Banner by Lilly Koerner

Paging Doctor…Who??

By Korrina Gidwani

“Incoming 62-year-old male patient... upper gastrointestinal bleeding...” echoes through the hallways of the Intensive Care Unit at a University of Pittsburgh Medical Center facility. Chaos ensues as dozens of healthcare workers prepare a new ICU room for the incoming patient. As clinicians rapidly perform a myriad of standardized medical exams and record pertinent information, the Intensive Care Unit (ICU) doctor on call is paged. The physician pores over the patient’s past medical history and notes from the patient’s time in the Emergency Room a few hours prior. Although the physician has never met this patient before and is unfamiliar with his medical history, she will have no trouble formulating a diagnosis and comprehensive treatment plan, all thanks to one unique piece of technology.

But... why? Isn’t the ICU physician required to pore over years of medical records for this patient and craft a treatment plan all within a mere few minutes? Oftentimes, yes, especially considering that it is a matter of life-or-death in the ICU. However, thanks to the Artificial Intelligence (AI) system being developed by Dr. Shyam Visweswaran, an MD/PhD at the University of Pittsburgh School of Medicine, the process of diagnosing and treating patients under pressure is undergoing a complete transformation.

In such instances, Dr. Visweswaran’s AI system, a “Learning Electronic Medical Record (L-EMR)” system, can examine a patient’s medical record and analyze thousands of data points and personal information within minutes. It can pinpoint subtle patterns and assemble a model that can guide future decision-making tasks. It may flag important trends, such as patterns in high blood pressure and severe allergies, and even predict disease progression rates based on existing information. The AI system can highlight pertinent information and apply statistical modeling to predict medical data, including laboratory test results, medication orders, vital signs, and physiological measurements. Ultimately, the AI system will aim to make this crucial data accessible to clinicians at the right time. With the assistance of an intelligent AI system, clinicians can effectively integrate a multitude of information while minimizing data overload, make comprehensive and well-researched medical decisions, boost efficiency in life-or-death scenarios, and eliminate medical errors.

With that in mind, how can an AI system intervene and assist clinicians in the case of the 62-year-old man with gastrointestinal bleeding? After the ICU team situates the patient, they must craft an appropriate treatment plan. They may opt to treat the patient with intravenous blood until laboratory and physiological results suggest otherwise.

After these results are recorded, an AI system could elucidate trends in hemoglobin values, heart rate, and blood administration. The system can even use statistical models that employ pattern-based learning using past patient data in order to predict a future trajectory for the patient’s condition. Clinicians can refer to these concise, comprehensive results to determine if the patient has stabilized and whether or not to cease intravenous blood administration. As a result, physicians can make informed decisions that are driven by large-scale trends and predictions generated by an AI system. Although AI systems consistently exceed expectations in clinical trials and appear to be a promising technology, their extensive autonomy and intelligence have sparked concern within the scientific community. Critics of AI remain apprehensive about the accuracy, efficiency, and general impact of AI systems within medicine. Are AI systems truly capable of overtaking the medical field, thus reducing the demand for healthcare workers and jeopardizing patient-physician relationships? Moreover, how can these novel systems be implemented in hospitals where they pose a risk of being hacked and compromising patient privacy? Well, let’s find out.

There are two common types of AI systems: expert systems and machine learning. Expert systems draw on information from a knowledge base, which consists of experiential information inputted by human experts. When presented with a novel scenario, expert systems access their knowledge base and craft a response based on past scenarios and experiences. Expert systems typically employ logical reasoning to accomplish this task. On the other hand, machine learning systems apply statistical methods to elucidate patterns from thousands of data points. When presented with a novel scenario, machine learning systems will align the new information with one or more familiar patterns. These patterns will dictate the system’s response to a given scenario. Due to the efficiency and accuracy of machine learning interfaces, it has demonstrated immense potential for use in the medical field. For instance, past studies have highlighted the importance of machine learning when diagnosing and treating breast cancer.

To accomplish this, one group of researchers crafted artificial neural networks, an interconnected group of computing units called “neurons”, and utilized them in conjunction with decision trees, algorithms that follow predetermined conditions when classifying a set of data. They trained a machine learning system using over 200,000 data points and later integrated the system into clinical settings. These data sets encompassed information on lymph node status, tumor size, and several other factors related to breast cancer. The machine learning system extracted trends and patterns after thoroughly poring over the existing data set. Then, the system referred to these trends when diagnosing breast cancer and modeling 10-year disease trajectories for patients. Similar studies have been conducted across the nation, with several promising projects based in Pittsburgh.

Within the Department of Biomedical Informatics at the University of Pittsburgh School of Medicine, Dr. Shyam Visweswaran and his team have dedicated their time to crafting unique AI systems that streamline clinical tasks. They frequently apply similar strategies, namely machine learning, when training AI systems for use in a clinical setting. Typically, healthcare professionals consult Dr. Visweswaran if they require an AI system to boost the efficiency of clinical tasks. He and his team usually collaborate for months in order to produce an AI system that is specialized to the clinical task at hand. After constructing an AI system, Dr. Visweswaran can implement the system into the medical field in hopes of addressing laborious clinical tasks.

Dr. Visweswaran stated that AI holds remarkable potential as a clinical tool due to its ability to solve three medical challenges: knowledge overload, patient data overload, and process overload.

“Medical knowledge is rapidly growing, especially in terms of new knowledge about molecular mechanisms. This increase in knowledge threatens to overwhelm human memory capacity,” Dr. Visweswaran explained.

Therefore, an AI system may intervene to streamline basic clinical tasks. For instance, EMR systems have been implemented in hospitals around the world. These EMR systems are rapidly expanding, with millions of data points from medical tests and imaging being added every single day. An AI system is capable of analyzing these data points to pinpoint patterns and trends within the data, as described earlier. Healthcare professionals can refer to these large-scale trends and cross-check their work as they formulate diagnoses and long-term treatment plans. They can utilize these resources to ensure that they do not overlook relevant data when evaluating and treating a patient. Moreover, process overload is especially prominent in the medical field since healthcare professionals must input vast amounts of patient information into the EMR system. This process tends to be tedious and time-consuming since the majority of the healthcare community relies on hand-written notes and diagrams. AI systems can streamline this process as well, namely through data input using speech recognition technology.

It is apparent that AI technology is impactful and possesses several advantages, but how does it specifically benefit healthcare professionals? At which points in the diagnosis and treatment process do AI systems appear and serve as a useful tool for clinicians?

To answer these questions, Dr. Visweswaran cited his personal research experiences and consultation projects. According to Dr. Visweswaran, medical AI systems can be developed to streamline clinical processes that are difficult or tedious for humans to complete. These systems are referred to as “intelligence augmentation” systems due to their efficacy in bolstering human efforts.

For instance, physicians may find it difficult to predict the trajectory of a patient’s hypertension, especially if the patient has minimal medical information in the EMR system. To remedy this issue, Dr. Visweswaran and his team crafted an AI system that utilizes patient data to predict the trajectory of patients’ hypertension. The system accomplishes this challenging task by highlighting relevant trends and making calculations using existing information. In addition, Dr. Visweswaran and his team constructed another AI system that can monitor a patient’s vital signs during lengthy surgeries. His AI tools can be programmed to alert surgeons if a patient’s vital signs exceed normal parameters or if a stroke starts to develop in the brain. These tasks are crucial to complete during a surgery, so it would be beneficial to reinforce human efforts using AI-based techniques. In general, these examples highlight the necessity of AI systems and the feasibility of using AI technology at any stage in the diagnosis and treatment process.

After explaining his day-to-day tasks and emphasizing the advantages of AI through the lens of his own research efforts, Dr. Visweswaran addressed the controversies associated with the use of AI systems in a medical setting. He believes that, although AI systems possess remarkable capabilities, they are “unlikely to replace clinicians because the human brain is still the most flexible, powerful tool.” Instead, AI systems aim to assist clinicians in formulating complex diagnoses, understanding new biomedical knowledge, and staying up to date with their rapidly evolving field, as seen in earlier examples. Other clinicians seem to agree with his remarks as dozens of UPMC workers have requested his team to craft AI systems that streamline medical processes. For the most part, clinicians have welcomed AI systems with open arms and do not feel threatened by the use of these autonomous technological systems. Lastly, Dr. Visweswaran discussed security measures taken by AI companies and research teams to safeguard patient information. These groups understand the importance of securing confidential patient information and work extensively to promote patient privacy and confidentiality before marketing an AI system. Based on this information, it is apparent that AI systems will positively impact the medical field.

If asked to envision the role of technology in our society 20 years from now, what would you imagine? Certainly, most of you can agree that technology will have a more versatile, pronounced role in our society, especially within the medical field. The remarkable efforts of Dr. Visweswaran and other researchers have put us one step closer to achieving this goal. As of now, AI systems are beginning to demonstrate their aptitude in aiding healthcare professionals with clinical tasks and addressing numerous issues in the medical field, including data overload, knowledge overload, and process overload. Furthermore, for the critics of AI, it may be comforting to know that AI systems are crafted to assist clinicians – not interfere with patient-physician relationships or jeopardize patient privacy.

As for the future of AI, Dr. Visweswaran believes it can be integrated into homes to monitor patients with chronic conditions and predict long-term trends for patients in the hospital, among many other uses. Dr. Visweswaran’s impressive research involving EMR systems, brain monitoring during surgery, and ICU tasks are also quite telling of AI’s potential to aid healthcare workers for years to come. Now, when you have your next wellness check-up, look out for an AI system working behind-the-scenes and actively transforming the field of medicine!