Interview with Dr. Marko Notar, Smart Blood Analytics

Good morning and welcome to today's interview with Dr Marko Notar, CEO of Smart Blood Analytics Research Foundation in Zurich, Switzerland. Dr Notar, thank you very much for joining us today to discuss setting a new standard for infection diagnostics and delivering precision medicine at scale through AI. 

Thank you very much for inviting us to interview as well. 

So, what makes rapid and accurate differentiation between bacterial and viral infections so important in the fight against antibiotic resistance? 

Yeah, well, thank you for the question. Well, I would say the fact is that every time we prescribe antibiotics unnecessarily, we somehow strengthen the enemy we are trying to fight, yes, which is antibiotic resistance. And today roughly up to half of antibiotics worldwide are still given to patients with viral infections where they provide absolutely no benefit. And this misuse not only exposes patients to side effects but also accelerates the rise of drug-resistant bacteria, which we all know is one of the greatest threats to modern medicine. And if I may say so, the urgency of this problem was underlined at the G20 Health Minister meeting in Bali in October 2022, when and where antimicrobial resistance was recognised as a global health and development priority. And at those meetings, the leaders highlighted the needs for stronger antibiotic control and innovative diagnostic solution to reduce misuse. And actually, that's exactly what rapid and accurate differentiation enables, for example. Yeah, it gives doctors the confidence to prescribe antibiotics only when they are truly needed and to protects patients from unnecessary treatment. And if I may, what I believe is the most important contributes directly to the global effort to slow the spread of the resistance. 

Fantastic. Thank you very much. Why are traditional markers like CRP and procalcitonin often reliable on their own, especially in certain CRP ranges? 

Yeah, well the most straightforward answer to your question is because the body's inflammatory response is messy and time dependent. But if I go a bit more into details, CRP and procalcitonin tell only part of the story. So, C-reactive protein is very general. It rises not only in infections but also in conditions like autoimmune disease, cancer or even trauma. And it also rises slowly which makes less useful early in infections. Yes. On the other hand, yeah, procalcitonin is somewhat more specific but it has a blind spot also that's worth to mention. It can be elevated after surgery. Trauma mentioned and threshold are differently by pathogen and settings, and together these limitations mean that both markers overlap somehow heavily between bacterial and viral infections, especially in the grey zone of CRP between 10 and 40 milligrams per litre. And that grey zone is exactly when clinicians are most uncertain. Yes, the doctors are uncertain in that in that range of CRP and why are more multi factorial host signature approach is needed to see the full diagnostic picture. 

Thank you very much. So moving on to the virus versus the bacteria solution. How does the virus versus bacteria software work and what makes it different from existing diagnostic tools? 

Yeah, again, the shortest answer to that question is that we give routine blood test a second life. Our virus versus software helps doctor quickly tell if an infection is caused by bacteria or virus. It uses the results of routine blood tests plus age and biological sex to make its prediction very easy. Instead of relying on just one marker like CRP or procalcitonin, it looks at many blood values together at the same time and this makes it much more accurate in spotting subtle difference between the two types of infections. And I would like to mention that the system was trained on more than 44,000 real patient cases, so it reflects actual medical practise and powered by advanced machine learning, it can reach about 83% accuracy which is significantly better than standard methods. And another point, its largest strength is in the so-called grey zone or dimensioned. Their traditional test often fails to give clear answers. What I need also to say is that by providing a clear diagnosis, the software helps doctors avoid unnecessary antibiotic prescriptions. And this is especially important in the global fight against my antibiotic resistant. Overall, in a short, if I may to answer, it turns everyday blood tests into a smart tool that improves care and protects public health. 

Yes, absolutely. Thank you. And why was XGBoost chosen as the core algorithm and what performance improvements has it delivered? 

Yeah, thank you. It's a bit of technical question maybe more related to our machine learning. But I'm trying to answer, we selected extreme grading boosting as the core of our learning algorithm because it's one of the most trusted high-performance approaches in modern machine learning. Yes, since gaining global recognition, if I'm not wrong in 2016, XGBoost has consistently delivered top results across scientific research and the real-world applications, including medicine where we are present. And for us, it has several decisive advantages. And I will list only a few of them. First is speed and efficiency. The fact is that XGBoost trains quickly, runs smoothly even on small devices and scales to large computing environments quite easily. The second topic is transparency. We built incompatibility for shapely explanations. Doctors and regulators can easily understand how the models make decisions. And the final things if I may need to mention regarding XGBoost is robustness with real data. Unlike many alternatives, it naturally handles missing information which is crucial feature when working with patient blood results when many times you don't have full access to the full blood parameters. 

And also, another point, compared to other popular methods such a random forest or deep neural networks, XGBoost gave us a clear performance edge in our virus, virus versus bacteria study. It produces compact, reliable models that were both faster and more accurate. Based on our scientific articles. I can give you just a few examples, up to 3% higher overall accuracy, up to 6% improvement in the crucial CRP range between 10 to 40 milligrams per litre and the definitely best in class like lowest Brier score and highest AVC. And if I meant to say these results were not just the numbers, they help us secure MDR medical device certification, which is very important. I mean, coming to the market proving that our technology meets the highest standards of safety and reliability in our healthcare. 

Fantastic, thank you very much. And then looking at performance and the clinical impact, how does explainable AI help clinicians understand and trust the model’s recommendations? 

If I may, I would start this way. Yeah, the paranoia and fear of using artificial intelligence in medicine is still on a very high level. And do you recognise that the trust is the foundation of medicine? That's a fact. After so many years, we seem so paranoid and I would say so different approaches in that field and that no clinician will rely on an AI system unless they understand why it made a specific recommendation. AI can feel like a black box and at some point of view it is a black box, but we deliberately opened that black box as much as possible. Well, as mentioned with Shapley explainability, our platform highlights which laboratory values and patient parameters had the greatest impact on the prediction. And this allows doctors to critically evaluate whether the AIs decision aligned with their knowledge. On the other hand, and importantly, the AI can also surface connections and early signs that are not yet taught in textbooks or covered in in any medical school. The fact is that the medicine evolves faster than traditional education cycles and our explainability framework ensures that when the AI introduced such novel insights, physician can clearly see how and why these new markers influence the outcome. This transparency transforms artificial intelligence from a mysterious algorithm into our best clinical partner and on that both validates current knowledge and responsibility and responsibly introduces the medicine of tomorrow. 

Thank you. And I think that leads quite nicely into the next question - can you share an example of how the software could change a treatment decision in a real-world clinical case? 

Yeah, for example, the autumn is coming in and that is also the period when we observe more viral and bacterial infection, right. For example, a patient arrives with fever, cough and CRP of 25 milligrams per litre. Today many doctors would prescribe antibiotics just in case to be a safe if I may to say so. Yeah, our software analysis the blood parameters and identifies a strong viral profile, for example, delivering a high confidence viral result. The doctor is then reassured to withhold antibiotics, sparing the patient unnecessary treatment and helping preserve antibiotics for future patients. You may ask me, yeah, one case may seem small, but multiplied thousands of times, the impact is transformative. 

Absolutely. Thank you very much for that. And then thinking further in terms of the implementation and adoption, how can the system be integrated into existing hospital or clinical workflows and what are the data protection measures that are in place? 

Yeah, maybe let me first answer to the integration and workflow. Our system is designed for seamless integration into hospital workflows. It can be used directly as a secure web application or connected to laboratory system and electronic health records via API so that results appear automatically without requiring extra steps from doctors, clinicians, whatever we say and answering about your question. Speaking about data protection and privacy, well, our system is fully GDPR compliant and built on strictest privacy and security standards. Importantly, the algorithm does not receive any personal patient identifiers. As I have mentioned during this interview, I've received only blood test values, biological sex and age are processed. So no names, no contact details or other identifying data are ever transferred. And definitely this ensures that patient privacy is safeguarded by the design. And what is also worth to mentioned, all transmitted data is encrypted, how would I say intransit and at rest anonymised where applicable and also protected by strict access control. And at the very end, security and privacy are not add-ons. But regarding that CE marking and certification have been implemented into the system from the day one. 

Fantastic, thank you very much. What role has CE certification under MDR Regulation (EU) 2017/745 played in supporting clinical adoption? 

Yeah, thank you for this question. And on this place, I think it's worth to mention that we were the first company in the field of blood test results interpretation to acquire the CE certification under the new MDR directive. And of course, we are very proud of that. CE certification is a powerful enabler and it's a matter of all further steps. That's true. And speaking about the market, it's also true. It shows that our software isn't just innovative. It means that our tool has met Europe's highest regulatory standards for safety, quality and clinical performance. 

Crucially, the CE mark also give us the legal right to place the product on the European market, bridging the gap from research to real world adoption and the I would say the most important, at least for hospital and clinicians. This certification removes uncertainty and provides confidence that the tool is both effective and official approved as medical device. When speaking about the other markets beyond Europe, the CE mark acts as a global quality seal, helping to accelerate approvals and adoptions in other regions. 

Great, thank you very much. And then thinking about the impact on the future outlook, how does this technology contribute to better antibiotic stewardship and improved patient outcomes? 

Very simple. It puts precision into every prescription. That's the fact. By accurately distinguishing bacteria from viral infections, clinicians can prescribe only when antibiotics are truly needed. Invalidation, for example, our approach reduced inappropriate antibiotic use by 20 to 30%. That's a huge that directly supports, I would say global stewardship goals, protects patients from unnecessary side effects, which we can easily say trigger other comorbidities and helps preserve the effectiveness of antibiotics for the future while improving care for today's patients. 

Fantastic, thank you. And then lastly, looking ahead, what are your priorities for further developing or expanding the virus versus bacteria model? 

Yeah, thank you for that question. It's a bit tricky, but I'm trying to answer. Our next frontier is tropical medicine. It's not a secret. Malaria, Dengue and other febrile illness remain the hardest diagnostic challenges and the places where our model can have the greatest impact. Yeah, high quality data in these conditions is short, that's also the fact. But we would like to build partnership with hospitals and research networks across endemic regions to somehow could create datasets needed to unlock reliable artificial intelligence support in these settings. At the same time, it's not a secret. We are preparing for clinical adoption beyond Europe and the United States where we are already present. And with the CE marking and certification as our foundation, we are engaging with regulators and health system in Africa, Asia and Latin America and also in the Middle East to ensure safe effective integration into local healthcare systems. I would say our vision is clear to make an advanced clinical decision support system availability to our clinician, especially where uncertainty is greatest and the stakes are the highest. 

Yes, fantastic. Thank you very much. And with that, we will close today's interview. A very big thanks again to Dr Notar for your time today for sharing such interesting insights into your very significant and impactful work at Smart Blood Analytics and wishing you every success for the future. Many thanks again and take care. 

Thank you very much. I highly appreciate that the discussion was fruitful also from my side. 

Thank you very much. 

Thank you.