Subscribe
Share
Share
Embed
Communicable takes on hot topics in infectious diseases and clinical microbiology. Hosted by the editors of CMI Communications, the open-access journal of ESCMID, the European Society of Clinical Microbiology & Infectious Diseases.
Communicable E61: Love PK/PD, part 3
===
[00:00:00]
Angela: hello and welcome back to Communicable, the podcast brought to you by CMI Communications, ESCMID's open access journal covering infectious diseases and clinical microbiology. My name is Angela Huttner, and I am the editor-in-chief of CMI Communications and an infectious diseases physician based at the Geneva University Hospital in Switzerland. I'm joined by two co-hosts today. The first is Amy Legg, assistant director of pharmacy and consultant pharmacist in infectious diseases and research in Brisbane, Australia, and editorial fellow at our sister journal, CMI.
Amy: Hi. Great to be here.
so glad you're back. And my second co-host is Rekha Pai Mangalore, ID physician from Melbourne, whom you also know and love.
Rekha: Hi, everyone. Good to be here
Amy: Thanks, Angela. So today we're back with our third installment of the Love PKPD [00:01:00] series. This time we will be unpacking MIPD, or for mere mortals, model informed precision dosing.
And we're delighted to have two expert guests with us. The first is Amanda Gwee, Associate Professor of Pediatrics at the University of Melbourne in Australia. She's a general pediatrician, infectious diseases physician, and clinical pharmacologist at the Royal Children's Hospital in Naarm, Melbourne.
She's the head of the clinical pharmacology unit at Royal Children's Hospital, and leader of the antimicrobials group at Murdoch Children's Research Institute. Amanda's research focuses on optimizing the dosing of anti-infective agents in children, an area of increasing importance given the emerging antimicrobial resistance.
And as part of her work, she's the creator of the beloved KidsCalc MIPD software. She's also a fellow editor working for the Pediatric Infectious Diseases Journal and Journal of Antimicrobial Chemotherapy. Amanda, welcome to Communicable.
Amanda: Thanks so much for having [00:02:00] me
Angela: and I'm delighted to introduce our second guest, Sebastian Wicher, professor of clinical pharmacy at the University of Hamburg in Germany, where he leads the research group in clinical pharmacy at the Institute of Pharmacy. His research is all about PK/PD and precision dosing of anti-infectives. His group develops open access web-based software to foster MIPD, such as TDMX, which is available freely online. He has served as president of the International Society of Anti-Infective Pharmacology, otherwise known as ISAP, and as education officer and chairperson of ESCMID's PK/PD Study Group.
And finally, he too is an editor at our potential competitor the International Journal of Antimicrobial Agents. Sebastian, it is great to have you here.
Sebastian: Thanks, Angela, for the kind, introduction, and thanks, for having me today.
Before we jump in, I'll just say that both of our guests, are developing software. They're working hard on this, But they have no financial conflicts of interest [00:03:00] to declare. Okay.
Angela: Leave it to you, Amy.
Amy: Yeah. I feel like the bigger conflict is that I use both of them and am desperate to keep them both, best friends with me. So as our listeners know, we start these episodes with a get to know you question for everyone. So today's, get to know you question is, if you weren't a pharmacist or a doctor, what career would you want?
Hosts go first to break the ice. So Angela, what's your alternative career?
Angela: All right. So I've done so many of these episodes that I do believe I've been asked this before. and I have to be consistent. I would be a travel writer. It's my dream job.
That's enough from me because we have so many people.
Amy: Rekha, what's yours?
Rekha: I would like to have been a chef. a baker actually, because I think that you can't go wrong when you mix chocolate and butter. just the most beautiful combination.
Angela: Amy, what about you?
Amy: well, I would have been a professional netballer.
I would now be a netball coach, probably coaching the Firebirds and the [00:04:00] Australian Diamonds, and then in the future I would be a netball commentator.
Angela: Oh my gosh. I love it.
I think I would understand that so much better if we had netball in Switzerland,
Amy: I didn't even think that people wouldn't know what netball is. It's a very popular sport in Australia. I don't know, maybe popular is the wrong word, but it's what you play through school, and, it's, like basketball where you can't bounce the ball, which sounds a little silly.
Angela: Oh. Well, she's nodding, so she's clearly got this.
Amy: Yeah, it's, definitely a sport where the girls are the best. You only ever hear- Ooh ... about the female netball team. Yeah, it's
Amanda: way harder than basketball.
Amy: Yeah, it's way harder than basketball.
Angela: Ooh. I like that. Yeah, Sebastian.
what is your alternate job, Gave this some thought. Mine is more boring, I would say. You know, I studied pharmacy, but I did my PhD, between mathematics and pharmacy in a graduate program, and I always envied my mathematics, fellow PhD students [00:05:00] who had a very good understanding of, statistics, nonlinear mixed effects modeling, and I had to work much harder to understand all of this. So at that time, I thought, "Damn, I should have studied mathematics." But probably they have thought the same about us, because of our pharmacology and PK understanding. So the grass is always greener on the other side, obviously, and I'm probably happy with where I'm at.
Angela: Aw. Okay, but you win the nerd award. You know that, right? Yeah. That's a good thing here. You can't be better than a nerd here, so.
Sebastian: No, you have to be a little bit nerdy for MIPD, at least if you develop those tools.
Angela: Yes, indeed. And this is where I think I should announce that, guys, I am, like, your mascot in this episode.
I am the person on this episode that has never been even close to real MIPD. So I will be like the, horse whisperer. Amanda, we didn't hear about your alternate life-
Amanda: Yeah. I'm a bit embarrassed of mine [00:06:00] after Sebastian's, but when I was young and fitter, I used to want to be a hip hop dancer, and specifically a backup dancer for Janet Jackson.
That was my dream. now, I think probably a
Speaker: chef like
Amanda: Rekha. Like, I'd love to be on a show Is It Cake?, you know?
Angela: I love it. Sebastian, we went from mathematics to Janet Jackson backup dancer. what breadth, what breadth we have here.
That is so cool.
Sebastian2: Indeed. I agree.
Okay, so getting down to serious work here. We are nerds, and therefore we are going to discuss model-informed precision dosing.
Angela: Amanda, can you briefly explain MIPD and how it can be used in practice?
Amanda: So MIPD is obviously model-informed precision dosing. It's when you use a population pharmacokinetic model, which is essentially a model that describes what happens to the drug in a population of patients, and you use that model to figure out a better dose for your patient.
and most [00:07:00] commonly, it's a model embedded in, some sort of interface, usually a web interface, and it allows the user to input specific patient characteristics, whether that be age or weight or kidney function, and it allows the user to generate individualized dose for that patient. So it's pretty handy tool.
when can we use it? I prefer to use it for drugs with a narrow therapeutic range, like vancomycin and gentamicin are wonderful examples where, you want to stay within that therapeutic range to avoid toxicity and make sure you have effective treatment. And I like to use it in patients where their, PK is likely to be very varied, such as, you know, if things are rapidly changing and due to developmental changes such as the neonatal period or in critically unwell children or those who have end organ dysfunction.
So when you need to really tailor dosing to that particular patient
Amy: Perfect. that's a great start. Sebastian, getting a little bit more [00:08:00] into the nitty-gritty, can you talk about the role of MIPD from the first dose as opposed to using it once you've got a drug concentration?
Sebastian: Yeah, for sure. So actually, this is something that is probably a little bit underused still. I think in most cases, people do use MIPD for dose adaptation once a, a treatment was already started.
but actually, that's only one of the things you can do with MIPD. You can actually start already, the treatment course before you do TDM. Traditionally, you start with a standard dose or you start with a dosing table, but, uh, there you can maybe consider one or two covariates.
Sebastian: Yeah, maybe it's body weight, maybe it's renal function, and then the table would become very complicated to read, and nobody could actually work with this. So, a pharmacometric model, has no limits. technically, any number of covariates can be implemented. Usually, you see maybe four, five, or six covariates that are implemented in pharmacometric models.
And then, when you want to use, [00:09:00] MIPD, so a pharmacometric model for calculation of the first dose, you would enter all those, values, into the user interface, as Amanda explained, or you would draw them from the health record if your software is connected to the electronic health record, so you wouldn't have to punch that in manually.
And then you ask the model, " Please calculate a dose, that hits a specific PK/PD target already before you start the treatment course." And, well, then it depends how the software works. Some softwares only will give you a straight line. That's the typical prediction. Some softwares also show you uncertainty bands around this prediction, and this prediction is usually called the typical patient.
That's what the PK profile looks like for a patient with those specific covariates. And then if you use an optimization algorithm, which is embedded in the MIPD software, this, PK/PD target attainment is already maximized after the first dose. This is, not yet used, regularly, but [00:10:00] there's one nice example I would like to mention, the BENEFICIAL trial from Peter de Cock, which was published very recently in June, which looked at, critically ill or severely ill neonates, and they actually implemented this. They individualized already the first dose, and they could clearly show that, target attainment was, quite good early on in the therapeutic course.
there's clearly value in this. so yeah, maybe this is something for the show notes, , a follow-up read for the listeners how to do that.
Amy: Yeah. Perfect. And is there any difference, like people might assume that it's less accurate to start from the first dose until waiting till the therapeutic drug monitoring has been undertaken and you've got a drug concentration?
Sebastian: Yeah. good catch, Amy. Of course. that's why we need to do a therapeutic drug monitoring for those drugs. So if you would look at those uncertainty bands, if you calculate the first dose, they are usually quite wide because, there is a lot of inter-patient variability, even if you have the covariates included in the model.
but still, it's better than [00:11:00] just a standard dose or, maybe a dose that doesn't consider all of the covariates. But there's little evidence, to be honest on the topic of first dose individualization. There's more on, the inclusion of TDM data into the MIPD workflow
Amy: Yeah. I think it's that it's a bigger spectrum than people think sometimes, isn't it?
we've got totally optimized precision dosing with TDM at one end, and then we've got a standard dose on the other. And at least, you know, the forecasting aspect empirically, is somewhere between.
Sebastian: Yeah. and maybe one more thing to add, on this first dose aspect. Even if you don't individualize the first dose and you start with a standard dose in your setting, with MIPD, you can actually sample after the first or the second dose already.
You do not have to wait for a steady state. so also this process is faster than the conventional process where you will have to wait five half-lives until the patient reach steady state.
Amanda: Can I add a couple things to that? just on the BENEFICIAL trial, like I totally agree the [00:12:00] trial is amazing.
And the other thing to note in that trial was even though they didn't find a significant difference, there was fewer, vancomycin related adverse effects in the model informed precision dosing group. Obviously, we need a larger cohort to find this significant difference. But, you know, it's, quite reassuring that, you know, it is a safer approach.
And then just on, you know, starting versus first dose adjustment, you know, when in our neonatal model informed precision dosing work, we find if we use, model informed dosing from the start, about eighty percent of children achieve therapeutic target. And then if we take a level after the first dose, we can increase that to over ninety percent.
Amanda: So we're just evaluating that second part in a new trial.
Wow. Very exciting. Yeah. We, covered the BENEFICIAL trial in our ESCMID Global, Late Breakers, and it's a fascinating trial, not least because of the fact that done in neonates. it is really one of the [00:13:00] areas where we need more clinical data paired with PK/PD.
Angela: You know, we're doing so much based on theory, and then finally we have something that really puts everything together so nicely and kind of puts us to shame, you know, some of us who are troglodytes and not yet doing MIPD in our hospitals.
hopefully later we'll have some time to talk about implementation of MIPD,
Amanda, can you give us, some practical advice in extreme situations like how do you think, about body size metrics and, how are these used within MIPD software?
Amanda: ultimately, the body size metrics that's included in the software depends on, you know, what was found to be most relevant to the model. as we develop models, we evaluate different body size m- metrics, whether that be ideal body weight, actual body weight, body surface area. I have to say that I'm a bit, biased towards using actual body weight.
Amy would know only as a [00:14:00] pediatrician, I can tell you it is so difficult to get an accurate recent height measurement to calculate body surface area in a critically unwell child. You just think about how quickly children grow. And so when I develop my models, I keep that in mind, and definitely if I'm not seeing a big difference between body surface area and body weight, I would always prefer body weight, just from a clinical, practical point of view.
Amy: I think people do get very confused, about body weight metrics, so I think that was a lovely, answer there, Amenda.
And then to follow up in another extreme situation, how do you think about kidney function in practice? so for a long time, pharmacists have been taught to use creatinine clearance, so taking a serum creatinine level and then using the Cockcroft-Gault equation, which takes into account things like body size, to get a very accurate marker of, or to get a supposedly accurate marker of renal function.
And then pathology results started coming out with standardized eGFRs to [00:15:00] 1.73 meters squared that was immediately visible on standard lab reports. So obviously, doctors became very familiar just looking at that number, but that was all standardized to 1.73 meters squared, which a lot of patients aren't.
And so then a recommendation, of course, came through to de-index the eGFR to try and modify the eGFR that was coming up on the standard lab tests to better reflect the size of the patient. so following that journey of the different ways that we think about renal function and therefore drug dosing, how do you think about renal function in your patients?
Amanda: in children, if we really want an accurate, GFR measurement, for example, for dosing chemotherapeutics, we would just get a nuclear medicine GFR scan because, we don't trust estimated GFR calculators.
So I'm talking about the creatinine-based calculators such as the modified Schwartz equation. Having said that, uh, the modified Schwartz has been shown to [00:16:00] outperform other calculators in the pediatric age group in children over two, but its accuracy in children under two hasn't been, that good.
And when you look across the creatinine-based equations, the accuracy of these equations in estimating GFR ranges anywhere between, like, 50% to 75%. So there is a lot of inaccuracy in those calculations. we actually have developed our own GFR calculator, and we've just had that paper accepted for publication.
It's called the GROW GFR Calculator, so it'll soon come out, and we'll be building it in KidsCalc. what I have to say, though, is that it is not standardized to, a measurement of 1.73 meters squared. So that standardization is from healthy adults from the 1920s, and there are studies showing that if your body surface area is less than 1.6 or greater than two, then you have greater discrepancies in [00:17:00] your, eGFR measurements.
So the standardizing to that, body surface area often leads to overestimation of pediatric, GFRs, and that's why our calculator doesn't standardize to that. But I appreciate that it does, uh, cause problems when you're grading around chronic kidney disease when, you know, the grading is all standardized.
Amy: That's one of the biggest problems, is how much we're expecting these to do. You know, we're trying to work out what their current renal function is, whether they've got chronic kidney disease, and, serum creatinine is, responsible for all of it.
Amanda: Yeah. And I think that, you know, there is a role for, using GFR equations that are not standardized to one point seven three meters squared in dose adjustment in the setting of renal impairment, and then you can use the, standardized to the body surface area for CKD grading.
Appreciating, though, that you are likely overestimating your GFR.
Amy: Do you teach this to your pharmacist, Sebastian?
Sebastian: Yeah, I would, I would add a shortcut actually to [00:18:00] simplify things for our listeners. I very much agree to what Amanda said.
I'm also not a big fan of the body surface area calculations. If you go back to the original equations, this is actually not precision dosing. It's okay for the CKD grading, I agree. but, the body surface area has, issues. The good thing about MIPD software is that you usually don't have to worry about this.
You enter the raw covariates. So you enter body weight, you enter body height, you enter serum creatinine. That's the usual values that go into this. And then depending on how the models were developed, they use different equations. Some use indeed the classic creatinine clearance Cockcroft-Gault equations.
Others use, the CKD-EPI equation, which is, more accurate, particularly in the, space around sixty milliliters per minute, for example. But as a user of an MIPD software, this is usually intrinsically considered. So you en-enter the raw covariates, and then the software uses the appropriate equation that was also [00:19:00] used to develop the respective model.
And this is truly important because if you would use another equation for, eGFR than, the one that went into the model, in the first place, you would introduce some bias. So actually for clinical practice, this is something where you don't have to worry so much. That's my two cents I would like to add here.
Sebastian, I spend so much of my life dealing with this question. a lot of us in Australia have been using, TDMx, but of all the MIPD, there's often a concentration time profile that's displayed that reflects the population PK parameters.
Amy: And then once we've got our individual patient parameters entered, there's a new line that represents our individual patient's PK parameters. Is it a problem if those two lines diverge?
Sebastian: Yeah, that's a good one. well, first of all, it's not a problem, it's a feature of the process. So if all the time your individual patient would be on the population profile, [00:20:00] that would mean all your patients would be the same, then we wouldn't have to do, TDM and dose individualization.
So it is expected that we have, some deviation from, the population PK profile. But, of course, I, perfectly understand where you are coming from. The question is, how far can you actually safely deviate from this population profile? there are two ways to diagnose this. The first one is, if you add the prediction intervals around the population PK profile, you would want your TDM measurement to sit within this prediction interval.
If it's outside, that's an indicator that this measurement probably behaves differently than your population prediction. The other indicator, that can be looked at and that is also implemented in some MIPD softwares, is that you can have a look at the PK parameter distribution. So there you have, the mode of the distribution, that's the typical clearance, and then you have some distribution around [00:21:00] it that shows you, for instance, how the clearance, should vary, in the population if this model was true.
And then if your clearance estimate is outside of this distribution, so the 95%, interval, for example, this is another indicator that maybe you are stretching things too much and you're trying to fit a model that doesn't really, fit to your patient. What would you do in these situations? You would probably try another model, if you have a set of models at your hand that you can choose.
if you don't have another model, well, you can continue, with the software, but you would very carefully want to inspect the model fit and also the predictions you make from that because you are outside of the safe space, I would say, where the model isn't very well-informed.
Amy: Did you want to add anything to that, Amanda?
Amanda: No, I feel like Sebastian answered that really well and actually probably answered my next question.
Sebastian: Apologies. Yeah. I touched it a little bit, but not fully.
[00:22:00] It was gonna be such a great segue, though. So Amanda, following on that, you are a pediatrician. children to me are like the ultimate individuals, right? The most varied population you can have inside one patient population. So how much tolerance do you have for this level of variation, when you're trying to fit a model, so to speak, on your patient?
Angela: What do you think?
Amanda: Yeah. I think there is no model that is going to perfectly describe your patient's, pharmacokinetics. I think that, there are going to be, lots of different models that may be able to describe it, and we definitely don't expect that an individual's estimates will definitely match the population average, estimates.
So you're trying to choose an appropriate model, if I'm trying to choose one for a child, I make sure that the model was developed from a pediatric population. I make sure that the disease states seem to match, that, the baseline covariates, and by that I mean like [00:23:00] looking at age, weight, and creatinine, all the relevant ones to the model, match to get a good idea of whether that model would apply to my patient population.
Then I look at some of the parameters to see whether I think, they would reflect. That's the way I would choose it for my population. It doesn't have to be exact, though. There will be no model that is exact to your patient.
Angela: So here's a question. Would you use an obese model on a non-obese patient if the other PK parameters match?
Amanda: No, I wouldn't, because as I said, you wanna make sure, like you cannot rely entirely on PK parameters matching. You have to make sure that the patient population is relevant to yours.
Angela: Mm-hmm. Yeah.
Amanda: And I think that is one of the most common problems that I see when people are actually applying the wrong models, to their patients.
You know, for example, I see a lot of people applying critically ill models to, non-critically ill populations where a lot of the critically ill children may have renal [00:24:00] hyperfiltration and other, changes that are taken into account
Sebastian: I agree to that. I think the population, and the patient really need to match because we're doing sparse sampling, and we may not be able to differentiate two models with sparse sampling. So it makes sense to have a perfect match there.
And to summarize, it's okay if the population PK lines and the individual PK lines merge because not all patients are average. but what we'd have to make sure is that our drug concentration is on the concentration time profile.
Sebastian: Great summary, Amy. We could have answered more briefly.
Rekha: So do, people need to develop their own models? Like for example, Amy and I have hospitals which have drug monitoring. So would it be useful for us to have our own models and our own sort of models informing dosing software? When we have a whole heap of models developed in different populations across the world, and I'm trying to apply it to my patient, and my patient really doesn't suit [00:25:00] that model, and I'm still trying to do it.
I'm trying to match the best covariates the best as possible. So is it better that if we have the capacity to develop our own?
I'm not sure. So it depends on what you want to do. So if you want to do a TDM, and MIPD, in a space where there are already a lot of models, for instance, for vancomycin, I think this is a very well-studied drug with, I think, now more than 50 models available. I would rather check whether my use case or your use case that you would want to study is covered by those models.
Sebastian: And once you start with the MIPD, maybe do a short evaluation, do your drug levels that you measure match, and rely on other people's work that have done external model evaluations in this space. I think building your own models, is of course a must if you have no evidence from the literature and if you want to explore something new in a research setting, then you will have to develop a new model.
But there you probably also need to do a clinical trial and do [00:26:00] proper sampling and, It's more difficult to piggyback on, top of clinical practice, I would say.
Amy: so Sebastian, all models are wrong, but some models are useful. What is an acceptable error amount when validating a pop PK model? And I think a lot of clinicians are worried about how much you can trust them, and so how much variability is there in these MIPD estimates?
Sebastian: Yeah, that's a good one as well, and the answer is, unfortunately, it depends because it depends on the PK/PD target.
I give an example. We're currently working on developing something for linezolid. There the target is two to eight milligrams per liter. So if you would aim, in the middle of this target range, you would aim for four milligrams per liter, you could actually allow an error margin of around 50%, if you have an unbiased model.
If you aim for four, your concentrations would still fall in that range. So that's quite a forgiving [00:27:00] drug in these regards. If you compare that to a non-ID drug, we're also working on busulfan. Busulfan, has a very tight, target. The AUC target is only 80 to 100. So here, the model, needs to be much more accurate, that you can really use them for precision dosing.
So it depends, and also the good news is most software packages will warn you if, there is a misfit, and you should clearly, acknowledge those warnings because that tells you that you are at risk of doing some biased calculations, which may not be precision doses in the end
Sebastian, I can't remember the, drug regulatory standards around residual standard error, but I think they say less than 20% or less than 30% parameters.
Sebastian: . For busulfan, 20 to 30% will be not enough. lenalidomide, it will be enough. so I would rather, look at the warnings that come back from the software if there is a, residual error that is too large for this very specific [00:28:00] patient.
Yeah. That's at least my view on these things. Yeah.
Amanda: But it really does depend on why you're developing the model. If it's for, drug regulatory reasons, obviously you have to abide by, what they want. If not, they won't accept your plan.
Sebastian: Fully agree. Yeah. This was a bit framed towards MIPD.
we just had to you know, try and check a voriconazole, performance. And it's so funny because the, you know, the percentage changes can be tiny, but if it goes over the sort of therapeutic range, then it's a fail. And it was just fascinating trying to do that with, you know, all the different statistical tests around error, and just realizing that it's just clinical.
Amy: Like, you- you've just gotta always come back to what's safe, what's best for the patient. It's, you know, all that technical stuff aside, don't ever let the computer talk you into something that you wouldn't otherwise do sort of thing. Or don't let the computer talk to you into something that you know in your heart is unsafe.
Sebastian: [00:29:00] Fully agree.
Rekha: Here's one for both of you, so in practice, would you ever modify recommendations from MIPD, before applying them to a patient? would you, you just follow it blindly or would you decide to do something different?
What should we do?
I think all of us think that MIPD should never replace clinical judgment. It's a clinical tool that, the treating clinician needs to interpret the results of. And yes, I either commonly modify or sometimes even reject, the findings of model-informed precision dosing, especially if from a clinical point of view, I think the results don't apply.
Amanda: So if the dose recommended is excessively high that I don't feel comfortable prescribing it, if the, child has an evolving clinical state like evolving renal impairment, or if we add in another drug which I think is going to cause a drug interaction, I'll often modify or not follow it. I think ultimately we have to apply the standard rules of prescribing when interpreting the [00:30:00] results of, MIPD
Yeah, I fully agree. I'm also a big fan of the human in the loop. maybe to add, to the clinical r- reasoning that you mentioned, Amanda, also sometimes there are just errors in the electronical health record. A peak is not a trough, a trough is not a peak, and so forth, and the software just doesn't know this.
Sebastian: and it's always good if a human being, a doctor or a clinical pharmacist inspects those plots and detects whether there's something fishy going on and then of course overrule those decisions made by the software. The human should be in the loop clearly
Yeah. And don't assume the computer's smarter than you, isn't it?
Amy: You know?
Rekha: Yeah, Clinical judgment always, Yeah.
Amy: So following on from that, we've been talking, you know, about error and, evolving patient issues. One thing that I think people fundamentally think is that the more drug concentrations you have and are able to include into MIPD, the more accurate your predictions will get.
But in practice, sometimes [00:31:00] actually patient's PK changes so much that the more drug concentrations you have, it's actually harder for the model to potentially find the exact PK parameters. So I just wondered, Sebastian, can you talk us through whether it's better to have lots of TDM over a long period of time as opposed to lots of TDM in one dosing interval?
Amy: and whether you might ever not include some TDM when you're, using MIPD on a patient.
Sebastian: Another good question, and unfortunately another "it depends" answer. if you do have a rather stable patient, a general ward patient, then, it's actually true what you said. So adding more TDM samples usually helps, to improve the predictive performance.
we have seen that in some of the external model evaluations, for example, for vancomycin in general ward patients that, this was true. In ICU patients, patients that are instable, this is not necessarily always true. Those patients are changing very rapidly, and if you see that, that your, PK profile, your [00:32:00] individual PK profile is diverging from your TDM samples, indeed, it may be, an idea to remove some of the older samples and focus more on the most recent ones because they are clearly the more relevant ones, and they are also the ones that showed up in our external model evaluations as the ones said to be, more informative.
some software packages also have some weighting functions included, so they actually automatically weigh down some, older, samples, so then the user doesn't have to take care of this. So instable patients are clearly interesting, and I think, also there, the algorithms, can be further developed.
We are, for instance, looking right now into using some machine learning algorithms on top of the pop PK models that may detect those instabilities that are not covered by, a covariate. this is truly an interesting field also for the nerds, in MIPD.
Yeah. I was just thinking about something when you said concentrations , depending where your patient is [00:33:00] situated, ward versus, ICU.
Rekha: I had a question about how do we make these sort of, personalized dosing tools accessible. For example, half the world, that needs them doesn't have them, Lots of, sepsis burden and infection burden and, resistance. Would these MIPD tools be useful without TDM, at all?
Is that something that you could think of with machine learning? how we can make these things more accessible to places because at the moment we have resource, settings that have access, whereas there's a massive sepsis burden on children and adults in,
Sebastian: Yeah
Rekha: in many parts of the world, so.
I think this is the holy grail. if we can develop models that become so accurate, that the TDM measurements are not necessary any longer, indeed, I could foresee a wonderful world, where the software could actually guide those doses. Unfortunately, w- with, just with covariates, we can explain some of the variability to date, but not all of it.
Sebastian: [00:34:00] So, this is something for the future, and we'll have to see whether, those new algorithms will, prove to work in this setting as well, that they can actually replace TDM.
Amanda: I really struggle with this when I was on a guideline committee for, an LMIC setting, and they really wanted to use our vancomycin dosing calculator, but they don't do therapeutic drug monitoring.
And I think the thing is that we are so comfortable as clinicians using dosing strategies where when you try to find the original study on which it's based, you can't find any data. And we sit comfortably, it's, it's a historical practice, and yet there's no data on it. And yet, um, we have these models that, yes, were created in other populations, but where there are more data, and yet we're more happy using this historical dosing strategy.
So I struggle with it ethically which one is better. I actually don't know. Like, in that instance, they said that they would prefer using the dosing [00:35:00] calculator, so we gave them access to it. Like, I think, in LMIC settings, they still have, internet access and can access, you know, a lot of free software.
The issue is the turnaround time for TDM. I don't know which is better, honestly, because I think, notoriously, especially in children, I can't speak for adult dosing, all our dosing is pretty poor. So anything better to me is still a little better until we can find the right answer for that population.
Rekha: Thanks. That's great.
Angela: So I have a question for both of you about implementation following on Rekha's comments.
There are hospitals who would really love to get in on this MIPD. We know, we do believe it is better for our patients. we do think we should use these tools. you, Sebastian, have developed TDMx, which from my understanding, it is an in-house tool. It is not commercially developed by any means, but it is open for use.
And there are, of course, tricky issues, particularly here in Europe, [00:36:00] where, MIPD is considered a medical device. So you actually need to, have a formal license. Before you can use it on patients. there's of course that barrier to uptake to implementation. But for countries where you don't have this extra regulatory layer, what are the steps to implementing, to getting people to be at ease?
do you have examples, Sebastian, I think you, do speak on this in other countries outside Europe how does it work, that people can get more access to these tools?
Sebastian: Yeah. Very good question, Angela, and I think this is a mission very close to my heart to make these things, accessible.
So TDMx, as you mentioned, we have it as an in-house medical device within our medical center, but it's also an educational tool, and it's freely available online. What I usually do is, it's important to familiarize, the users with, uh, MIPD.
If you haven't heard of population PK before, and if you don't know what a pharmacometric model is, the hurdle may be quite high. So there needs to be [00:37:00] some education at the first place to also show, um, what, uh, the model can do and what are the benefits, but what are also the potential pitfalls. We talked about the human in the loop.
This is clearly important, so people shouldn't say, "Okay, this is the magic dosing tool. This will do everything for me." No, you need to be educated to use those tools. So that's the first step. And then I would run an implementation pilot, where, the, local infrastructure is set up. Also, the, behaviors will have to change a little bit.
So in many settings, a trough sample is just more or less a random blood draw, but, to make this work, in an MIPD workflow, this has to be very well documented. The timing of dosing needs to be documented. The timing of the sample draw needs to be documented. So it's kind of a change of the workflow, and it needs buy-in, not only from those that prescribe the drugs but also from, the nurses, and from other people involved in this [00:38:00] process.
And then I would run a pilot and expand from there, and I would really focus on, those indications that benefit, from the MIPD process. It's very clear that this is vancomycin, for example, uh, with AUC guided dosing. If you want to implement this for beta-lactams, for example, I would also focus, not on TDM for everyone but those that have a higher risk of an elevated MIC, for example, or those that have very high renal function where the individualized dosing has a chance to make a difference.
Sebastian: So to use your resources wisely in those settings.
Amanda: Yeah, I just wanted to add to Sebastian's points. I agree with all of them. one of the issues that came up when we implemented, Kids Calc at our hospital was how were the pharmacists going to check the dosing strategy when all the doses are individualized.
And, we needed a way that people could check the inputs, into the calculator versus the output. That's why, in the calculator, we ended up modifying, [00:39:00] it and having a copy paste function so that at least then there was the ability to check. Because obviously, in a hospital setting, if everyone's having individualized doses, it does
It's harder for pharmacists to check that the dosing is correct.
Amy: Oh, Amanda, you're an ally, and we appreciate you. Actually, that's one thing that we haven't really talked about. Another sort of benefit of MIPD is just that it's hard to calculate AUCs, and MIPD does it for you. It's one of the main ways that people get into it is they wanna have a vanc AUC and they don't know how to calculate it manually.
but there's so much more to it than that, of course.
Yeah, good point.
Angela: Yeah, yeah. Start simple. So finally, a quick lightning round on targets. This is for both of you. What PD target do you aim for, for beta-lactams? Are you purists? Do you go for four times the MIC for the whole interval?
what do you do for beta-lactams? What's your target? Amanda, you first
Amanda: I'm a purist, [00:40:00] so I go 50% time above MIC, and in CSF I would go 100% time above MIC only, 'cause obviously it's higher stakes and you wanna make sure you have effective concentrations that Hmm. Are facing into it. Hmm.
Angela: 100%, but not four times MIC for 100%. Okay, fair enough. Sebastian, what about you?
Sebastian: Yeah, I'm a fan of continuous infusions actually, or extended infusions, and there it's quite easy to reach four times, the MIC.
and it's quite easy to implement also. It's quite easy to dose adjust. So, under these settings I would use four, four to six times the MIC actually to also suppress potential resistance development. But I agree in a, general ward patient you can use lower targets of course. And for intermittent dosing you will not reach four times above the MIC.
That's quite unrealistic. if you use the ECOFFs, for example, for Pseudomonas.
Oh, no. Yeah. there you'll be stretching.
Amy: And we should grab Rekha's, thoughts on that. What's [00:41:00] your beta-lactam target?
Rekha: So ICU patients it's 100% time over MIC. Ward patients it varies 70 to 100%, time over MIC.
Ah. Yeah. So I, I, I haven't aimed for 50% time over MIC for, for some time now. Really? Yeah.
Yeah, totally agree about continuous infusions. So easy to get to higher targets. we've implemented continuous infusions in our ICU and it's like four times over MIC easy. Six times I'm not so sure, Sebastian. I don't think I've aimed for six times at all.
Angela: Hmm.
Rekha: Yeah.
Angela: I'm actually impressed, Rekha, that you accept 70% of time over MIC. I thought you'd go for 100%.
Rekha: No, I actually-
Angela: All the time ... do
Rekha: go for 100% most of the times. I can accept 70% on wards. Depends- Uh-huh ... on, it depends. Yeah. That's, that's the, that's the word
Angela: Yeah. so obviously the MIC is very pathogen specific but do you sort of go even beyond just the MIC in the sense that, you know, let's say you have a bacterium that's encapsulated. Do [00:42:00] you say, "Okay, well for that bacterium, you know, for Klebsiella I wanna go four times above the MIC"? Do you get really pathogen specific or do you just say MIC across the board?
Rekha: I do, generally it is, uh, bug, drug, and, uh, infection syndrome that I use before I decide on my dosing. it's not just the MIC alone. -
there are more data for gram-negative bacteraemia for higher percentage time above MIC than there are for data on gram-positive bacteraemia, I know from Amy's excellent reviews are really poor on the pharmacokinetic/pharmacodynamic target. I think the answer is that these targets all stem from pre-clinical data. and at the moment we don't know. The, if you have a critically unwell patient, it makes sense to achieve a higher percentage time above MIC.
Amanda: But I wanna give also a slightly opposing view as a pediatrician on extended infusion. You know, like if you have a child on a general [00:43:00] ward, like a toddler, on an extended infusion where the dosing is at three or four times a day, it's a lot of less time that that toddler can walk around the ward or other things, and it's a lot harder for families.
Like, I understand it in critically ill patients, but in the average ward patient, actually it raises a lot of practical issues, Yeah ... that, Mm ... does affect the quality of life of our patients. Yeah.
Rekha: Amanda, I have to add to that. It's not just the, children, it's adults too. when we think about extending infusions on the wards, we have to think carefully because it's a similar situation where patients want to go out, they want to go down to the café, and they really don't wanna be confined to an IV pole, for that longer period of time.
So We use it sparingly on the wards. ICU is of course easier because, uh, it's a different cohort of patients.
Angela: So Sebastian, any thoughts?
Sebastian: No, I agree to that. Mm-hmm. Gotcha. I mean, I think, ICU harder, continuous general ward can be also the shorter infusions, yeah, and lower targets that are achieved.
So that's fine.
Angela: So this was supposed to be a [00:44:00] lightning round, but we are nerds, so it is not a very fast lightning round. Next up was, what are your PD targets for aminoglycosides? do you go for Cmax over MIC, or are you leaning more towards AUC over MIC? Sebastian, you first this time.
Sebastian: Yeah, the Cmax over MIC, eight to 10, that's classically ... That was developed for three times daily dosing originally. So if you use once daily dosing, actually you should also have the, keep an eye on AUC over MIC of 80 to 100 because, just the Cmax over MIC with once daily dosing or extended interval dosing will not give you enough.
So it's actually both. I would keep an eye on the AUC all the time. But Cmax over MIC is clearly a valid target, yeah.
Angela: Hmm. Hmm. He wants it both ways. Amanda, what about you?
Amanda: I was gonna say the same thing. I agree. I think there's both. I think, there's clear evidence that Cmax over MIC correlates with efficacy.
but AUC as a total [00:45:00] drug measure is also associated with both efficacy and toxicity, and I think that, both are relevant
Angela: Amy? Rekha?
I think I've been really fascinated watching, how the AUC target has sort of destroyed aminoglycosides because once we move to once daily dosing, their pharmacokinetic properties in that they're low protein binding and quickly cleared renally.
Amy: Like Sebastian says, once we move to once daily dosing, it's very difficult to achieve the AUC, Mm ... that you can often achieve the Cmax target that you would like to because we give them as a push, and then they get rapidly cleared. So I completely agree that both would be ideal.
but unfortunately, what we see in papers coming out is that if you only look at their efficacy when trying to achieve an AUC target, then you often find that they fail, and it's a bit of a throwback to their pharmacokinetic parameters. [00:46:00] And so we've just gotta ask ourselves if we're happy to just keep having them as a, drug class in our armamentarium, knowing that we'll be much more likely to achieve a Cmax target than an AUC target with once daily dosing.
Mm.
Amanda: And can I just add that we shouldn't ignore troughs as well for aminoglycosides- Yeah ... just, you know, in terms of its toxicodynamics and risk of, you know, cochlear toxicity and nephrotoxicity.
it's an indicator of drug accumulation, and I don't think we can completely remove that at this point in time. It makes it for a very tricky therapeutic drug monitoring when, peak AUC and troughs are all relevant, but, frustratingly, they all are to some degree.
Angela: Yeah. I mean, I think just for listeners, we rarely now use aminoglycosides more than once daily, right?
When we're using them therapeutically and not in this sort of voodoo-esque synergy for endocarditis and whatnot, we use them once daily. And if you're using them once daily, you don't have to do [00:47:00] peak levels, right? But you must do trough levels because you wanna make sure they're not accumulating. And, yeah, I think, People stumble over that, because we've made the switch to once daily. trough levels still matter, but they're really for toxicity.
okay. Finally, vancomycin.
Angela: What MIC denominator do you use for dosing? Amanda?
Amanda: I think everybody is going to say one milligram per liter, I don't know. mainly 'cause that's the default easy target and it gives you your target range of, an area under the concentration time curve for 400, which is the number that everybody uses as their therapeutic cutoff.
Having said that, for coagulase negative staph, the susceptibility cutoff is up to four. and we did a, PK/PD study in young infants looking at CONS and we didn't find a clear relationship with the MIC. We just found that, higher microbiological cure, if you got an AUC greater than 400.
So [00:48:00] I don't know for CONS. Yeah.
Sebastian: Agreed. maybe one thing to add, um, for MRSA, clearly the broth microdilution MIC
is usually one or less, so I would also use one. There is one thing if you would use the E test there, even with an MRSA, you sometimes get readings of 1.5 or two, but the target is for broth microdilution MIC. So that's why a one is, I think, a good pick in those cases
Rekha: Can I ask a quick question? about targets because, you know, you said 400. A question for everyone. what AUC targets- Mm ... are you using, for CNS infections, for vancomycin?
Would you go up to 700, for example?
I just took a very, pragmatic approach that we used to aim for a higher trough, so we can aim for a similarly higher AUC.
Rekha: Yep.
Amy: cause kids technically I think you're a bit happier to go up to 650 in kids, Amanda?
Amanda: Yeah. We're actually just doing a study looking at vancomycin associated nephrotoxicity in [00:49:00] AUC. But I definitely wouldn't tend to, for a CNS infection, go up to an AUC of 600 to 700 just because of the risk of nephrotoxicity.
In general, if, we've got a, reasonably high MIC, I'd switch drugs. Yeah.
I think that nephrotoxicity is sort of time-dependent, right? If you're only using vancomycin for a few days, then it's sort of, you know, go hard and go home, right?
Angela: you have less fear about toxicity if you know you're just gonna be giving it for a few days or, you know, obviously you're not gonna let it linger at such high doses for a really long time.
Sebastian: Yeah, I agree. I mean, many of those toxicities develop over time.
It's not only vancomycin, it's also linezolid toxicity, it's aminoglycoside toxicity. So if it's really just one or two or three doses, the risk is lower than if you have a longer treatment course, yeah. And, again, putting priorities which patients should undergo MIPD, those that are at higher risks, yeah.
So there, should be the investment.
Amy: Before we wrap up, any final messages you wanna [00:50:00] give to our audience about MIPD?
Amanda?
Amanda: I think it's just time that we use it. I think, you know, there is now good evidence for it, and our patients deserve to be, dosed correctly when they've got serious infections, and I think it is the safest, most effective way to give them treatment.
That was inspiring.
Amy: Sebastian, any final messages from you for our audience?
Sebastian: Yeah, similar as Amanda. if you still do troughs for vancomycin, it's probably time to change. the tools are there, and it's not difficult to implement this.
There are free tools outside, and it's just about making a difference and making, for vancomycin for example, the, the treatment more safe.
Amanda: Sebastian and I would be happy for any of the listeners to reach out about TDMX or KidsCalc if they're interested in hearing more about it.
You know, the reason why, our tools are freely out there is because they're educational tools, and we want, people to [00:51:00] learn how to, dose their patients accurately. Actually, on the KidsCalc website, there's a contact us email if you have any questions. yeah, we're always here.
Rekha: I can attest that both Sebastian and Amanda are extremely accessible and have been wonderful throughout my last few years of fumbling through MIPD or any TDM questions.
Amy: And for anyone looking to upskill, I think maybe we can also plug, ESCMID
yeah, that's a good point. We, every year or every second year do a course on TDM, PKPD, and therapeutic drug monitoring.
Sebastian: and this is available in the, online library of ESCMID. And, uh, of course, there will be new courses, organized, for instance, by EPASG, in the future, which is probably a very good place, to be to learn about this and get some hands-on experience.
Amy: And I am prepared to come to Europe and help with that if you need.
and similarly, I should also plug that in Brisbane, although it does move around, [00:52:00] there's a population PK seminar every year that's, both to teach people how to model but also how to understand MIPD.
Thank you so much for the conversation, and thank you for listening to Communicable, the CMI Comms podcast. This episode was edited by Katie Hostettler-Oi.
Rekha: Theme music was composed and conducted by Joseph McDade. Any published literature we've discussed today can be found in the show notes. You can subscribe to Communicable on Spotify, Apple, wherever you get your podcasts, or you can find it on ESCMID's website for the CMI Comms Journal. Thanks for listening and helping CMI Comms and ESCMID move the conversation in ID and clinical microbiology further along.