Hello Friends đ
We have a new episode of Tech Tuesday for you this week where Daphna and I chat with Thorsten Waloschek, CEO of NeoPredix a company building predictive tools for the NICU. We have been trialing their bilirubin prediction app (Neopredix.b1) in our unit and have been blown away by how innovative this application is and by how much value it provides our team. We hope you enjoy this conversation. All the info you may need about Neopredix are tagged below.
Enjoy!
----
You can get in touch with Thorsten Waloschek via email: thorsten.waloschek@neopredix.com
Or contact Lia Butler, US director of sales at: lia.butler@neopredix.com
----
The transcript of today's episode can be found below đ
Ben 0:54
Welcome. Hello, everybody. Welcome back to the incubator podcast, we have another episode of Tech Tuesday. Ready for you Daphna? How are you this morning?
Daphna 1:07
I'm doing really well. You know, we have been trialing this, this exciting product. So I'm glad we're getting a chance to talk.
Ben 1:18
Yeah, we're a little we're a little biased. We are a little biased. And so
Daphna 1:22
we always told me told our, our listeners that we will be totally transparent about the things that we bring onto the podcast. Yeah.
Ben 1:31
We're never we're never bringing on anybody, especially on Tech Tuesday, where we have a vested interest in how a company is doing. But for example, in this case, we are we have the pleasure of having with us today, Torsten Wilczek, who is the CEO of a company called Neo predicts. And it is it is a company that has a very interesting tool to to make predictions about bilirubin levels. We'll talk about that. Torsten, thank you so much for being on with us this morning.
Unknown Speaker 1:58
My pleasure. Hi, Daphna. Hi, Ben.
Ben 2:01
How are you? So for people who are not familiar with Neil predicts? I'm going to give you the mic to give us like, a quick intro as to what it is and what is its intended use?
Speaker 3 2:15
Absolutely. So Neo predicts, we are a Swiss German digital health startup, we are a spinoff from the University of Basel in Switzerland. And your project is built on the foundation of uniting the most up to date technologies, with clinical decision making to deliver precision medicine to all patients under the maternal neonatal and Perinatal umbrella. And so we are very focused on the first 1000 days of life. And here in Europe, it's not a standard expression, what we mean by the first 1000 days of life is from conception all the way through the second birthday. But we have an even more special focus on the time around birth. And we are very focused on that. And we want to help clinicians to make more informed decisions. So we claim to have the leading platform in predictive maternal and neonatal clinical decision support tools.
Ben 3:16
I think that's very exciting, because there's a lot of talk about a lot of talks about chargee, PT, artificial intelligence. And I think in the medical community, there is a sense of like, well, how, what does that mean for us on a day to day basis in terms of how can we leverage these tools for taking better care of our neonates? And I think, as we've talked on the podcast before, some of these tools are like the early onset sepsis Risk Calculator and things like that. But how do we actually bring in predictive algorithms into the NICU? I think Neil predicts is the company that is trying to do that on multiple fronts. Now, for the sake of brevity and for clarity, I think it's important to mention that new predicts is a large umbrella company that's going to try to focus on predictive algorithms for multiple aspects of neonatal care. But at this time, if people are interested in your predicts the tool that you have available, which is a tool that Daphna and I are using in our NICU is the tool to predict levels of bilirubin Is that Is that correct?
Speaker 3 4:19
Yes. So the way I like to put it as we can predict the dynamic progression of bilirubin and how we do this, and I think this is part of your question, and you mentioned sheds, GPT and other tools. In my mind, honestly, there's a lot of hype around that not only in the medic community, but all over the place, right. So I would like to kind of scale it back a little bit because in my mind, we didn't use we didn't call it artificial intelligence 20 years ago. We called it fuzzy logic, for example, but decision tools, whether in the medical environment or outside the medical environment has been around for quite some time. Now And I think that that hype will come down again. And then we are able to really focus on the on the outcome, which in our case is what we want to bring to the table, to our customers to clinicians in the end to the benefit of patients, is to really have more informed decisions. And our approach is a combination of physiological data, biological data, and pharmacological understanding, plus mathematical expertise. And of course, to make sense out of all of these tons of data points, we need to use tools like artificial intelligence, machine based learning. But in the end, it's math. And by combining all of this expertise, we are able to provide our customers to clinicians, a sneak preview into the future. And if we and I'm totally convinced that as an added value, if we can help you clinicians to understand what will happen within the next 24 or 48, maybe 60 hours, we hope to put you into a more proactive situation, because then you don't need to wait for symptoms, you can proactively addressed situations that will occur in the near term future.
Daphna 6:18
And I think that's really the difference between say units, or maybe using the Billy tool website, that this can show you data points over time for a particular patient, but then it actually shows you what is anticipated to occur over the subsequent 60 hours. Can you tell us more specifically like what that looks like on the internet? Absolutely.
Speaker 3 6:42
So we and the first tool that we started a few months ago to commercialize in United States, that's the one you're referring to is what we call new predicts be one bilirubin prediction. So we can predict the dynamic progression of bilirubin for up to 66 Zero 60 hours into the future based on the last bilirubin measurement in the hospital. Now there's two ways to measure bilirubin nowadays, at least, there is total serum. So blood based and that's Transcutaneous. Measurement of bilirubin through the skin, we can use both types of data points, we just need to know what it is if it's TSB or TCB Transcutaneous or total serum. And at the minimum, we need two data points. If we have three or four, it's a little better, but we can work with two. And then we need just a handful of physiological data from the specific patient. birth weight, date of birth. If that is not available, point zero, we don't we don't care about the date. It's just we need to zero our somewhere. And the birth weight I mentioned gestational age at birth. And very important, the mode of delivery C section versus vaginal makes a huge difference plus two or more bilirubin data points. And putting that together, we can actually forecast predict the next six hours and we don't predict an endpoint, we really show you a dynamic curve. Basically think about it as a set in six minute increments. And of course we put a curve through it. But it's a step by step by step prediction over the next 60 hours into the future. So if you have to make a discharge decision, after let's say, I asked you what is the typical discharge age have a, you know, 38 weaker? vaginal delivery? Everything's fine.
Ben 8:28
Yeah, I would say 48 to 72 hours of life is when that discharge will usually happen.
Speaker 3 8:33
Yeah, so let's say at 48 hours approximately, you you are at the point where you think about a discharge decision. And at that point, if there's two measurements available, we can show you how for this particular patient will ruin levels will develop into the future. And based on that you can make a discharge or not discharge decision that's yours. That's not ours. We don't indicate anything, we just show you how that will develop and potentially prepare the patient, the parents in our case for why a follow up within the next 24 to 48 hours. And we absolutely follow the new AAP guidelines, but hyperbilirubinemia. So according to that you can actually also better plan the the at what point of time parents should bring back the baby, because you pretty much know what will happen in the next 48 hours. Right?
Ben 9:29
I think you're you're describing a tool that that sounds really good. And I think it's important for me to mention that as someone who's used it. The tool looks really good too. So for example, the graph is so comprehensive because you you'll have your data points that that you've collected, whether it was the CRM or the cutaneous measurements, and then you'll have this this forecast with with this for this forecast, which has like a range of where the prediction model will tell you the bilirubin should fall into it. coming days. And, and what's interesting is that you can overlay on top of that the newest phototherapy guidelines. So you will be able to see whether a baby technically has a high risk of crossing the threshold and really needing for therapy, you can generate a PDF out of this, you can bring it into the room. And I think when you're when parents are asking you like, well, but why why are you not discharging us today? Or why are you discharging us? If it's not safe, then when you bring this visually, it is so easy to explain, and so easy to understand that for us, at least, it's made conversation with families, much more streamlined. I guess, the follow up question is, how do you guys get this prediction? What kind of data is this based on? And? Yeah, I'm curious if you can answer that question. Yeah.
Speaker 3 10:50
And first of all, I would like to say thank you, because you just basically confirmed what we intended to do. And in the industry, and as a as a as a tool provider, if you will, that as a medical device, or as a tool provider. Of course, it's very rewarding to get that feedback that because we can develop whatever we want to what we think is great. But if it doesn't help you in your daily practice, it's worthless. So I'm very, very happy for the feedback. Thank you for that, too, to answer your question. And so of course, we had to train first to train the algorithm and then to validate the algorithm. That to send a practice, of course, in the industry, and also outside medical. So we use the initial datasets that came from to university hospitals in Germany, and in Switzerland, to really understand the dynamics of bilirubin. And I think one of the key findings was that there's dozens of influencing factors. And you'll know better than I do, which ones those are dozens of influencing factors that have an impact on the bilirubin levels. But based on deep data research, we came up with this very short list that I just mentioned, gestational age, birth, weight, etc, etc. That really make a difference. Pretty much the rest is only background noise. So I think that was the first huge step in our development that we were able to, you know, don't look at all of the noise in the background, but really look at the data that is important that really makes a difference. So that was step one. Again, we used a couple of 1000 data sets from babies in Germany and in Switzerland, then to get a better understanding. Because Germany, Switzerland, of course, has a very similar patient population. We looked at the middle Mediterranean population in Europe, because now I have to think slowly and speak slowly, because I always mix it up in the Mediterranean population, there's a significantly higher incidence of G six PD levels, right? Correct. Then in the European or even the American population, so we connected with a university hospital in Greece, and received a very large dataset of Greek babies, making sure that we cover that specific incidence. And then because we just wanted to make sure, and especially because of the Transcutaneous measurements through the skin, the skin color can have an impact not on bilirubin itself, but on the measured value. So we wanted to make sure that we also cover that. And that's we connected with a hospital in Kenya actually, to make sure to also have data from that population. And we very strictly separated the datasets which we use to train the population from the datasets we use to validate the popular the algorithm. Last but not least, because the regulatory authorities in Europe versus the United States, when it comes to predictive analytics have very different approaches at the moment. So here in Europe, we had to run a prospective clinical trial, which we did, which generated a new dataset, and was a multicenter clinical trial in Germany, which has been published which is available, which again, we used for new validation. So actually, it was trained with various datasets. With new data sets, we ran the first validation and with the new prospective study that generated completely new data. We validated the second time. And now since we are active in United States and have first customers and reference sites, we collected American data and use that again, for another validation, as we say validation never ends. So that was the step by step approach, making sure that we have a really large sample size, a couple of 1000 babies from various population populations around the world.
Ben 14:56
And I think that's a very important point because as clinicians that validation is something that That puts our minds at ease. And I think these these articles are readily available if you if you look for them, I think, in frontiers journal called frontiers is where you'll find that the latest one, there's also publications in the Journal of Pediatric research. So there are publications that have looked at this in sequential timeframe. So it's very, very reassuring. Absolutely, I
Daphna 15:28
actually appreciate you walking us through that process. I feel like I learned a lot about just how models are validated. So I think that's really neat. As someone who's actually not that tech savvy, I find the tool very easy to use, also, you know, easy to kind of store the data, maybe that is a concern that people might have about storing patient data. And maybe you can speak a little bit to that as
Speaker 3 15:58
well. Of course, that is a very valid concern, and we should be concerned. And so and that also brings me and I will answer your question. And I will add a second answer to that, because that's also an important part when we all currently in this hype, as we address it, as I called it, think about artificial intelligence. And But to answer your question directly. So we need we want to store data to for a certain amount of time. So you as the user, as clinicians can go back to it and look at it again, because you're discharged a baby, and maybe for whatever reason might be readmitted and that data is gone, that wouldn't be helpful. So for a certain amount of time, we we store the data on our service servers, HIPAA compliant servers in the United States. But and that is, depending on customer hospital preference, in general, we suggest to delete the data, 14 days after discharge, unless the customer tells us differently. But this is the way that we propose it. And that's what we do. And we are contractually obliged to do so. So we don't store data in the, in the long run, not even in the midterm, two weeks is not long. And that also shows that we don't use any patient data to continuously change our algorithm. And that makes a huge difference. As I mentioned, we use all of these various tools to develop the algorithm and to train the algorithm. But once we decided, okay, that's, that's it, and then it was successfully validated, and we and our customers, and the regulatory authorities are happy with the result, we as I said, we close the algorithm, it's not a self learning tool. And that makes all the difference, because it will always behave the same way. So you are nobody else needs to be concerned, anybody else doesn't need to be concerned about Okay, today, this algorithm behaves this way. Tomorrow, 55 more babies are in the algorithm. And that might change the algorithm will not happen without us. If we feel that we have any indication that we can further improve our algorithm, make it even more specific, or maybe extend the prediction timeline. I mean, there are various ways to continuously improve it, we will release a new version of the algorithm, but it will not be a self learning algorithm. And that makes a difference. And I can only suggest that not only us specifically, but anybody who deal is working with algorithms always ask yourself the question, well, is that self learning, which doesn't mean it's bad, it's bad, but it means that it will behave the next 10 minutes different than 10 minutes before because it's self learning continually changing. What we do is it's closed, it's it's stable. And if we feel we can improve it, we will and then release a new version.
Daphna 18:54
Yeah, and maybe to clarify from the user, and the value and quote unquote, storing data is really those bilirubin points for a specific patient. So you open it, you document it, at one time point, you come back six hours later, the time point lives is there, you can put in the next time point. So without having to re input all of the original information. And you can use a bed space number you could use a you know, a random patient identifier. So there are lots of opportunities that way to keep patients you know, straight on on on the app.
Speaker 3 19:36
Absolutely. And what we also offer is integration if wanted, if requested into electronic medical records. So we started working with epic and we approved third party supplier to Epic. So if a customer is using epic and wants to integrate as their various levels of integration, or actually my IT team taught me many times don't say integration, say connectivity. So there's various ways of connecting to Epic, or similar systems, which then really is a very smooth integrate integration, in this case, into your daily workflow. Because the moment you look at the patient data, you just basically say, I want this baby to be to see the baby room prediction, we pull the data unnecessary and push back the result to you and show it. Again, there's various ways to do that, in your workspace, in your screen, the way you're used to it, we started working with epic, the next step will be Cerner. In Europe, we started working with SAP, there's very many different versions of electronic medical records here in Europe, which makes it a little more difficult. And on top of if that is not wanted, or if it's not possible, because some customers use some unique systems. Then we also offer a standalone platform, which you can access from any internet connected device. Again, as I mentioned, everything's hosted on HIPAA compliant servers in the United States.
Daphna 21:11
Well, I know we're getting to the, the end of our session with you, we've learned so much, I guess maybe without divulging any major company secrets, you know, what's, what's next, like? What's on the horizon for predictive models? In medicine? And, you know, especially our listeners in neonatal care,
Speaker 3 21:32
yeah, I love that question. And my team always tells me okay to us, and don't say too much, but I want to because I really love I'm, I dedicated really the last 20 years of my professional life, to newborn care, or specifically didn't really need to care. And I developed over the time, my view will personal mission. And this is to make sure that every baby has the best possible start into life. And and whenever I get engaged in projects and companies in ideas that that way, I see, well, this is the long side this mission, and then I'm just super excited and constantly want to talk about it. So of course, I will answer your question. And we are a participant in the Mayo Clinic accelerate platform program. And with support from the Mayo Clinic, we are currently developing a maternal predictive algorithm. And as I mentioned, we are looking at the first 1000 days of life, which include from the point of conception, and our focus at this point is at pre on preeclampsia. So I cannot talk about how that will work. And for exactly, exactly how long we can predict and what exactly we can predict. But this is our next big project we are currently working on to develop a predictive algorithm on the topic of preeclampsia. And the interesting thing is that this idea of using pharmacometrics, AI, machine based learning, and really deep medical understanding, and the founders of the companies, moreover, medical doctors is that it can be applied to so many different areas. Again, we focus on the first 1000 days of life. And we'll make sure that step by step we are providing a platform, not all of the tools will be predictive. But all of the tools will be applicable for this very special time of life, for human being. And all of the tools will be clinical decision support tools that help you to understand data points better to sometimes look into the future, but always with the idea to help you to improve medical outcome, and to put you into a proactive situation to make decisions. So next step, preeclampsia. And we are working on a handful of additional projects. As a small startup company, it's a phased approach, step by step, we can't do everything at once, despite the fact that I would love to do that. But if you think about the top 10 challenges around the time of birth, or specifically for newborn babies, think about the top 10 on five of those top 10 topics we're working on. Saving.
Ben 24:22
Yeah, and and you alluded to this, your your track record in the in the not just the healthcare space, but the neonatal health care spaces is impeccable. And I think you have a long track record of excellence in that in that area. So I think it only adds to, I think the credibility of the work you're doing and the quality of the work that you're doing as well. I think the question that I just wanted to close with is, how easy is it? Let's say somebody is listening and says I'm actually quite interested. We have so many babies where we're always unsure what Should we be doing at around the time of discharge? This tool could be helping me? How do I get this in the hands of my clinician? What is the best way to get this done? Torsten?
Speaker 3 25:10
Yeah, very simple. Our website new products.com offers a very simple site where you can sign up for a free of charge demo. Since this is all digital. And of course, in the digital age, we don't need to travel for that, we can set up an online appointment, we have a team in United States as well as in Europe. So even time difference doesn't matter that to us. So whenever whatever is convenient for our customers, we can set up a time within 48 hours really short notice to show whomever was interested in an online demo by via Microsoft Teams or Zoom again, we are very flexible to give a first Outlook to give a first taste. And if this is still of interest after that first demo, which even with question answer takes maximum 30 minutes, we can also do it in 20 minutes. If the timing is very tight. If then there's more interest, we can set up that specific clinic with a demo account, give clinicians access to our system, we offer a minimum four weeks free of charge testing of our system, which is absolutely Unlimited, there is no no no loops, no hoops you have to go through, you get an access code. And it's in the protective environment. Again, HIPAA compliant, no storage, no long term storage of data. And if a clinic decides for whatever reason, that is not not for them, all data will be deleted, no harm done and no expenses, except for time invested. Unnecessary. And if after a few weeks of a demo access, the customer decides this is something that is of interest, and they want to use long term, then of course we make an offer. Again, depending on the level of connectivity to local EMR systems, it might take a little bit of time to get it up and running. But to give an access code and give access to the system can be done in less than 24 hours. So very straightforward, very easy and very flexible.
Ben 27:16
We can confirm that it is very easy and straightforward. Yeah, Torsten, thank you so much for making the time to chat with us today. I think this is this was very helpful. Congratulations on on the really a beautifully designed platform that is hitting it on the dedicated hitting the nail on the head with what clinicians are looking for. We'll have all the information that we talked about on the episode page. And please, when you anybody interested, please feel free to contact Torsten and his team. To find out more about Neil predicts be one first and thank you so much.
Speaker 3 27:47
Thank you so much was my pleasure. And as I mentioned, I'm super happy to get positive feedback. Thank you so much. And thank you for taking care of babies. This is really in my heart and my soul. Thank you. Thank you.
Ben 28:02
Thank you for listening to the incubator podcast. If you liked this episode, please leave us a review on Apple podcast or the Apple podcast website. You can find other episodes of the show on Apple podcast, Spotify, Google podcasts, or the podcast app of your choice. We would love to hear from you. So feel free to send us questions, comments or suggestions to our email address NICU podcast@gmail.com. You can also message the show on Instagram or Twitter at NICU podcast or through our website at WWW dot d dash incubator that org This podcast is intended to be purely for entertainment and informational purposes and should not be construed as medical advice. If you have any medical concerns. Please see your primary care professional. Thank you
Transcribed by https://otter.ai
ăłăĄăłă