prompt antiviral therapy. This matters because early treatment, particularly for HSV encephalitis, improves outcomes. Diagnostic stewardship should promote timely, appropriate testing, not create barriers based on outdated assumptions…

Laboratory protocols should always be responsive to new data. When high-quality evidence emerges that challenges existing practice, policies must be reviewed and revised. While a 5 × 10⁶/L CSF leucocyte threshold may remain reasonable in the investigation of suspected meningitis, it is no longer valid in the setting of suspected encephalitis, where pleocytosis may be absent in a substantial proportion of cases. Stewardship frameworks should incorporate these distinctions, and most importantly, allow flexibility in order to optimise patient safety.

Laboratory practice must evolve with emerging data to ensure that diagnostic stewardship supports, rather than hinders, accurate and timely diagnosis. I am a diagnostic stewardship enthusiast, but I am the first to admit that we don’t get it right all the time.

Michael

p.s. Check out this great editorial on this topic!

References