Conversations with the Editors: Artificial Intelligence–Based Technologies Leading the Innovation in Surgical Care

Thomas Scheeren, MD, PhD: In almost 30 years of clinical activity, Dr Scheeren has worked at 3 academic university hospitals in 2 European countries, getting a broad overview and being involved in multiple teaching activities in the fields of anesthesiology and intensive care medicine. In 1999, Dr Scheeren obtained his PhD from the University Hospital in Düsseldorf, Germany. His PhD thesis included numeral studies on cardiovascular physiology. Consequently, he obtained his qualification for professorship (associate professor, Privat-Dozent). In 2004, he was appointed as a full professor of clinical and experimental anesthesia at the University of Rostock, Rostock, Germany. In 2010, he was appointed full professor of anaesthesia and cardiovascular physiology at the University Medical Center in Groningen, the Netherlands. He currently leads an active research group on cardiovascular physiology and hemodynamic monitoring at this Dutch university hospital. His long-lasting interest in the fields of cardiovascular anesthesiology and physiology, particularly focusing on hemodynamic monitoring, including microcirculation and tissue oxygenation, has led to >180 peer-reviewed publications in international scientific journals and numerous lectures at national and international scientific meetings. He is editor-in-chief of the Journal of Clinical Monitoring and Computing and associate editor of the journal Anesthesia & Analgesia.

Luciano Ravera, PhD, MBA: Dr Luciano Ravera graduated cum laude in economics at Bocconi University of Milan and holds a master's degree in business administration from Harvard Business School. He became part of the Humanitas Group in 1994, after working at Barclays Bank and Booz Allen & Hamilton. From 1999 to 2005 he held the position as managing director of Humanitas Gavazzeni (Bergamo) and managing director of Istituto di Ricovero e Cura a Carattere Scientifico (IRCCS) Humanitas Clinical Institute in Milan, Italy. From 2005 to 2009 he held the position of managing director of Humanitas Mater Domini (Castellanza – Varese) and director of strategic planning and business development of the Humanitas Group Spa. He has been member of the board of directors since 2006 and chief executive officer at IRCCS Humanitas Clinical Institute since 2010.

1. Carlo Federici, PhD: Professor Scheeren and Professor Ravera, thank you both for your time and for agreeing to this interview. I would like to start by asking you a general question. In your opinion, what are the most relevant benefits for the patients of adopting AI [artificial intelligence]–based medical technologies in healthcare?

Thomas Scheeren, MD, PhD: “Well, I have experience with 2 AI-based technologies. The first one is the Hypotension Prediction Index (HPI) and the other one is the assisted fluid management. Hypotension, also known as low blood pressure, is a problem. It frequently occurs in perioperative settings both in the operating rooms and in the intensive care units. This condition is associated with adverse outcomes for the patient, such as myocardial injury, acute kidney injury, and cerebral injury. Thus, many vital organs are affected and death cannot be excluded. The HPI may reduce this frequently occurring hypotensive burden.

Starting with the way the care is delivered, it would be a paradigm change because as I have just said, I'm an expert in hemodynamic monitoring, which entails describing or measuring the current hemodynamic state of the patient (eg, blood pressure in this moment). Now, with this new technology, we can look into the future, the near future, and predict what will happen in the next minutes. This is a game changer, just like weather forecast is, because it allows to know whether to take an umbrella with us or not. In this case, we can predict the onset of hypotension, and, consequently, we gain time to think about the underlying mechanism, an appropriate treatment, or call for extra help and so on. In this way, we are ahead of the potential complications. Secondly, this technology allows us to change from a reactive approach (ie, waiting until hypotension occurs to treat it) to a proactive one, aimed at preventing or limiting it by prophylactic treatment. Thirdly, this technology allows the shift from a specific blood pressure management to one that looks after the etiology of such condition. For example, hypotension could be due to hypovolemia, loss of vasomotor tone, or a heart dysfunction. Each cause has its specific treatment. Therefore, knowing the cause is essential to deliver the right treatment.

To this regard, along with the HPI, there is the so-called secondary screen, which gives you information about the reasons behind the hypotensive event in that specific case. As I said, this allows a specific treatment of the underlying cause of hypotension instead of an unspecific one-size-fits-all approach, which could be inappropriate and deleterious for the patient. As a consequence, the incidence of the related adverse outcomes/complications may be decreased. In a recent study by Davies et al.Ability of an arterial waveform analysis–derived hypotension prediction index to predict future hypotensive events in surgical patients. (see Figure 1 for a summary of the results of the study), we found that there was almost a linear relationship linking the HPI to the onset of a hypotensive event. Also, the higher the HPI, the shorter the time to the hypotensive event. In other words, with high numbers, you can be very sure that the event will occur in the next minutes, and you have to act right away. HPI proved to be a better predictor for hypotension than any other hemodynamic variable. However, there are no outcome data as of yet, and the link between the use of HPI and a reduction of patients’ complications are still to be clinically demonstrated. Thus, the question under investigation is, if we can reduce the hypotensive burden, will it actually improve the outcomes of our patients? These data are missing as of today, but it is intuitive to believe so.Figure 1

Figure 1Summary of the clinical study by Davies et al. on the diagnostic ability of an Hypotension Prediction Index in predicting impending intraoperative hypotension.

The other technology I was mentioning is the assisted fluid management. Also, this one is AI based and allows reducing the huge variability in fluid administration that is currently happening. The amount of fluids provided in the same procedure varies considerably among hospitals, but even among practitioners within the same hospital. Such variability can be mitigated using this AI-based technology, which makes individual predictions based on the previous experience of an earlier fluid administration. This may allow giving fluids only if the patient is in need of fluids, thereby reducing this huge variability that I just mentioned because variability, as we all know, is the enemy of quality.

Luciano Ravera, PhD, MBA: I believe that the benefits for patients are several. I would start from the diagnostic phase. Thanks to the use of artificial intelligence, we are and will increasingly be able to provide patients with a diagnosis more quickly and more reliably, thus an increased certainty on the most appropriate diagnostic pathway for the patient. Certainly, this also entails an advantage in terms of safety and accuracy. Even more so in the cases where patients would need surgical intervention. Currently, the use of artificial intelligence in the operating room offers benefits in terms of the precision of the surgery, as well as speed and rapidity of the intervention in some cases. Artificial intelligence can also be helpful to manage patient flow more advantageously, both outside and inside the hospital. In fact, the patient flow would ameliorate through the use of artificial intelligence. I believe that we will be able to see its effects also in the patient's homecare. Once dismissed from the hospital, the patient can be telemonitored, thus always virtually remaining within the hospital perimeter. Artificial intelligence combines well with the great theme of personalized medicine. It would allow being more accurate not only in terms of diagnosis but also in terms of surgical and pharmacological treatment, treatment in general. There are 2 areas, in which we have progressed and which we have led as Humanitas. Firstly, a project concerned with the use of artificial intelligence in endoscopy, allowing the early detection of colon cancer. In this case, we managed to have higher accuracy of detection, about 13% higher compared to a traditional endoscopy without the support of artificial intelligence. Thus, in this case, we are able to obtain a more accurate and earlier diagnosis, with the aforementioned advantages. During the COVID-19 pandemic, we were able to use artificial intelligence to support the radiologist in the interpretation of lung CT [computed tomography] scans. This allowed us to predict the intensity and progression of the disease in its various stages. This was of uttermost importance, even more so for COVID-19 patients. In fact, this allowed a more accurate diagnosis and early therapy from the beginning for patients potentially more at risk.

2. Dr Federici: “Beyond the benefits to the patients, how are these medical technologies affecting the way care is delivered, its costs and/or efficiency gains in the process of care?

Dr Scheeren: In the case of HPI, for example, there certainly are additional costs relative to this additional monitoring. The industry charges a certain price for that, but I think that these are more than compensated. If we can avoid postoperative complications, then we could easily calculate what the costs of these complications are and, consequently, the savings related to the use of this technology. Indeed, complications can be very expensive. There is a studySynergistic implications of multiple postoperative outcomes. showing that one complication costs around $6350, 2 complications cost about $12,800, and 3 cost $42,800. Thus, this is an exponential increase in costs. If you have complications, the length of stay will be prolonged from 5 days (normal length of stay). One complication can prolong the length of stay to 8 days, 2 to 11 days, and 3 to 60 days. These, of course, are associated with costs. It is then easy to understand what the use of this technology could save up. Probably, on a national level, it would let you save huge amounts of money.Potential return on investment for implementation of perioperative goal-directed fluid therapy in major surgery: a nationwide database study. In this way, we can prove its cost-effectiveness. In my opinion, this technology is cost-effective, at least in high-risk patients (ie, patients with comorbidities or involved in high-risk procedures or high-risk surgeries because these are associated with a high incidence of complications). Nonetheless, as I mentioned, evidence is still missing on the direct effect of HPI on the reduction of complications. This effect is still to be demonstrated.

Dr Ravera: I believe that as Humanitas we have always focused on the theme of innovation and, more recently, that of digitalization among our core values. What does this mean? In my opinion, for a hospital it means, as I was saying, to follow patients in their complete treatment pathways and not just per single service. There is a need to perceive the hospital as a continuum. Hence, the possibility of following the patient at a distance, even at home. Overall, it should allow us, I say “should” since we are talking about technologies that in some cases are still in their infancy stage and are still used partially, to reach the target goals of the mission of Humanitas (i.e., high quality of care at a sustainable cost for the National Health Service). In respect to this, I believe that the great challenge is being able to ensure safe, quality health care at reasonable costs. What do we (ie, Humanitas) mean by reasonable cost? If we manage to diagnose patients more accurately, we can ensure greater compliance with better treatment and probably save time, resources, and eventually costs. We usually refer to this as a sustainable approach. What does it mean? It means that we must somehow ensure that the available treatments are, respectively, accessible to a large share of the population at a cost that is the same one of the National Health Service for us Italians.

2.1.B Dr Federici: Thus, could the use of AI-based technologies reduce costs by improving the processes of patient care?”

2.1.B Dr Ravera: Yes! I can say this is our bet. However, I must say that in some cases we already have some evidence about its diagnosing capabilities. For instance, we can promptly take the patient to surgery as soon as the need arises. Therefore, we guarantee a greater probability of successful recovery. For example, in most cases, operating ahead of time on a patient that suffers from colon cancer means having the possibility and probability to guarantee greater survival rates. This is certainly related to the concept of quality. Most of the times early diagnoses are linked to easier interventions (ie, less expensive surgeries or pharmacological treatments). This well fits the concept of sustainability”.

2.2.B Dr Federici: Professor Ravera, other than these diagnostic and artificial intelligence tools, as Humanitas, do you have any experience with AI-based technologies mainly focused on physiological monitoring of patients? For example, in operating rooms or intensive care units (ICUs).

2.2.B Dr Ravera: So what we're doing in this ongoing project is improving patient monitoring systems in the operating room and ICU by adopting artificial intelligence techniques, I would say today with the presumption—I would not be in a position to say with certainty—to reduce postoperative complications. Using AI-based technologies before and during surgery, we should be able to ensure better predictability of the intervention, which also means the possibility of limiting and reducing complications both during and after the procedure, be it in ICUs or in hospital stay. In fact, these novel systems allow us to better plan and perform the surgical procedure. Thanks to these technologies, we tend to have a surgical plan, which is better prepared and more tailored to the specific patient's needs. Thus, a more accurate (ie, lower rates of postoperative complications) surgical procedure can be performed. In respect to all this, frankly, overall, we are in a phase of work in progress.

2.1A Dr Federici: Professor Scheeren, you mentioned you have experience with the HPI, among other technologies. HPI is one example of the potentiality of AI-based technologies to improve hemodynamic stability and physiological monitoring. What are the most relevant clinical outcomes that should be considered when evaluating similar medical technologies?

2.1A Dr Scheeren: Yes, currently there are a couple of studies in the literature showing that HPI can effectively reduce the incidence of hypotension. This is the first important step. However, as you said, there is also the need to show that if we reduce such incidence, patient outcomes are improved (ie, complications are reduced). Of course, also measuring the impact on costs would be very important because we, as doctors, always have difficulties to convince our hospital administration that this additional monitoring, which implies additional costs, is justified and needed and maybe counterbalanced by downstream savings.

Moreover, we will have to redistribute the costs that are associated with it. For example, if anesthesiologists have to implement this technology, they have to buy the monitors and disposables, which are quite an investment for their budget. However, the effects (ie, the reduction in complications) will also allow less impact on the surgical ward or on the ICU, reducing the length of stay and so forth. These are usually allocated different budgets in most hospitals. This is why a redistribution is needed. The costs that will be saved up in one place should be balanced with the investment that has to be done in a different place.”

3. Dr Federici: In your experience, what are the main barriers to adoption of AI-based technologies?

Dr Scheeren: “Following on the example of HPI, we have an ongoing study, in which we are looking at the possible barriers of implementing this technology into clinical practice. So far, we found four barriers:1.

Lack of knowledge. In this case, it means that they underestimate and do not believe in the entity of the long-term complications of hypotension. However, this association between hypotension and adverse outcomes was proven on big cohorts of more than 100,000 patients. However, it has not been proven yet that this is a causal relation.

2.

Human capabilities. Preventing hypotension can be quite challenging in some patients. For example, if the surgeon does something wrong, can we still prevent hypotension despite a copious blood loss or similar?

3.

Social influences. The surgeon and the surgical team did not necessarily encourage the anesthesiologist to avoid or treat hypotension. They were not so much interested in or afraid of the hypotension. This is again connected to the first points or lack of knowledge relative to the fact that this is an actual health problem.

4.

Action planning and flexibility. It is believed that the critical thresholds could be different for each patient. For example, a blood pressure threshold of 65 mm Hg could be good for one patient and not good for another one.

Overall, these were the barriers that we identified when we tried to implement the HPI in our clinical practice.”

Dr Ravera: I think there are several factors that could do so. I acknowledge that so far I've given the impression that it's all smooth and easy. In reality, the barriers to overcome are numerous.1.

Data management, property, and privacy. The interoperability of data is essential: modern medicine will not be made by a large single hospital but by N hospitals and N research centers, working together. It is, therefore, necessary to be able to exchange data with ease. I think that the theme of data management, data property, and data privacy is certainly an important issue. Currently, it is often a barrier. However, we are hoping to progress more and more towards easier data management in the future, at least at the European level.

2.

Culture. There is still a factor that is partly generational and partly simply cultural. Not all doctors, or nurses, are nowadays ready to use new technologies. It's not properly a barrier, but there is an issue in training. Moreover, it is worth reminding that the issue applies also to the patient because before we were talking about remote monitoring, of home care, telemedicine, and the new technologies bring a burden, in terms of information, both to the patient and the health care provider, so this is a topic that I would take into account.

3.

Costs. Often, especially at an early development stage, some new technologies may have increased costs. So, if in a medium-term perspective we can calculate benefits that allow higher costs, in some areas, the investment in technological innovation may not be trivial, especially the initial phase. From this point of view, the investment is probably in terms of privacy and reimbursement policies. Certainly, an approach that is more oriented to the so-called value. The so-called value-based care is a term that we often talk about and as of today is only partly practiced.

Returning to my initial point, we follow the patient flow and not only the single performance. Thus, a health system that follows an approach based on value and less on silos, where we see revenues and costs, will be helpful in the future, but, as of today, it is likely to be a barrier.

I mean this in terms of the life science sector in general, for us health care providers, but also in regard to pharma companies and providers of medical devices. We would need a whole ecosystem that follows a value approach overcoming the historical silos of costs and revenues or different vertical sectors, keeping in mind the patient's journey, so the outcome and the patient's health. This is something that is being discussed by many people of the field. However, as of today, it is still to be considered a barrier as it is reachable only in a few pilot experiences. Nonetheless, we can only reach it in a few cases, a few pilot experiences. I think we are in the right direction though.

3.1.B Dr Federici: Professor Ravera, regarding the care processes, in addition to the cultural change that you mentioned, do you think that the adoption of these technologies requires an important internal reorganization? Of course, depending on the technology.

3.1.B Dr Ravera: I believe that in the moment in which, regarding the training, we will have operators trained on new technologies, any changes in clinical pathways will come consequently. The organization adapts when the patients that are involved know what awaits them in terms of care path and the operators are prepared in terms of processes and skills.

3.2.B Dr Federici: However, in your experience, was the acceptance of new techniques by the patients and the operators positive?

3.2.B Dr Ravera: Yes, I am optimistic, but it will take time. I reiterate the need for skills training. But I do not think that these are unachievable barriers.

4. Dr Federici: Are there any other factors that may affect the successful deployment and (cost-)effectiveness of these technologies in hospitals (eg, how technologies are perceived and used by clinicians or the need to reorganize the process of care and the roles of health care providers)?

Dr Scheeren: Yes, as I said we need education, we need to create awareness of the incidence and the consequences of hypotension and of the variability in fluid management, which is connected to the other technology I mentioned at the beginning. Of course, we have to be aware of the fact that a monitoring technology itself cannot improve the patient outcomes unless it is interpreted correctly by the doctors and associated with a treatment algorithm. However, we have to make clear to people that just using the monitor does not help the patient. Based on the monitor readings you have to draw the right conclusions and initiate the right treatment. And this is maybe something for which we should develop a specific decision support tool. I know that there are some regulatory issues with that since it goes beyond monitoring. Finally, we should involve the nurses. In fact, they are also at the bedside. This is true for the operating room but also and particularly for the ICU, where there are fewer doctors and not a one-to-one relation like in the operating room. The nurses, though, are there, and they should be able to interpret the readings from the monitor and to initiate the right treatment without consulting a doctor in every instance. There should be dedicated protocols for this in the hospitals.

Dr Ravera: So, I mentioned the endoscopy case study that we conducted in collaboration with others. In that case, we are certainly not the only ones to reach interesting results. There was a diagnostic accuracy that increased by 13%; this is a positive and solid result. Still in a trial but replicable on big numbers. I think that in the medium term, not always in the short term, you can get to show that these equations reach better quality. I reiterate the fact that certain short-term incentives, in terms of technological innovation, would help this new health system in its entirety. Thus, certainly, some economic incentives could accelerate the adoption of new technologies, probably not only the adoption but also increase the speed of propagation. Here, I think it is important because today we have, we not as Humanitas but as research centers, cutting-edge hospitals; we all have some good and sometimes excellent experiences to discuss, regarding artificial intelligence, always more and more. However, we must create a movement. To do so we need more than an experiment; we need thousands of patients who can experience technological innovation. Very often, by putting together the data and practices of many hospitals, we can more easily reach our goal. So, working together and sharing more easily data and procedures could be a further boost to the growth of technological innovation. From this point of view, I must say that some research funds, at the Italian level and especially at the European level, can also be considered an interesting incentive, specifically to this work of dissemination of knowledge and attention to new technologies.

4.1.B Dr Federici: Furthermore, from the point of view of payers, when you talk about incentives, do you also refer to the fact that the value of these technologies is recognized and then reimbursed?

4.1.B Dr Ravera: Yes, when I talk about value-based care, in some cases, we should think about introducing diagnosis related group tariffs (DRG) of the path rather than payments; this would provide a bonus according to the look of care achieved by the technology. So, I could think of the fact that we could link more quality to economic reimbursement. I believe that many hospitals today would accept the bet when presented with a proven better clinical trial; the state or the region can pay more or less, depending on the result of the trial, linking pricing policies and reimbursement policies in some way to the outcomes of care. It applies to technological innovation but could also apply in general; this is something that we in Humanitas would look at favorably. We are very keen to say, “Let's measure the outcome of what we do.” There are already many performance indicators both at the Italian level, such as the project outcomes Agenas, and at the European and international levels. In short, we have many examples of good measurements of the quality of care provided by hospitals. Linking it to public pricing policies could be interesting for everyone, including the patient.

留言 (0)

沒有登入
gif